Apply clang-format on repository
Code is formatted as per a revised clang format configuration
file(not part of this delivery). Version 14.0.6 is used.
Exclusion List:
- files with .cl extension
- files that are not strictly C/C++ (e.g. Android.bp, Sconscript ...)
And the following directories
- compute_kernel_writer/validation/
- tests/
- include/
- src/core/NEON/kernels/convolution/
- src/core/NEON/kernels/arm_gemm/
- src/core/NEON/kernels/arm_conv/
- data/
There will be a follow up for formatting of .cl files and the
files under tests/ and compute_kernel_writer/validation/.
Signed-off-by: Felix Thomasmathibalan <felixjohnny.thomasmathibalan@arm.com>
Change-Id: Ib7eb1fcf4e7537b9feaefcfc15098a804a3fde0a
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10391
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
diff --git a/src/core/CL/CLCommandBuffer.cpp b/src/core/CL/CLCommandBuffer.cpp
index 7fcfdf2..d094dcd 100644
--- a/src/core/CL/CLCommandBuffer.cpp
+++ b/src/core/CL/CLCommandBuffer.cpp
@@ -38,7 +38,7 @@
const auto &cl_device = CLKernelLibrary::get().get_device();
const auto has_mutable_dispatch = command_buffer_mutable_dispatch_supported(cl_device);
- if(has_mutable_dispatch)
+ if (has_mutable_dispatch)
{
return std::make_unique<CLMutableCommandBuffer>(queue);
}
diff --git a/src/core/CL/CLCommandBuffer.h b/src/core/CL/CLCommandBuffer.h
index 8a94e38..90e4341 100644
--- a/src/core/CL/CLCommandBuffer.h
+++ b/src/core/CL/CLCommandBuffer.h
@@ -87,7 +87,8 @@
* @param[in] global The global work size.
* @param[in] local The local work size.
*/
- virtual void add_kernel(cl_kernel kernel, const cl::NDRange &offset, const cl::NDRange &global, const cl::NDRange &local) = 0;
+ virtual void
+ add_kernel(cl_kernel kernel, const cl::NDRange &offset, const cl::NDRange &global, const cl::NDRange &local) = 0;
/** Add the mutable argument to the current kernel enqueue command.
*
@@ -154,7 +155,7 @@
CLCommandBuffer &state(State state);
private:
- State _state{ State::Created };
+ State _state{State::Created};
};
} // namespace arm_compute
diff --git a/src/core/CL/CLCompatCommandBuffer.cpp b/src/core/CL/CLCompatCommandBuffer.cpp
index f1a902c..242fd77 100644
--- a/src/core/CL/CLCompatCommandBuffer.cpp
+++ b/src/core/CL/CLCompatCommandBuffer.cpp
@@ -31,8 +31,7 @@
namespace arm_compute
{
-CLCompatCommandBuffer::CLCompatCommandBuffer(cl_command_queue queue)
- : _queue(queue)
+CLCompatCommandBuffer::CLCompatCommandBuffer(cl_command_queue queue) : _queue(queue)
{
}
@@ -40,11 +39,14 @@
{
}
-void CLCompatCommandBuffer::add_kernel(cl_kernel kernel, const cl::NDRange &offset, const cl::NDRange &global, const cl::NDRange &local)
+void CLCompatCommandBuffer::add_kernel(cl_kernel kernel,
+ const cl::NDRange &offset,
+ const cl::NDRange &global,
+ const cl::NDRange &local)
{
ARM_COMPUTE_ERROR_ON(state() != State::Created);
- _kernel_cmds.push_back(KernelCommand{ kernel, offset, global, local, {} });
+ _kernel_cmds.push_back(KernelCommand{kernel, offset, global, local, {}});
}
void CLCompatCommandBuffer::add_mutable_argument_generic(cl_uint arg_idx, const void *value, size_t size)
@@ -52,7 +54,7 @@
ARM_COMPUTE_ERROR_ON(state() != State::Created);
ARM_COMPUTE_ERROR_ON(_kernel_cmds.empty());
- _kernel_cmds.back().mutable_args.push_back(cl_mutable_dispatch_arg_khr{ arg_idx, size, value });
+ _kernel_cmds.back().mutable_args.push_back(cl_mutable_dispatch_arg_khr{arg_idx, size, value});
}
void CLCompatCommandBuffer::finalize()
@@ -61,7 +63,7 @@
_kernel_cmds.shrink_to_fit();
- for(auto &cmd : _kernel_cmds)
+ for (auto &cmd : _kernel_cmds)
{
cmd.mutable_args.shrink_to_fit();
}
@@ -80,25 +82,19 @@
{
ARM_COMPUTE_ERROR_ON(state() != State::Finalized);
- for(const auto &cmd : _kernel_cmds)
+ for (const auto &cmd : _kernel_cmds)
{
- for(const auto &arg : cmd.mutable_args)
+ for (const auto &arg : cmd.mutable_args)
{
const auto error = clSetKernelArg(cmd.kernel, arg.arg_index, arg.arg_size, arg.arg_value);
handle_cl_error("clSetKernelArg", error);
}
- const auto error = clEnqueueNDRangeKernel(
- _queue,
- cmd.kernel,
- static_cast<cl_uint>(cmd.global.dimensions()),
- cmd.offset.dimensions() != 0 ? cmd.offset.get() : nullptr,
- cmd.global.get(),
- cmd.local.dimensions() != 0 ? cmd.local.get() : nullptr,
- 0,
- nullptr,
- nullptr);
+ const auto error =
+ clEnqueueNDRangeKernel(_queue, cmd.kernel, static_cast<cl_uint>(cmd.global.dimensions()),
+ cmd.offset.dimensions() != 0 ? cmd.offset.get() : nullptr, cmd.global.get(),
+ cmd.local.dimensions() != 0 ? cmd.local.get() : nullptr, 0, nullptr, nullptr);
handle_cl_error("clEnqueueNDRangeKernel", error);
}
diff --git a/src/core/CL/CLCompatCommandBuffer.h b/src/core/CL/CLCompatCommandBuffer.h
index e91d52d..d5df106 100644
--- a/src/core/CL/CLCompatCommandBuffer.h
+++ b/src/core/CL/CLCompatCommandBuffer.h
@@ -57,7 +57,10 @@
/** Disallow move assignment. */
CLCompatCommandBuffer &operator=(CLCompatCommandBuffer &&) = delete;
- void add_kernel(cl_kernel kernel, const cl::NDRange &offset, const cl::NDRange &global, const cl::NDRange &local) override;
+ void add_kernel(cl_kernel kernel,
+ const cl::NDRange &offset,
+ const cl::NDRange &global,
+ const cl::NDRange &local) override;
void finalize() override;
diff --git a/src/core/CL/CLCompileContext.cpp b/src/core/CL/CLCompileContext.cpp
index 2d024f9..9bbc326 100644
--- a/src/core/CL/CLCompileContext.cpp
+++ b/src/core/CL/CLCompileContext.cpp
@@ -22,19 +22,19 @@
* SOFTWARE.
*/
#include "arm_compute/core/CL/CLCompileContext.h"
-#include "arm_compute/core/CL/OpenCL.h"
#include "arm_compute/core/CL/CLHelpers.h"
+#include "arm_compute/core/CL/OpenCL.h"
#include "arm_compute/core/Error.h"
#include "arm_compute/core/Utils.h"
+
#include "support/StringSupport.h"
#include <regex>
namespace arm_compute
{
-CLBuildOptions::CLBuildOptions()
- : _build_opts()
+CLBuildOptions::CLBuildOptions() : _build_opts()
{
}
@@ -45,7 +45,7 @@
void CLBuildOptions::add_option_if(bool cond, std::string option)
{
- if(cond)
+ if (cond)
{
add_option(std::move(option));
}
@@ -63,7 +63,7 @@
void CLBuildOptions::add_options_if(bool cond, const StringSet &options)
{
- if(cond)
+ if (cond)
{
add_options(options);
}
@@ -79,26 +79,35 @@
return _build_opts == other._build_opts;
}
-Program::Program()
- : _context(), _device(), _is_binary(false), _name(), _source(), _binary()
+Program::Program() : _context(), _device(), _is_binary(false), _name(), _source(), _binary()
{
}
Program::Program(cl::Context context, std::string name, std::string source)
- : _context(std::move(context)), _device(), _is_binary(false), _name(std::move(name)), _source(std::move(source)), _binary()
+ : _context(std::move(context)),
+ _device(),
+ _is_binary(false),
+ _name(std::move(name)),
+ _source(std::move(source)),
+ _binary()
{
}
Program::Program(cl::Context context, cl::Device device, std::string name, std::vector<unsigned char> binary)
- : _context(std::move(context)), _device(std::move(device)), _is_binary(true), _name(std::move(name)), _source(), _binary(std::move(binary))
+ : _context(std::move(context)),
+ _device(std::move(device)),
+ _is_binary(true),
+ _name(std::move(name)),
+ _source(),
+ _binary(std::move(binary))
{
}
Program::operator cl::Program() const
{
- if(_is_binary)
+ if (_is_binary)
{
- return cl::Program(_context, { _device }, { _binary });
+ return cl::Program(_context, {_device}, {_binary});
}
else
{
@@ -112,12 +121,12 @@
{
return program.build(build_options.c_str()) == CL_SUCCESS;
}
- catch(const cl::Error &e)
+ catch (const cl::Error &e)
{
cl_int err = CL_SUCCESS;
const auto build_info = program.getBuildInfo<CL_PROGRAM_BUILD_LOG>(&err);
- for(auto &pair : build_info)
+ for (auto &pair : build_info)
{
std::cerr << pair.second << std::endl;
}
@@ -133,14 +142,12 @@
return cl_program;
}
-Kernel::Kernel()
- : _name(), _kernel()
+Kernel::Kernel() : _name(), _kernel()
{
}
Kernel::Kernel(std::string name, const cl::Program &program)
- : _name(std::move(name)),
- _kernel(cl::Kernel(program, _name.c_str()))
+ : _name(std::move(name)), _kernel(cl::Kernel(program, _name.c_str()))
{
}
CLCompileContext::CLCompileContext()
@@ -156,15 +163,19 @@
_is_wbsm_supported = get_wbsm_support_info(device);
}
-Kernel CLCompileContext::create_kernel(const std::string &kernel_name, const std::string &program_name, const std::string &program_source,
- const std::string &kernel_path, const StringSet &build_options_set, bool is_binary) const
+Kernel CLCompileContext::create_kernel(const std::string &kernel_name,
+ const std::string &program_name,
+ const std::string &program_source,
+ const std::string &kernel_path,
+ const StringSet &build_options_set,
+ bool is_binary) const
{
const std::string build_options = generate_build_options(build_options_set, kernel_path);
const std::string built_program_name = program_name + "_" + build_options;
auto built_program_it = _built_programs_map.find(built_program_name);
cl::Program cl_program;
- if(_built_programs_map.end() != built_program_it)
+ if (_built_programs_map.end() != built_program_it)
{
// If program has been built, retrieve to create kernel from it
cl_program = built_program_it->second;
@@ -184,11 +195,12 @@
return Kernel(kernel_name, cl_program);
}
-const Program &CLCompileContext::load_program(const std::string &program_name, const std::string &program_source, bool is_binary) const
+const Program &
+CLCompileContext::load_program(const std::string &program_name, const std::string &program_source, bool is_binary) const
{
const auto program_it = _programs_map.find(program_name);
- if(program_it != _programs_map.end())
+ if (program_it != _programs_map.end())
{
return program_it->second;
}
@@ -199,9 +211,10 @@
ARM_COMPUTE_UNUSED(is_binary);
program = Program(_context, program_name, program_source);
#else /* EMBEDDED_KERNELS */
- if(is_binary)
+ if (is_binary)
{
- program = Program(_context, _device.cl_device(), program_name, std::vector<unsigned char>(program_source.begin(), program_source.end()));
+ program = Program(_context, _device.cl_device(), program_name,
+ std::vector<unsigned char>(program_source.begin(), program_source.end()));
}
else
{
@@ -218,18 +231,19 @@
void CLCompileContext::set_context(cl::Context context)
{
_context = std::move(context);
- if(_context.get() != nullptr)
+ if (_context.get() != nullptr)
{
const auto cl_devices = _context.getInfo<CL_CONTEXT_DEVICES>();
- if(!cl_devices.empty())
+ if (!cl_devices.empty())
{
_device = CLDevice(cl_devices[0]);
}
}
}
-std::string CLCompileContext::generate_build_options(const StringSet &build_options_set, const std::string &kernel_path) const
+std::string CLCompileContext::generate_build_options(const StringSet &build_options_set,
+ const std::string &kernel_path) const
{
std::string concat_str;
bool ext_supported = false;
@@ -241,27 +255,27 @@
#endif // defined(ARM_COMPUTE_DEBUG_ENABLED)
GPUTarget gpu_arch = get_arch_from_target(_device.target());
- concat_str += " -DGPU_ARCH=" + support::cpp11::to_string(
- static_cast<std::underlying_type<GPUTarget>::type>(gpu_arch));
+ concat_str +=
+ " -DGPU_ARCH=" + support::cpp11::to_string(static_cast<std::underlying_type<GPUTarget>::type>(gpu_arch));
- if(_device.supported("cl_khr_fp16"))
+ if (_device.supported("cl_khr_fp16"))
{
concat_str += " -DARM_COMPUTE_OPENCL_FP16_ENABLED=1 ";
}
- if(_device.supported("cl_arm_integer_dot_product_int8") || _device.supported("cl_khr_integer_dot_product"))
+ if (_device.supported("cl_arm_integer_dot_product_int8") || _device.supported("cl_khr_integer_dot_product"))
{
concat_str += " -DARM_COMPUTE_OPENCL_DOT8_ENABLED=1 ";
}
- if(_device.supported("cl_arm_integer_dot_product_accumulate_int8"))
+ if (_device.supported("cl_arm_integer_dot_product_accumulate_int8"))
{
concat_str += " -DARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED=1 ";
}
std::tie(ext_supported, ext_buildopts) = _device.is_non_uniform_workgroup_supported();
- if(ext_supported)
+ if (ext_supported)
{
concat_str += ext_buildopts;
}
@@ -270,7 +284,7 @@
ARM_COMPUTE_ERROR("Non uniform workgroup size is not supported!!");
}
- if(gpu_arch != GPUTarget::UNKNOWN && gpu_arch != GPUTarget::MIDGARD && get_ddk_version() >= 11)
+ if (gpu_arch != GPUTarget::UNKNOWN && gpu_arch != GPUTarget::MIDGARD && get_ddk_version() >= 11)
{
concat_str += " -DUNROLL_WITH_PRAGMA ";
}
@@ -295,7 +309,7 @@
#endif /* EMBEDDED_KERNELS */
// Concatenate set
- for(const auto &el : s)
+ for (const auto &el : s)
{
concat_set += " " + el;
}
@@ -340,7 +354,7 @@
GPUTarget _target = get_target_from_device(_device.cl_device());
cl::NDRange default_range;
- switch(_target)
+ switch (_target)
{
case GPUTarget::MIDGARD:
case GPUTarget::T600:
@@ -370,7 +384,8 @@
size_t result;
size_t err = kernel.getWorkGroupInfo(_device.cl_device(), CL_KERNEL_WORK_GROUP_SIZE, &result);
- ARM_COMPUTE_ERROR_ON_MSG(err != 0, "clGetKernelWorkGroupInfo failed to return the maximum workgroup size for the kernel");
+ ARM_COMPUTE_ERROR_ON_MSG(err != 0,
+ "clGetKernelWorkGroupInfo failed to return the maximum workgroup size for the kernel");
ARM_COMPUTE_UNUSED(err);
return result;
@@ -392,7 +407,7 @@
const std::regex ddk_regex("r([0-9]*)p[0-9]");
std::smatch ddk_match;
- if(std::regex_search(device_version, ddk_match, ddk_regex))
+ if (std::regex_search(device_version, ddk_match, ddk_regex))
{
return std::stoi(ddk_match[1]);
}
diff --git a/src/core/CL/CLHelpers.cpp b/src/core/CL/CLHelpers.cpp
index 78f3610..5ea99d3 100644
--- a/src/core/CL/CLHelpers.cpp
+++ b/src/core/CL/CLHelpers.cpp
@@ -22,14 +22,15 @@
* SOFTWARE.
*/
#include "arm_compute/core/CL/CLHelpers.h"
+
#include "arm_compute/core/CL/CLKernelLibrary.h"
#include "arm_compute/core/CL/CLTypes.h"
-#include "arm_compute/core/utils/DataTypeUtils.h"
#include "arm_compute/core/Error.h"
#include "arm_compute/core/Log.h"
#include "arm_compute/core/Types.h"
-#include "src/gpu/cl/ClCompileContext.h"
+#include "arm_compute/core/utils/DataTypeUtils.h"
+#include "src/gpu/cl/ClCompileContext.h"
#include "src/gpu/cl/ClKernelLibrary.h"
#include <utility>
@@ -39,7 +40,7 @@
{
std::string get_cl_type_from_data_type(const DataType &dt)
{
- switch(dt)
+ switch (dt)
{
case DataType::U8:
case DataType::QASYMM8:
@@ -75,7 +76,7 @@
std::string get_cl_promoted_type_from_data_type(const DataType &dt)
{
- switch(dt)
+ switch (dt)
{
case DataType::U8:
case DataType::QASYMM8:
@@ -105,7 +106,7 @@
std::string get_cl_unsigned_type_from_element_size(size_t element_size)
{
- switch(element_size)
+ switch (element_size)
{
case 1:
return "uchar";
@@ -123,7 +124,7 @@
std::string get_cl_signed_type_from_element_size(size_t element_size)
{
- switch(element_size)
+ switch (element_size)
{
case 1:
return "char";
@@ -141,7 +142,7 @@
std::string get_cl_select_type_from_data_type(const DataType &dt)
{
- switch(dt)
+ switch (dt)
{
case DataType::U8:
case DataType::QASYMM8:
@@ -174,7 +175,7 @@
std::string get_cl_dot8_acc_type_from_data_type(const DataType &dt)
{
- switch(dt)
+ switch (dt)
{
case DataType::U8:
case DataType::QASYMM8:
@@ -192,7 +193,7 @@
std::string get_data_size_from_data_type(const DataType &dt)
{
- switch(dt)
+ switch (dt)
{
case DataType::U8:
case DataType::S8:
@@ -244,8 +245,9 @@
const GPUTarget gpu_target = get_target_from_name(device_name);
// SW_WORKAROUND: Workaround for DDK revision r14p0.to enable cl_arm_integer_dot_product_int8
- std::set<GPUTarget> sw_workaround_issue = { GPUTarget::G76 };
- return (device_supports_extension(device, "cl_arm_integer_dot_product_int8") || sw_workaround_issue.count(gpu_target) != 0);
+ std::set<GPUTarget> sw_workaround_issue = {GPUTarget::G76};
+ return (device_supports_extension(device, "cl_arm_integer_dot_product_int8") ||
+ sw_workaround_issue.count(gpu_target) != 0);
}
bool dot8_acc_supported(const cl::Device &device)
@@ -256,23 +258,23 @@
CLVersion get_cl_version(const cl::Device &device)
{
std::string version_str = device.getInfo<CL_DEVICE_VERSION>();
- if(version_str.find("OpenCL 3") != std::string::npos)
+ if (version_str.find("OpenCL 3") != std::string::npos)
{
return CLVersion::CL30;
}
- else if(version_str.find("OpenCL 2") != std::string::npos)
+ else if (version_str.find("OpenCL 2") != std::string::npos)
{
return CLVersion::CL20;
}
- else if(version_str.find("OpenCL 1.2") != std::string::npos)
+ else if (version_str.find("OpenCL 1.2") != std::string::npos)
{
return CLVersion::CL12;
}
- else if(version_str.find("OpenCL 1.1") != std::string::npos)
+ else if (version_str.find("OpenCL 1.1") != std::string::npos)
{
return CLVersion::CL11;
}
- else if(version_str.find("OpenCL 1.0") != std::string::npos)
+ else if (version_str.find("OpenCL 1.0") != std::string::npos)
{
return CLVersion::CL10;
}
@@ -287,14 +289,15 @@
return (pos != std::string::npos);
}
-bool cl_winograd_convolution_layer_supported(const Size2D &output_tile, const Size2D &kernel_size, DataLayout data_layout)
+bool cl_winograd_convolution_layer_supported(const Size2D &output_tile,
+ const Size2D &kernel_size,
+ DataLayout data_layout)
{
ARM_COMPUTE_ERROR_ON(data_layout == DataLayout::UNKNOWN);
using WinogradConfiguration = std::pair<std::pair<int, int>, std::pair<int, int>>;
- std::vector<WinogradConfiguration> winograd_configs_nchw =
- {
+ std::vector<WinogradConfiguration> winograd_configs_nchw = {
WinogradConfiguration(std::pair<int, int>(1, 2), std::pair<int, int>(1, 3)),
WinogradConfiguration(std::pair<int, int>(1, 4), std::pair<int, int>(1, 3)),
WinogradConfiguration(std::pair<int, int>(2, 1), std::pair<int, int>(3, 1)),
@@ -303,11 +306,9 @@
WinogradConfiguration(std::pair<int, int>(4, 4), std::pair<int, int>(3, 3)),
WinogradConfiguration(std::pair<int, int>(4, 4), std::pair<int, int>(5, 5)),
WinogradConfiguration(std::pair<int, int>(4, 1), std::pair<int, int>(5, 1)),
- WinogradConfiguration(std::pair<int, int>(1, 4), std::pair<int, int>(1, 5))
- };
+ WinogradConfiguration(std::pair<int, int>(1, 4), std::pair<int, int>(1, 5))};
- std::vector<WinogradConfiguration> winograd_configs_nhwc =
- {
+ std::vector<WinogradConfiguration> winograd_configs_nhwc = {
WinogradConfiguration(std::pair<int, int>(2, 2), std::pair<int, int>(3, 3)),
WinogradConfiguration(std::pair<int, int>(1, 4), std::pair<int, int>(1, 3)),
WinogradConfiguration(std::pair<int, int>(4, 1), std::pair<int, int>(3, 1)),
@@ -324,19 +325,21 @@
std::pair<int, int>(kernel_size.width, kernel_size.height));
// Return true if supported
- if(data_layout == DataLayout::NCHW)
+ if (data_layout == DataLayout::NCHW)
{
- return (std::find(winograd_configs_nchw.begin(), winograd_configs_nchw.end(), p) != winograd_configs_nchw.end());
+ return (std::find(winograd_configs_nchw.begin(), winograd_configs_nchw.end(), p) !=
+ winograd_configs_nchw.end());
}
else
{
- return (std::find(winograd_configs_nhwc.begin(), winograd_configs_nhwc.end(), p) != winograd_configs_nhwc.end());
+ return (std::find(winograd_configs_nhwc.begin(), winograd_configs_nhwc.end(), p) !=
+ winograd_configs_nhwc.end());
}
}
size_t preferred_vector_width(const cl::Device &device, const DataType dt)
{
- switch(dt)
+ switch (dt)
{
case DataType::U8:
case DataType::S8:
@@ -382,7 +385,7 @@
cl_int err = clGetDeviceInfo(device(), CL_DEVICE_IMAGE_PITCH_ALIGNMENT, sizeof(cl_uint), &pixel_aligment, nullptr);
- if(err == CL_SUCCESS)
+ if (err == CL_SUCCESS)
{
return pixel_aligment;
}
@@ -396,12 +399,14 @@
{
cl_bool supported = CL_FALSE;
- cl_int err = clGetDeviceInfo(device(), CL_DEVICE_NON_UNIFORM_WORK_GROUP_SUPPORT, sizeof(cl_bool), &supported, nullptr);
+ cl_int err =
+ clGetDeviceInfo(device(), CL_DEVICE_NON_UNIFORM_WORK_GROUP_SUPPORT, sizeof(cl_bool), &supported, nullptr);
return (err == CL_SUCCESS && supported == CL_TRUE);
}
-cl::Kernel create_kernel(const CLCompileContext &ctx, const std::string &kernel_name, const std::set<std::string> &build_opts)
+cl::Kernel
+create_kernel(const CLCompileContext &ctx, const std::string &kernel_name, const std::set<std::string> &build_opts)
{
opencl::ClKernelLibrary &klib = opencl::ClKernelLibrary::get();
@@ -409,7 +414,8 @@
auto kernel_src = klib.program(program_name);
const std::string kernel_path = klib.kernel_path();
- return static_cast<cl::Kernel>(ctx.create_kernel(kernel_name, program_name, kernel_src.program, kernel_path, build_opts, kernel_src.is_binary));
+ return static_cast<cl::Kernel>(ctx.create_kernel(kernel_name, program_name, kernel_src.program, kernel_path,
+ build_opts, kernel_src.is_binary));
}
cl::NDRange create_lws_hint_parallel_implementations(unsigned int input_dimension, unsigned int vector_size)
@@ -423,8 +429,9 @@
bool get_wbsm_support_info(const cl::Device &device)
{
cl_bitfield capabilities = 0;
- cl_int err = clGetDeviceInfo(device.get(), CL_DEVICE_SCHEDULING_CONTROLS_CAPABILITIES_ARM, sizeof(cl_bitfield), &capabilities, nullptr);
- if((err == CL_SUCCESS) && (capabilities & CL_KERNEL_EXEC_INFO_WORKGROUP_BATCH_SIZE_MODIFIER_ARM))
+ cl_int err = clGetDeviceInfo(device.get(), CL_DEVICE_SCHEDULING_CONTROLS_CAPABILITIES_ARM, sizeof(cl_bitfield),
+ &capabilities, nullptr);
+ if ((err == CL_SUCCESS) && (capabilities & CL_KERNEL_EXEC_INFO_WORKGROUP_BATCH_SIZE_MODIFIER_ARM))
{
return true;
}
@@ -433,35 +440,33 @@
void set_wbsm(cl::Kernel &kernel, cl_int wbsm_hint)
{
- cl_int err = clSetKernelExecInfo(kernel.get(),
- CL_KERNEL_EXEC_INFO_WORKGROUP_BATCH_SIZE_MODIFIER_ARM,
- sizeof(cl_int),
- &wbsm_hint);
+ cl_int err = clSetKernelExecInfo(kernel.get(), CL_KERNEL_EXEC_INFO_WORKGROUP_BATCH_SIZE_MODIFIER_ARM,
+ sizeof(cl_int), &wbsm_hint);
ARM_COMPUTE_UNUSED(err);
ARM_COMPUTE_ERROR_ON(err != CL_SUCCESS);
}
bool export_to_cl_image(const ITensorInfo *tensor)
{
- if(tensor->tensor_shape()[0] % 4 != 0)
+ if (tensor->tensor_shape()[0] % 4 != 0)
{
return false;
}
// If not floating point
- if(!is_data_type_float(tensor->data_type()))
+ if (!is_data_type_float(tensor->data_type()))
{
return false;
}
// Check if the cl_khr_image2d_from_buffer extension is supported on the target platform
- if(!image2d_from_buffer_supported(CLKernelLibrary::get().get_device()))
+ if (!image2d_from_buffer_supported(CLKernelLibrary::get().get_device()))
{
return false;
}
// Check cl image pitch alignment
- if(get_cl_image_pitch_alignment(CLKernelLibrary::get().get_device()) == 0)
+ if (get_cl_image_pitch_alignment(CLKernelLibrary::get().get_device()) == 0)
{
return false;
}
@@ -471,7 +476,7 @@
const size_t max_image_w = CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_WIDTH>();
const size_t max_image_h = CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_HEIGHT>();
- if(image_w > max_image_w || image_h > max_image_h)
+ if (image_w > max_image_w || image_h > max_image_h)
{
return false;
}
@@ -481,9 +486,9 @@
void set_unroll_with_pragma(CLBuildOptions &built_opts, std::initializer_list<int> values)
{
- for(const int value : values)
+ for (const int value : values)
{
- if(value > max_manual_loop_unrolling)
+ if (value > max_manual_loop_unrolling)
{
built_opts.add_option("-DUNROLL_WITH_PRAGMA");
return;
diff --git a/src/core/CL/CLKernelLibrary.cpp b/src/core/CL/CLKernelLibrary.cpp
index c5a0796..e69d006 100644
--- a/src/core/CL/CLKernelLibrary.cpp
+++ b/src/core/CL/CLKernelLibrary.cpp
@@ -22,8 +22,11 @@
* SOFTWARE.
*/
#include "arm_compute/core/CL/CLKernelLibrary.h"
+
#include "arm_compute/core/Error.h"
+
#include "src/gpu/cl/ClKernelLibrary.h"
+
#include <algorithm>
#include <array>
#include <fstream>
@@ -31,8 +34,7 @@
#include <vector>
namespace arm_compute
{
-CLKernelLibrary::CLKernelLibrary()
- : _compile_context()
+CLKernelLibrary::CLKernelLibrary() : _compile_context()
{
opencl_is_available(); // Make sure the OpenCL symbols are initialised *before* the CLKernelLibrary is built
}
@@ -41,13 +43,15 @@
static CLKernelLibrary _kernel_library;
return _kernel_library;
}
-Kernel CLKernelLibrary::create_kernel(const std::string &kernel_name, const std::set<std::string> &build_options_set) const
+Kernel CLKernelLibrary::create_kernel(const std::string &kernel_name,
+ const std::set<std::string> &build_options_set) const
{
const opencl::ClKernelLibrary &klib = opencl::ClKernelLibrary::get();
const std::string program_name = klib.program_name(kernel_name);
auto program = klib.program(program_name);
const std::string &kernel_path = CLKernelLibrary::get().get_kernel_path();
- return _compile_context.create_kernel(kernel_name, program_name, program.program, kernel_path, build_options_set, program.is_binary);
+ return _compile_context.create_kernel(kernel_name, program_name, program.program, kernel_path, build_options_set,
+ program.is_binary);
}
std::string CLKernelLibrary::get_program_name(const std::string &kernel_name) const
{
@@ -131,4 +135,4 @@
{
return _compile_context;
}
-} // namespace arm_compute
\ No newline at end of file
+} // namespace arm_compute
diff --git a/src/core/CL/CLMutableCommandBuffer.cpp b/src/core/CL/CLMutableCommandBuffer.cpp
index b9c59ac..05b351f 100644
--- a/src/core/CL/CLMutableCommandBuffer.cpp
+++ b/src/core/CL/CLMutableCommandBuffer.cpp
@@ -31,8 +31,7 @@
namespace arm_compute
{
-CLMutableCommandBuffer::CLMutableCommandBuffer(cl_command_queue queue)
- : CLCommandBuffer()
+CLMutableCommandBuffer::CLMutableCommandBuffer(cl_command_queue queue) : CLCommandBuffer()
{
cl_int status = CL_SUCCESS;
@@ -52,7 +51,10 @@
handle_cl_error("clReleaseCommandBufferKHR", status);
}
-void CLMutableCommandBuffer::add_kernel(cl_kernel kernel, const cl::NDRange &offset, const cl::NDRange &global, const cl::NDRange &local)
+void CLMutableCommandBuffer::add_kernel(cl_kernel kernel,
+ const cl::NDRange &offset,
+ const cl::NDRange &global,
+ const cl::NDRange &local)
{
ARM_COMPUTE_ERROR_ON(state() != State::Created);
@@ -65,18 +67,8 @@
};
const auto error = clCommandNDRangeKernelKHR(
- _cb,
- nullptr,
- properties,
- kernel,
- global.dimensions(),
- offset.dimensions() != 0 ? offset.get() : nullptr,
- global.get(),
- local.dimensions() != 0 ? local.get() : nullptr,
- 0,
- nullptr,
- nullptr,
- &mutable_handle);
+ _cb, nullptr, properties, kernel, global.dimensions(), offset.dimensions() != 0 ? offset.get() : nullptr,
+ global.get(), local.dimensions() != 0 ? local.get() : nullptr, 0, nullptr, nullptr, &mutable_handle);
handle_cl_error("clCommandNDRangeKernelKHR", error);
@@ -114,7 +106,7 @@
size_t arg_no = 0;
- for(auto &mut_dispatch_cfg : _mut_dispatch_cfgs)
+ for (auto &mut_dispatch_cfg : _mut_dispatch_cfgs)
{
ARM_COMPUTE_ERROR_ON(arg_no >= _mut_arg_cfgs.size());
mut_dispatch_cfg.arg_list = &_mut_arg_cfgs[arg_no];
@@ -132,9 +124,7 @@
{
ARM_COMPUTE_ERROR_ON(state() != State::Finalized);
- const auto error = clUpdateMutableCommandsKHR(
- _cb,
- &_mut_cfg);
+ const auto error = clUpdateMutableCommandsKHR(_cb, &_mut_cfg);
handle_cl_error("clUpdateMutableCommandsKHR", error);
}
@@ -143,13 +133,7 @@
{
ARM_COMPUTE_ERROR_ON(state() != State::Finalized);
- const auto error = clEnqueueCommandBufferKHR(
- 0,
- nullptr,
- _cb,
- 0,
- nullptr,
- nullptr);
+ const auto error = clEnqueueCommandBufferKHR(0, nullptr, _cb, 0, nullptr, nullptr);
handle_cl_error("clEnqueueCommandBufferKHR", error);
}
diff --git a/src/core/CL/CLMutableCommandBuffer.h b/src/core/CL/CLMutableCommandBuffer.h
index 04e94b0..8997d7d 100644
--- a/src/core/CL/CLMutableCommandBuffer.h
+++ b/src/core/CL/CLMutableCommandBuffer.h
@@ -57,7 +57,10 @@
/** Disallow move assignment. */
CLMutableCommandBuffer &operator=(CLMutableCommandBuffer &&) = delete;
- void add_kernel(cl_kernel kernel, const cl::NDRange &offset, const cl::NDRange &global, const cl::NDRange &local) override;
+ void add_kernel(cl_kernel kernel,
+ const cl::NDRange &offset,
+ const cl::NDRange &global,
+ const cl::NDRange &local) override;
void finalize() override;
diff --git a/src/core/CL/CLUtils.cpp b/src/core/CL/CLUtils.cpp
index 289300b..290ed32 100644
--- a/src/core/CL/CLUtils.cpp
+++ b/src/core/CL/CLUtils.cpp
@@ -26,9 +26,10 @@
#include "arm_compute/core/CL/CLCompileContext.h"
#include "arm_compute/core/CL/CLKernelLibrary.h"
#include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/Validate.h"
#include "arm_compute/core/utils/ActivationFunctionUtils.h"
#include "arm_compute/core/utils/StringUtils.h"
+#include "arm_compute/core/Validate.h"
+
#include "support/StringSupport.h"
namespace arm_compute
@@ -38,15 +39,15 @@
ARM_COMPUTE_ERROR_ON_NULLPTR(tensor);
const cl::Context &ctx = CLKernelLibrary::get().context();
- const cl::Buffer &buffer = tensor->cl_buffer();
+ const cl::Buffer &buffer = tensor->cl_buffer();
const ITensorInfo *info = tensor->info();
- ARM_COMPUTE_ERROR_ON_MSG(info->lock_paddings(),
- "Tensor paddings must not be locked to allow extending paddings to satisfy cl_image pitch alignment requirement");
+ ARM_COMPUTE_ERROR_ON_MSG(info->lock_paddings(), "Tensor paddings must not be locked to allow extending paddings to "
+ "satisfy cl_image pitch alignment requirement");
- const size_t image_w{ info->dimension(0) / 4 };
- const size_t image_h{ info->tensor_shape().total_size() / info->dimension(0) };
- const size_t max_image_w{ CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_WIDTH>() };
- const size_t max_image_h{ CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_HEIGHT>() };
+ const size_t image_w{info->dimension(0) / 4};
+ const size_t image_h{info->tensor_shape().total_size() / info->dimension(0)};
+ const size_t max_image_w{CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_WIDTH>()};
+ const size_t max_image_h{CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_HEIGHT>()};
ARM_COMPUTE_UNUSED(max_image_w, max_image_h);
ARM_COMPUTE_ERROR_ON_MSG(image_w > max_image_w, "Image width exceeds maximum width for exporting to cl_image");
@@ -58,18 +59,22 @@
return create_image2d_from_buffer(ctx, buffer, shape2d, info->data_type(), image_row_pitch, image_type);
}
-cl::Image2D create_image2d_from_buffer(const cl::Context &ctx, const cl::Buffer &buffer, const TensorShape &shape2d, DataType data_type, size_t image_row_pitch, CLImage2DType image_type)
+cl::Image2D create_image2d_from_buffer(const cl::Context &ctx,
+ const cl::Buffer &buffer,
+ const TensorShape &shape2d,
+ DataType data_type,
+ size_t image_row_pitch,
+ CLImage2DType image_type)
{
ARM_COMPUTE_ERROR_ON_MSG(!image2d_from_buffer_supported(CLKernelLibrary::get().get_device()),
"The extension cl_khr_image2d_from_buffer is not supported on the target platform");
ARM_COMPUTE_ERROR_ON_MSG(get_cl_image_pitch_alignment(CLKernelLibrary::get().get_device()) == 0,
"Impossible to retrieve the cl_image pitch alignment");
- ARM_COMPUTE_ERROR_ON_MSG(buffer.get() == nullptr,
- "Cannot create cl_image from empty cl_buffer");
+ ARM_COMPUTE_ERROR_ON_MSG(buffer.get() == nullptr, "Cannot create cl_image from empty cl_buffer");
cl_channel_type cl_data_type;
- switch(data_type)
+ switch (data_type)
{
case DataType::F32:
cl_data_type = CL_FLOAT;
@@ -84,7 +89,7 @@
cl_mem cl_image;
cl_int err = CL_SUCCESS;
- const cl_image_format format = { CL_RGBA, cl_data_type };
+ const cl_image_format format = {CL_RGBA, cl_data_type};
cl_image_desc desc;
memset(&desc, 0, sizeof(desc));
@@ -94,7 +99,7 @@
desc.image_width = shape2d[0];
desc.image_height = shape2d[1];
- switch(image_type)
+ switch (image_type)
{
case CLImage2DType::ReadOnly:
cl_image = clCreateImage(ctx(), CL_MEM_READ_ONLY, &format, &desc, nullptr, &err);
@@ -114,7 +119,7 @@
void handle_cl_error(const std::string &function_name, cl_int error_code)
{
- if(error_code != CL_SUCCESS)
+ if (error_code != CL_SUCCESS)
{
std::string error_message = function_name + " - Error code: " + std::to_string(error_code);
ARM_COMPUTE_ERROR(error_message.c_str());
diff --git a/src/core/CL/CLUtils.h b/src/core/CL/CLUtils.h
index de9c1b3..f9dcfea 100644
--- a/src/core/CL/CLUtils.h
+++ b/src/core/CL/CLUtils.h
@@ -72,7 +72,12 @@
*
* @return cl::Image2D object
*/
-cl::Image2D create_image2d_from_buffer(const cl::Context &ctx, const cl::Buffer &buffer, const TensorShape &shape2d, DataType data_type, size_t image_row_pitch, CLImage2DType image_type);
+cl::Image2D create_image2d_from_buffer(const cl::Context &ctx,
+ const cl::Buffer &buffer,
+ const TensorShape &shape2d,
+ DataType data_type,
+ size_t image_row_pitch,
+ CLImage2DType image_type);
/** Check for CL error code and throw exception accordingly.
*
diff --git a/src/core/CL/CLValidate.h b/src/core/CL/CLValidate.h
index 7b5294e..50d224f 100644
--- a/src/core/CL/CLValidate.h
+++ b/src/core/CL/CLValidate.h
@@ -29,11 +29,13 @@
namespace arm_compute
{
-#define ARM_COMPUTE_ERROR_ON_F16_UNSUPPORTED(tensor) \
- ARM_COMPUTE_ERROR_THROW_ON(::arm_compute::error_on_unsupported_fp16(__func__, __FILE__, __LINE__, tensor, CLKernelLibrary::get().fp16_supported()))
+#define ARM_COMPUTE_ERROR_ON_F16_UNSUPPORTED(tensor) \
+ ARM_COMPUTE_ERROR_THROW_ON(::arm_compute::error_on_unsupported_fp16(__func__, __FILE__, __LINE__, tensor, \
+ CLKernelLibrary::get().fp16_supported()))
-#define ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(tensor) \
- ARM_COMPUTE_RETURN_ON_ERROR(::arm_compute::error_on_unsupported_fp16(__func__, __FILE__, __LINE__, tensor, CLKernelLibrary::get().fp16_supported()))
+#define ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(tensor) \
+ ARM_COMPUTE_RETURN_ON_ERROR(::arm_compute::error_on_unsupported_fp16(__func__, __FILE__, __LINE__, tensor, \
+ CLKernelLibrary::get().fp16_supported()))
/** Return an error if int64_base_atomics extension is not supported by the device.
*
@@ -43,11 +45,13 @@
*
* @return Status
*/
-inline arm_compute::Status error_on_unsupported_int64_base_atomics(const char *function, const char *file, const int line)
+inline arm_compute::Status
+error_on_unsupported_int64_base_atomics(const char *function, const char *file, const int line)
{
- if(!CLKernelLibrary::get().int64_base_atomics_supported())
+ if (!CLKernelLibrary::get().int64_base_atomics_supported())
{
- return ARM_COMPUTE_CREATE_ERROR_LOC(arm_compute::ErrorCode::UNSUPPORTED_EXTENSION_USE, function, file, line, "Atomic functions are not supported");
+ return ARM_COMPUTE_CREATE_ERROR_LOC(arm_compute::ErrorCode::UNSUPPORTED_EXTENSION_USE, function, file, line,
+ "Atomic functions are not supported");
}
return arm_compute::Status{};
}
diff --git a/src/core/CL/DefaultLWSHeuristics.cpp b/src/core/CL/DefaultLWSHeuristics.cpp
index a53fdbb..f96b24d 100644
--- a/src/core/CL/DefaultLWSHeuristics.cpp
+++ b/src/core/CL/DefaultLWSHeuristics.cpp
@@ -31,13 +31,13 @@
{
ARM_COMPUTE_UNUSED(gws_y);
- if(gws_z != 1)
+ if (gws_z != 1)
{
return cl::NDRange(4, 4, 2);
}
else
{
- if(gws_x > 256)
+ if (gws_x > 256)
{
return cl::NDRange(2, 16, 1);
}
@@ -59,9 +59,9 @@
{
ARM_COMPUTE_UNUSED(gws_z);
- if(gws_x < gws_y)
+ if (gws_x < gws_y)
{
- if(gws_x < 4)
+ if (gws_x < 4)
{
return cl::NDRange(std::min(gws_x, static_cast<size_t>(2u)), 32, 1);
}
@@ -81,7 +81,7 @@
ARM_COMPUTE_UNUSED(gws_y);
ARM_COMPUTE_UNUSED(gws_z);
- if(gws_x < 32)
+ if (gws_x < 32)
{
return cl::NDRange(gws_x, 4, 4);
}
@@ -100,7 +100,7 @@
const size_t gws_y = gws[1];
const size_t gws_z = gws[2];
- switch(kernel_type)
+ switch (kernel_type)
{
case CLKernelType::GEMM:
{
@@ -124,4 +124,4 @@
}
}
}
-} // namespace arm_compute
\ No newline at end of file
+} // namespace arm_compute
diff --git a/src/core/CL/ICLKernel.cpp b/src/core/CL/ICLKernel.cpp
index dc3a86a..ac53e7f 100644
--- a/src/core/CL/ICLKernel.cpp
+++ b/src/core/CL/ICLKernel.cpp
@@ -25,18 +25,23 @@
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/Helpers.h"
+
#include "src/core/helpers/Utils.h"
#include <cstddef>
-void arm_compute::enqueue(cl::CommandQueue &queue, ICLKernel &kernel, const Window &window, const cl::NDRange &lws_hint, bool use_dummy_work_items)
+void arm_compute::enqueue(cl::CommandQueue &queue,
+ ICLKernel &kernel,
+ const Window &window,
+ const cl::NDRange &lws_hint,
+ bool use_dummy_work_items)
{
- if(kernel.kernel()() == nullptr)
+ if (kernel.kernel()() == nullptr)
{
return;
}
- for(unsigned int i = 0; i < Coordinates::num_max_dimensions; ++i)
+ for (unsigned int i = 0; i < Coordinates::num_max_dimensions; ++i)
{
ARM_COMPUTE_ERROR_ON(window[i].step() == 0);
// Make sure that dimensions > Z are 1
@@ -46,7 +51,7 @@
cl::NDRange gws = ICLKernel::gws_from_window(window, use_dummy_work_items);
// Check for empty NDRange
- if(gws.dimensions() == 0)
+ if (gws.dimensions() == 0)
{
return;
}
@@ -54,7 +59,7 @@
kernel.cache_gws(gws);
cl::NDRange valid_lws;
- if(lws_hint[0] * lws_hint[1] * lws_hint[2] > kernel.get_max_workgroup_size())
+ if (lws_hint[0] * lws_hint[1] * lws_hint[2] > kernel.get_max_workgroup_size())
{
valid_lws = cl::NullRange;
}
@@ -65,12 +70,12 @@
cl::NDRange lws = cl::NullRange;
- if((valid_lws[0] <= gws[0]) && (valid_lws[1] <= gws[1]) && (valid_lws[2] <= gws[2]))
+ if ((valid_lws[0] <= gws[0]) && (valid_lws[1] <= gws[1]) && (valid_lws[2] <= gws[2]))
{
lws = valid_lws;
}
- if(CLKernelLibrary::get().is_wbsm_supported())
+ if (CLKernelLibrary::get().is_wbsm_supported())
{
set_wbsm(kernel.kernel(), kernel.wbsm_hint());
}
@@ -90,7 +95,7 @@
// Calculate offset to the start of the window
unsigned int offset_first_element = info->offset_first_element_in_bytes();
- for(unsigned int n = 0; n < info->num_dimensions(); ++n)
+ for (unsigned int n = 0; n < info->num_dimensions(); ++n)
{
offset_first_element += (window.is_broadcasted(n) ? 0 : window[n].start()) * strides[n];
}
@@ -98,7 +103,7 @@
unsigned int idx_start = idx;
_kernel.setArg(idx++, tensor->cl_buffer());
- for(unsigned int d = 0; d < dimension_size; ++d)
+ for (unsigned int d = 0; d < dimension_size; ++d)
{
_kernel.setArg<cl_uint>(idx++, window.is_broadcasted(d) ? 0 : strides[d]);
_kernel.setArg<cl_uint>(idx++, window.is_broadcasted(d) ? 0 : (strides[d] * window[d].step()));
@@ -107,7 +112,8 @@
_kernel.setArg<cl_uint>(idx++, offset_first_element);
ARM_COMPUTE_ERROR_ON_MSG_VAR(idx_start + num_arguments_per_tensor<dimension_size>() != idx,
- "add_%dD_tensor_argument() is supposed to add exactly %d arguments to the kernel", dimension_size, num_arguments_per_tensor<dimension_size>());
+ "add_%dD_tensor_argument() is supposed to add exactly %d arguments to the kernel",
+ dimension_size, num_arguments_per_tensor<dimension_size>());
ARM_COMPUTE_UNUSED(idx_start);
}
@@ -178,7 +184,7 @@
size_t ICLKernel::get_max_workgroup_size()
{
- if(_max_workgroup_size == 0)
+ if (_max_workgroup_size == 0)
{
_max_workgroup_size = CLKernelLibrary::get().max_local_workgroup_size(_kernel);
}
@@ -187,7 +193,7 @@
cl::NDRange ICLKernel::gws_from_window(const Window &window, bool use_dummy_work_items)
{
- if((window.x().end() - window.x().start()) == 0 || (window.y().end() - window.y().start()) == 0)
+ if ((window.x().end() - window.x().start()) == 0 || (window.y().end() - window.y().start()) == 0)
{
return cl::NullRange;
}
@@ -196,7 +202,7 @@
(window.y().end() - window.y().start()) / window.y().step(),
(window.z().end() - window.z().start()) / window.z().step());
- if(use_dummy_work_items)
+ if (use_dummy_work_items)
{
gws.get()[0] = get_next_power_two(gws[0]);
gws.get()[1] = get_next_power_two(gws[1]);
diff --git a/src/core/CL/ICLKernel.h b/src/core/CL/ICLKernel.h
index c82809c..6aebef1 100644
--- a/src/core/CL/ICLKernel.h
+++ b/src/core/CL/ICLKernel.h
@@ -27,10 +27,10 @@
#include "arm_compute/core/CL/CLKernelLibrary.h"
#include "arm_compute/core/CL/CLTypes.h"
#include "arm_compute/core/CL/OpenCL.h"
+#include "arm_compute/core/experimental/Types.h"
#include "arm_compute/core/GPUTarget.h"
#include "arm_compute/core/IKernel.h"
#include "arm_compute/core/Validate.h"
-#include "arm_compute/core/experimental/Types.h"
#include "arm_compute/runtime/CL/CLTuningParams.h"
#include "src/core/CL/DefaultLWSHeuristics.h"
@@ -43,14 +43,14 @@
{
bool is_same_lws(cl::NDRange lws0, cl::NDRange lws1)
{
- if(lws0.dimensions() != lws1.dimensions())
+ if (lws0.dimensions() != lws1.dimensions())
{
return false;
}
- for(size_t i = 0; i < lws0.dimensions(); ++i)
+ for (size_t i = 0; i < lws0.dimensions(); ++i)
{
- if(lws0.get()[i] != lws1.get()[i])
+ if (lws0.get()[i] != lws1.get()[i])
{
return false;
}
@@ -71,7 +71,7 @@
*
* @return The number of arguments enqueued per array object.
*/
- template <unsigned int dimension_size>
+ template <unsigned int dimension_size>
constexpr static unsigned int num_arguments_per_array()
{
return num_arguments_per_tensor<dimension_size>();
@@ -80,7 +80,7 @@
*
* @return The number of arguments enqueued per tensor object.
*/
- template <unsigned int dimension_size>
+ template <unsigned int dimension_size>
constexpr static unsigned int num_arguments_per_tensor()
{
return 2 + 2 * dimension_size;
@@ -116,11 +116,13 @@
* @param[in] window The maximum window which will be returned by window()
* @param[in] tuning_params_hint (Optional) Tuning parameters to use.
*/
- void configure_internal(const Window &window, CLTuningParams tuning_params_hint = CLTuningParams(CLKernelLibrary::get().default_ndrange(), 0))
+ void configure_internal(const Window &window,
+ CLTuningParams tuning_params_hint = CLTuningParams(CLKernelLibrary::get().default_ndrange(),
+ 0))
{
_tuning_params_hint = tuning_params_hint;
- if(is_same_lws(_tuning_params_hint.get_lws(), CLKernelLibrary::get().default_ndrange()))
+ if (is_same_lws(_tuning_params_hint.get_lws(), CLKernelLibrary::get().default_ndrange()))
{
// Disable use_dummy_work_items at configure time. Because dummy work items only affect gws size, which
// will be recalculated with use_dummy_work_items flag at run time again anyway.
@@ -133,7 +135,13 @@
public:
/** Constructor */
ICLKernel()
- : _kernel(nullptr), _target(GPUTarget::MIDGARD), _config_id(arm_compute::default_config_id), _max_workgroup_size(0), _type(CLKernelType::UNKNOWN), _tuning_params_hint(), _cached_gws(cl::NullRange)
+ : _kernel(nullptr),
+ _target(GPUTarget::MIDGARD),
+ _config_id(arm_compute::default_config_id),
+ _max_workgroup_size(0),
+ _type(CLKernelType::UNKNOWN),
+ _tuning_params_hint(),
+ _cached_gws(cl::NullRange)
{
}
/** Returns a reference to the OpenCL kernel of this object.
@@ -161,7 +169,11 @@
* @param[in] window Window the kernel will be executed on.
*/
template <typename T>
- void add_1D_array_argument(unsigned int &idx, const ICLArray<T> *array, const Strides &strides, unsigned int num_dimensions, const Window &window)
+ void add_1D_array_argument(unsigned int &idx,
+ const ICLArray<T> *array,
+ const Strides &strides,
+ unsigned int num_dimensions,
+ const Window &window)
{
add_array_argument<T, 1>(idx, array, strides, num_dimensions, window);
}
@@ -184,7 +196,7 @@
*/
void add_1D_tensor_argument_if(bool cond, unsigned int &idx, const ICLTensor *tensor, const Window &window)
{
- if(cond)
+ if (cond)
{
add_1D_tensor_argument(idx, tensor, window);
}
@@ -208,7 +220,7 @@
*/
void add_2D_tensor_argument_if(bool cond, unsigned int &idx, const ICLTensor *tensor, const Window &window)
{
- if(cond)
+ if (cond)
{
add_2D_tensor_argument(idx, tensor, window);
}
@@ -469,7 +481,11 @@
* @param[in] window Window the kernel will be executed on.
*/
template <typename T, unsigned int dimension_size>
- void add_array_argument(unsigned int &idx, const ICLArray<T> *array, const Strides &strides, unsigned int num_dimensions, const Window &window);
+ void add_array_argument(unsigned int &idx,
+ const ICLArray<T> *array,
+ const Strides &strides,
+ unsigned int num_dimensions,
+ const Window &window);
/** Add the passed tensor's parameters to the object's kernel's arguments starting from the index idx.
*
* @param[in,out] idx Index at which to start adding the tensor's arguments. Will be incremented by the number of kernel arguments set.
@@ -505,7 +521,11 @@
*
* @note If any dimension of the lws is greater than the global workgroup size then no lws will be passed.
*/
-void enqueue(cl::CommandQueue &queue, ICLKernel &kernel, const Window &window, const cl::NDRange &lws_hint = CLKernelLibrary::get().default_ndrange(), bool use_dummy_work_items = false);
+void enqueue(cl::CommandQueue &queue,
+ ICLKernel &kernel,
+ const Window &window,
+ const cl::NDRange &lws_hint = CLKernelLibrary::get().default_ndrange(),
+ bool use_dummy_work_items = false);
/** Add the passed array's parameters to the object's kernel's arguments starting from the index idx.
*
@@ -516,14 +536,15 @@
* @param[in] window Window the kernel will be executed on.
*/
template <typename T, unsigned int dimension_size>
-void ICLKernel::add_array_argument(unsigned &idx, const ICLArray<T> *array, const Strides &strides, unsigned int num_dimensions, const Window &window)
+void ICLKernel::add_array_argument(
+ unsigned &idx, const ICLArray<T> *array, const Strides &strides, unsigned int num_dimensions, const Window &window)
{
ARM_COMPUTE_ERROR_ON(array == nullptr);
// Calculate offset to the start of the window
unsigned int offset_first_element = 0;
- for(unsigned int n = 0; n < num_dimensions; ++n)
+ for (unsigned int n = 0; n < num_dimensions; ++n)
{
offset_first_element += window[n].start() * strides[n];
}
@@ -531,7 +552,7 @@
unsigned int idx_start = idx;
_kernel.setArg(idx++, array->cl_buffer());
- for(unsigned int dimension = 0; dimension < dimension_size; dimension++)
+ for (unsigned int dimension = 0; dimension < dimension_size; dimension++)
{
_kernel.setArg<cl_uint>(idx++, strides[dimension]);
_kernel.setArg<cl_uint>(idx++, strides[dimension] * window[dimension].step());
@@ -540,8 +561,9 @@
_kernel.setArg<cl_uint>(idx++, offset_first_element);
ARM_COMPUTE_ERROR_ON_MSG_VAR(idx_start + num_arguments_per_array<dimension_size>() != idx,
- "add_%dD_array_argument() is supposed to add exactly %d arguments to the kernel", dimension_size, num_arguments_per_array<dimension_size>());
+ "add_%dD_array_argument() is supposed to add exactly %d arguments to the kernel",
+ dimension_size, num_arguments_per_array<dimension_size>());
ARM_COMPUTE_UNUSED(idx_start);
}
-}
+} // namespace arm_compute
#endif /*ARM_COMPUTE_ICLKERNEL_H */
diff --git a/src/core/CL/ICLSimple2DKernel.cpp b/src/core/CL/ICLSimple2DKernel.cpp
index 5d8295b..3f7edbb 100644
--- a/src/core/CL/ICLSimple2DKernel.cpp
+++ b/src/core/CL/ICLSimple2DKernel.cpp
@@ -40,6 +40,5 @@
add_2D_tensor_argument(idx, _input, slice);
add_2D_tensor_argument(idx, _output, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(window.slide_window_slice_2D(slice));
+ } while (window.slide_window_slice_2D(slice));
}
diff --git a/src/core/CL/ICLSimple2DKernel.h b/src/core/CL/ICLSimple2DKernel.h
index 5246492..97bc1e5 100644
--- a/src/core/CL/ICLSimple2DKernel.h
+++ b/src/core/CL/ICLSimple2DKernel.h
@@ -37,5 +37,5 @@
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
};
-}
+} // namespace arm_compute
#endif /*ARM_COMPUTE_ICLSIMPLE2DKERNEL_H */
diff --git a/src/core/CL/ICLSimple3DKernel.cpp b/src/core/CL/ICLSimple3DKernel.cpp
index fef1a86..71d7d1f 100644
--- a/src/core/CL/ICLSimple3DKernel.cpp
+++ b/src/core/CL/ICLSimple3DKernel.cpp
@@ -42,6 +42,5 @@
add_3D_tensor_argument(idx, _input, slice);
add_3D_tensor_argument(idx, _output, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(window.slide_window_slice_3D(slice));
+ } while (window.slide_window_slice_3D(slice));
}
diff --git a/src/core/CL/ICLSimple3DKernel.h b/src/core/CL/ICLSimple3DKernel.h
index ff0b274..5071b6b 100644
--- a/src/core/CL/ICLSimple3DKernel.h
+++ b/src/core/CL/ICLSimple3DKernel.h
@@ -39,5 +39,5 @@
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
};
-}
+} // namespace arm_compute
#endif /*ARM_COMPUTE_ICLSIMPLE3DKERNEL_H */
diff --git a/src/core/CL/ICLSimpleKernel.cpp b/src/core/CL/ICLSimpleKernel.cpp
index d67fefd..c31db83 100644
--- a/src/core/CL/ICLSimpleKernel.cpp
+++ b/src/core/CL/ICLSimpleKernel.cpp
@@ -22,30 +22,35 @@
* SOFTWARE.
*/
#include "src/core/CL/ICLSimpleKernel.h"
+
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/IAccessWindow.h"
#include "arm_compute/core/Validate.h"
#include "arm_compute/core/Window.h"
+
#include "src/core/helpers/WindowHelpers.h"
using namespace arm_compute;
-ICLSimpleKernel::ICLSimpleKernel()
- : _input(nullptr), _output(nullptr)
+ICLSimpleKernel::ICLSimpleKernel() : _input(nullptr), _output(nullptr)
{
}
-void ICLSimpleKernel::configure(const ICLTensor *input, ICLTensor *output, unsigned int num_elems_processed_per_iteration, bool border_undefined, const BorderSize &border_size)
+void ICLSimpleKernel::configure(const ICLTensor *input,
+ ICLTensor *output,
+ unsigned int num_elems_processed_per_iteration,
+ bool border_undefined,
+ const BorderSize &border_size)
{
_input = input;
_output = output;
// Configure kernel window
- Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration), border_undefined, border_size);
+ Window win =
+ calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration), border_undefined, border_size);
AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration);
- update_window_and_padding(win,
- AccessWindowHorizontal(input->info(), 0, num_elems_processed_per_iteration),
+ update_window_and_padding(win, AccessWindowHorizontal(input->info(), 0, num_elems_processed_per_iteration),
output_access);
output_access.set_valid_region(win, input->info()->valid_region(), border_undefined, border_size);
diff --git a/src/core/CL/ICLSimpleKernel.h b/src/core/CL/ICLSimpleKernel.h
index b35547a..6afd730 100644
--- a/src/core/CL/ICLSimpleKernel.h
+++ b/src/core/CL/ICLSimpleKernel.h
@@ -26,6 +26,7 @@
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/Helpers.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -55,12 +56,16 @@
* @param[in] border_undefined (Optional) True if the border mode is undefined. False if it's replicate or constant.
* @param[in] border_size (Optional) Size of the border.
*/
- void configure(const ICLTensor *input, ICLTensor *output, unsigned int num_elems_processed_per_iteration, bool border_undefined = false, const BorderSize &border_size = BorderSize());
+ void configure(const ICLTensor *input,
+ ICLTensor *output,
+ unsigned int num_elems_processed_per_iteration,
+ bool border_undefined = false,
+ const BorderSize &border_size = BorderSize());
protected:
const ICLTensor *_input;
ICLTensor *_output;
};
-}
+} // namespace arm_compute
#endif /*ARM_COMPUTE_ICLSIMPLEKERNEL_H */
diff --git a/src/core/CL/ICLTensor.cpp b/src/core/CL/ICLTensor.cpp
index b541bff..0771db7 100644
--- a/src/core/CL/ICLTensor.cpp
+++ b/src/core/CL/ICLTensor.cpp
@@ -27,8 +27,7 @@
using namespace arm_compute;
-ICLTensor::ICLTensor()
- : _mapping(nullptr)
+ICLTensor::ICLTensor() : _mapping(nullptr)
{
}
diff --git a/src/core/CL/OpenCL.cpp b/src/core/CL/OpenCL.cpp
index b092dfb..35421d0 100644
--- a/src/core/CL/OpenCL.cpp
+++ b/src/core/CL/OpenCL.cpp
@@ -36,11 +36,7 @@
namespace arm_compute
{
-CLSymbols::CLSymbols() noexcept(false)
- : _loaded(
-{
- false, false
-})
+CLSymbols::CLSymbols() noexcept(false) : _loaded({false, false})
{
}
@@ -52,9 +48,9 @@
bool CLSymbols::load_default()
{
- static const std::vector<std::string> libraries_filenames{ "libOpenCL.so", "libGLES_mali.so", "libmali.so" };
+ static const std::vector<std::string> libraries_filenames{"libOpenCL.so", "libGLES_mali.so", "libmali.so"};
- if(_loaded.first)
+ if (_loaded.first)
{
return _loaded.second;
}
@@ -62,34 +58,32 @@
// Indicate that default loading has been tried
_loaded.first = true;
- if(load(libraries_filenames, /* use_loader */ false))
+ if (load(libraries_filenames, /* use_loader */ false))
{
- ARM_COMPUTE_ERROR_ON_MSG(this->clBuildProgram_ptr == nullptr, "Failed to load OpenCL symbols from shared library");
+ ARM_COMPUTE_ERROR_ON_MSG(this->clBuildProgram_ptr == nullptr,
+ "Failed to load OpenCL symbols from shared library");
return true;
}
#ifdef __ANDROID__
// When running in NDK environment, the above libraries are not accessible.
- static const std::vector<std::string> android_libraries_filenames{ "libOpenCL-pixel.so", "libOpenCL-car.so" };
+ static const std::vector<std::string> android_libraries_filenames{"libOpenCL-pixel.so", "libOpenCL-car.so"};
- if(load(android_libraries_filenames, /* use_loader */ true))
+ if (load(android_libraries_filenames, /* use_loader */ true))
{
- ARM_COMPUTE_ERROR_ON_MSG(this->clBuildProgram_ptr == nullptr, "Failed to load OpenCL symbols from android shared library");
+ ARM_COMPUTE_ERROR_ON_MSG(this->clBuildProgram_ptr == nullptr,
+ "Failed to load OpenCL symbols from android shared library");
return true;
}
#endif // __ANDROID__
// If not returned till here then libraries not found
std::stringstream ss;
- std::for_each(libraries_filenames.begin(), libraries_filenames.end(), [&ss](const std::string & s)
- {
- ss << s << " ";
- });
+ std::for_each(libraries_filenames.begin(), libraries_filenames.end(),
+ [&ss](const std::string &s) { ss << s << " "; });
#ifdef __ANDROID__
- std::for_each(android_libraries_filenames.begin(), android_libraries_filenames.end(), [&ss](const std::string & s)
- {
- ss << s << " ";
- });
+ std::for_each(android_libraries_filenames.begin(), android_libraries_filenames.end(),
+ [&ss](const std::string &s) { ss << s << " "; });
#endif // __ANDROID__
std::cerr << "Couldn't find any of the following OpenCL library: " << ss.str() << std::endl;
return false;
@@ -99,15 +93,15 @@
{
void *handle = nullptr;
unsigned int index = 0;
- for(index = 0; index < libraries_filenames.size(); ++index)
+ for (index = 0; index < libraries_filenames.size(); ++index)
{
handle = dlopen(libraries_filenames[index].c_str(), RTLD_LAZY | RTLD_LOCAL);
- if(handle != nullptr)
+ if (handle != nullptr)
{
break;
}
}
- if(index == libraries_filenames.size())
+ if (index == libraries_filenames.size())
{
// Set status of loading to failed
_loaded.second = false;
@@ -115,22 +109,23 @@
}
#ifdef __ANDROID__
- typedef void* (*loadOpenCLPointer_t)(const char* name);
+ typedef void *(*loadOpenCLPointer_t)(const char *name);
loadOpenCLPointer_t loadOpenCLPointer;
- if (use_loader) {
+ if (use_loader)
+ {
typedef void (*enableOpenCL_t)();
- enableOpenCL_t enableOpenCL =
- reinterpret_cast<enableOpenCL_t>(dlsym(handle, "enableOpenCL"));
+ enableOpenCL_t enableOpenCL = reinterpret_cast<enableOpenCL_t>(dlsym(handle, "enableOpenCL"));
enableOpenCL();
- loadOpenCLPointer = reinterpret_cast<loadOpenCLPointer_t>(
- dlsym(handle, "loadOpenCLPointer"));
- } else {
+ loadOpenCLPointer = reinterpret_cast<loadOpenCLPointer_t>(dlsym(handle, "loadOpenCLPointer"));
+ }
+ else
+ {
loadOpenCLPointer = nullptr;
}
-#define LOAD_FUNCTION_PTR(func_name, _handle) \
- func_name##_ptr = reinterpret_cast<decltype(func_name) *>( use_loader ? \
- loadOpenCLPointer(#func_name) : dlsym(handle, #func_name));
+#define LOAD_FUNCTION_PTR(func_name, _handle) \
+ func_name##_ptr = reinterpret_cast<decltype(func_name) *>(use_loader ? loadOpenCLPointer(#func_name) \
+ : dlsym(handle, #func_name));
#else /* __ANDROID__ */
(void)use_loader; // Avoid unused warning
#define LOAD_FUNCTION_PTR(func_name, handle) \
@@ -234,12 +229,11 @@
}
} // namespace arm_compute
-cl_int clEnqueueMarker(cl_command_queue command_queue,
- cl_event *event)
+cl_int clEnqueueMarker(cl_command_queue command_queue, cl_event *event)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clEnqueueMarker_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_queue, event);
}
@@ -249,12 +243,11 @@
}
}
-cl_int clWaitForEvents(cl_uint num_events,
- const cl_event *event_list)
+cl_int clWaitForEvents(cl_uint num_events, const cl_event *event_list)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clWaitForEvents_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(num_events, event_list);
}
@@ -264,12 +257,18 @@
}
}
-cl_int clEnqueueSVMMap(cl_command_queue command_queue, cl_bool blocking_map, cl_map_flags flags, void *svm_ptr,
- size_t size, cl_uint num_events_in_wait_list, const cl_event *event_wait_list, cl_event *event)
+cl_int clEnqueueSVMMap(cl_command_queue command_queue,
+ cl_bool blocking_map,
+ cl_map_flags flags,
+ void *svm_ptr,
+ size_t size,
+ cl_uint num_events_in_wait_list,
+ const cl_event *event_wait_list,
+ cl_event *event)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clEnqueueSVMMap_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_queue, blocking_map, flags, svm_ptr, size, num_events_in_wait_list, event_wait_list, event);
}
@@ -279,12 +278,15 @@
}
}
-cl_int clEnqueueSVMUnmap(cl_command_queue command_queue, void *svm_ptr, cl_uint num_events_in_wait_list,
- const cl_event *event_wait_list, cl_event *event)
+cl_int clEnqueueSVMUnmap(cl_command_queue command_queue,
+ void *svm_ptr,
+ cl_uint num_events_in_wait_list,
+ const cl_event *event_wait_list,
+ cl_event *event)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clEnqueueSVMUnmap_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_queue, svm_ptr, num_events_in_wait_list, event_wait_list, event);
}
@@ -298,7 +300,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clSVMAlloc_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(context, flags, size, alignment);
}
@@ -312,7 +314,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clSVMFree_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
func(context, svm_pointer);
}
@@ -326,7 +328,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clGetContextInfo_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(context, param_name, param_value_size, param_value, param_value_size_ret);
}
@@ -343,7 +345,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clCreateCommandQueue_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(context, device, properties, errcode_ret);
}
@@ -360,7 +362,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clCreateCommandQueueWithProperties_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(context, device, properties, errcode_ret);
}
@@ -370,17 +372,16 @@
}
}
-cl_context clCreateContext(
- const cl_context_properties *properties,
- cl_uint num_devices,
- const cl_device_id *devices,
- void (*pfn_notify)(const char *, const void *, size_t, void *),
- void *user_data,
- cl_int *errcode_ret)
+cl_context clCreateContext(const cl_context_properties *properties,
+ cl_uint num_devices,
+ const cl_device_id *devices,
+ void (*pfn_notify)(const char *, const void *, size_t, void *),
+ void *user_data,
+ cl_int *errcode_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clCreateContext_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(properties, num_devices, devices, pfn_notify, user_data, errcode_ret);
}
@@ -398,7 +399,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clCreateContextFromType_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(properties, device_type, pfn_notify, user_data, errcode_ret);
}
@@ -408,17 +409,16 @@
}
}
-cl_int clBuildProgram(
- cl_program program,
- cl_uint num_devices,
- const cl_device_id *device_list,
- const char *options,
- void(CL_CALLBACK *pfn_notify)(cl_program program, void *user_data),
- void *user_data)
+cl_int clBuildProgram(cl_program program,
+ cl_uint num_devices,
+ const cl_device_id *device_list,
+ const char *options,
+ void(CL_CALLBACK *pfn_notify)(cl_program program, void *user_data),
+ void *user_data)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clBuildProgram_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(program, num_devices, device_list, options, pfn_notify, user_data);
}
@@ -428,22 +428,22 @@
}
}
-cl_int clEnqueueNDRangeKernel(
- cl_command_queue command_queue,
- cl_kernel kernel,
- cl_uint work_dim,
- const size_t *global_work_offset,
- const size_t *global_work_size,
- const size_t *local_work_size,
- cl_uint num_events_in_wait_list,
- const cl_event *event_wait_list,
- cl_event *event)
+cl_int clEnqueueNDRangeKernel(cl_command_queue command_queue,
+ cl_kernel kernel,
+ cl_uint work_dim,
+ const size_t *global_work_offset,
+ const size_t *global_work_size,
+ const size_t *local_work_size,
+ cl_uint num_events_in_wait_list,
+ const cl_event *event_wait_list,
+ cl_event *event)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clEnqueueNDRangeKernel_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
- return func(command_queue, kernel, work_dim, global_work_offset, global_work_size, local_work_size, num_events_in_wait_list, event_wait_list, event);
+ return func(command_queue, kernel, work_dim, global_work_offset, global_work_size, local_work_size,
+ num_events_in_wait_list, event_wait_list, event);
}
else
{
@@ -451,15 +451,11 @@
}
}
-cl_int clSetKernelArg(
- cl_kernel kernel,
- cl_uint arg_index,
- size_t arg_size,
- const void *arg_value)
+cl_int clSetKernelArg(cl_kernel kernel, cl_uint arg_index, size_t arg_size, const void *arg_value)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clSetKernelArg_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(kernel, arg_index, arg_size, arg_value);
}
@@ -473,7 +469,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clRetainMemObject_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(memobj);
}
@@ -487,7 +483,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clReleaseMemObject_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(memobj);
}
@@ -497,17 +493,16 @@
}
}
-cl_int clEnqueueUnmapMemObject(
- cl_command_queue command_queue,
- cl_mem memobj,
- void *mapped_ptr,
- cl_uint num_events_in_wait_list,
- const cl_event *event_wait_list,
- cl_event *event)
+cl_int clEnqueueUnmapMemObject(cl_command_queue command_queue,
+ cl_mem memobj,
+ void *mapped_ptr,
+ cl_uint num_events_in_wait_list,
+ const cl_event *event_wait_list,
+ cl_event *event)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clEnqueueUnmapMemObject_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_queue, memobj, mapped_ptr, num_events_in_wait_list, event_wait_list, event);
}
@@ -521,7 +516,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clRetainCommandQueue_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_queue);
}
@@ -535,7 +530,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clReleaseContext_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(context);
}
@@ -548,7 +543,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clReleaseEvent_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(event);
}
@@ -558,22 +553,22 @@
}
}
-cl_int clEnqueueWriteBuffer(
- cl_command_queue command_queue,
- cl_mem buffer,
- cl_bool blocking_write,
- size_t offset,
- size_t size,
- const void *ptr,
- cl_uint num_events_in_wait_list,
- const cl_event *event_wait_list,
- cl_event *event)
+cl_int clEnqueueWriteBuffer(cl_command_queue command_queue,
+ cl_mem buffer,
+ cl_bool blocking_write,
+ size_t offset,
+ size_t size,
+ const void *ptr,
+ cl_uint num_events_in_wait_list,
+ const cl_event *event_wait_list,
+ cl_event *event)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clEnqueueWriteBuffer_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
- return func(command_queue, buffer, blocking_write, offset, size, ptr, num_events_in_wait_list, event_wait_list, event);
+ return func(command_queue, buffer, blocking_write, offset, size, ptr, num_events_in_wait_list, event_wait_list,
+ event);
}
else
{
@@ -581,22 +576,22 @@
}
}
-cl_int clEnqueueReadBuffer(
- cl_command_queue command_queue,
- cl_mem buffer,
- cl_bool blocking_read,
- size_t offset,
- size_t size,
- void *ptr,
- cl_uint num_events_in_wait_list,
- const cl_event *event_wait_list,
- cl_event *event)
+cl_int clEnqueueReadBuffer(cl_command_queue command_queue,
+ cl_mem buffer,
+ cl_bool blocking_read,
+ size_t offset,
+ size_t size,
+ void *ptr,
+ cl_uint num_events_in_wait_list,
+ const cl_event *event_wait_list,
+ cl_event *event)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clEnqueueReadBuffer_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
- return func(command_queue, buffer, blocking_read, offset, size, ptr, num_events_in_wait_list, event_wait_list, event);
+ return func(command_queue, buffer, blocking_read, offset, size, ptr, num_events_in_wait_list, event_wait_list,
+ event);
}
else
{
@@ -604,17 +599,16 @@
}
}
-cl_int clGetProgramBuildInfo(
- cl_program program,
- cl_device_id device,
- cl_program_build_info param_name,
- size_t param_value_size,
- void *param_value,
- size_t *param_value_size_ret)
+cl_int clGetProgramBuildInfo(cl_program program,
+ cl_device_id device,
+ cl_program_build_info param_name,
+ size_t param_value_size,
+ void *param_value,
+ size_t *param_value_size_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clGetProgramBuildInfo_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(program, device, param_name, param_value_size, param_value, param_value_size_ret);
}
@@ -628,7 +622,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clRetainProgram_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(program);
}
@@ -638,27 +632,27 @@
}
}
-void *clEnqueueMapBuffer(
- cl_command_queue command_queue,
- cl_mem buffer,
- cl_bool blocking_map,
- cl_map_flags map_flags,
- size_t offset,
- size_t size,
- cl_uint num_events_in_wait_list,
- const cl_event *event_wait_list,
- cl_event *event,
- cl_int *errcode_ret)
+void *clEnqueueMapBuffer(cl_command_queue command_queue,
+ cl_mem buffer,
+ cl_bool blocking_map,
+ cl_map_flags map_flags,
+ size_t offset,
+ size_t size,
+ cl_uint num_events_in_wait_list,
+ const cl_event *event_wait_list,
+ cl_event *event,
+ cl_int *errcode_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clEnqueueMapBuffer_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
- return func(command_queue, buffer, blocking_map, map_flags, offset, size, num_events_in_wait_list, event_wait_list, event, errcode_ret);
+ return func(command_queue, buffer, blocking_map, map_flags, offset, size, num_events_in_wait_list,
+ event_wait_list, event, errcode_ret);
}
else
{
- if(errcode_ret != nullptr)
+ if (errcode_ret != nullptr)
{
*errcode_ret = CL_OUT_OF_RESOURCES;
}
@@ -670,7 +664,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clReleaseCommandQueue_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_queue);
}
@@ -680,24 +674,23 @@
}
}
-cl_program clCreateProgramWithBinary(
- cl_context context,
- cl_uint num_devices,
- const cl_device_id *device_list,
- const size_t *lengths,
- const unsigned char **binaries,
- cl_int *binary_status,
- cl_int *errcode_ret)
+cl_program clCreateProgramWithBinary(cl_context context,
+ cl_uint num_devices,
+ const cl_device_id *device_list,
+ const size_t *lengths,
+ const unsigned char **binaries,
+ cl_int *binary_status,
+ cl_int *errcode_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clCreateProgramWithBinary_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(context, num_devices, device_list, lengths, binaries, binary_status, errcode_ret);
}
else
{
- if(errcode_ret != nullptr)
+ if (errcode_ret != nullptr)
{
*errcode_ret = CL_OUT_OF_RESOURCES;
}
@@ -709,7 +702,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clRetainContext_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(context);
}
@@ -723,7 +716,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clReleaseProgram_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(program);
}
@@ -737,7 +730,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clFlush_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_queue);
}
@@ -751,7 +744,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clFinish_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_queue);
}
@@ -761,16 +754,15 @@
}
}
-cl_int clGetProgramInfo(
- cl_program program,
- cl_program_info param_name,
- size_t param_value_size,
- void *param_value,
- size_t *param_value_size_ret)
+cl_int clGetProgramInfo(cl_program program,
+ cl_program_info param_name,
+ size_t param_value_size,
+ void *param_value,
+ size_t *param_value_size_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clGetProgramInfo_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(program, param_name, param_value_size, param_value, param_value_size_ret);
}
@@ -780,20 +772,17 @@
}
}
-cl_kernel clCreateKernel(
- cl_program program,
- const char *kernel_name,
- cl_int *errcode_ret)
+cl_kernel clCreateKernel(cl_program program, const char *kernel_name, cl_int *errcode_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clCreateKernel_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(program, kernel_name, errcode_ret);
}
else
{
- if(errcode_ret != nullptr)
+ if (errcode_ret != nullptr)
{
*errcode_ret = CL_OUT_OF_RESOURCES;
}
@@ -805,7 +794,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clRetainKernel_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(kernel);
}
@@ -815,22 +804,17 @@
}
}
-cl_mem clCreateBuffer(
- cl_context context,
- cl_mem_flags flags,
- size_t size,
- void *host_ptr,
- cl_int *errcode_ret)
+cl_mem clCreateBuffer(cl_context context, cl_mem_flags flags, size_t size, void *host_ptr, cl_int *errcode_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clCreateBuffer_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(context, flags, size, host_ptr, errcode_ret);
}
else
{
- if(errcode_ret != nullptr)
+ if (errcode_ret != nullptr)
{
*errcode_ret = CL_OUT_OF_RESOURCES;
}
@@ -839,21 +823,17 @@
}
cl_program clCreateProgramWithSource(
- cl_context context,
- cl_uint count,
- const char **strings,
- const size_t *lengths,
- cl_int *errcode_ret)
+ cl_context context, cl_uint count, const char **strings, const size_t *lengths, cl_int *errcode_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clCreateProgramWithSource_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(context, count, strings, lengths, errcode_ret);
}
else
{
- if(errcode_ret != nullptr)
+ if (errcode_ret != nullptr)
{
*errcode_ret = CL_OUT_OF_RESOURCES;
}
@@ -865,7 +845,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clReleaseKernel_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(kernel);
}
@@ -878,12 +858,12 @@
cl_int clGetDeviceIDs(cl_platform_id platform,
cl_device_type device_type,
cl_uint num_entries,
- cl_device_id *devices,
+ cl_device_id *devices,
cl_uint *num_devices)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clGetDeviceIDs_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(platform, device_type, num_entries, devices, num_devices);
}
@@ -901,7 +881,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clGetDeviceInfo_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(device, param_name, param_value_size, param_value, param_value_size_ret);
}
@@ -911,15 +891,12 @@
}
}
-cl_int clGetMemObjectInfo(cl_mem memobj,
- cl_mem_info param_name,
- size_t param_value_size,
- void *param_value,
- size_t *param_value_size_ret)
+cl_int clGetMemObjectInfo(
+ cl_mem memobj, cl_mem_info param_name, size_t param_value_size, void *param_value, size_t *param_value_size_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clGetMemObjectInfo_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(memobj, param_name, param_value_size, param_value, param_value_size_ret);
}
@@ -933,7 +910,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clRetainEvent_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(event);
}
@@ -951,7 +928,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clGetPlatformInfo_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(platform, param_name, param_value_size, param_value, param_value_size_ret);
}
@@ -965,7 +942,7 @@
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clGetPlatformIDs_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(num_entries, platforms, num_platforms);
}
@@ -975,17 +952,16 @@
}
}
-cl_int
-clGetKernelWorkGroupInfo(cl_kernel kernel,
- cl_device_id device,
- cl_kernel_work_group_info param_name,
- size_t param_value_size,
- void *param_value,
- size_t *param_value_size_ret)
+cl_int clGetKernelWorkGroupInfo(cl_kernel kernel,
+ cl_device_id device,
+ cl_kernel_work_group_info param_name,
+ size_t param_value_size,
+ void *param_value,
+ size_t *param_value_size_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clGetKernelWorkGroupInfo_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(kernel, device, param_name, param_value_size, param_value, param_value_size_ret);
}
@@ -995,16 +971,15 @@
}
}
-cl_int
-clGetCommandQueueInfo(cl_command_queue command_queue,
- cl_command_queue_info param_name,
- size_t param_value_size,
- void *param_value,
- size_t *param_value_size_ret)
+cl_int clGetCommandQueueInfo(cl_command_queue command_queue,
+ cl_command_queue_info param_name,
+ size_t param_value_size,
+ void *param_value,
+ size_t *param_value_size_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clGetCommandQueueInfo_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_queue, param_name, param_value_size, param_value, param_value_size_ret);
}
@@ -1014,16 +989,15 @@
}
}
-cl_int
-clGetKernelInfo(cl_kernel kernel,
- cl_kernel_info param_name,
- size_t param_value_size,
- void *param_value,
- size_t *param_value_size_ret)
+cl_int clGetKernelInfo(cl_kernel kernel,
+ cl_kernel_info param_name,
+ size_t param_value_size,
+ void *param_value,
+ size_t *param_value_size_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clGetKernelInfo_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(kernel, param_name, param_value_size, param_value, param_value_size_ret);
}
@@ -1033,16 +1007,15 @@
}
}
-cl_int
-clGetEventProfilingInfo(cl_event event,
- cl_profiling_info param_name,
- size_t param_value_size,
- void *param_value,
- size_t *param_value_size_ret)
+cl_int clGetEventProfilingInfo(cl_event event,
+ cl_profiling_info param_name,
+ size_t param_value_size,
+ void *param_value,
+ size_t *param_value_size_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clGetEventProfilingInfo_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(event, param_name, param_value_size, param_value, param_value_size_ret);
}
@@ -1052,23 +1025,22 @@
}
}
-cl_mem
-clCreateImage(cl_context context,
- cl_mem_flags flags,
- const cl_image_format *image_format,
- const cl_image_desc *image_desc,
- void *host_ptr,
- cl_int *errcode_ret)
+cl_mem clCreateImage(cl_context context,
+ cl_mem_flags flags,
+ const cl_image_format *image_format,
+ const cl_image_desc *image_desc,
+ void *host_ptr,
+ cl_int *errcode_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clCreateImage_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(context, flags, image_format, image_desc, host_ptr, errcode_ret);
}
else
{
- if(errcode_ret != nullptr)
+ if (errcode_ret != nullptr)
{
*errcode_ret = CL_OUT_OF_RESOURCES;
}
@@ -1076,14 +1048,12 @@
}
}
-cl_int clSetKernelExecInfo(cl_kernel kernel,
- cl_kernel_exec_info param_name,
- size_t param_value_size,
- const void *param_value)
+cl_int
+clSetKernelExecInfo(cl_kernel kernel, cl_kernel_exec_info param_name, size_t param_value_size, const void *param_value)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clSetKernelExecInfo_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(kernel, param_name, param_value_size, param_value);
}
@@ -1093,22 +1063,21 @@
}
}
-cl_command_buffer_khr clCreateCommandBufferKHR(
- cl_uint num_queues,
- const cl_command_queue* queues,
- const cl_command_buffer_properties_khr* properties,
- cl_int* errcode_ret)
+cl_command_buffer_khr clCreateCommandBufferKHR(cl_uint num_queues,
+ const cl_command_queue *queues,
+ const cl_command_buffer_properties_khr *properties,
+ cl_int *errcode_ret)
{
arm_compute::CLSymbols::get().load_default();
const auto func = arm_compute::CLSymbols::get().clCreateCommandBufferKHR_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(num_queues, queues, properties, errcode_ret);
}
else
{
- if(errcode_ret != nullptr)
+ if (errcode_ret != nullptr)
{
*errcode_ret = CL_INVALID_OPERATION;
}
@@ -1122,7 +1091,7 @@
arm_compute::CLSymbols::get().load_default();
const auto func = arm_compute::CLSymbols::get().clFinalizeCommandBufferKHR_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_buffer);
}
@@ -1137,7 +1106,7 @@
arm_compute::CLSymbols::get().load_default();
const auto func = arm_compute::CLSymbols::get().clRetainCommandBufferKHR_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_buffer);
}
@@ -1152,7 +1121,7 @@
arm_compute::CLSymbols::get().load_default();
const auto func = arm_compute::CLSymbols::get().clReleaseCommandBufferKHR_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_buffer);
}
@@ -1162,18 +1131,17 @@
}
}
-cl_int clEnqueueCommandBufferKHR(
- cl_uint num_queues,
- cl_command_queue* queues,
- cl_command_buffer_khr command_buffer,
- cl_uint num_events_in_wait_list,
- const cl_event* event_wait_list,
- cl_event* event)
+cl_int clEnqueueCommandBufferKHR(cl_uint num_queues,
+ cl_command_queue *queues,
+ cl_command_buffer_khr command_buffer,
+ cl_uint num_events_in_wait_list,
+ const cl_event *event_wait_list,
+ cl_event *event)
{
arm_compute::CLSymbols::get().load_default();
const auto func = arm_compute::CLSymbols::get().clEnqueueCommandBufferKHR_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(num_queues, queues, command_buffer, num_events_in_wait_list, event_wait_list, event);
}
@@ -1183,27 +1151,26 @@
}
}
-
-cl_int clCommandNDRangeKernelKHR(
- cl_command_buffer_khr command_buffer,
- cl_command_queue command_queue,
- const cl_ndrange_kernel_command_properties_khr* properties,
- cl_kernel kernel,
- cl_uint work_dim,
- const size_t* global_work_offset,
- const size_t* global_work_size,
- const size_t* local_work_size,
- cl_uint num_sync_points_in_wait_list,
- const cl_sync_point_khr* sync_point_wait_list,
- cl_sync_point_khr* sync_point,
- cl_mutable_command_khr* mutable_handle)
+cl_int clCommandNDRangeKernelKHR(cl_command_buffer_khr command_buffer,
+ cl_command_queue command_queue,
+ const cl_ndrange_kernel_command_properties_khr *properties,
+ cl_kernel kernel,
+ cl_uint work_dim,
+ const size_t *global_work_offset,
+ const size_t *global_work_size,
+ const size_t *local_work_size,
+ cl_uint num_sync_points_in_wait_list,
+ const cl_sync_point_khr *sync_point_wait_list,
+ cl_sync_point_khr *sync_point,
+ cl_mutable_command_khr *mutable_handle)
{
arm_compute::CLSymbols::get().load_default();
const auto func = arm_compute::CLSymbols::get().clCommandNDRangeKernelKHR_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
- return func(command_buffer, command_queue, properties, kernel, work_dim, global_work_offset, global_work_size, local_work_size, num_sync_points_in_wait_list, sync_point_wait_list, sync_point, mutable_handle);
+ return func(command_buffer, command_queue, properties, kernel, work_dim, global_work_offset, global_work_size,
+ local_work_size, num_sync_points_in_wait_list, sync_point_wait_list, sync_point, mutable_handle);
}
else
{
@@ -1211,14 +1178,13 @@
}
}
-cl_int clUpdateMutableCommandsKHR(
- cl_command_buffer_khr command_buffer,
- const cl_mutable_base_config_khr* mutable_config)
+cl_int clUpdateMutableCommandsKHR(cl_command_buffer_khr command_buffer,
+ const cl_mutable_base_config_khr *mutable_config)
{
arm_compute::CLSymbols::get().load_default();
const auto func = arm_compute::CLSymbols::get().clUpdateMutableCommandsKHR_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(command_buffer, mutable_config);
}
@@ -1228,23 +1194,22 @@
}
}
-cl_mem
-clImportMemoryARM(cl_context context,
- cl_mem_flags flags,
- const cl_import_properties_arm *properties,
- void *memory,
- size_t size,
- cl_int *errcode_ret)
+cl_mem clImportMemoryARM(cl_context context,
+ cl_mem_flags flags,
+ const cl_import_properties_arm *properties,
+ void *memory,
+ size_t size,
+ cl_int *errcode_ret)
{
arm_compute::CLSymbols::get().load_default();
auto func = arm_compute::CLSymbols::get().clImportMemoryARM_ptr;
- if(func != nullptr)
+ if (func != nullptr)
{
return func(context, flags, properties, memory, size, errcode_ret);
}
else
{
- if(errcode_ret != nullptr)
+ if (errcode_ret != nullptr)
{
*errcode_ret = CL_OUT_OF_RESOURCES;
}
diff --git a/src/core/CL/cl_kernels/activation_float_helpers.h b/src/core/CL/cl_kernels/activation_float_helpers.h
index 3f93c8d..02faae2 100644
--- a/src/core/CL/cl_kernels/activation_float_helpers.h
+++ b/src/core/CL/cl_kernels/activation_float_helpers.h
@@ -31,7 +31,8 @@
#endif // GPU_ARCH == GPU_ARCH_BIFROST
// Hard-Swish
-#define hard_swish_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (x * ((min(max((x + (DATA_TYPE)3.0), (DATA_TYPE)0.0), (DATA_TYPE)6.0)) * (DATA_TYPE)0.166666667))
+#define hard_swish_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) \
+ (x * ((min(max((x + (DATA_TYPE)3.0), (DATA_TYPE)0.0), (DATA_TYPE)6.0)) * (DATA_TYPE)0.166666667))
// Logistic Activation
#define logistic_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) ((DATA_TYPE)1.0 / ((DATA_TYPE)1.0 + exp(-x)))
@@ -49,13 +50,16 @@
#define lu_brelu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (min(max(x, (DATA_TYPE)B_VAL), (DATA_TYPE)A_VAL))
// Leaky RELU Activation
-#define lrelu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) ((min(x, (DATA_TYPE)0.0) * (DATA_TYPE)A_VAL) + max(x, (DATA_TYPE)0.0))
+#define lrelu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) \
+ ((min(x, (DATA_TYPE)0.0) * (DATA_TYPE)A_VAL) + max(x, (DATA_TYPE)0.0))
// Soft RELU Activation
#define srelu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (log((DATA_TYPE)1.0 + exp(x)))
// ELU Activation
-#define elu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (select(((DATA_TYPE)A_VAL * (exp(x) - (DATA_TYPE)1.0)), x, (SELECT_VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE))isgreaterequal(x, (DATA_TYPE)0.0)))
+#define elu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) \
+ (select(((DATA_TYPE)A_VAL * (exp(x) - (DATA_TYPE)1.0)), x, \
+ (SELECT_VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE))isgreaterequal(x, (DATA_TYPE)0.0)))
// Absolute Activation
#define abs_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (fabs(x))
@@ -70,7 +74,8 @@
#define linear_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (MLA((DATA_TYPE)B_VAL, (DATA_TYPE)A_VAL, x))
// GELU Activation
-#define gelu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (x * (DATA_TYPE)0.5 * ((DATA_TYPE)1.0 + erf(x / (DATA_TYPE)1.41421356237)))
+#define gelu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) \
+ (x * (DATA_TYPE)0.5 * ((DATA_TYPE)1.0 + erf(x / (DATA_TYPE)1.41421356237)))
// Identity Activation
#define identity_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (x)
diff --git a/src/core/CL/cl_kernels/activation_quant_helpers.h b/src/core/CL/cl_kernels/activation_quant_helpers.h
index c420578..c758ff1 100644
--- a/src/core/CL/cl_kernels/activation_quant_helpers.h
+++ b/src/core/CL/cl_kernels/activation_quant_helpers.h
@@ -60,17 +60,17 @@
}
#define ACTIVATION_OP2(op, x) op##_op(x)
-#define ACTIVATION_OP(op, x) ACTIVATION_OP2(op, x)
+#define ACTIVATION_OP(op, x) ACTIVATION_OP2(op, x)
#if defined(S1_VAL) && defined(S2_VAL)
#if defined(O1_VAL) && defined(O2_VAL)
#define PERFORM_ACTIVATION_QUANT(act, data) \
({ \
data = ACTIVATION_OP(act, data); \
- \
+ \
VEC_DATA_TYPE(float, VEC_SIZE) \
fdata = CONVERT(data, VEC_DATA_TYPE(float, VEC_SIZE)); \
- \
+ \
fdata = round((fdata - (float)O1_VAL) * ((float)S1_VAL / (float)S2_VAL) + (float)O2_VAL); \
data = CONVERT_SAT(fdata, VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)); \
})
@@ -78,17 +78,14 @@
#define PERFORM_ACTIVATION_QUANT(act, data) \
({ \
data = ACTIVATION_OP(act, data); \
- \
+ \
VEC_DATA_TYPE(float, VEC_SIZE) \
fdata = CONVERT(data, VEC_DATA_TYPE(float, VEC_SIZE)); \
- \
+ \
fdata = round((fdata) * ((float)S1_VAL / (float)S2_VAL)); \
data = CONVERT_SAT(fdata, VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)); \
})
#endif /* defined(O1_VAL) && defined(O2_VAL) */
#else /* defined(S1_VAL) && defined(S2_VAL) */
-#define PERFORM_ACTIVATION_QUANT(act, data) \
- ({ \
- data = ACTIVATION_OP(act, data); \
- })
+#define PERFORM_ACTIVATION_QUANT(act, data) ({ data = ACTIVATION_OP(act, data); })
#endif /* defined(S1_VAL) && defined(S2_VAL) */
diff --git a/src/core/CL/cl_kernels/gemm_helpers.h b/src/core/CL/cl_kernels/gemm_helpers.h
index 0e938cb..4bef023 100644
--- a/src/core/CL/cl_kernels/gemm_helpers.h
+++ b/src/core/CL/cl_kernels/gemm_helpers.h
@@ -34,14 +34,14 @@
*
*/
#define SCALAR_ACCESS_STR(offset, n0, x) scalar_access_##offset##_##n0(x)
-#define SCALAR_ACCESS(offset, n0, x) SCALAR_ACCESS_STR(offset, n0, x)
+#define SCALAR_ACCESS(offset, n0, x) SCALAR_ACCESS_STR(offset, n0, x)
// offset == 0
-#define scalar_access_0_1(x) ((x).s0)
-#define scalar_access_0_2(x) ((x).s01)
-#define scalar_access_0_3(x) ((x).s012)
-#define scalar_access_0_4(x) ((x).s0123)
-#define scalar_access_0_8(x) ((x).s01234567)
+#define scalar_access_0_1(x) ((x).s0)
+#define scalar_access_0_2(x) ((x).s01)
+#define scalar_access_0_3(x) ((x).s012)
+#define scalar_access_0_4(x) ((x).s0123)
+#define scalar_access_0_8(x) ((x).s01234567)
#define scalar_access_0_16(x) ((x).s0123456789ABCDEF)
// offset == 1
@@ -100,8 +100,7 @@
* @param[in] Z The z-axis offset vector
* @{
*/
-#define LOAD_TENSOR_ROW_0(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
- ({})
+#define LOAD_TENSOR_ROW_0(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) ({})
#define LOAD_TENSOR_ROW_1(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##0) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));
@@ -186,8 +185,10 @@
* @param[in] Z The z-axis offset vector
* @{
*/
-#define LOAD_TENSOR_STR(M0, N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) LOAD_TENSOR_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)
-#define LOAD_TENSOR(M0, N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) LOAD_TENSOR_STR(M0, N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)
+#define LOAD_TENSOR_STR(M0, N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
+ LOAD_TENSOR_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)
+#define LOAD_TENSOR(M0, N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
+ LOAD_TENSOR_STR(M0, N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)
/** @} */ // end of group LOAD_TENSOR
/** Load 2D tensor (consecutive rows and columns) with Z offset.
@@ -202,8 +203,7 @@
* @param[in] Z The z-axis offset vector
* @{
*/
-#define LOAD_TENSOR_M0X0(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
- ({})
+#define LOAD_TENSOR_M0X0(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) ({})
#define LOAD_TENSOR_M0X1(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
LOAD_TENSOR(M0, N0, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);
@@ -279,8 +279,10 @@
* @param[in] Z The z-axis offset vector
* @{
*/
-#define LOAD_TENSOR_M0XN0_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) LOAD_TENSOR_M0X##N0(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
-#define LOAD_TENSOR_M0XN0(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) LOAD_TENSOR_M0XN0_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+#define LOAD_TENSOR_M0XN0_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ LOAD_TENSOR_M0X##N0(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+#define LOAD_TENSOR_M0XN0(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ LOAD_TENSOR_M0XN0_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
/** @}*/ // end of group LOAD_TENSOR_M0XN0
/** Loads the rows from 0 to n-1 in the given variables (BASENAME0 to BASENAMEn-1).
@@ -395,8 +397,10 @@
* @param[in] Z The z-axis offset vector
* @{
*/
-#define LOAD_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) LOAD_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)
-#define LOAD_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) LOAD_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)
+#define LOAD_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
+ LOAD_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)
+#define LOAD_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
+ LOAD_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)
/** @} */ // end of group LOAD_BLOCK
/** Partially load the 0 to (n-1)th rows of the given variables
@@ -517,8 +521,10 @@
* @param[in] Z The offset in z-axis direction
* @{
*/
-#define LOAD_BLOCK_PARTIAL_STR(LOAD_M0, LOAD_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) LOAD_ROW_PARTIAL_##LOAD_M0(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)
-#define LOAD_BLOCK_PARTIAL(LOAD_M0, LOAD_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) LOAD_BLOCK_PARTIAL_STR(LOAD_M0, LOAD_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)
+#define LOAD_BLOCK_PARTIAL_STR(LOAD_M0, LOAD_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
+ LOAD_ROW_PARTIAL_##LOAD_M0(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)
+#define LOAD_BLOCK_PARTIAL(LOAD_M0, LOAD_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
+ LOAD_BLOCK_PARTIAL_STR(LOAD_M0, LOAD_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)
/** Load a block that can be partial in both x and y dimensions
*
* @note in cases @p PARTIAL_STORE_N0 != 1, 2, 3, 4, 8, 16, extra vload(s) will be invoked, thus incurring small performance penalty.
@@ -541,22 +547,23 @@
* @param[in] PARTIAL_COND_Y Condition on the y axis to perform the partial load Y. True to use PARTIAL_STORE_M0 rather than M0.
* @param[in] PARTIAL_COND_X Condition on the x axis to perform the partial load X. True to use PARTIAL_STORE_N0 rather than N0.
*/
-#define LOAD_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
- if(!(PARTIAL_COND_X) && !(PARTIAL_COND_Y)) \
- { \
- LOAD_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
- } \
- else if((PARTIAL_COND_Y) && !(PARTIAL_COND_X)) \
- { \
- LOAD_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
- } \
- else if(!(PARTIAL_COND_Y) && (PARTIAL_COND_X)) \
- { \
- LOAD_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
- } \
- else \
- { \
- LOAD_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
+#define LOAD_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, \
+ PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+ if (!(PARTIAL_COND_X) && !(PARTIAL_COND_Y)) \
+ { \
+ LOAD_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
+ } \
+ else if ((PARTIAL_COND_Y) && !(PARTIAL_COND_X)) \
+ { \
+ LOAD_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
+ } \
+ else if (!(PARTIAL_COND_Y) && (PARTIAL_COND_X)) \
+ { \
+ LOAD_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
+ } \
+ else \
+ { \
+ LOAD_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
}
/** Load a block that can only be partial in x but not y.
*
@@ -578,14 +585,15 @@
* @param[in] PARTIAL_STORE_N0 The partial size in x, for partial blocks. Supported range: [1, @p N0)
* @param[in] PARTIAL_COND_X Condition on the x axis to perform the partial load X. True to use PARTIAL_STORE_N0 rather than N0.
*/
-#define LOAD_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X) \
- if(!(PARTIAL_COND_X)) \
- { \
- LOAD_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
- } \
- else \
- { \
- LOAD_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
+#define LOAD_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_N0, \
+ PARTIAL_COND_X) \
+ if (!(PARTIAL_COND_X)) \
+ { \
+ LOAD_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
+ } \
+ else \
+ { \
+ LOAD_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
}
/** Load a block that can only be partial in y but not x.
*
@@ -607,14 +615,15 @@
* @param[in] PARTIAL_STORE_M0 The partial size in y, for partial blocks. Supported range: [1, @p M0)
* @param[in] PARTIAL_COND_Y Condition on the y axis to perform the partial store Y. True to use PARTIAL_STORE_M0 rather than M0.
*/
-#define LOAD_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y) \
- if(!(PARTIAL_COND_Y)) \
- { \
- LOAD_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
- } \
- else \
- { \
- LOAD_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
+#define LOAD_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, \
+ PARTIAL_COND_Y) \
+ if (!(PARTIAL_COND_Y)) \
+ { \
+ LOAD_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
+ } \
+ else \
+ { \
+ LOAD_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z); \
}
/** @} */ // end of group LOAD_BLOCK_PARTIAL
/** Boundary-aware GeMM block load
@@ -676,28 +685,33 @@
*/
#if PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
// Case1: No partial blocks in either x or y
-#define LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+#define LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, \
+ PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
LOAD_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)
#elif PARTIAL_STORE_M0 > 0 && PARTIAL_STORE_N0 == 0
// Case2: Partial blocks in y
-#define LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
- REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), BASENAME, 0); \
+#define LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, \
+ PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+ REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), BASENAME, 0); \
LOAD_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y)
#elif PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 > 0
// Case3: Partial blocks in x
-#define LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
- REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), BASENAME, 0); \
+#define LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, \
+ PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+ REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), BASENAME, 0); \
LOAD_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X)
#else // PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
// Case4: Partial blocks in both x and y
-#define LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
- REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), BASENAME, 0); \
- LOAD_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X)
+#define LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, \
+ PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+ REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), BASENAME, 0); \
+ LOAD_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, \
+ PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X)
-#endif // PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
+#endif // PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
/** @} */ // end of group LOAD_BLOCK_BOUNDARY_AWARE
/** Loads the rows from 0 to n-1 in the given variables (BASENAME0 to BASENAMEn-1).
@@ -795,8 +809,10 @@
* @param[in] Y_STEP_ROW The incremental step row for the y coordinate (in pixels)
* @{
*/
-#define LOAD_TEXTURE2D_STR(M0, N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) LOAD_TEXTURE2D_ROW_##M0(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)
-#define LOAD_TEXTURE2D(M0, N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) LOAD_TEXTURE2D_STR(M0, N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)
+#define LOAD_TEXTURE2D_STR(M0, N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
+ LOAD_TEXTURE2D_ROW_##M0(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)
+#define LOAD_TEXTURE2D(M0, N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
+ LOAD_TEXTURE2D_STR(M0, N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)
/** @} */ // end of group LOAD_TEXTURE2D
/** Loads the rows from 0 to n-1 in the given variables (BASENAME0 to BASENAMEn-1) passing the Y index for each row to be loaded.
@@ -815,7 +831,7 @@
#define LOAD_ROW_INDIRECT_1(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##0; \
- if(Y_MASK##0 != 0) \
+ if (Y_MASK##0 != 0) \
BASENAME##0 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##0 * STRIDE_Y)); \
else \
BASENAME##0 = 0;
@@ -824,7 +840,7 @@
LOAD_ROW_INDIRECT_1(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##1; \
- if(Y_MASK##1 != 0) \
+ if (Y_MASK##1 != 0) \
BASENAME##1 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##1 * STRIDE_Y)); \
else \
BASENAME##1 = 0;
@@ -833,7 +849,7 @@
LOAD_ROW_INDIRECT_2(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##2; \
- if(Y_MASK##2 != 0) \
+ if (Y_MASK##2 != 0) \
BASENAME##2 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##2 * STRIDE_Y)); \
else \
BASENAME##2 = 0;
@@ -842,7 +858,7 @@
LOAD_ROW_INDIRECT_3(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##3; \
- if(Y_MASK##3 != 0) \
+ if (Y_MASK##3 != 0) \
BASENAME##3 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##3 * STRIDE_Y)); \
else \
BASENAME##3 = 0;
@@ -851,7 +867,7 @@
LOAD_ROW_INDIRECT_4(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##4; \
- if(Y_MASK##4 != 0) \
+ if (Y_MASK##4 != 0) \
BASENAME##4 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##4 * STRIDE_Y)); \
else \
BASENAME##4 = 0;
@@ -860,7 +876,7 @@
LOAD_ROW_INDIRECT_5(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##5; \
- if(Y_MASK##5 != 0) \
+ if (Y_MASK##5 != 0) \
BASENAME##5 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##5 * STRIDE_Y)); \
else \
BASENAME##5 = 0;
@@ -869,7 +885,7 @@
LOAD_ROW_INDIRECT_6(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##6; \
- if(Y_MASK##6 != 0) \
+ if (Y_MASK##6 != 0) \
BASENAME##6 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##6 * STRIDE_Y)); \
else \
BASENAME##6 = 0;
@@ -878,7 +894,7 @@
LOAD_ROW_INDIRECT_7(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##7; \
- if(Y_MASK##7 != 0) \
+ if (Y_MASK##7 != 0) \
BASENAME##7 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##7 * STRIDE_Y)); \
else \
BASENAME##7 = 0;
@@ -887,7 +903,7 @@
LOAD_ROW_INDIRECT_8(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##8; \
- if(Y_MASK##8 != 0) \
+ if (Y_MASK##8 != 0) \
BASENAME##8 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##8 * STRIDE_Y)); \
else \
BASENAME##8 = 0;
@@ -896,7 +912,7 @@
LOAD_ROW_INDIRECT_9(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##9; \
- if(Y_MASK##9 != 0) \
+ if (Y_MASK##9 != 0) \
BASENAME##9 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##9 * STRIDE_Y)); \
else \
BASENAME##9 = 0;
@@ -905,7 +921,7 @@
LOAD_ROW_INDIRECT_10(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##A; \
- if(Y_MASK##A != 0) \
+ if (Y_MASK##A != 0) \
BASENAME##A = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##A * STRIDE_Y)); \
else \
BASENAME##A = 0;
@@ -914,7 +930,7 @@
LOAD_ROW_INDIRECT_11(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##B; \
- if(Y_MASK##B != 0) \
+ if (Y_MASK##B != 0) \
BASENAME##B = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##B * STRIDE_Y)); \
else \
BASENAME##B = 0;
@@ -923,7 +939,7 @@
LOAD_ROW_INDIRECT_12(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##C; \
- if(Y_MASK##C != 0) \
+ if (Y_MASK##C != 0) \
BASENAME##C = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##C * STRIDE_Y)); \
else \
BASENAME##C = 0;
@@ -932,7 +948,7 @@
LOAD_ROW_INDIRECT_13(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##D; \
- if(Y_MASK##D != 0) \
+ if (Y_MASK##D != 0) \
BASENAME##D = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##D * STRIDE_Y)); \
else \
BASENAME##D = 0;
@@ -941,7 +957,7 @@
LOAD_ROW_INDIRECT_14(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##E; \
- if(Y_MASK##E != 0) \
+ if (Y_MASK##E != 0) \
BASENAME##E = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##E * STRIDE_Y)); \
else \
BASENAME##E = 0;
@@ -950,7 +966,7 @@
LOAD_ROW_INDIRECT_15(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
VEC_DATA_TYPE(DATA_TYPE, N0) \
BASENAME##F; \
- if(Y_MASK##F != 0) \
+ if (Y_MASK##F != 0) \
BASENAME##F = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##F * STRIDE_Y)); \
else \
BASENAME##F = 0;
@@ -976,8 +992,10 @@
* @param[in] Y_MASK The y-axis mask vector. If 0, forces BASENAMEn to 0
* @{
*/
-#define LOAD_BLOCK_INDIRECT_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) LOAD_ROW_INDIRECT_##M0(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)
-#define LOAD_BLOCK_INDIRECT(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) LOAD_BLOCK_INDIRECT_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)
+#define LOAD_BLOCK_INDIRECT_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
+ LOAD_ROW_INDIRECT_##M0(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)
+#define LOAD_BLOCK_INDIRECT(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) \
+ LOAD_BLOCK_INDIRECT_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)
/** @} */ // end of group LOAD_BLOCK_INDIRECT
/** Loads the elements from 0 to n-1 in the given variables (BASENAME0 to BASENAMEn-1).
@@ -1088,8 +1106,10 @@
* @param[in] STRIDE_Y The stride in y-axis direction
* @{
*/
-#define LOAD_SCALAR_AS_VECTOR_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) LOAD_ELEMENT_##M0(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)
-#define LOAD_SCALAR_AS_VECTOR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) LOAD_SCALAR_AS_VECTOR_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)
+#define LOAD_SCALAR_AS_VECTOR_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
+ LOAD_ELEMENT_##M0(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)
+#define LOAD_SCALAR_AS_VECTOR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
+ LOAD_SCALAR_AS_VECTOR_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)
/** @} */ // end of group LOAD_SCALAR_AS_VECTOR
/** Basic macros to calculate Z offset values from Z0 to Zn-1
@@ -1187,8 +1207,10 @@
* @param[in] STRIDE_Y The stride value in y-axis direction
* @{
*/
-#define CALCULATE_Z_OFFSET_STR(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) CALCULATE_Z_OFFSET_##M0(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)
-#define CALCULATE_Z_OFFSET(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) CALCULATE_Z_OFFSET_STR(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)
+#define CALCULATE_Z_OFFSET_STR(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) \
+ CALCULATE_Z_OFFSET_##M0(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)
+#define CALCULATE_Z_OFFSET(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) \
+ CALCULATE_Z_OFFSET_STR(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)
/** @} */ // end of group CALCULATE_Z_OFFSET
/** Scale the rows in the given variables (BASENAME0 to BASENAMEn-1)
@@ -1199,8 +1221,7 @@
* @param[in] SCALE The scale factor
* @{
*/
-#define SCALE_ROW_1(DATA_TYPE, BASENAME, SCALE) \
- BASENAME##0 *= (DATA_TYPE)SCALE;
+#define SCALE_ROW_1(DATA_TYPE, BASENAME, SCALE) BASENAME##0 *= (DATA_TYPE)SCALE;
#define SCALE_ROW_2(DATA_TYPE, BASENAME, SCALE) \
SCALE_ROW_1(DATA_TYPE, BASENAME, SCALE) \
@@ -1275,7 +1296,7 @@
* @{
*/
#define SCALE_BLOCK_STR(N, DATA_TYPE, BASENAME, SCALE) SCALE_ROW_##N(DATA_TYPE, BASENAME, SCALE)
-#define SCALE_BLOCK(N, DATA_TYPE, BASENAME, SCALE) SCALE_BLOCK_STR(N, DATA_TYPE, BASENAME, SCALE)
+#define SCALE_BLOCK(N, DATA_TYPE, BASENAME, SCALE) SCALE_BLOCK_STR(N, DATA_TYPE, BASENAME, SCALE)
/** @} */ // end of group SCALE_BLOCK
/** Create a new vector containing the values at the given index for a set of given vectors
@@ -1287,8 +1308,7 @@
* @param[in] TYPE The data type of the destination vectors
* @{
*/
-#define COLUMN_VECTOR1(IDX_COL, BASENAME, X, TYPE) \
- TYPE BASENAME##IDX_COL = (TYPE)((X##0).s##IDX_COL);
+#define COLUMN_VECTOR1(IDX_COL, BASENAME, X, TYPE) TYPE BASENAME##IDX_COL = (TYPE)((X##0).s##IDX_COL);
#define COLUMN_VECTOR2(IDX_COL, BASENAME, X, TYPE) \
VEC_DATA_TYPE(TYPE, 2) \
BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 2))((X##0).s##IDX_COL, (X##1).s##IDX_COL);
@@ -1297,13 +1317,20 @@
BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 3))((X##0).s##IDX_COL, (X##1).s##IDX_COL, (X##2).s##IDX_COL);
#define COLUMN_VECTOR4(IDX_COL, BASENAME, X, TYPE) \
VEC_DATA_TYPE(TYPE, 4) \
- BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 4))((X##0).s##IDX_COL, (X##1).s##IDX_COL, (X##2).s##IDX_COL, (X##3).s##IDX_COL);
-#define COLUMN_VECTOR8(IDX_COL, BASENAME, X, TYPE) \
- VEC_DATA_TYPE(TYPE, 8) \
- BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 8))((X##0).s##IDX_COL, (X##1).s##IDX_COL, (X##2).s##IDX_COL, (X##3).s##IDX_COL, (X##4).s##IDX_COL, (X##5).s##IDX_COL, (X##6).s##IDX_COL, (X##7).s##IDX_COL);
-#define COLUMN_VECTOR16(IDX_COL, BASENAME, X, TYPE) \
- VEC_DATA_TYPE(TYPE, 16) \
- BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 16))((X##0).s##IDX_COL, (X##1).s##IDX_COL, (X##2).s##IDX_COL, (X##3).s##IDX_COL, (X##4).s##IDX_COL, (X##5).s##IDX_COL, (X##6).s##IDX_COL, (X##7).s##IDX_COL, (X##8).s##IDX_COL, (X##9).s##IDX_COL, (X##A).s##IDX_COL, (X##B).s##IDX_COL, (X##C).s##IDX_COL, (X##D).s##IDX_COL, (X##E).s##IDX_COL, (X##F).s##IDX_COL);
+ BASENAME##IDX_COL = \
+ (VEC_DATA_TYPE(TYPE, 4))((X##0).s##IDX_COL, (X##1).s##IDX_COL, (X##2).s##IDX_COL, (X##3).s##IDX_COL);
+#define COLUMN_VECTOR8(IDX_COL, BASENAME, X, TYPE) \
+ VEC_DATA_TYPE(TYPE, 8) \
+ BASENAME##IDX_COL = \
+ (VEC_DATA_TYPE(TYPE, 8))((X##0).s##IDX_COL, (X##1).s##IDX_COL, (X##2).s##IDX_COL, (X##3).s##IDX_COL, \
+ (X##4).s##IDX_COL, (X##5).s##IDX_COL, (X##6).s##IDX_COL, (X##7).s##IDX_COL);
+#define COLUMN_VECTOR16(IDX_COL, BASENAME, X, TYPE) \
+ VEC_DATA_TYPE(TYPE, 16) \
+ BASENAME##IDX_COL = \
+ (VEC_DATA_TYPE(TYPE, 16))((X##0).s##IDX_COL, (X##1).s##IDX_COL, (X##2).s##IDX_COL, (X##3).s##IDX_COL, \
+ (X##4).s##IDX_COL, (X##5).s##IDX_COL, (X##6).s##IDX_COL, (X##7).s##IDX_COL, \
+ (X##8).s##IDX_COL, (X##9).s##IDX_COL, (X##A).s##IDX_COL, (X##B).s##IDX_COL, \
+ (X##C).s##IDX_COL, (X##D).s##IDX_COL, (X##E).s##IDX_COL, (X##F).s##IDX_COL);
/** @} */ // end of group COLUMN_VECTORn
/** Create a new vector containing the values at the given index. Utility macros for transposing a colum-vector
@@ -1315,8 +1342,7 @@
* @param[in] TYPE The data type of the destination vectors
* @{
*/
-#define COLUMN_VECTOR_SCALAR1(IDX_COL, BASENAME, X, TYPE) \
- TYPE BASENAME##IDX_COL = (TYPE)((X##0));
+#define COLUMN_VECTOR_SCALAR1(IDX_COL, BASENAME, X, TYPE) TYPE BASENAME##IDX_COL = (TYPE)((X##0));
#define COLUMN_VECTOR_SCALAR2(IDX_COL, BASENAME, X, TYPE) \
VEC_DATA_TYPE(TYPE, 2) \
BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 2))((X##0), (X##1));
@@ -1329,9 +1355,10 @@
#define COLUMN_VECTOR_SCALAR8(IDX_COL, BASENAME, X, TYPE) \
VEC_DATA_TYPE(TYPE, 8) \
BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 8))((X##0), (X##1), (X##2), (X##3), (X##4), (X##5), (X##6), (X##7));
-#define COLUMN_VECTOR_SCALAR16(IDX_COL, BASENAME, X, TYPE) \
- VEC_DATA_TYPE(TYPE, 16) \
- BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 16))((X##0), (X##1), (X##2), (X##3), (X##4), (X##5), (X##6), (X##7), (X##8), (X##9), (X##A), (X##B), (X##C), (X##D), (X##E), (X##F));
+#define COLUMN_VECTOR_SCALAR16(IDX_COL, BASENAME, X, TYPE) \
+ VEC_DATA_TYPE(TYPE, 16) \
+ BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 16))((X##0), (X##1), (X##2), (X##3), (X##4), (X##5), (X##6), (X##7), \
+ (X##8), (X##9), (X##A), (X##B), (X##C), (X##D), (X##E), (X##F));
/** @} */ // end of group COLUMN_VECTOR_SCALARn
/** Create transposed vectors of the given vectors
@@ -1343,8 +1370,7 @@
* @param[in] TYPE The data type of the transposed vectors
* @{
*/
-#define TRANSPOSE_K0X1(K0, BASENAME, BS, TYPE) \
- COLUMN_VECTOR_SCALAR(K0, 0, BASENAME, BS, TYPE);
+#define TRANSPOSE_K0X1(K0, BASENAME, BS, TYPE) COLUMN_VECTOR_SCALAR(K0, 0, BASENAME, BS, TYPE);
#define TRANSPOSE_K0X2(K0, BASENAME, BS, TYPE) \
COLUMN_VECTOR(K0, 0, BASENAME, BS, TYPE); \
COLUMN_VECTOR(K0, 1, BASENAME, BS, TYPE);
@@ -1417,8 +1443,7 @@
* @param[in] BIAS The basename of the added variables
* @{
*/
-#define ADD_ROW_1(BASENAME, BIAS) \
- BASENAME##0 += BIAS##0;
+#define ADD_ROW_1(BASENAME, BIAS) BASENAME##0 += BIAS##0;
#define ADD_ROW_2(BASENAME, BIAS) \
ADD_ROW_1(BASENAME, BIAS) \
@@ -1493,7 +1518,7 @@
* @{
*/
#define ADD_BLOCK_STR(N, BASENAME, BIAS) ADD_ROW_##N(BASENAME, BIAS)
-#define ADD_BLOCK(N, BASENAME, BIAS) ADD_BLOCK_STR(N, BASENAME, BIAS)
+#define ADD_BLOCK(N, BASENAME, BIAS) ADD_BLOCK_STR(N, BASENAME, BIAS)
/** @} */ // end of group ADD_BLOCK
/** Broadcast (add single value) to the each element of the destination variables
@@ -1503,8 +1528,7 @@
* @param[in] BIAS The variable containing the value to add
* @{
*/
-#define ADD_ROW_BROADCAST_1(BASENAME, BIAS) \
- BASENAME##0 += BIAS;
+#define ADD_ROW_BROADCAST_1(BASENAME, BIAS) BASENAME##0 += BIAS;
#define ADD_ROW_BROADCAST_2(BASENAME, BIAS) \
ADD_ROW_BROADCAST_1(BASENAME, BIAS) \
@@ -1578,7 +1602,7 @@
* @{
*/
#define ADD_BLOCK_BROADCAST_STR(N, BASENAME, BIAS) ADD_ROW_BROADCAST_##N(BASENAME, BIAS)
-#define ADD_BLOCK_BROADCAST(N, BASENAME, BIAS) ADD_BLOCK_BROADCAST_STR(N, BASENAME, BIAS)
+#define ADD_BLOCK_BROADCAST(N, BASENAME, BIAS) ADD_BLOCK_BROADCAST_STR(N, BASENAME, BIAS)
/** @} */ // end of group ADD_BLOCK_BROADCAST
/** Apply activation to the given variables
@@ -1668,8 +1692,10 @@
* @param[in] B_VAL Additional value required by the activation
* @{
*/
-#define ACTIVATION_BLOCK_STR(N, ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) ACTIVATION_ROW_##N(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)
-#define ACTIVATION_BLOCK(N, ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) ACTIVATION_BLOCK_STR(N, ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)
+#define ACTIVATION_BLOCK_STR(N, ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
+ ACTIVATION_ROW_##N(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)
+#define ACTIVATION_BLOCK(N, ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
+ ACTIVATION_BLOCK_STR(N, ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)
/** @} */ // end of group ACTIVATION_BLOCK
/** Apply convert_<data_type> to the given variables
@@ -1773,6 +1799,8 @@
* @param[in] BASENAME_DST The basename of the destination variables
* @{
*/
-#define CONVERT_BLOCK_STR(M, N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) CONVERT_ROW_##M(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)
-#define CONVERT_BLOCK(M, N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) CONVERT_BLOCK_STR(M, N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)
+#define CONVERT_BLOCK_STR(M, N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
+ CONVERT_ROW_##M(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)
+#define CONVERT_BLOCK(M, N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
+ CONVERT_BLOCK_STR(M, N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)
/** @} */ // end of group CONVERT_BLOCK
diff --git a/src/core/CL/cl_kernels/helpers.h b/src/core/CL/cl_kernels/helpers.h
index b2ceaf9..87a1875 100644
--- a/src/core/CL/cl_kernels/helpers.h
+++ b/src/core/CL/cl_kernels/helpers.h
@@ -81,11 +81,11 @@
* @return The reversed vector
* @{
*/
-#define REV1(x) ((x))
-#define REV2(x) ((x).s10)
-#define REV3(x) ((x).s210)
-#define REV4(x) ((x).s3210)
-#define REV8(x) ((x).s76543210)
+#define REV1(x) ((x))
+#define REV2(x) ((x).s10)
+#define REV3(x) ((x).s210)
+#define REV4(x) ((x).s3210)
+#define REV8(x) ((x).s76543210)
#define REV16(x) ((x).sFEDCBA9876543210)
/** @} */ // end of group REVn
@@ -99,7 +99,7 @@
* @{
*/
#define REVERSE_STR(x, s) REV##s((x))
-#define REVERSE(x, s) REVERSE_STR(x, s)
+#define REVERSE(x, s) REVERSE_STR(x, s)
/** @} */ // end of group REVERSE
/** Circular-right-shift (rotate-right) the vector of size s by the amount of n.
@@ -138,16 +138,16 @@
#define ROT8_7(x) ((x).s12345670)
#define ROT8_8(x) ((x))
-#define ROT16_0(x) ((x))
-#define ROT16_1(x) ((x).sF0123456789ABCDE)
-#define ROT16_2(x) ((x).sEF0123456789ABCD)
-#define ROT16_3(x) ((x).sDEF0123456789ABC)
-#define ROT16_4(x) ((x).sCDEF0123456789AB)
-#define ROT16_5(x) ((x).sBCDEF0123456789A)
-#define ROT16_6(x) ((x).sABCDEF0123456789)
-#define ROT16_7(x) ((x).s9ABCDEF012345678)
-#define ROT16_8(x) ((x).s89ABCDEF01234567)
-#define ROT16_9(x) ((x).s789ABCDEF0123456)
+#define ROT16_0(x) ((x))
+#define ROT16_1(x) ((x).sF0123456789ABCDE)
+#define ROT16_2(x) ((x).sEF0123456789ABCD)
+#define ROT16_3(x) ((x).sDEF0123456789ABC)
+#define ROT16_4(x) ((x).sCDEF0123456789AB)
+#define ROT16_5(x) ((x).sBCDEF0123456789A)
+#define ROT16_6(x) ((x).sABCDEF0123456789)
+#define ROT16_7(x) ((x).s9ABCDEF012345678)
+#define ROT16_8(x) ((x).s89ABCDEF01234567)
+#define ROT16_9(x) ((x).s789ABCDEF0123456)
#define ROT16_10(x) ((x).s6789ABCDEF012345)
#define ROT16_11(x) ((x).s56789ABCDEF01234)
#define ROT16_12(x) ((x).s456789ABCDEF0123)
@@ -168,7 +168,7 @@
* @{
*/
#define ROTATE_STR(x, s, n) ROT##s##_##n(x)
-#define ROTATE(x, s, n) ROTATE_STR(x, s, n)
+#define ROTATE(x, s, n) ROTATE_STR(x, s, n)
/** @} */ // end of group ROTATE
/** Creates a vector of size n filled with offset values corresponding to the location of each element.
@@ -179,11 +179,11 @@
* @return The vector filled with offset values
* @{
*/
-#define V_OFFS1(dt) (dt##1)(0)
-#define V_OFFS2(dt) (dt##2)(0, 1)
-#define V_OFFS3(dt) (dt##3)(0, 1, 2)
-#define V_OFFS4(dt) (dt##4)(0, 1, 2, 3)
-#define V_OFFS8(dt) (dt##8)(0, 1, 2, 3, 4, 5, 6, 7)
+#define V_OFFS1(dt) (dt##1)(0)
+#define V_OFFS2(dt) (dt##2)(0, 1)
+#define V_OFFS3(dt) (dt##3)(0, 1, 2)
+#define V_OFFS4(dt) (dt##4)(0, 1, 2, 3)
+#define V_OFFS8(dt) (dt##8)(0, 1, 2, 3, 4, 5, 6, 7)
#define V_OFFS16(dt) (dt##16)(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)
/** @} */ // end of group V_OFFSn
@@ -197,11 +197,11 @@
* @{
*/
#define VEC_OFFS_STR(dt, s) V_OFFS##s(dt)
-#define VEC_OFFS(dt, s) VEC_OFFS_STR(dt, s)
+#define VEC_OFFS(dt, s) VEC_OFFS_STR(dt, s)
/** @} */ // end of group VEC_OFFS
#define VLOAD_STR(size) vload##size
-#define VLOAD(size) VLOAD_STR(size)
+#define VLOAD(size) VLOAD_STR(size)
/** Extended partial vload that correctly handles scalar values as well.
* Load the **lower** 0 to (n-1)th elements of the given vector while minimising the amount of load ops
@@ -219,23 +219,23 @@
* @{
*/
#define VLOAD_PARTIAL_STR(size, load_size) vload_partial_##size##_##load_size
-#define VLOAD_PARTIAL(size, load_size) VLOAD_PARTIAL_STR(size, load_size)
+#define VLOAD_PARTIAL(size, load_size) VLOAD_PARTIAL_STR(size, load_size)
#define NO_LOAD(data, offs, ptr) \
{ \
}
// Size == 1 (scalar)
-#define vload_partial_1_0 NO_LOAD
-#define vload_partial_1_1 vload1
-#define vload_partial_1_2 NO_LOAD
-#define vload_partial_1_3 NO_LOAD
-#define vload_partial_1_4 NO_LOAD
-#define vload_partial_1_5 NO_LOAD
-#define vload_partial_1_6 NO_LOAD
-#define vload_partial_1_7 NO_LOAD
-#define vload_partial_1_8 NO_LOAD
-#define vload_partial_1_9 NO_LOAD
+#define vload_partial_1_0 NO_LOAD
+#define vload_partial_1_1 vload1
+#define vload_partial_1_2 NO_LOAD
+#define vload_partial_1_3 NO_LOAD
+#define vload_partial_1_4 NO_LOAD
+#define vload_partial_1_5 NO_LOAD
+#define vload_partial_1_6 NO_LOAD
+#define vload_partial_1_7 NO_LOAD
+#define vload_partial_1_8 NO_LOAD
+#define vload_partial_1_9 NO_LOAD
#define vload_partial_1_10 NO_LOAD
#define vload_partial_1_11 NO_LOAD
#define vload_partial_1_12 NO_LOAD
@@ -244,16 +244,16 @@
#define vload_partial_1_15 NO_LOAD
#define vload_partial_1_16 NO_LOAD
// Size == 2
-#define vload_partial_2_0 NO_LOAD
-#define vload_partial_2_1 vload_partial_1
-#define vload_partial_2_2 vload_partial_2
-#define vload_partial_2_3 NO_LOAD
-#define vload_partial_2_4 NO_LOAD
-#define vload_partial_2_5 NO_LOAD
-#define vload_partial_2_6 NO_LOAD
-#define vload_partial_2_7 NO_LOAD
-#define vload_partial_2_8 NO_LOAD
-#define vload_partial_2_9 NO_LOAD
+#define vload_partial_2_0 NO_LOAD
+#define vload_partial_2_1 vload_partial_1
+#define vload_partial_2_2 vload_partial_2
+#define vload_partial_2_3 NO_LOAD
+#define vload_partial_2_4 NO_LOAD
+#define vload_partial_2_5 NO_LOAD
+#define vload_partial_2_6 NO_LOAD
+#define vload_partial_2_7 NO_LOAD
+#define vload_partial_2_8 NO_LOAD
+#define vload_partial_2_9 NO_LOAD
#define vload_partial_2_10 NO_LOAD
#define vload_partial_2_11 NO_LOAD
#define vload_partial_2_12 NO_LOAD
@@ -262,16 +262,16 @@
#define vload_partial_2_15 NO_LOAD
#define vload_partial_2_16 NO_LOAD
// Size == 3
-#define vload_partial_3_0 NO_LOAD
-#define vload_partial_3_1 vload_partial_1
-#define vload_partial_3_2 vload_partial_2
-#define vload_partial_3_3 vload_partial_3
-#define vload_partial_3_4 NO_LOAD
-#define vload_partial_3_5 NO_LOAD
-#define vload_partial_3_6 NO_LOAD
-#define vload_partial_3_7 NO_LOAD
-#define vload_partial_3_8 NO_LOAD
-#define vload_partial_3_9 NO_LOAD
+#define vload_partial_3_0 NO_LOAD
+#define vload_partial_3_1 vload_partial_1
+#define vload_partial_3_2 vload_partial_2
+#define vload_partial_3_3 vload_partial_3
+#define vload_partial_3_4 NO_LOAD
+#define vload_partial_3_5 NO_LOAD
+#define vload_partial_3_6 NO_LOAD
+#define vload_partial_3_7 NO_LOAD
+#define vload_partial_3_8 NO_LOAD
+#define vload_partial_3_9 NO_LOAD
#define vload_partial_3_10 NO_LOAD
#define vload_partial_3_11 NO_LOAD
#define vload_partial_3_12 NO_LOAD
@@ -280,16 +280,16 @@
#define vload_partial_3_15 NO_LOAD
#define vload_partial_3_16 NO_LOAD
// Size == 4
-#define vload_partial_4_0 NO_LOAD
-#define vload_partial_4_1 vload_partial_1
-#define vload_partial_4_2 vload_partial_2
-#define vload_partial_4_3 vload_partial_3
-#define vload_partial_4_4 vload_partial_4
-#define vload_partial_4_5 NO_LOAD
-#define vload_partial_4_6 NO_LOAD
-#define vload_partial_4_7 NO_LOAD
-#define vload_partial_4_8 NO_LOAD
-#define vload_partial_4_9 NO_LOAD
+#define vload_partial_4_0 NO_LOAD
+#define vload_partial_4_1 vload_partial_1
+#define vload_partial_4_2 vload_partial_2
+#define vload_partial_4_3 vload_partial_3
+#define vload_partial_4_4 vload_partial_4
+#define vload_partial_4_5 NO_LOAD
+#define vload_partial_4_6 NO_LOAD
+#define vload_partial_4_7 NO_LOAD
+#define vload_partial_4_8 NO_LOAD
+#define vload_partial_4_9 NO_LOAD
#define vload_partial_4_10 NO_LOAD
#define vload_partial_4_11 NO_LOAD
#define vload_partial_4_12 NO_LOAD
@@ -298,16 +298,16 @@
#define vload_partial_4_15 NO_LOAD
#define vload_partial_4_16 NO_LOAD
// Size == 8
-#define vload_partial_8_0 NO_LOAD
-#define vload_partial_8_1 vload_partial_1
-#define vload_partial_8_2 vload_partial_2
-#define vload_partial_8_3 vload_partial_3
-#define vload_partial_8_4 vload_partial_4
-#define vload_partial_8_5 vload_partial_5
-#define vload_partial_8_6 vload_partial_6
-#define vload_partial_8_7 vload_partial_7
-#define vload_partial_8_8 vload_partial_8
-#define vload_partial_8_9 NO_LOAD
+#define vload_partial_8_0 NO_LOAD
+#define vload_partial_8_1 vload_partial_1
+#define vload_partial_8_2 vload_partial_2
+#define vload_partial_8_3 vload_partial_3
+#define vload_partial_8_4 vload_partial_4
+#define vload_partial_8_5 vload_partial_5
+#define vload_partial_8_6 vload_partial_6
+#define vload_partial_8_7 vload_partial_7
+#define vload_partial_8_8 vload_partial_8
+#define vload_partial_8_9 NO_LOAD
#define vload_partial_8_10 NO_LOAD
#define vload_partial_8_11 NO_LOAD
#define vload_partial_8_12 NO_LOAD
@@ -316,16 +316,16 @@
#define vload_partial_8_15 NO_LOAD
#define vload_partial_8_16 NO_LOAD
// Size == 16
-#define vload_partial_16_0 NO_LOAD
-#define vload_partial_16_1 vload_partial_1
-#define vload_partial_16_2 vload_partial_2
-#define vload_partial_16_3 vload_partial_3
-#define vload_partial_16_4 vload_partial_4
-#define vload_partial_16_5 vload_partial_5
-#define vload_partial_16_6 vload_partial_6
-#define vload_partial_16_7 vload_partial_7
-#define vload_partial_16_8 vload_partial_8
-#define vload_partial_16_9 vload_partial_9
+#define vload_partial_16_0 NO_LOAD
+#define vload_partial_16_1 vload_partial_1
+#define vload_partial_16_2 vload_partial_2
+#define vload_partial_16_3 vload_partial_3
+#define vload_partial_16_4 vload_partial_4
+#define vload_partial_16_5 vload_partial_5
+#define vload_partial_16_6 vload_partial_6
+#define vload_partial_16_7 vload_partial_7
+#define vload_partial_16_8 vload_partial_8
+#define vload_partial_16_9 vload_partial_9
#define vload_partial_16_10 vload_partial_10
#define vload_partial_16_11 vload_partial_11
#define vload_partial_16_12 vload_partial_12
@@ -351,17 +351,13 @@
* @param[in] PTR The base pointer
* @{
*/
-#define vload_partial_1(DATA, OFFSET, PTR) \
- DATA.s0 = vload1(OFFSET, PTR);
+#define vload_partial_1(DATA, OFFSET, PTR) DATA.s0 = vload1(OFFSET, PTR);
-#define vload_partial_2(DATA, OFFSET, PTR) \
- DATA.s01 = vload2(OFFSET, PTR);
+#define vload_partial_2(DATA, OFFSET, PTR) DATA.s01 = vload2(OFFSET, PTR);
-#define vload_partial_3(DATA, OFFSET, PTR) \
- DATA.s012 = vload3(OFFSET, PTR);
+#define vload_partial_3(DATA, OFFSET, PTR) DATA.s012 = vload3(OFFSET, PTR);
-#define vload_partial_4(DATA, OFFSET, PTR) \
- DATA.s0123 = vload4(OFFSET, PTR);
+#define vload_partial_4(DATA, OFFSET, PTR) DATA.s0123 = vload4(OFFSET, PTR);
#define vload_partial_5(DATA, OFFSET, PTR) \
vload_partial_4(DATA.s0123, OFFSET, PTR); \
@@ -375,8 +371,7 @@
vload_partial_4(DATA.s0123, OFFSET, PTR); \
vload_partial_3(DATA.s456, OFFSET, PTR + 4);
-#define vload_partial_8(DATA, OFFSET, PTR) \
- DATA.s01234567 = vload8(OFFSET, PTR);
+#define vload_partial_8(DATA, OFFSET, PTR) DATA.s01234567 = vload8(OFFSET, PTR);
#define vload_partial_9(DATA, OFFSET, PTR) \
vload_partial_8(DATA.s01234567, OFFSET, PTR); \
@@ -406,13 +401,12 @@
vload_partial_8(DATA.s01234567, OFFSET, PTR); \
vload_partial_7(DATA.s89ABCDEF, OFFSET, PTR + 8);
-#define vload_partial_16(DATA, OFFSET, PTR) \
- DATA = vload16(OFFSET, PTR);
+#define vload_partial_16(DATA, OFFSET, PTR) DATA = vload16(OFFSET, PTR);
/** @} */ // end of groupd vload_partial_n
/** @} */ // end of groupd VLOAD_PARTIAL
-#define PIXEL_UNIT4 1
-#define PIXEL_UNIT8 2
+#define PIXEL_UNIT4 1
+#define PIXEL_UNIT8 2
#define PIXEL_UNIT16 4
/** Utility macro to convert a vector size in pixel unit.
@@ -425,27 +419,45 @@
* @{
*/
#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size) PIXEL_UNIT##vec_size
-#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(vec_size) CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size)
+#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(vec_size) CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size)
/** @} */ // end of group CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT
#define read_image2d_floatx1(img, x_coord, y_coord) (float4)(read_imagef(img, (int2)(x_coord, y_coord)));
-#define read_image2d_floatx2(img, x_coord, y_coord) (float8)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)));
-#define read_image2d_floatx4(img, x_coord, y_coord) (float16)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)), read_imagef(img, (int2)(x_coord + 2, y_coord)), read_imagef(img, (int2)(x_coord + 3, y_coord)));
+#define read_image2d_floatx2(img, x_coord, y_coord) \
+ (float8)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)));
+#define read_image2d_floatx4(img, x_coord, y_coord) \
+ (float16)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)), \
+ read_imagef(img, (int2)(x_coord + 2, y_coord)), read_imagef(img, (int2)(x_coord + 3, y_coord)));
#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#define read_image2d_halfx1(img, x_coord, y_coord) (half4)(read_imageh(img, (int2)(x_coord, y_coord)));
-#define read_image2d_halfx2(img, x_coord, y_coord) (half8)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)));
-#define read_image2d_halfx4(img, x_coord, y_coord) (half16)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)), read_imageh(img, (int2)(x_coord + 2, y_coord)), read_imageh(img, (int2)(x_coord + 3, y_coord)));
+#define read_image2d_halfx2(img, x_coord, y_coord) \
+ (half8)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)));
+#define read_image2d_halfx4(img, x_coord, y_coord) \
+ (half16)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)), \
+ read_imageh(img, (int2)(x_coord + 2, y_coord)), read_imageh(img, (int2)(x_coord + 3, y_coord)));
#endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#define write_image2d_floatx1(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values));
-#define write_image2d_floatx2(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567));
-#define write_image2d_floatx4(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567), write_imagef(img, (int2)(x_coord + 2, y_coord), values.s89AB), write_imagef(img, (int2)(x_coord + 3, y_coord), values.sCDEF));
+#define write_image2d_floatx2(img, x_coord, y_coord, values) \
+ (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), \
+ write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567));
+#define write_image2d_floatx4(img, x_coord, y_coord, values) \
+ (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), \
+ write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567), \
+ write_imagef(img, (int2)(x_coord + 2, y_coord), values.s89AB), \
+ write_imagef(img, (int2)(x_coord + 3, y_coord), values.sCDEF));
#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#define write_image2d_halfx1(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values));
-#define write_image2d_halfx2(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567));
-#define write_image2d_halfx4(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567), write_imageh(img, (int2)(x_coord + 2, y_coord), values.s89AB), write_imageh(img, (int2)(x_coord + 3, y_coord), values.sCDEF));
+#define write_image2d_halfx2(img, x_coord, y_coord, values) \
+ (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), \
+ write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567));
+#define write_image2d_halfx4(img, x_coord, y_coord, values) \
+ (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), \
+ write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567), \
+ write_imageh(img, (int2)(x_coord + 2, y_coord), values.s89AB), \
+ write_imageh(img, (int2)(x_coord + 3, y_coord), values.sCDEF));
#endif // defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
/** Utility macro to read a 2D OpenCL image object.
@@ -462,7 +474,7 @@
* @{
*/
#define READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord) read_image2d_##data_type##x##n0(img, x_coord, y_coord)
-#define READ_IMAGE2D(data_type, n0, img, x_coord, y_coord) READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord)
+#define READ_IMAGE2D(data_type, n0, img, x_coord, y_coord) READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord)
/** @} */
/** Utility macro to write a 2D OpenCL image object.
@@ -478,26 +490,28 @@
*
* @{
*/
-#define WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values) write_image2d_##data_type##x##n0(img, x_coord, y_coord, values)
-#define WRITE_IMAGE2D(data_type, n0, img, x_coord, y_coord, values) WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values)
+#define WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values) \
+ write_image2d_##data_type##x##n0(img, x_coord, y_coord, values)
+#define WRITE_IMAGE2D(data_type, n0, img, x_coord, y_coord, values) \
+ WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values)
/** @} */
#define VSTORE_STR(size) vstore##size
-#define VSTORE(size) VSTORE_STR(size)
+#define VSTORE(size) VSTORE_STR(size)
-#define float1 float
-#define half1 half
-#define char1 char
-#define uchar1 uchar
-#define short1 short
+#define float1 float
+#define half1 half
+#define char1 char
+#define uchar1 uchar
+#define short1 short
#define ushort1 ushort
-#define int1 int
-#define uint1 uint
-#define long1 long
-#define ulong1 ulong
+#define int1 int
+#define uint1 uint
+#define long1 long
+#define ulong1 ulong
#define double1 double
-#define vload1(OFFSET, PTR) *(OFFSET + PTR)
+#define vload1(OFFSET, PTR) *(OFFSET + PTR)
#define vstore1(DATA, OFFSET, PTR) *(OFFSET + PTR) = DATA
/** Extended partial vstore that correctly handles scalar values as well.
@@ -516,23 +530,23 @@
* @{
*/
#define VSTORE_PARTIAL_STR(size, store_size) vstore_partial_##size##_##store_size
-#define VSTORE_PARTIAL(size, store_size) VSTORE_PARTIAL_STR(size, store_size)
+#define VSTORE_PARTIAL(size, store_size) VSTORE_PARTIAL_STR(size, store_size)
#define NO_STORE(data, offs, ptr) \
{ \
}
// Size == 1 (scalar)
-#define vstore_partial_1_0 NO_STORE
-#define vstore_partial_1_1 vstore1
-#define vstore_partial_1_2 NO_STORE
-#define vstore_partial_1_3 NO_STORE
-#define vstore_partial_1_4 NO_STORE
-#define vstore_partial_1_5 NO_STORE
-#define vstore_partial_1_6 NO_STORE
-#define vstore_partial_1_7 NO_STORE
-#define vstore_partial_1_8 NO_STORE
-#define vstore_partial_1_9 NO_STORE
+#define vstore_partial_1_0 NO_STORE
+#define vstore_partial_1_1 vstore1
+#define vstore_partial_1_2 NO_STORE
+#define vstore_partial_1_3 NO_STORE
+#define vstore_partial_1_4 NO_STORE
+#define vstore_partial_1_5 NO_STORE
+#define vstore_partial_1_6 NO_STORE
+#define vstore_partial_1_7 NO_STORE
+#define vstore_partial_1_8 NO_STORE
+#define vstore_partial_1_9 NO_STORE
#define vstore_partial_1_10 NO_STORE
#define vstore_partial_1_11 NO_STORE
#define vstore_partial_1_12 NO_STORE
@@ -541,16 +555,16 @@
#define vstore_partial_1_15 NO_STORE
#define vstore_partial_1_16 NO_STORE
// Size == 2
-#define vstore_partial_2_0 NO_STORE
-#define vstore_partial_2_1 vstore_partial_1
-#define vstore_partial_2_2 vstore_partial_2
-#define vstore_partial_2_3 NO_STORE
-#define vstore_partial_2_4 NO_STORE
-#define vstore_partial_2_5 NO_STORE
-#define vstore_partial_2_6 NO_STORE
-#define vstore_partial_2_7 NO_STORE
-#define vstore_partial_2_8 NO_STORE
-#define vstore_partial_2_9 NO_STORE
+#define vstore_partial_2_0 NO_STORE
+#define vstore_partial_2_1 vstore_partial_1
+#define vstore_partial_2_2 vstore_partial_2
+#define vstore_partial_2_3 NO_STORE
+#define vstore_partial_2_4 NO_STORE
+#define vstore_partial_2_5 NO_STORE
+#define vstore_partial_2_6 NO_STORE
+#define vstore_partial_2_7 NO_STORE
+#define vstore_partial_2_8 NO_STORE
+#define vstore_partial_2_9 NO_STORE
#define vstore_partial_2_10 NO_STORE
#define vstore_partial_2_11 NO_STORE
#define vstore_partial_2_12 NO_STORE
@@ -559,16 +573,16 @@
#define vstore_partial_2_15 NO_STORE
#define vstore_partial_2_16 NO_STORE
// Size == 3
-#define vstore_partial_3_0 NO_STORE
-#define vstore_partial_3_1 vstore_partial_1
-#define vstore_partial_3_2 vstore_partial_2
-#define vstore_partial_3_3 vstore_partial_3
-#define vstore_partial_3_4 NO_STORE
-#define vstore_partial_3_5 NO_STORE
-#define vstore_partial_3_6 NO_STORE
-#define vstore_partial_3_7 NO_STORE
-#define vstore_partial_3_8 NO_STORE
-#define vstore_partial_3_9 NO_STORE
+#define vstore_partial_3_0 NO_STORE
+#define vstore_partial_3_1 vstore_partial_1
+#define vstore_partial_3_2 vstore_partial_2
+#define vstore_partial_3_3 vstore_partial_3
+#define vstore_partial_3_4 NO_STORE
+#define vstore_partial_3_5 NO_STORE
+#define vstore_partial_3_6 NO_STORE
+#define vstore_partial_3_7 NO_STORE
+#define vstore_partial_3_8 NO_STORE
+#define vstore_partial_3_9 NO_STORE
#define vstore_partial_3_10 NO_STORE
#define vstore_partial_3_11 NO_STORE
#define vstore_partial_3_12 NO_STORE
@@ -577,16 +591,16 @@
#define vstore_partial_3_15 NO_STORE
#define vstore_partial_3_16 NO_STORE
// Size == 4
-#define vstore_partial_4_0 NO_STORE
-#define vstore_partial_4_1 vstore_partial_1
-#define vstore_partial_4_2 vstore_partial_2
-#define vstore_partial_4_3 vstore_partial_3
-#define vstore_partial_4_4 vstore_partial_4
-#define vstore_partial_4_5 NO_STORE
-#define vstore_partial_4_6 NO_STORE
-#define vstore_partial_4_7 NO_STORE
-#define vstore_partial_4_8 NO_STORE
-#define vstore_partial_4_9 NO_STORE
+#define vstore_partial_4_0 NO_STORE
+#define vstore_partial_4_1 vstore_partial_1
+#define vstore_partial_4_2 vstore_partial_2
+#define vstore_partial_4_3 vstore_partial_3
+#define vstore_partial_4_4 vstore_partial_4
+#define vstore_partial_4_5 NO_STORE
+#define vstore_partial_4_6 NO_STORE
+#define vstore_partial_4_7 NO_STORE
+#define vstore_partial_4_8 NO_STORE
+#define vstore_partial_4_9 NO_STORE
#define vstore_partial_4_10 NO_STORE
#define vstore_partial_4_11 NO_STORE
#define vstore_partial_4_12 NO_STORE
@@ -595,16 +609,16 @@
#define vstore_partial_4_15 NO_STORE
#define vstore_partial_4_16 NO_STORE
// Size == 8
-#define vstore_partial_8_0 NO_STORE
-#define vstore_partial_8_1 vstore_partial_1
-#define vstore_partial_8_2 vstore_partial_2
-#define vstore_partial_8_3 vstore_partial_3
-#define vstore_partial_8_4 vstore_partial_4
-#define vstore_partial_8_5 vstore_partial_5
-#define vstore_partial_8_6 vstore_partial_6
-#define vstore_partial_8_7 vstore_partial_7
-#define vstore_partial_8_8 vstore_partial_8
-#define vstore_partial_8_9 NO_STORE
+#define vstore_partial_8_0 NO_STORE
+#define vstore_partial_8_1 vstore_partial_1
+#define vstore_partial_8_2 vstore_partial_2
+#define vstore_partial_8_3 vstore_partial_3
+#define vstore_partial_8_4 vstore_partial_4
+#define vstore_partial_8_5 vstore_partial_5
+#define vstore_partial_8_6 vstore_partial_6
+#define vstore_partial_8_7 vstore_partial_7
+#define vstore_partial_8_8 vstore_partial_8
+#define vstore_partial_8_9 NO_STORE
#define vstore_partial_8_10 NO_STORE
#define vstore_partial_8_11 NO_STORE
#define vstore_partial_8_12 NO_STORE
@@ -613,16 +627,16 @@
#define vstore_partial_8_15 NO_STORE
#define vstore_partial_8_16 NO_STORE
// Size == 16
-#define vstore_partial_16_0 NO_STORE
-#define vstore_partial_16_1 vstore_partial_1
-#define vstore_partial_16_2 vstore_partial_2
-#define vstore_partial_16_3 vstore_partial_3
-#define vstore_partial_16_4 vstore_partial_4
-#define vstore_partial_16_5 vstore_partial_5
-#define vstore_partial_16_6 vstore_partial_6
-#define vstore_partial_16_7 vstore_partial_7
-#define vstore_partial_16_8 vstore_partial_8
-#define vstore_partial_16_9 vstore_partial_9
+#define vstore_partial_16_0 NO_STORE
+#define vstore_partial_16_1 vstore_partial_1
+#define vstore_partial_16_2 vstore_partial_2
+#define vstore_partial_16_3 vstore_partial_3
+#define vstore_partial_16_4 vstore_partial_4
+#define vstore_partial_16_5 vstore_partial_5
+#define vstore_partial_16_6 vstore_partial_6
+#define vstore_partial_16_7 vstore_partial_7
+#define vstore_partial_16_8 vstore_partial_8
+#define vstore_partial_16_9 vstore_partial_9
#define vstore_partial_16_10 vstore_partial_10
#define vstore_partial_16_11 vstore_partial_11
#define vstore_partial_16_12 vstore_partial_12
@@ -648,17 +662,13 @@
* @param[in] PTR The base pointer
* @{
*/
-#define vstore_partial_1(DATA, OFFSET, PTR) \
- vstore1(DATA.s0, OFFSET, PTR);
+#define vstore_partial_1(DATA, OFFSET, PTR) vstore1(DATA.s0, OFFSET, PTR);
-#define vstore_partial_2(DATA, OFFSET, PTR) \
- vstore2(DATA.s01, OFFSET, PTR);
+#define vstore_partial_2(DATA, OFFSET, PTR) vstore2(DATA.s01, OFFSET, PTR);
-#define vstore_partial_3(DATA, OFFSET, PTR) \
- vstore3(DATA.s012, OFFSET, PTR);
+#define vstore_partial_3(DATA, OFFSET, PTR) vstore3(DATA.s012, OFFSET, PTR);
-#define vstore_partial_4(DATA, OFFSET, PTR) \
- vstore4(DATA.s0123, OFFSET, PTR);
+#define vstore_partial_4(DATA, OFFSET, PTR) vstore4(DATA.s0123, OFFSET, PTR);
#define vstore_partial_5(DATA, OFFSET, PTR) \
vstore_partial_4(DATA.s0123, OFFSET, PTR); \
@@ -672,8 +682,7 @@
vstore_partial_4(DATA.s0123, OFFSET, PTR); \
vstore_partial_3(DATA.s456, OFFSET, PTR + 4);
-#define vstore_partial_8(DATA, OFFSET, PTR) \
- vstore8(DATA.s01234567, OFFSET, PTR);
+#define vstore_partial_8(DATA, OFFSET, PTR) vstore8(DATA.s01234567, OFFSET, PTR);
#define vstore_partial_9(DATA, OFFSET, PTR) \
vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
@@ -703,186 +712,156 @@
vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
vstore_partial_7(DATA.s89abcdef, OFFSET, PTR + 8);
-#define vstore_partial_16(DATA, OFFSET, PTR) \
- vstore16(DATA, OFFSET, PTR);
+#define vstore_partial_16(DATA, OFFSET, PTR) vstore16(DATA, OFFSET, PTR);
/** @} */ // end of groupd vstore_partial_n
/** @} */ // end of groupd VSTORE_PARTIAL
// Convert built-in functions with _sat modifier are not supported in floating point so we create defines
// without _sat to overcome this issue
-#define convert_float_sat convert_float
-#define convert_float1_sat convert_float
-#define convert_float2_sat convert_float2
-#define convert_float3_sat convert_float3
-#define convert_float4_sat convert_float4
-#define convert_float8_sat convert_float8
+#define convert_float_sat convert_float
+#define convert_float1_sat convert_float
+#define convert_float2_sat convert_float2
+#define convert_float3_sat convert_float3
+#define convert_float4_sat convert_float4
+#define convert_float8_sat convert_float8
#define convert_float16_sat convert_float16
-#define convert_half_sat convert_float
-#define convert_half1_sat convert_half
-#define convert_half2_sat convert_half2
-#define convert_half3_sat convert_half3
-#define convert_half4_sat convert_half4
-#define convert_half8_sat convert_half8
-#define convert_half16_sat convert_half16
+#define convert_half_sat convert_float
+#define convert_half1_sat convert_half
+#define convert_half2_sat convert_half2
+#define convert_half3_sat convert_half3
+#define convert_half4_sat convert_half4
+#define convert_half8_sat convert_half8
+#define convert_half16_sat convert_half16
-#define convert_float1 convert_float
-#define convert_half1 convert_half
-#define convert_char1 convert_char
-#define convert_uchar1 convert_uchar
-#define convert_short1 convert_short
+#define convert_float1 convert_float
+#define convert_half1 convert_half
+#define convert_char1 convert_char
+#define convert_uchar1 convert_uchar
+#define convert_short1 convert_short
#define convert_ushort1 convert_ushort
-#define convert_int1 convert_int
-#define convert_uint1 convert_uint
-#define convert_long1 convert_long
-#define convert_ulong1 convert_ulong
+#define convert_int1 convert_int
+#define convert_uint1 convert_uint
+#define convert_long1 convert_long
+#define convert_ulong1 convert_ulong
#define convert_double1 convert_double
-#define convert_char1_sat convert_char_sat
-#define convert_uchar1_sat convert_uchar_sat
-#define convert_uchar2_sat convert_uchar2_sat
-#define convert_uchar3_sat convert_uchar3_sat
-#define convert_uchar4_sat convert_uchar4_sat
-#define convert_uchar8_sat convert_uchar8_sat
+#define convert_char1_sat convert_char_sat
+#define convert_uchar1_sat convert_uchar_sat
+#define convert_uchar2_sat convert_uchar2_sat
+#define convert_uchar3_sat convert_uchar3_sat
+#define convert_uchar4_sat convert_uchar4_sat
+#define convert_uchar8_sat convert_uchar8_sat
#define convert_uchar16_sat convert_uchar16_sat
-#define convert_short1_sat convert_short_sat
+#define convert_short1_sat convert_short_sat
#define convert_ushort1_sat convert_ushort_sat
-#define convert_int1_sat convert_int_sat
-#define convert_uint1_sat convert_uint_sat
-#define convert_long1_sat convert_long_sat
-#define convert_ulong1_sat convert_ulong_sat
+#define convert_int1_sat convert_int_sat
+#define convert_uint1_sat convert_uint_sat
+#define convert_long1_sat convert_long_sat
+#define convert_ulong1_sat convert_ulong_sat
#define convert_double1_sat convert_double_sat
#define VEC_DATA_TYPE_STR(type, size) type##size
-#define VEC_DATA_TYPE(type, size) VEC_DATA_TYPE_STR(type, size)
+#define VEC_DATA_TYPE(type, size) VEC_DATA_TYPE_STR(type, size)
#define CONVERT_STR(x, type) (convert_##type((x)))
-#define CONVERT(x, type) CONVERT_STR(x, type)
+#define CONVERT(x, type) CONVERT_STR(x, type)
#define CONVERT_SAT_STR(x, type) (convert_##type##_sat((x)))
-#define CONVERT_SAT(x, type) CONVERT_SAT_STR(x, type)
+#define CONVERT_SAT(x, type) CONVERT_SAT_STR(x, type)
#define CONVERT_SAT_ROUND_STR(x, type, round) (convert_##type##_sat_##round((x)))
-#define CONVERT_SAT_ROUND(x, type, round) CONVERT_SAT_ROUND_STR(x, type, round)
+#define CONVERT_SAT_ROUND(x, type, round) CONVERT_SAT_ROUND_STR(x, type, round)
-#define select_vec_dt_uchar(size) uchar##size
-#define select_vec_dt_char(size) char##size
+#define select_vec_dt_uchar(size) uchar##size
+#define select_vec_dt_char(size) char##size
#define select_vec_dt_ushort(size) ushort##size
-#define select_vec_dt_short(size) short##size
-#define select_vec_dt_half(size) short##size
-#define select_vec_dt_uint(size) uint##size
-#define select_vec_dt_int(size) int##size
-#define select_vec_dt_float(size) int##size
-#define select_vec_dt_ulong(size) ulong##size
-#define select_vec_dt_long(size) long##size
+#define select_vec_dt_short(size) short##size
+#define select_vec_dt_half(size) short##size
+#define select_vec_dt_uint(size) uint##size
+#define select_vec_dt_int(size) int##size
+#define select_vec_dt_float(size) int##size
+#define select_vec_dt_ulong(size) ulong##size
+#define select_vec_dt_long(size) long##size
#define SELECT_VEC_DATA_TYPE_STR(type, size) select_vec_dt_##type(size)
-#define SELECT_VEC_DATA_TYPE(type, size) SELECT_VEC_DATA_TYPE_STR(type, size)
-#define SELECT_DATA_TYPE(type) SELECT_VEC_DATA_TYPE_STR(type, 1)
+#define SELECT_VEC_DATA_TYPE(type, size) SELECT_VEC_DATA_TYPE_STR(type, size)
+#define SELECT_DATA_TYPE(type) SELECT_VEC_DATA_TYPE_STR(type, 1)
-#define signed_int_vec_dt_uchar(size) char##size
-#define signed_int_vec_dt_char(size) char##size
+#define signed_int_vec_dt_uchar(size) char##size
+#define signed_int_vec_dt_char(size) char##size
#define signed_int_vec_dt_ushort(size) short##size
-#define signed_int_vec_dt_short(size) short##size
-#define signed_int_vec_dt_half(size) short##size
-#define signed_int_vec_dt_uint(size) int##size
-#define signed_int_vec_dt_int(size) int##size
-#define signed_int_vec_dt_float(size) int##size
-#define signed_int_vec_dt_ulong(size) long##size
-#define signed_int_vec_dt_long(size) long##size
+#define signed_int_vec_dt_short(size) short##size
+#define signed_int_vec_dt_half(size) short##size
+#define signed_int_vec_dt_uint(size) int##size
+#define signed_int_vec_dt_int(size) int##size
+#define signed_int_vec_dt_float(size) int##size
+#define signed_int_vec_dt_ulong(size) long##size
+#define signed_int_vec_dt_long(size) long##size
#define SIGNED_INT_VEC_DATA_TYPE_STR(type, size) signed_int_vec_dt_##type(size)
-#define SIGNED_INT_VEC_DATA_TYPE(type, size) SIGNED_INT_VEC_DATA_TYPE_STR(type, size)
-#define SIGNED_INT_DATA_TYPE(type) SIGNED_INT_VEC_DATA_TYPE_STR(type, 1)
+#define SIGNED_INT_VEC_DATA_TYPE(type, size) SIGNED_INT_VEC_DATA_TYPE_STR(type, size)
+#define SIGNED_INT_DATA_TYPE(type) SIGNED_INT_VEC_DATA_TYPE_STR(type, 1)
-#define sum_reduce_1(x) (x)
-#define sum_reduce_2(x) ((x).s0) + ((x).s1)
-#define sum_reduce_3(x) sum_reduce_2((x).s01) + ((x).s2)
-#define sum_reduce_4(x) sum_reduce_2((x).s01) + sum_reduce_2((x).s23)
-#define sum_reduce_8(x) sum_reduce_4((x).s0123) + sum_reduce_4((x).s4567)
+#define sum_reduce_1(x) (x)
+#define sum_reduce_2(x) ((x).s0) + ((x).s1)
+#define sum_reduce_3(x) sum_reduce_2((x).s01) + ((x).s2)
+#define sum_reduce_4(x) sum_reduce_2((x).s01) + sum_reduce_2((x).s23)
+#define sum_reduce_8(x) sum_reduce_4((x).s0123) + sum_reduce_4((x).s4567)
#define sum_reduce_16(x) sum_reduce_8((x).s01234567) + sum_reduce_8((x).s89ABCDEF)
#define SUM_REDUCE_STR(x, size) sum_reduce_##size(x)
-#define SUM_REDUCE(x, size) SUM_REDUCE_STR(x, size)
+#define SUM_REDUCE(x, size) SUM_REDUCE_STR(x, size)
-#define prod_reduce_1(x) (x)
-#define prod_reduce_2(x) ((x).s0) * ((x).s1)
-#define prod_reduce_3(x) prod_reduce_2((x).s01) * ((x).s2)
-#define prod_reduce_4(x) prod_reduce_2((x).s01) * prod_reduce_2((x).s23)
-#define prod_reduce_8(x) prod_reduce_4((x).s0123) * prod_reduce_4((x).s4567)
+#define prod_reduce_1(x) (x)
+#define prod_reduce_2(x) ((x).s0) * ((x).s1)
+#define prod_reduce_3(x) prod_reduce_2((x).s01) * ((x).s2)
+#define prod_reduce_4(x) prod_reduce_2((x).s01) * prod_reduce_2((x).s23)
+#define prod_reduce_8(x) prod_reduce_4((x).s0123) * prod_reduce_4((x).s4567)
#define prod_reduce_16(x) prod_reduce_8((x).s01234567) * prod_reduce_8((x).s89ABCDEF)
#define PROD_REDUCE_STR(x, size) prod_reduce_##size(x)
-#define PROD_REDUCE(x, size) PROD_REDUCE_STR(x, size)
+#define PROD_REDUCE(x, size) PROD_REDUCE_STR(x, size)
-#define max_reduce_1(x) (x)
-#define max_reduce_2(x) max(((x).s0), ((x).s1))
-#define max_reduce_3(x) max(max_reduce_2((x).s01), ((x).s2))
-#define max_reduce_4(x) max(max_reduce_2((x).s01), max_reduce_2((x).s23))
-#define max_reduce_8(x) max(max_reduce_4((x).s0123), max_reduce_4((x).s4567))
+#define max_reduce_1(x) (x)
+#define max_reduce_2(x) max(((x).s0), ((x).s1))
+#define max_reduce_3(x) max(max_reduce_2((x).s01), ((x).s2))
+#define max_reduce_4(x) max(max_reduce_2((x).s01), max_reduce_2((x).s23))
+#define max_reduce_8(x) max(max_reduce_4((x).s0123), max_reduce_4((x).s4567))
#define max_reduce_16(x) max(max_reduce_8((x).s01234567), max_reduce_8((x).s89ABCDEF))
#define MAX_REDUCE_STR(x, size) max_reduce_##size(x)
-#define MAX_REDUCE(x, size) MAX_REDUCE_STR(x, size)
+#define MAX_REDUCE(x, size) MAX_REDUCE_STR(x, size)
-#define min_reduce_1(x) (x)
-#define min_reduce_2(x) min(((x).s0), ((x).s1))
-#define min_reduce_3(x) min(min_reduce_2((x).s01), ((x).s2))
-#define min_reduce_4(x) min(min_reduce_2((x).s01), min_reduce_2((x).s23))
-#define min_reduce_8(x) min(min_reduce_4((x).s0123), min_reduce_4((x).s4567))
+#define min_reduce_1(x) (x)
+#define min_reduce_2(x) min(((x).s0), ((x).s1))
+#define min_reduce_3(x) min(min_reduce_2((x).s01), ((x).s2))
+#define min_reduce_4(x) min(min_reduce_2((x).s01), min_reduce_2((x).s23))
+#define min_reduce_8(x) min(min_reduce_4((x).s0123), min_reduce_4((x).s4567))
#define min_reduce_16(x) min(min_reduce_8((x).s01234567), min_reduce_8((x).s89ABCDEF))
#define MIN_REDUCE_STR(x, size) min_reduce_##size(x)
-#define MIN_REDUCE(x, size) MIN_REDUCE_STR(x, size)
+#define MIN_REDUCE(x, size) MIN_REDUCE_STR(x, size)
-#define VECTOR_DECLARATION(name) \
- __global uchar *name##_ptr, \
- uint name##_stride_x, \
- uint name##_step_x, \
- uint name##_offset_first_element_in_bytes
+#define VECTOR_DECLARATION(name) \
+ __global uchar *name##_ptr, uint name##_stride_x, uint name##_step_x, uint name##_offset_first_element_in_bytes
-#define IMAGE_DECLARATION(name) \
- __global uchar *name##_ptr, \
- uint name##_stride_x, \
- uint name##_step_x, \
- uint name##_stride_y, \
- uint name##_step_y, \
- uint name##_offset_first_element_in_bytes
+#define IMAGE_DECLARATION(name) \
+ __global uchar *name##_ptr, uint name##_stride_x, uint name##_step_x, uint name##_stride_y, uint name##_step_y, \
+ uint name##_offset_first_element_in_bytes
-#define TENSOR3D_DECLARATION(name) \
- __global uchar *name##_ptr, \
- uint name##_stride_x, \
- uint name##_step_x, \
- uint name##_stride_y, \
- uint name##_step_y, \
- uint name##_stride_z, \
- uint name##_step_z, \
- uint name##_offset_first_element_in_bytes
+#define TENSOR3D_DECLARATION(name) \
+ __global uchar *name##_ptr, uint name##_stride_x, uint name##_step_x, uint name##_stride_y, uint name##_step_y, \
+ uint name##_stride_z, uint name##_step_z, uint name##_offset_first_element_in_bytes
-#define TENSOR4D_DECLARATION(name) \
- __global uchar *name##_ptr, \
- uint name##_stride_x, \
- uint name##_step_x, \
- uint name##_stride_y, \
- uint name##_step_y, \
- uint name##_stride_z, \
- uint name##_step_z, \
- uint name##_stride_w, \
- uint name##_step_w, \
- uint name##_offset_first_element_in_bytes
+#define TENSOR4D_DECLARATION(name) \
+ __global uchar *name##_ptr, uint name##_stride_x, uint name##_step_x, uint name##_stride_y, uint name##_step_y, \
+ uint name##_stride_z, uint name##_step_z, uint name##_stride_w, uint name##_step_w, \
+ uint name##_offset_first_element_in_bytes
-#define TENSOR5D_DECLARATION(name) \
- __global uchar *name##_ptr, \
- uint name##_stride_x, \
- uint name##_step_x, \
- uint name##_stride_y, \
- uint name##_step_y, \
- uint name##_stride_z, \
- uint name##_step_z, \
- uint name##_stride_w, \
- uint name##_step_w, \
- uint name##_stride_v, \
- uint name##_step_v, \
- uint name##_offset_first_element_in_bytes
+#define TENSOR5D_DECLARATION(name) \
+ __global uchar *name##_ptr, uint name##_stride_x, uint name##_step_x, uint name##_stride_y, uint name##_step_y, \
+ uint name##_stride_z, uint name##_step_z, uint name##_stride_w, uint name##_step_w, uint name##_stride_v, \
+ uint name##_step_v, uint name##_offset_first_element_in_bytes
#define CONVERT_TO_VECTOR_STRUCT(name) \
update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x)
@@ -890,38 +869,47 @@
#define CONVERT_TO_VECTOR_STRUCT_NO_STEP(name) \
update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0)
-#define CONVERT_TO_IMAGE_STRUCT(name) \
- update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y)
+#define CONVERT_TO_IMAGE_STRUCT(name) \
+ update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, \
+ name##_stride_y, name##_step_y)
#define CONVERT_TO_IMAGE_STRUCT_NO_STEP(name) \
update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0)
-#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
- update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)
+#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
+ update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, \
+ name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, \
+ name##_step_z)
-#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP(name) \
- update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, name##_step_z)
+#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP(name) \
+ update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, \
+ name##_stride_y, 0, name##_stride_z, name##_step_z)
-#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
- update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)
+#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
+ update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, \
+ name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, \
+ name##_step_z)
-#define CONVERT_TO_TENSOR3D_STRUCT(name) \
- update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
- name##_stride_z, name##_step_z)
+#define CONVERT_TO_TENSOR3D_STRUCT(name) \
+ update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, \
+ name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)
-#define CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(name) \
- update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0)
+#define CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(name) \
+ update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, \
+ name##_stride_y, 0, name##_stride_z, 0)
-#define CONVERT_TO_TENSOR4D_STRUCT(name, mod_size) \
- update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
- name##_stride_z, name##_step_z, name##_stride_w, name##_step_w, mod_size)
+#define CONVERT_TO_TENSOR4D_STRUCT(name, mod_size) \
+ update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, \
+ name##_stride_y, name##_step_y, name##_stride_z, name##_step_z, name##_stride_w, \
+ name##_step_w, mod_size)
-#define CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(name, mod_size) \
- update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0, name##_stride_w, 0, mod_size)
+#define CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(name, mod_size) \
+ update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, \
+ name##_stride_y, 0, name##_stride_z, 0, name##_stride_w, 0, mod_size)
-#define CONVERT_TO_TENSOR3D_STRUCT_NO_UPDATE_PTR(name) \
- tensor3D_ptr_no_update(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
- name##_stride_z, name##_step_z)
+#define CONVERT_TO_TENSOR3D_STRUCT_NO_UPDATE_PTR(name) \
+ tensor3D_ptr_no_update(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, \
+ name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)
/** Structure to hold Vector information */
typedef struct Vector
@@ -970,10 +958,10 @@
*
* @return An image object
*/
-inline Vector update_vector_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x)
+inline Vector
+update_vector_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x)
{
- Vector vector =
- {
+ Vector vector = {
.ptr = ptr,
.offset_first_element_in_bytes = offset_first_element_in_bytes,
.stride_x = stride_x,
@@ -993,15 +981,13 @@
*
* @return An image object
*/
-inline Image update_image_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y)
+inline Image update_image_workitem_ptr(
+ __global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y)
{
- Image img =
- {
- .ptr = ptr,
- .offset_first_element_in_bytes = offset_first_element_in_bytes,
- .stride_x = stride_x,
- .stride_y = stride_y
- };
+ Image img = {.ptr = ptr,
+ .offset_first_element_in_bytes = offset_first_element_in_bytes,
+ .stride_x = stride_x,
+ .stride_y = stride_y};
img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y;
return img;
}
@@ -1019,16 +1005,21 @@
*
* @return A 3D tensor object
*/
-inline Image update_image_from_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
+inline Image update_image_from_tensor3D_workitem_ptr(__global uchar *ptr,
+ uint offset_first_element_in_bytes,
+ uint stride_x,
+ uint step_x,
+ uint stride_y,
+ uint step_y,
+ uint stride_z,
+ uint step_z)
{
- Image img =
- {
- .ptr = ptr,
- .offset_first_element_in_bytes = offset_first_element_in_bytes,
- .stride_x = stride_x,
- .stride_y = stride_y
- };
- img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z;
+ Image img = {.ptr = ptr,
+ .offset_first_element_in_bytes = offset_first_element_in_bytes,
+ .stride_x = stride_x,
+ .stride_y = stride_y};
+ img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y +
+ get_global_id(2) * step_z;
return img;
}
@@ -1045,17 +1036,22 @@
*
* @return A 3D tensor object
*/
-inline Tensor3D update_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
+inline Tensor3D update_tensor3D_workitem_ptr(__global uchar *ptr,
+ uint offset_first_element_in_bytes,
+ uint stride_x,
+ uint step_x,
+ uint stride_y,
+ uint step_y,
+ uint stride_z,
+ uint step_z)
{
- Tensor3D tensor =
- {
- .ptr = ptr,
- .offset_first_element_in_bytes = offset_first_element_in_bytes,
- .stride_x = stride_x,
- .stride_y = stride_y,
- .stride_z = stride_z
- };
- tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z;
+ Tensor3D tensor = {.ptr = ptr,
+ .offset_first_element_in_bytes = offset_first_element_in_bytes,
+ .stride_x = stride_x,
+ .stride_y = stride_y,
+ .stride_z = stride_z};
+ tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y +
+ get_global_id(2) * step_z;
return tensor;
}
@@ -1072,34 +1068,44 @@
*
* @return A 3D tensor object
*/
-inline Tensor3D tensor3D_ptr_no_update(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
+inline Tensor3D tensor3D_ptr_no_update(__global uchar *ptr,
+ uint offset_first_element_in_bytes,
+ uint stride_x,
+ uint step_x,
+ uint stride_y,
+ uint step_y,
+ uint stride_z,
+ uint step_z)
{
- Tensor3D tensor =
- {
- .ptr = ptr,
- .offset_first_element_in_bytes = offset_first_element_in_bytes,
- .stride_x = stride_x,
- .stride_y = stride_y,
- .stride_z = stride_z
- };
+ Tensor3D tensor = {.ptr = ptr,
+ .offset_first_element_in_bytes = offset_first_element_in_bytes,
+ .stride_x = stride_x,
+ .stride_y = stride_y,
+ .stride_z = stride_z};
return tensor;
}
-inline Tensor4D update_tensor4D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z, uint stride_w,
- uint step_w,
- uint mod_size)
+inline Tensor4D update_tensor4D_workitem_ptr(__global uchar *ptr,
+ uint offset_first_element_in_bytes,
+ uint stride_x,
+ uint step_x,
+ uint stride_y,
+ uint step_y,
+ uint stride_z,
+ uint step_z,
+ uint stride_w,
+ uint step_w,
+ uint mod_size)
{
- Tensor4D tensor =
- {
- .ptr = ptr,
- .offset_first_element_in_bytes = offset_first_element_in_bytes,
- .stride_x = stride_x,
- .stride_y = stride_y,
- .stride_z = stride_z,
- .stride_w = stride_w
- };
+ Tensor4D tensor = {.ptr = ptr,
+ .offset_first_element_in_bytes = offset_first_element_in_bytes,
+ .stride_x = stride_x,
+ .stride_y = stride_y,
+ .stride_z = stride_z,
+ .stride_w = stride_w};
- tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + (get_global_id(2) % mod_size) * step_z + (get_global_id(2) / mod_size) * step_w;
+ tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y +
+ (get_global_id(2) % mod_size) * step_z + (get_global_id(2) / mod_size) * step_w;
return tensor;
}
@@ -1171,7 +1177,8 @@
const uint x = index;
- return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + tensor->offset_first_element_in_bytes;
+ return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z +
+ tensor->offset_first_element_in_bytes;
}
#endif // _HELPER_H
diff --git a/src/core/CL/cl_kernels/helpers_asymm.h b/src/core/CL/cl_kernels/helpers_asymm.h
index 562c5d3..166260a 100644
--- a/src/core/CL/cl_kernels/helpers_asymm.h
+++ b/src/core/CL/cl_kernels/helpers_asymm.h
@@ -34,7 +34,7 @@
* @return The converted vector
*/
#define CONVERT_DOWN_RTE_STR(x, type) (convert_##type##_rte((x)))
-#define CONVERT_DOWN_RTE(x, type) CONVERT_DOWN_RTE_STR(x, type)
+#define CONVERT_DOWN_RTE(x, type) CONVERT_DOWN_RTE_STR(x, type)
/** Quantize a floating-point scalar value to 8-bit asymmetric
*
@@ -84,14 +84,15 @@
*
* @return quantized values
*/
-#define QUANTIZE_IMPL(type, size) \
- inline VEC_DATA_TYPE(type, size) quantize_##type##size(VEC_DATA_TYPE(float, size) input, float offset, float scale) \
- { \
- VEC_DATA_TYPE(float, size) \
- out_f32 = input / (VEC_DATA_TYPE(float, size))(scale) + (VEC_DATA_TYPE(float, size))(offset); \
- VEC_DATA_TYPE(type, size) \
- res = CONVERT_SAT(CONVERT_DOWN_RTE(out_f32, VEC_DATA_TYPE(int, size)), VEC_DATA_TYPE(type, size)); \
- return res; \
+#define QUANTIZE_IMPL(type, size) \
+ inline VEC_DATA_TYPE(type, size) \
+ quantize_##type##size(VEC_DATA_TYPE(float, size) input, float offset, float scale) \
+ { \
+ VEC_DATA_TYPE(float, size) \
+ out_f32 = input / (VEC_DATA_TYPE(float, size))(scale) + (VEC_DATA_TYPE(float, size))(offset); \
+ VEC_DATA_TYPE(type, size) \
+ res = CONVERT_SAT(CONVERT_DOWN_RTE(out_f32, VEC_DATA_TYPE(int, size)), VEC_DATA_TYPE(type, size)); \
+ return res; \
}
/** Dequantize a vector of values to floating-point
@@ -101,10 +102,11 @@
*
* @return dequantized values in floating point
*/
-#define DEQUANTIZE_IMPL(type, size) \
- inline VEC_DATA_TYPE(float, size) dequantize_##type##size(VEC_DATA_TYPE(type, size) input, float offset, float scale) \
- { \
- return (CONVERT(input, VEC_DATA_TYPE(float, size)) - offset) * scale; \
+#define DEQUANTIZE_IMPL(type, size) \
+ inline VEC_DATA_TYPE(float, size) \
+ dequantize_##type##size(VEC_DATA_TYPE(type, size) input, float offset, float scale) \
+ { \
+ return (CONVERT(input, VEC_DATA_TYPE(float, size)) - offset) * scale; \
}
/** Correctly-rounded-to-nearest division by a power-of-two.
@@ -113,18 +115,17 @@
*
* @return Correctly-rounded-to-nearest division by a power-of-two.
*/
-#define ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(size) \
- inline VEC_DATA_TYPE(int, size) asymm_rounding_divide_by_POW2_##size(VEC_DATA_TYPE(int, size) x, VEC_DATA_TYPE(int, size) exponent) \
- { \
- const VEC_DATA_TYPE(int, size) \
- zero = (VEC_DATA_TYPE(int, size))0; \
- const VEC_DATA_TYPE(int, size) \
- one = (VEC_DATA_TYPE(int, size))1; \
- VEC_DATA_TYPE(int, size) \
- mask = (one << exponent) - one; \
- VEC_DATA_TYPE(int, size) \
- threshold = (mask >> 1) + select(zero, one, (SELECT_VEC_DATA_TYPE(int, size))(x < 0)); \
- return (x >> exponent) + select(zero, one, (SELECT_VEC_DATA_TYPE(int, size))((x & mask) > threshold)); \
+#define ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(size) \
+ inline VEC_DATA_TYPE(int, size) \
+ asymm_rounding_divide_by_POW2_##size(VEC_DATA_TYPE(int, size) x, VEC_DATA_TYPE(int, size) exponent) \
+ { \
+ const VEC_DATA_TYPE(int, size) zero = (VEC_DATA_TYPE(int, size))0; \
+ const VEC_DATA_TYPE(int, size) one = (VEC_DATA_TYPE(int, size))1; \
+ VEC_DATA_TYPE(int, size) \
+ mask = (one << exponent) - one; \
+ VEC_DATA_TYPE(int, size) \
+ threshold = (mask >> 1) + select(zero, one, (SELECT_VEC_DATA_TYPE(int, size))(x < 0)); \
+ return (x >> exponent) + select(zero, one, (SELECT_VEC_DATA_TYPE(int, size))((x & mask) > threshold)); \
}
/** Product of two numbers, interpreting them as fixed-point values in the interval [-1, 1),
@@ -167,27 +168,29 @@
*
* @return Result in fixed-point format Q0.
*/
-#define ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(size) \
- inline VEC_DATA_TYPE(int, size) asymm_exp_on_interval_between_negative_one_quarter_and_0_excl##size(VEC_DATA_TYPE(int, size) a) \
- { \
- const VEC_DATA_TYPE(int, size) constant_term = 1895147668; \
- const VEC_DATA_TYPE(int, size) constant_1_over_3 = 715827883; \
- const int k_fractional_bits = 31; \
- VEC_DATA_TYPE(int, size) \
- x = a + (1 << (k_fractional_bits - 3)); \
- VEC_DATA_TYPE(int, size) \
- x2 = ASYMM_MULT(x, x, size); \
- VEC_DATA_TYPE(int, size) \
- x3 = ASYMM_MULT(x2, x, size); \
- VEC_DATA_TYPE(int, size) \
- x4 = ASYMM_MULT(x2, x2, size); \
- VEC_DATA_TYPE(int, size) \
- x4_over_4 = ASYMM_ROUNDING_DIVIDE_BY_POW2(x4, 2, size); \
- VEC_DATA_TYPE(int, size) \
- x4_over_24_plus_x3_over_6_plus_x2 = ASYMM_MULT((x4_over_4 + x3), constant_1_over_3, size) + x2; \
- VEC_DATA_TYPE(int, size) \
- x4_over_24_plus_x3_over_6_plus_x2_over_2 = ASYMM_ROUNDING_DIVIDE_BY_POW2(x4_over_24_plus_x3_over_6_plus_x2, 1, size); \
- return constant_term + ASYMM_MULT(constant_term, x + x4_over_24_plus_x3_over_6_plus_x2_over_2, size); \
+#define ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(size) \
+ inline VEC_DATA_TYPE(int, size) \
+ asymm_exp_on_interval_between_negative_one_quarter_and_0_excl##size(VEC_DATA_TYPE(int, size) a) \
+ { \
+ const VEC_DATA_TYPE(int, size) constant_term = 1895147668; \
+ const VEC_DATA_TYPE(int, size) constant_1_over_3 = 715827883; \
+ const int k_fractional_bits = 31; \
+ VEC_DATA_TYPE(int, size) \
+ x = a + (1 << (k_fractional_bits - 3)); \
+ VEC_DATA_TYPE(int, size) \
+ x2 = ASYMM_MULT(x, x, size); \
+ VEC_DATA_TYPE(int, size) \
+ x3 = ASYMM_MULT(x2, x, size); \
+ VEC_DATA_TYPE(int, size) \
+ x4 = ASYMM_MULT(x2, x2, size); \
+ VEC_DATA_TYPE(int, size) \
+ x4_over_4 = ASYMM_ROUNDING_DIVIDE_BY_POW2(x4, 2, size); \
+ VEC_DATA_TYPE(int, size) \
+ x4_over_24_plus_x3_over_6_plus_x2 = ASYMM_MULT((x4_over_4 + x3), constant_1_over_3, size) + x2; \
+ VEC_DATA_TYPE(int, size) \
+ x4_over_24_plus_x3_over_6_plus_x2_over_2 = \
+ ASYMM_ROUNDING_DIVIDE_BY_POW2(x4_over_24_plus_x3_over_6_plus_x2, 1, size); \
+ return constant_term + ASYMM_MULT(constant_term, x + x4_over_24_plus_x3_over_6_plus_x2_over_2, size); \
}
/** Each bit of the result is set to the corresponding bit of either then_val or
@@ -198,10 +201,11 @@
*
* @returns Result contaning bits from @p then_val or from @p else_val depending on corresponding bit in @p if_mask is set or not.
*/
-#define ASYMM_SELECT_USING_MASK_IMPL(size) \
- inline VEC_DATA_TYPE(int, size) asymm_select_using_mask##size(VEC_DATA_TYPE(int, size) if_mask, VEC_DATA_TYPE(int, size) then_val, VEC_DATA_TYPE(int, size) else_val) \
- { \
- return (if_mask & then_val) ^ (~if_mask & else_val); \
+#define ASYMM_SELECT_USING_MASK_IMPL(size) \
+ inline VEC_DATA_TYPE(int, size) asymm_select_using_mask##size( \
+ VEC_DATA_TYPE(int, size) if_mask, VEC_DATA_TYPE(int, size) then_val, VEC_DATA_TYPE(int, size) else_val) \
+ { \
+ return (if_mask & then_val) ^ (~if_mask & else_val); \
}
/** For each element of input vector, the corresponding bits of the result item are set
@@ -234,18 +238,19 @@
return select(all_zeros, all_ones, (SELECT_VEC_DATA_TYPE(int, size))(a != 0)); \
}
-#define EXP_BARREL_SHIFTER_IMPL(size) \
- inline VEC_DATA_TYPE(int, size) exp_barrel_shifter##size(VEC_DATA_TYPE(int, size) result, int exponent, int fp_multiplier, int k_integer_bits, int k_fractional_bits, VEC_DATA_TYPE(int, size) remainder) \
- { \
- if(k_integer_bits > exponent) \
- { \
- const int k_shift_amount = k_integer_bits > exponent ? k_fractional_bits + exponent : 0; \
- return ASYMM_SELECT_USING_MASK( \
- ASYMM_MASK_IF_NON_ZERO(remainder & (1 << k_shift_amount), size), \
- ASYMM_MULT(result, fp_multiplier, size), result, size); \
- } \
- \
- return result; \
+#define EXP_BARREL_SHIFTER_IMPL(size) \
+ inline VEC_DATA_TYPE(int, size) \
+ exp_barrel_shifter##size(VEC_DATA_TYPE(int, size) result, int exponent, int fp_multiplier, int k_integer_bits, \
+ int k_fractional_bits, VEC_DATA_TYPE(int, size) remainder) \
+ { \
+ if (k_integer_bits > exponent) \
+ { \
+ const int k_shift_amount = k_integer_bits > exponent ? k_fractional_bits + exponent : 0; \
+ return ASYMM_SELECT_USING_MASK(ASYMM_MASK_IF_NON_ZERO(remainder & (1 << k_shift_amount), size), \
+ ASYMM_MULT(result, fp_multiplier, size), result, size); \
+ } \
+ \
+ return result; \
}
/** Calculates \f$ exp(x) \f$ for x < 0.
@@ -254,39 +259,40 @@
*
* @return Result in fixed-point format Q0.
*/
-#define ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(size) \
- inline VEC_DATA_TYPE(int, size) asymm_exp_on_negative_values##size(VEC_DATA_TYPE(int, size) a, int k_integer_bits) \
- { \
- const int k_fractional_bits = 31 - k_integer_bits; \
- VEC_DATA_TYPE(int, size) \
- k_one_quarter = 1 << (k_fractional_bits - 2); \
- VEC_DATA_TYPE(int, size) \
- mask = k_one_quarter - 1; \
- VEC_DATA_TYPE(int, size) \
- a_mod_quarter_minus_one_quarter = (a & mask) - k_one_quarter; \
- VEC_DATA_TYPE(int, size) \
- a_mod_quarter_minus_one_quarter_scaled = a_mod_quarter_minus_one_quarter << k_integer_bits; \
- VEC_DATA_TYPE(int, size) \
- result = ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL(a_mod_quarter_minus_one_quarter_scaled, size); \
- VEC_DATA_TYPE(int, size) \
- remainder = a_mod_quarter_minus_one_quarter - a; \
- \
- result = EXP_BARREL_SHIFTER(result, -2, 1672461947, k_integer_bits, k_fractional_bits, remainder, size); \
- result = EXP_BARREL_SHIFTER(result, -1, 1302514674, k_integer_bits, k_fractional_bits, remainder, size); \
- result = EXP_BARREL_SHIFTER(result, +0, 790015084, k_integer_bits, k_fractional_bits, remainder, size); \
- result = EXP_BARREL_SHIFTER(result, +1, 290630308, k_integer_bits, k_fractional_bits, remainder, size); \
- result = EXP_BARREL_SHIFTER(result, +2, 39332535, k_integer_bits, k_fractional_bits, remainder, size); \
- result = EXP_BARREL_SHIFTER(result, +3, 720401, k_integer_bits, k_fractional_bits, remainder, size); \
- result = EXP_BARREL_SHIFTER(result, +4, 242, k_integer_bits, k_fractional_bits, remainder, size); \
- \
- if(k_integer_bits > 5) \
- { \
- const VEC_DATA_TYPE(int, size) clamp = -(1 << (k_fractional_bits + 5)); \
- result = ASYMM_SELECT_USING_MASK(ASYMM_MASK_IF_NON_ZERO(a < clamp, size), 0, result, size); \
- } \
- \
- const VEC_DATA_TYPE(int, size) Q0_one = INT_MAX; \
- return ASYMM_SELECT_USING_MASK(ASYMM_MASK_IF_ZERO(a, size), Q0_one, result, size); \
+#define ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(size) \
+ inline VEC_DATA_TYPE(int, size) asymm_exp_on_negative_values##size(VEC_DATA_TYPE(int, size) a, int k_integer_bits) \
+ { \
+ const int k_fractional_bits = 31 - k_integer_bits; \
+ VEC_DATA_TYPE(int, size) \
+ k_one_quarter = 1 << (k_fractional_bits - 2); \
+ VEC_DATA_TYPE(int, size) \
+ mask = k_one_quarter - 1; \
+ VEC_DATA_TYPE(int, size) \
+ a_mod_quarter_minus_one_quarter = (a & mask) - k_one_quarter; \
+ VEC_DATA_TYPE(int, size) \
+ a_mod_quarter_minus_one_quarter_scaled = a_mod_quarter_minus_one_quarter << k_integer_bits; \
+ VEC_DATA_TYPE(int, size) \
+ result = ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL(a_mod_quarter_minus_one_quarter_scaled, \
+ size); \
+ VEC_DATA_TYPE(int, size) \
+ remainder = a_mod_quarter_minus_one_quarter - a; \
+ \
+ result = EXP_BARREL_SHIFTER(result, -2, 1672461947, k_integer_bits, k_fractional_bits, remainder, size); \
+ result = EXP_BARREL_SHIFTER(result, -1, 1302514674, k_integer_bits, k_fractional_bits, remainder, size); \
+ result = EXP_BARREL_SHIFTER(result, +0, 790015084, k_integer_bits, k_fractional_bits, remainder, size); \
+ result = EXP_BARREL_SHIFTER(result, +1, 290630308, k_integer_bits, k_fractional_bits, remainder, size); \
+ result = EXP_BARREL_SHIFTER(result, +2, 39332535, k_integer_bits, k_fractional_bits, remainder, size); \
+ result = EXP_BARREL_SHIFTER(result, +3, 720401, k_integer_bits, k_fractional_bits, remainder, size); \
+ result = EXP_BARREL_SHIFTER(result, +4, 242, k_integer_bits, k_fractional_bits, remainder, size); \
+ \
+ if (k_integer_bits > 5) \
+ { \
+ const VEC_DATA_TYPE(int, size) clamp = -(1 << (k_fractional_bits + 5)); \
+ result = ASYMM_SELECT_USING_MASK(ASYMM_MASK_IF_NON_ZERO(a < clamp, size), 0, result, size); \
+ } \
+ \
+ const VEC_DATA_TYPE(int, size) Q0_one = INT_MAX; \
+ return ASYMM_SELECT_USING_MASK(ASYMM_MASK_IF_ZERO(a, size), Q0_one, result, size); \
}
/** Calculates the product of a integer value by a power of two, with either a positive exponent
@@ -297,26 +303,27 @@
*
* @return Arithmetic left or right shift.
*/
-#define ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(size) \
- inline VEC_DATA_TYPE(int, size) asymm_saturating_rounding_mult_by_pow2##size(VEC_DATA_TYPE(int, size) x, int exponent) \
- { \
- if(exponent < 0) \
- { \
- return ASYMM_ROUNDING_DIVIDE_BY_POW2(x, -exponent, size); \
- } \
- \
- const VEC_DATA_TYPE(int, size) min = INT_MIN; \
- const VEC_DATA_TYPE(int, size) max = INT_MAX; \
- int threshold = ((1 << (31 - exponent)) - 1); \
- VEC_DATA_TYPE(int, size) \
- positive_mask = ASYMM_MASK_IF_NON_ZERO(x > threshold, size); \
- VEC_DATA_TYPE(int, size) \
- negative_mask = ASYMM_MASK_IF_NON_ZERO(x < -threshold, size); \
- VEC_DATA_TYPE(int, size) \
- result = x << exponent; \
- result = ASYMM_SELECT_USING_MASK(positive_mask, max, result, size); \
- result = ASYMM_SELECT_USING_MASK(negative_mask, min, result, size); \
- return result; \
+#define ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(size) \
+ inline VEC_DATA_TYPE(int, size) \
+ asymm_saturating_rounding_mult_by_pow2##size(VEC_DATA_TYPE(int, size) x, int exponent) \
+ { \
+ if (exponent < 0) \
+ { \
+ return ASYMM_ROUNDING_DIVIDE_BY_POW2(x, -exponent, size); \
+ } \
+ \
+ const VEC_DATA_TYPE(int, size) min = INT_MIN; \
+ const VEC_DATA_TYPE(int, size) max = INT_MAX; \
+ int threshold = ((1 << (31 - exponent)) - 1); \
+ VEC_DATA_TYPE(int, size) \
+ positive_mask = ASYMM_MASK_IF_NON_ZERO(x > threshold, size); \
+ VEC_DATA_TYPE(int, size) \
+ negative_mask = ASYMM_MASK_IF_NON_ZERO(x < -threshold, size); \
+ VEC_DATA_TYPE(int, size) \
+ result = x << exponent; \
+ result = ASYMM_SELECT_USING_MASK(positive_mask, max, result, size); \
+ result = ASYMM_SELECT_USING_MASK(negative_mask, min, result, size); \
+ return result; \
}
/** Calculates (a+b)/2, rounded to the nearest integer.
@@ -326,20 +333,21 @@
*
* @return (a+b)/2, rounded to the nearest integer.
*/
-#define ASYMM_ROUNDING_HALF_SUM_IMPL(size) \
- inline VEC_DATA_TYPE(int, size) asymm_rounding_half_sum##size(VEC_DATA_TYPE(int, size) a, VEC_DATA_TYPE(int, size) b) \
- { \
- VEC_DATA_TYPE(long, size) \
- a64 = convert_long##size(a); \
- VEC_DATA_TYPE(long, size) \
- b64 = convert_long##size(b); \
- VEC_DATA_TYPE(long, size) \
- sum = a64 + b64; \
- const VEC_DATA_TYPE(long, size) one = 1; \
- const VEC_DATA_TYPE(long, size) minus_one = -1; \
- VEC_DATA_TYPE(long, size) \
- sign = select(minus_one, one, (SELECT_VEC_DATA_TYPE(long, size))(sum >= 0)); \
- return convert_int##size((sum + sign) / 2); \
+#define ASYMM_ROUNDING_HALF_SUM_IMPL(size) \
+ inline VEC_DATA_TYPE(int, size) \
+ asymm_rounding_half_sum##size(VEC_DATA_TYPE(int, size) a, VEC_DATA_TYPE(int, size) b) \
+ { \
+ VEC_DATA_TYPE(long, size) \
+ a64 = convert_long##size(a); \
+ VEC_DATA_TYPE(long, size) \
+ b64 = convert_long##size(b); \
+ VEC_DATA_TYPE(long, size) \
+ sum = a64 + b64; \
+ const VEC_DATA_TYPE(long, size) one = 1; \
+ const VEC_DATA_TYPE(long, size) minus_one = -1; \
+ VEC_DATA_TYPE(long, size) \
+ sign = select(minus_one, one, (SELECT_VEC_DATA_TYPE(long, size))(sum >= 0)); \
+ return convert_int##size((sum + sign) / 2); \
}
/** Calculates \f$ 1 / (1 + x) \f$ for x in (0, 1).
@@ -354,12 +362,12 @@
const VEC_DATA_TYPE(int, size) Q0_one = INT_MAX; \
const VEC_DATA_TYPE(int, size) Q2_one = 1 << (31 - 2); \
VEC_DATA_TYPE(int, size) \
- half_denominator = ASYMM_ROUNDING_HALF_SUM(a, Q0_one, size); \
+ half_denominator = ASYMM_ROUNDING_HALF_SUM(a, Q0_one, size); \
const VEC_DATA_TYPE(int, size) Q2_48_over_17 = 1515870810; \
const VEC_DATA_TYPE(int, size) Q2_neg_32_over_17 = -1010580540; \
VEC_DATA_TYPE(int, size) \
x = Q2_48_over_17 + ASYMM_MULT(half_denominator, Q2_neg_32_over_17, size); \
- for(int i = 0; i < 3; i++) \
+ for (int i = 0; i < 3; i++) \
{ \
VEC_DATA_TYPE(int, size) \
half_denominator_times_x = ASYMM_MULT(half_denominator, x, size); \
@@ -378,48 +386,57 @@
*
* @return Rescaled value.
*/
-#define ASYMM_RESCALE_IMPL(size) \
- inline VEC_DATA_TYPE(int, size) asymm_rescale##size(VEC_DATA_TYPE(int, size) value, int src_integer_bits, int dst_integer_bits) \
- { \
- int exponent = src_integer_bits - dst_integer_bits; \
- return ASYMM_SATURATING_ROUNDING_MULT_BY_POW2(value, exponent, size); \
+#define ASYMM_RESCALE_IMPL(size) \
+ inline VEC_DATA_TYPE(int, size) \
+ asymm_rescale##size(VEC_DATA_TYPE(int, size) value, int src_integer_bits, int dst_integer_bits) \
+ { \
+ int exponent = src_integer_bits - dst_integer_bits; \
+ return ASYMM_SATURATING_ROUNDING_MULT_BY_POW2(value, exponent, size); \
}
-#define QUANTIZE_STR(input, offset, scale, type, size) quantize_##type##size(input, offset, scale)
-#define QUANTIZE(input, offset, scale, type, size) QUANTIZE_STR(input, offset, scale, type, size)
+#define QUANTIZE_STR(input, offset, scale, type, size) quantize_##type##size(input, offset, scale)
+#define QUANTIZE(input, offset, scale, type, size) QUANTIZE_STR(input, offset, scale, type, size)
#define DEQUANTIZE_STR(input, offset, scale, type, size) dequantize_##type##size(input, offset, scale)
-#define DEQUANTIZE(input, offset, scale, type, size) DEQUANTIZE_STR(input, offset, scale, type, size)
+#define DEQUANTIZE(input, offset, scale, type, size) DEQUANTIZE_STR(input, offset, scale, type, size)
#define ASYMM_ROUNDING_DIVIDE_BY_POW2_STR(x, exponent, size) asymm_rounding_divide_by_POW2_##size(x, exponent)
-#define ASYMM_ROUNDING_DIVIDE_BY_POW2(x, exponent, size) ASYMM_ROUNDING_DIVIDE_BY_POW2_STR(x, exponent, size)
-#define ASYMM_MULT_STR(a, b, size) asymm_mult##size(a, b)
-#define ASYMM_MULT(a, b, size) ASYMM_MULT_STR(a, b, size)
+#define ASYMM_ROUNDING_DIVIDE_BY_POW2(x, exponent, size) ASYMM_ROUNDING_DIVIDE_BY_POW2_STR(x, exponent, size)
+#define ASYMM_MULT_STR(a, b, size) asymm_mult##size(a, b)
+#define ASYMM_MULT(a, b, size) ASYMM_MULT_STR(a, b, size)
#define ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(x, quantized_multiplier, left_shift, size) \
ASYMM_MULT(x *((VEC_DATA_TYPE(int, size))(1) << (-left_shift)), quantized_multiplier, size)
#define ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(x, quantized_multiplier, right_shift, size) \
ASYMM_ROUNDING_DIVIDE_BY_POW2(ASYMM_MULT(x, quantized_multiplier, size), right_shift, size)
-#define ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL(a, size) asymm_exp_on_interval_between_negative_one_quarter_and_0_excl##size(a)
-#define ASYMM_SELECT_USING_MASK(if_mask, then_val, else_val, size) asymm_select_using_mask##size(if_mask, then_val, else_val)
-#define ASYMM_MASK_IF_ZERO(a, size) asymm_mask_if_zero##size(a)
+#define ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL(a, size) \
+ asymm_exp_on_interval_between_negative_one_quarter_and_0_excl##size(a)
+#define ASYMM_SELECT_USING_MASK(if_mask, then_val, else_val, size) \
+ asymm_select_using_mask##size(if_mask, then_val, else_val)
+#define ASYMM_MASK_IF_ZERO(a, size) asymm_mask_if_zero##size(a)
#define ASYMM_MASK_IF_NON_ZERO(a, size) asymm_mask_if_non_zero##size(a)
-#define EXP_BARREL_SHIFTER(result, exponent, fp_multiplier, k_integer_bits, k_fractional_bits, remainder, size) exp_barrel_shifter##size(result, exponent, fp_multiplier, k_integer_bits, k_fractional_bits, remainder)
+#define EXP_BARREL_SHIFTER(result, exponent, fp_multiplier, k_integer_bits, k_fractional_bits, remainder, size) \
+ exp_barrel_shifter##size(result, exponent, fp_multiplier, k_integer_bits, k_fractional_bits, remainder)
#define ASYMM_EXP_ON_NEGATIVE_VALUES_STR(a, k_integer_bits, size) asymm_exp_on_negative_values##size(a, k_integer_bits)
-#define ASYMM_EXP_ON_NEGATIVE_VALUES(a, k_integer_bits, size) ASYMM_EXP_ON_NEGATIVE_VALUES_STR(a, k_integer_bits, size)
-#define ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_STR(a, size) asymm_one_over_one_plus_x_for_x_in_0_1##size(a)
-#define ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1(a, size) ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_STR(a, size)
-#define ASYMM_SATURATING_ROUNDING_MULT_BY_POW2(x, exponent, size) asymm_saturating_rounding_mult_by_pow2##size(x, exponent)
+#define ASYMM_EXP_ON_NEGATIVE_VALUES(a, k_integer_bits, size) ASYMM_EXP_ON_NEGATIVE_VALUES_STR(a, k_integer_bits, size)
+#define ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_STR(a, size) asymm_one_over_one_plus_x_for_x_in_0_1##size(a)
+#define ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1(a, size) ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_STR(a, size)
+#define ASYMM_SATURATING_ROUNDING_MULT_BY_POW2(x, exponent, size) \
+ asymm_saturating_rounding_mult_by_pow2##size(x, exponent)
#define ASYMM_ROUNDING_HALF_SUM(a, b, size) asymm_rounding_half_sum##size(a, b)
-#define ASYMM_RESCALE_STR(value, src_integer_bits, dst_integer_bits, size) asymm_rescale##size(value, src_integer_bits, dst_integer_bits)
-#define ASYMM_RESCALE(value, src_integer_bits, dst_integer_bits, size) ASYMM_RESCALE_STR(value, src_integer_bits, dst_integer_bits, size)
+#define ASYMM_RESCALE_STR(value, src_integer_bits, dst_integer_bits, size) \
+ asymm_rescale##size(value, src_integer_bits, dst_integer_bits)
+#define ASYMM_RESCALE(value, src_integer_bits, dst_integer_bits, size) \
+ ASYMM_RESCALE_STR(value, src_integer_bits, dst_integer_bits, size)
-#define MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(size) \
- inline VEC_DATA_TYPE(int, size) multiply_by_quantized_multiplier##size(VEC_DATA_TYPE(int, size) input, int qmul, int shift) \
- { \
- const int left_shift = shift > 0 ? shift : 0; \
- const int right_shift = shift > 0 ? 0 : -shift; \
- return ASYMM_ROUNDING_DIVIDE_BY_POW2(ASYMM_MULT(input * (1 << left_shift), qmul, size), right_shift, size); \
+#define MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(size) \
+ inline VEC_DATA_TYPE(int, size) \
+ multiply_by_quantized_multiplier##size(VEC_DATA_TYPE(int, size) input, int qmul, int shift) \
+ { \
+ const int left_shift = shift > 0 ? shift : 0; \
+ const int right_shift = shift > 0 ? 0 : -shift; \
+ return ASYMM_ROUNDING_DIVIDE_BY_POW2(ASYMM_MULT(input * (1 << left_shift), qmul, size), right_shift, size); \
}
-#define MULTIPLY_BY_QUANTIZED_MULTIPLIER(input, qmul, shift, size) multiply_by_quantized_multiplier##size(input, qmul, shift)
+#define MULTIPLY_BY_QUANTIZED_MULTIPLIER(input, qmul, shift, size) \
+ multiply_by_quantized_multiplier##size(input, qmul, shift)
QUANTIZE_IMPL(uchar, 1)
QUANTIZE_IMPL(char, 1)
diff --git a/src/core/CL/cl_kernels/load_store_utility.h b/src/core/CL/cl_kernels/load_store_utility.h
index 4ba2b2c..4daf0ad 100644
--- a/src/core/CL/cl_kernels/load_store_utility.h
+++ b/src/core/CL/cl_kernels/load_store_utility.h
@@ -223,8 +223,10 @@
* @param[in] Z The offset in z-axis direction
* @{
*/
-#define STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
-#define STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+#define STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+#define STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
/** @} */ // end of group STORE_BLOCK
/** Convert and store a block of the given size M0xN0
@@ -245,8 +247,10 @@
* @param[in] Z The offset in z-axis direction
* @{
*/
-#define CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
-#define CONVERT_STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+#define CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+#define CONVERT_STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
/** @} */ // end of group CONVERT_STORE_BLOCK
/** Partially store the 0 to (n-1)th rows of the given variables
@@ -365,8 +369,10 @@
* @param[in] Z The offset in z-axis direction
* @{
*/
-#define STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_PARTIAL_##STORE_M0(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
-#define STORE_BLOCK_PARTIAL(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+#define STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_ROW_PARTIAL_##STORE_M0(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
+#define STORE_BLOCK_PARTIAL(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
+ STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
/** Store a block that can be partial in both x and y dimensions
*
* @note in cases @p PARTIAL_STORE_N0 != 1, 2, 3, 4, 8, 16, extra vstore(s) will be invoked, thus incurring small performance penalty.
@@ -388,22 +394,23 @@
* @param[in] PARTIAL_COND_Y Condition on the y axis to perform the partial store Y. True to use PARTIAL_STORE_M0 rather than M0.
* @param[in] PARTIAL_COND_X Condition on the x axis to perform the partial store X. True to use PARTIAL_STORE_N0 rather than N0.
*/
-#define STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
- if(!(PARTIAL_COND_X) && !(PARTIAL_COND_Y)) \
- { \
- STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
- } \
- else if((PARTIAL_COND_Y) && !(PARTIAL_COND_X)) \
- { \
- STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
- } \
- else if(!(PARTIAL_COND_Y) && (PARTIAL_COND_X)) \
- { \
- STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
- } \
- else \
- { \
- STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
+#define STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, \
+ PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+ if (!(PARTIAL_COND_X) && !(PARTIAL_COND_Y)) \
+ { \
+ STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
+ } \
+ else if ((PARTIAL_COND_Y) && !(PARTIAL_COND_X)) \
+ { \
+ STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
+ } \
+ else if (!(PARTIAL_COND_Y) && (PARTIAL_COND_X)) \
+ { \
+ STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
+ } \
+ else \
+ { \
+ STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
}
/** Store a block that can only be partial in x but not y.
*
@@ -425,7 +432,7 @@
* @param[in] PARTIAL_COND_X Condition on the x axis to perform the partial store X. True to use PARTIAL_STORE_N0 rather than N0.
*/
#define STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X) \
- if(!(PARTIAL_COND_X)) \
+ if (!(PARTIAL_COND_X)) \
{ \
STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
} \
@@ -453,7 +460,7 @@
* @param[in] PARTIAL_COND_Y Condition on the y axis to perform the partial store Y. True to use PARTIAL_STORE_M0 rather than M0.
*/
#define STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y) \
- if(!(PARTIAL_COND_Y)) \
+ if (!(PARTIAL_COND_Y)) \
{ \
STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z); \
} \
@@ -517,23 +524,28 @@
#if defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0)
#if PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
// Case1: No partial blocks in either x or y
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, \
+ PARTIAL_COND_Y, PARTIAL_COND_X) \
STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#elif PARTIAL_STORE_M0 > 0 && PARTIAL_STORE_N0 == 0
// Case2: Partial blocks in y
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, \
+ PARTIAL_COND_Y, PARTIAL_COND_X) \
STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y)
#elif PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 > 0
// Case3: Partial blocks in x
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, \
+ PARTIAL_COND_Y, PARTIAL_COND_X) \
STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X)
#else // PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
// Case4: Partial blocks in both x and y
-#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
- STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X)
+#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, \
+ PARTIAL_COND_Y, PARTIAL_COND_X) \
+ STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, \
+ PARTIAL_COND_Y, PARTIAL_COND_X)
#endif // PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
@@ -560,8 +572,7 @@
#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \
((uint)(max(0, (int)(y * M0) - (int)((M0 - PARTIAL_STORE_M0) % M0))))
#else // defined(PARTIAL_STORE_M0)
-#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \
- ((uint)(y * M0))
+#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) ((uint)(y * M0))
#endif // defined(PARTIAL_STORE_M0)
/** @} */ // end of group COMPUTE_M0_START_ROW
diff --git a/src/core/CL/cl_kernels/repeat.h b/src/core/CL/cl_kernels/repeat.h
index bed94a7..cb2f4b0 100644
--- a/src/core/CL/cl_kernels/repeat.h
+++ b/src/core/CL/cl_kernels/repeat.h
@@ -75,7 +75,9 @@
P_X##_DEF(F, P_A, P_B, P_C); \
REPEAT_3_15(P_X, P_A, P_B, P_C)
-#define REPEAT_DEF_3_N(P_NUM, P_OP, P_A, P_B, P_C) REPEAT_3_##P_NUM(P_OP, P_A, P_B, P_C) //One level of indirection to ensure order of expansion does not affect preprocessing P_NUM
+#define REPEAT_DEF_3_N(P_NUM, P_OP, P_A, P_B, P_C) \
+ REPEAT_3_##P_NUM(P_OP, P_A, P_B, \
+ P_C) //One level of indirection to ensure order of expansion does not affect preprocessing P_NUM
#define REPEAT_3_N(P_NUM, P_OP, P_A, P_B, P_C) REPEAT_DEF_3_N(P_NUM, P_OP, P_A, P_B, P_C)
// Repeat macros with 4 param, excluding the implicit ID param
@@ -126,52 +128,59 @@
P_X##_DEF(F, P_A, P_B, P_C, P_D); \
REPEAT_4_15(P_X, P_A, P_B, P_C, P_D)
-#define REPEAT_DEF_4_N(P_NUM, P_OP, P_A, P_B, P_C, P_D) REPEAT_4_##P_NUM(P_OP, P_A, P_B, P_C, P_D) //One level of indirection to ensure order of expansion does not affect preprocessing P_NUM
+#define REPEAT_DEF_4_N(P_NUM, P_OP, P_A, P_B, P_C, P_D) \
+ REPEAT_4_##P_NUM(P_OP, P_A, P_B, P_C, \
+ P_D) //One level of indirection to ensure order of expansion does not affect preprocessing P_NUM
#define REPEAT_4_N(P_NUM, P_OP, P_A, P_B, P_C, P_D) REPEAT_DEF_4_N(P_NUM, P_OP, P_A, P_B, P_C, P_D)
// Macro for initializing N variables. Generates N statements that defines VAR##N = RHS_ACCESSOR_DEF(...)
-#define VAR_INIT_TO_CONST_DEF(ID, TYPE, VAR, VAL) TYPE VAR##ID = VAL
+#define VAR_INIT_TO_CONST_DEF(ID, TYPE, VAR, VAL) TYPE VAR##ID = VAL
#define REPEAT_VAR_INIT_TO_CONST(N, TYPE, VAR, VAL) REPEAT_3_N(N, VAR_INIT_TO_CONST, TYPE, VAR, VAL)
// Macro for initializing N variables by converting the data type. Generates N statements that defines VAR##N = RHS_ACCESSOR_DEF(...)
-#define VAR_INIT_CONVERT_DEF(ID, TYPE_OUT, VAR_IN, VAR_OUT) TYPE_OUT VAR_OUT##ID = CONVERT(VAR_IN##ID, TYPE_OUT)
+#define VAR_INIT_CONVERT_DEF(ID, TYPE_OUT, VAR_IN, VAR_OUT) TYPE_OUT VAR_OUT##ID = CONVERT(VAR_IN##ID, TYPE_OUT)
#define REPEAT_VAR_INIT_CONVERT(N, TYPE_OUT, VAR_IN, VAR_OUT) REPEAT_3_N(N, VAR_INIT_CONVERT, TYPE_OUT, VAR_IN, VAR_OUT)
// Macro for initializing N variables by converting the data type with saturation. Generates N statements that defines VAR##N = RHS_ACCESSOR_DEF(...)
#define VAR_INIT_CONVERT_SAT_DEF(ID, TYPE_OUT, VAR_IN, VAR_OUT) TYPE_OUT VAR_OUT##ID = CONVERT_SAT(VAR_IN##ID, TYPE_OUT)
-#define REPEAT_VAR_INIT_CONVERT_SAT(N, TYPE_OUT, VAR_IN, VAR_OUT) REPEAT_3_N(N, VAR_INIT_CONVERT_SAT, TYPE_OUT, VAR_IN, VAR_OUT)
+#define REPEAT_VAR_INIT_CONVERT_SAT(N, TYPE_OUT, VAR_IN, VAR_OUT) \
+ REPEAT_3_N(N, VAR_INIT_CONVERT_SAT, TYPE_OUT, VAR_IN, VAR_OUT)
// Macro for adding a constant to N variables. Generates N statements that defines VAR##N =RHS_ACCESSOR_DEF(...)
-#define ADD_CONST_TO_VAR_DEF(ID, TYPE, VAR, VAL) VAR##ID += (TYPE)VAL
+#define ADD_CONST_TO_VAR_DEF(ID, TYPE, VAR, VAL) VAR##ID += (TYPE)VAL
#define REPEAT_ADD_CONST_TO_VAR(N, TYPE, VAR, VAL) REPEAT_3_N(N, ADD_CONST_TO_VAR, TYPE, VAR, VAL)
// Macro for multiplying N variables (VAR_B) by a constant (VAL) and adding to other N variables (VAR_A). Generates N statements that defines VAR_A##N =RHS_ACCESSOR_DEF(...)
-#define MLA_VAR_WITH_CONST_VEC_DEF(ID, VAR_A, VAR_B, VAL) VAR_A##ID += VAR_B##ID * VAL
+#define MLA_VAR_WITH_CONST_VEC_DEF(ID, VAR_A, VAR_B, VAL) VAR_A##ID += VAR_B##ID * VAL
#define REPEAT_MLA_VAR_WITH_CONST_VEC(N, VAR_A, VAR_B, VAL) REPEAT_3_N(N, MLA_VAR_WITH_CONST_VEC, VAR_A, VAR_B, VAL)
// Macro for adding a vector to N-variables. Generates N statements that defines VAR##N =RHS_ACCESSOR_DEF(...)
#define ADD_VECTOR_TO_VAR_DEF(ID, TYPE, VAR, VEC) VAR##ID += VEC
-#define REPEAT_ADD_VECTOR_TO_VAR(N, VAR, VEC) REPEAT_3_N(N, ADD_VECTOR_TO_VAR, "", VAR, VEC)
+#define REPEAT_ADD_VECTOR_TO_VAR(N, VAR, VEC) REPEAT_3_N(N, ADD_VECTOR_TO_VAR, "", VAR, VEC)
// Macro for adding a two N-variables. Generates N statements that defines VAR##N =RHS_ACCESSOR_DEF(...)
#define ADD_TWO_VARS_DEF(ID, TYPE, VAR_A, VAR_B) VAR_A##ID += VAR_B##ID
-#define REPEAT_ADD_TWO_VARS(N, VAR_A, VAR_B) REPEAT_3_N(N, ADD_TWO_VARS, "", VAR_A, VAR_B)
+#define REPEAT_ADD_TWO_VARS(N, VAR_A, VAR_B) REPEAT_3_N(N, ADD_TWO_VARS, "", VAR_A, VAR_B)
// Macro for performing Max between a constant and N variables. Generates N statements that defines VAR##N =RHS_ACCESSOR_DEF(...)
-#define MAX_CONST_VAR_DEF(ID, TYPE, VAR, VAL) VAR##ID = max(VAR##ID, (TYPE)VAL)
+#define MAX_CONST_VAR_DEF(ID, TYPE, VAR, VAL) VAR##ID = max(VAR##ID, (TYPE)VAL)
#define REPEAT_MAX_CONST_VAR(N, TYPE, VAR, VAL) REPEAT_3_N(N, MAX_CONST_VAR, TYPE, VAR, VAL)
// Macro for performing Min between a constant and N variables. Generates N statements that defines VAR##N =RHS_ACCESSOR_DEF(...)
-#define MIN_CONST_VAR_DEF(ID, TYPE, VAR, VAL) VAR##ID = min(VAR##ID, (TYPE)VAL)
+#define MIN_CONST_VAR_DEF(ID, TYPE, VAR, VAL) VAR##ID = min(VAR##ID, (TYPE)VAL)
#define REPEAT_MIN_CONST_VAR(N, TYPE, VAR, VAL) REPEAT_3_N(N, MIN_CONST_VAR, TYPE, VAR, VAL)
// Macro for performing ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE to N variables. Generates N statements that defines VAR##N =RHS_ACCESSOR_DEF(...)
-#define ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE_DEF(ID, SIZE, VAR, RES_MUL, RES_SHIFT) VAR##ID = ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(VAR##ID, RES_MUL, RES_SHIFT, SIZE)
-#define REPEAT_ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(N, SIZE, VAR, RES_MUL, RES_SHIFT) REPEAT_4_N(N, ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE, SIZE, VAR, RES_MUL, RES_SHIFT)
+#define ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE_DEF(ID, SIZE, VAR, RES_MUL, RES_SHIFT) \
+ VAR##ID = ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(VAR##ID, RES_MUL, RES_SHIFT, SIZE)
+#define REPEAT_ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(N, SIZE, VAR, RES_MUL, RES_SHIFT) \
+ REPEAT_4_N(N, ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE, SIZE, VAR, RES_MUL, RES_SHIFT)
// Macro for performing ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE to N variables. Generates N statements that defines VAR##N =RHS_ACCESSOR_DEF(...)
-#define ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE_DEF(ID, SIZE, VAR, RES_MUL, RES_SHIFT) VAR##ID = ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(VAR##ID, RES_MUL, RES_SHIFT, SIZE)
-#define REPEAT_ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(N, SIZE, VAR, RES_MUL, RES_SHIFT) REPEAT_4_N(N, ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE, SIZE, VAR, RES_MUL, RES_SHIFT)
+#define ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE_DEF(ID, SIZE, VAR, RES_MUL, RES_SHIFT) \
+ VAR##ID = ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(VAR##ID, RES_MUL, RES_SHIFT, SIZE)
+#define REPEAT_ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(N, SIZE, VAR, RES_MUL, RES_SHIFT) \
+ REPEAT_4_N(N, ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE, SIZE, VAR, RES_MUL, RES_SHIFT)
// Macro for performing per-channel ASYMM_MULT_BY_QUANT_MULTIPLIER to N variables.
#define ASYMM_MULT_BY_QUANT_MULTIPLIER_PER_CHANNEL_DEF(ID, SIZE, VAR, RES_MUL, RES_SHIFT) \
@@ -182,6 +191,7 @@
VAR##ID_shift_gt0 = ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(VAR##ID, RES_MUL, RES_SHIFT, N0); \
VAR##ID = select(VAR##ID_shift_lt0, VAR##ID_shift_gt0, RES_SHIFT >= 0); \
})
-#define REPEAT_ASYMM_MULT_BY_QUANT_MULTIPLIER_PER_CHANNEL(N, SIZE, VAR, RES_MUL, RES_SHIFT) REPEAT_4_N(N, ASYMM_MULT_BY_QUANT_MULTIPLIER_PER_CHANNEL, SIZE, VAR, RES_MUL, RES_SHIFT)
+#define REPEAT_ASYMM_MULT_BY_QUANT_MULTIPLIER_PER_CHANNEL(N, SIZE, VAR, RES_MUL, RES_SHIFT) \
+ REPEAT_4_N(N, ASYMM_MULT_BY_QUANT_MULTIPLIER_PER_CHANNEL, SIZE, VAR, RES_MUL, RES_SHIFT)
#endif // ARM_COMPUTE_REPEAT_H
diff --git a/src/core/CL/cl_kernels/warp_helpers.h b/src/core/CL/cl_kernels/warp_helpers.h
index 642483a..6595bd1 100644
--- a/src/core/CL/cl_kernels/warp_helpers.h
+++ b/src/core/CL/cl_kernels/warp_helpers.h
@@ -31,11 +31,13 @@
* @param[in] border_size Border size of the image
*
*/
-inline const float8 clamp_to_border_with_size(float8 coords, const float width, const float height, const float border_size)
+inline const float8
+clamp_to_border_with_size(float8 coords, const float width, const float height, const float border_size)
{
const float4 clamped_x = clamp(coords.even, 0.0f - border_size, width - 1 + border_size);
const float4 clamped_y = clamp(coords.odd, 0.0f - border_size, height - 1 + border_size);
- return (float8)(clamped_x.s0, clamped_y.s0, clamped_x.s1, clamped_y.s1, clamped_x.s2, clamped_y.s2, clamped_x.s3, clamped_y.s3);
+ return (float8)(clamped_x.s0, clamped_y.s0, clamped_x.s1, clamped_y.s1, clamped_x.s2, clamped_y.s2, clamped_x.s3,
+ clamped_y.s3);
}
/** Clamps the given coordinates to the borders.
@@ -74,7 +76,8 @@
*/
inline const float8 get_neighbour_coords(const float2 coord)
{
- return (float8)(/*tl*/ coord.s0, coord.s1, /*tr*/ coord.s0 + 1, coord.s1, /*bl*/ coord.s0, coord.s1 + 1, /*br*/ coord.s0 + 1, coord.s1 + 1);
+ return (float8)(/*tl*/ coord.s0, coord.s1, /*tr*/ coord.s0 + 1, coord.s1, /*bl*/ coord.s0, coord.s1 + 1,
+ /*br*/ coord.s0 + 1, coord.s1 + 1);
}
/** Computes the bilinear interpolation for each set of coordinates in the vector coords and returns the values
@@ -85,37 +88,38 @@
* @param[in] height Height of the image
* @param[in] border_size Border size
*/
-inline const VEC_DATA_TYPE(DATA_TYPE, 4) bilinear_interpolate_with_border(const Image *in, const float8 coords, const float width, const float height, const float border_size)
+inline const VEC_DATA_TYPE(DATA_TYPE, 4) bilinear_interpolate_with_border(
+ const Image *in, const float8 coords, const float width, const float height, const float border_size)
{
// If any of the 4 texels is out of the image's boundaries we use the border value (REPLICATE or CONSTANT) for any texel out of the image.
// Sets the 4x4 coordinates for each of the four input texels
const float8 fc = floor(coords);
- const float16 c1 = (float16)(
- clamp_to_border_with_size(get_neighbour_coords((float2)(fc.s0, fc.s1)), width, height, border_size),
- clamp_to_border_with_size(get_neighbour_coords((float2)(fc.s2, fc.s3)), width, height, border_size));
- const float16 c2 = (float16)(
- clamp_to_border_with_size(get_neighbour_coords((float2)(fc.s4, fc.s5)), width, height, border_size),
- clamp_to_border_with_size(get_neighbour_coords((float2)(fc.s6, fc.s7)), width, height, border_size));
+ const float16 c1 =
+ (float16)(clamp_to_border_with_size(get_neighbour_coords((float2)(fc.s0, fc.s1)), width, height, border_size),
+ clamp_to_border_with_size(get_neighbour_coords((float2)(fc.s2, fc.s3)), width, height, border_size));
+ const float16 c2 =
+ (float16)(clamp_to_border_with_size(get_neighbour_coords((float2)(fc.s4, fc.s5)), width, height, border_size),
+ clamp_to_border_with_size(get_neighbour_coords((float2)(fc.s6, fc.s7)), width, height, border_size));
// Loads the values from the input image
const float16 t = (float16)(
- /* tl, tr, bl, br */
- * ((__global DATA_TYPE *)offset(in, c1.s0, c1.s1)), *((__global DATA_TYPE *)offset(in, c1.s2, c1.s3)),
- *((__global DATA_TYPE *)offset(in, c1.s4, c1.s5)), *((__global DATA_TYPE *)offset(in, c1.s6, c1.s7)),
- *((__global DATA_TYPE *)offset(in, c1.s8, c1.s9)), *((__global DATA_TYPE *)offset(in, c1.sa, c1.sb)),
- *((__global DATA_TYPE *)offset(in, c1.sc, c1.sd)), *((__global DATA_TYPE *)offset(in, c1.se, c1.sf)),
- *((__global DATA_TYPE *)offset(in, c2.s0, c2.s1)), *((__global DATA_TYPE *)offset(in, c2.s2, c2.s3)),
- *((__global DATA_TYPE *)offset(in, c2.s4, c2.s5)), *((__global DATA_TYPE *)offset(in, c2.s6, c2.s7)),
- *((__global DATA_TYPE *)offset(in, c2.s8, c2.s9)), *((__global DATA_TYPE *)offset(in, c2.sa, c2.sb)),
- *((__global DATA_TYPE *)offset(in, c2.sc, c2.sd)), *((__global DATA_TYPE *)offset(in, c2.se, c2.sf)));
- const float8 a = coords - fc;
- const float8 b = ((float8)(1.f)) - a;
- const float4 fr = (float4)(
- ((t.s0 * b.s0 * b.s1) + (t.s1 * a.s0 * b.s1) + (t.s2 * b.s0 * a.s1) + (t.s3 * a.s0 * a.s1)),
- ((t.s4 * b.s2 * b.s3) + (t.s5 * a.s2 * b.s3) + (t.s6 * b.s2 * a.s3) + (t.s7 * a.s2 * a.s3)),
- ((t.s8 * b.s4 * b.s5) + (t.s9 * a.s4 * b.s5) + (t.sa * b.s4 * a.s5) + (t.sb * a.s4 * a.s5)),
- ((t.sc * b.s6 * b.s7) + (t.sd * a.s6 * b.s7) + (t.se * b.s6 * a.s7) + (t.sf * a.s6 * a.s7)));
+ /* tl, tr, bl, br */
+ *((__global DATA_TYPE *)offset(in, c1.s0, c1.s1)), *((__global DATA_TYPE *)offset(in, c1.s2, c1.s3)),
+ *((__global DATA_TYPE *)offset(in, c1.s4, c1.s5)), *((__global DATA_TYPE *)offset(in, c1.s6, c1.s7)),
+ *((__global DATA_TYPE *)offset(in, c1.s8, c1.s9)), *((__global DATA_TYPE *)offset(in, c1.sa, c1.sb)),
+ *((__global DATA_TYPE *)offset(in, c1.sc, c1.sd)), *((__global DATA_TYPE *)offset(in, c1.se, c1.sf)),
+ *((__global DATA_TYPE *)offset(in, c2.s0, c2.s1)), *((__global DATA_TYPE *)offset(in, c2.s2, c2.s3)),
+ *((__global DATA_TYPE *)offset(in, c2.s4, c2.s5)), *((__global DATA_TYPE *)offset(in, c2.s6, c2.s7)),
+ *((__global DATA_TYPE *)offset(in, c2.s8, c2.s9)), *((__global DATA_TYPE *)offset(in, c2.sa, c2.sb)),
+ *((__global DATA_TYPE *)offset(in, c2.sc, c2.sd)), *((__global DATA_TYPE *)offset(in, c2.se, c2.sf)));
+ const float8 a = coords - fc;
+ const float8 b = ((float8)(1.f)) - a;
+ const float4 fr =
+ (float4)(((t.s0 * b.s0 * b.s1) + (t.s1 * a.s0 * b.s1) + (t.s2 * b.s0 * a.s1) + (t.s3 * a.s0 * a.s1)),
+ ((t.s4 * b.s2 * b.s3) + (t.s5 * a.s2 * b.s3) + (t.s6 * b.s2 * a.s3) + (t.s7 * a.s2 * a.s3)),
+ ((t.s8 * b.s4 * b.s5) + (t.s9 * a.s4 * b.s5) + (t.sa * b.s4 * a.s5) + (t.sb * a.s4 * a.s5)),
+ ((t.sc * b.s6 * b.s7) + (t.sd * a.s6 * b.s7) + (t.se * b.s6 * a.s7) + (t.sf * a.s6 * a.s7)));
return CONVERT(fr, VEC_DATA_TYPE(DATA_TYPE, 4));
}
@@ -126,7 +130,8 @@
* @param[in] width Width of the image
* @param[in] height Height of the image
*/
-inline const VEC_DATA_TYPE(DATA_TYPE, 4) bilinear_interpolate(const Image *in, const float8 coords, const float width, const float height)
+inline const VEC_DATA_TYPE(DATA_TYPE, 4)
+ bilinear_interpolate(const Image *in, const float8 coords, const float width, const float height)
{
return bilinear_interpolate_with_border(in, coords, width, height, 1);
}
diff --git a/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp b/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp
index 2728958..5b72354 100644
--- a/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp
+++ b/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp
@@ -31,6 +31,7 @@
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
#include "arm_compute/core/Validate.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -44,16 +45,20 @@
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
- ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::S32, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
+ DataType::S32, DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::S32, DataType::S64);
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(op != ReductionOperation::ARG_IDX_MAX && op != ReductionOperation::ARG_IDX_MIN, "Only ARG_IDX_MAX and ARG_IDX_MIN are supported");
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis >= TensorShape::num_max_dimensions, "Reduction axis greater than max number of dimensions");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(op != ReductionOperation::ARG_IDX_MAX && op != ReductionOperation::ARG_IDX_MIN,
+ "Only ARG_IDX_MAX and ARG_IDX_MIN are supported");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis >= TensorShape::num_max_dimensions,
+ "Reduction axis greater than max number of dimensions");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis > 3, "Unsupported reduction axis");
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
- ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U32, DataType::S32, DataType::S64, DataType::U64);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U32, DataType::S32, DataType::S64,
+ DataType::U64);
}
return Status{};
@@ -66,22 +71,34 @@
_type = CLKernelType::ELEMENTWISE;
}
-void CLArgMinMaxLayerKernel::configure(const ICLTensor *input, ICLTensor *output, unsigned int axis, ReductionOperation op)
+void CLArgMinMaxLayerKernel::configure(const ICLTensor *input,
+ ICLTensor *output,
+ unsigned int axis,
+ ReductionOperation op)
{
configure(CLKernelLibrary::get().get_compile_context(), input, output, axis, op);
}
-void CLArgMinMaxLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, unsigned int axis, ReductionOperation op)
+void CLArgMinMaxLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ unsigned int axis,
+ ReductionOperation op)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- TensorShape output_shape{ input->info()->tensor_shape() };
+ TensorShape output_shape{input->info()->tensor_shape()};
output_shape.set(axis, 1);
- auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape).set_data_type(DataType::S32).reset_padding().set_is_resizable(true));
+ auto_init_if_empty(*output->info(), input->info()
+ ->clone()
+ ->set_tensor_shape(output_shape)
+ .set_data_type(DataType::S32)
+ .reset_padding()
+ .set_is_resizable(true));
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), axis, op));
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
_input = input;
_output = output;
@@ -90,11 +107,14 @@
// Set build options
const auto adjusted_vector_size = adjust_vec_size(16U, input->info()->dimension(0));
- const auto vector_size = (adjusted_vector_size == 3U && axis == 0U) ? 2U : adjusted_vector_size; // the opencl kernel only supports sizes 2, 4, 8 and 16.
+ const auto vector_size = (adjusted_vector_size == 3U && axis == 0U)
+ ? 2U
+ : adjusted_vector_size; // the opencl kernel only supports sizes 2, 4, 8 and 16.
CLBuildOptions build_opts;
build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
- build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(input->info()->dimension(0) % vector_size));
+ build_opts.add_option("-DVEC_SIZE_LEFTOVER=" +
+ support::cpp11::to_string(input->info()->dimension(0) % vector_size));
build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(vector_size));
build_opts.add_option_if(is_data_type_float(input->info()->data_type()), "-DFLOAT_DATA_TYPE");
build_opts.add_option_if_else(op == ReductionOperation::ARG_IDX_MAX, "-DARG_MAX", "-DARG_MIN");
@@ -104,7 +124,7 @@
// Create kernel
std::string kernel_axis_name;
- switch(axis)
+ switch (axis)
{
case 0:
build_opts.add_option("-DWIDTH=" + support::cpp11::to_string(input->info()->dimension(0)));
@@ -135,7 +155,10 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLArgMinMaxLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int axis, ReductionOperation op)
+Status CLArgMinMaxLayerKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ unsigned int axis,
+ ReductionOperation op)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, axis, op));
return Status{};
@@ -146,7 +169,7 @@
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window);
- switch(_reduction_axis)
+ switch (_reduction_axis)
{
case 0:
{
@@ -154,7 +177,8 @@
Window out_window(window);
Window in_window(window);
out_window.set(Window::DimX, Window::Dimension(0, 0, 0));
- in_window.set(Window::DimX, Window::Dimension(0, _input->info()->dimension(0), _input->info()->dimension(0)));
+ in_window.set(Window::DimX,
+ Window::Dimension(0, _input->info()->dimension(0), _input->info()->dimension(0)));
in_window.set(Window::DimY, Window::Dimension(0, _input->info()->dimension(1), 1u));
// Get first input and output slices
@@ -166,15 +190,15 @@
add_2D_tensor_argument(idx, _input, in_slice);
add_2D_tensor_argument(idx, _output, out_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(in_window.slide_window_slice_2D(in_slice) && out_window.slide_window_slice_2D(out_slice));
+ } while (in_window.slide_window_slice_2D(in_slice) && out_window.slide_window_slice_2D(out_slice));
}
break;
case 1:
{
// Get first input and output slices
- Window window_in{ window };
- window_in.set(Window::DimY, Window::Dimension(0, _input->info()->dimension(1), _input->info()->dimension(1)));
+ Window window_in{window};
+ window_in.set(Window::DimY,
+ Window::Dimension(0, _input->info()->dimension(1), _input->info()->dimension(1)));
Window in_slice = window_in.first_slice_window_2D();
Window out_slice = window.first_slice_window_2D();
@@ -184,15 +208,15 @@
add_2D_tensor_argument(idx, _input, in_slice);
add_2D_tensor_argument(idx, _output, out_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(window_in.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(out_slice));
+ } while (window_in.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(out_slice));
}
break;
case 2:
{
// Get first input and output slices
- Window window_in{ window };
- window_in.set(Window::DimZ, Window::Dimension(0, _input->info()->dimension(2), _input->info()->dimension(2)));
+ Window window_in{window};
+ window_in.set(Window::DimZ,
+ Window::Dimension(0, _input->info()->dimension(2), _input->info()->dimension(2)));
Window in_slice = window_in.first_slice_window_3D();
Window out_slice = window.first_slice_window_3D();
@@ -202,14 +226,13 @@
add_3D_tensor_argument(idx, _input, in_slice);
add_3D_tensor_argument(idx, _output, out_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(window_in.slide_window_slice_3D(in_slice) && window.slide_window_slice_3D(out_slice));
+ } while (window_in.slide_window_slice_3D(in_slice) && window.slide_window_slice_3D(out_slice));
}
break;
case 3:
{
// Get first input and output slices
- Window window_in{ window };
+ Window window_in{window};
window_in.set(3, Window::Dimension(0, 1, 1));
Window in_slice = window_in.first_slice_window_4D();
Window out_slice = window.first_slice_window_4D();
@@ -220,8 +243,7 @@
add_4D_tensor_argument(idx, _input, in_slice);
add_4D_tensor_argument(idx, _output, out_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(window_in.slide_window_slice_4D(in_slice) && window.slide_window_slice_4D(out_slice));
+ } while (window_in.slide_window_slice_4D(in_slice) && window.slide_window_slice_4D(out_slice));
}
break;
default:
diff --git a/src/core/CL/kernels/CLArgMinMaxLayerKernel.h b/src/core/CL/kernels/CLArgMinMaxLayerKernel.h
index 5f36bdf..fb3b41b 100644
--- a/src/core/CL/kernels/CLArgMinMaxLayerKernel.h
+++ b/src/core/CL/kernels/CLArgMinMaxLayerKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLARGMINMAXLAYERKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -72,7 +73,11 @@
* @param[in] axis Axis along which to reduce. Supported reduction axis : 0,1,2,3
* @param[in] op Reduction operation to perform. Only ArgMin and ArgMax are supported.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, unsigned int axis, ReductionOperation op);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ unsigned int axis,
+ ReductionOperation op);
/** Static function to check if given info will lead to a valid configuration of @ref CLArgMinMaxLayerKernel.
*
@@ -84,7 +89,8 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int axis, ReductionOperation op);
+ static Status
+ validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int axis, ReductionOperation op);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLBatchNormalizationLayerKernel.cpp b/src/core/CL/kernels/CLBatchNormalizationLayerKernel.cpp
index 3fa8a8e..c88a852 100644
--- a/src/core/CL/kernels/CLBatchNormalizationLayerKernel.cpp
+++ b/src/core/CL/kernels/CLBatchNormalizationLayerKernel.cpp
@@ -23,58 +23,64 @@
*/
#include "src/core/CL/kernels/CLBatchNormalizationLayerKernel.h"
-#include "arm_compute/core/utils/ActivationFunctionUtils.h"
#include "arm_compute/core/CL/CLHelpers.h"
#include "arm_compute/core/CL/CLKernelLibrary.h"
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
+#include "arm_compute/core/utils/ActivationFunctionUtils.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
-
#include "support/StringSupport.h"
using namespace arm_compute;
namespace
{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output,
- const ITensorInfo *mean, const ITensorInfo *var,
- const ITensorInfo *beta, const ITensorInfo *gamma,
- float epsilon, ActivationLayerInfo act_info)
+Status validate_arguments(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const ITensorInfo *mean,
+ const ITensorInfo *var,
+ const ITensorInfo *beta,
+ const ITensorInfo *gamma,
+ float epsilon,
+ ActivationLayerInfo act_info)
{
ARM_COMPUTE_UNUSED(epsilon);
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(mean, var);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, mean, var);
- ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::CHANNEL)) != mean->dimension(0));
- if(beta != nullptr)
+ ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(get_data_layout_dimension_index(
+ input->data_layout(), DataLayoutDimension::CHANNEL)) != mean->dimension(0));
+ if (beta != nullptr)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(mean, beta);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, beta);
}
- if(gamma != nullptr)
+ if (gamma != nullptr)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(mean, gamma);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, gamma);
}
- if(act_info.enabled())
+ if (act_info.enabled())
{
ActivationLayerInfo::ActivationFunction act = act_info.activation();
ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() != DataType::F32 && input->data_type() != DataType::F16);
- ARM_COMPUTE_RETURN_ERROR_ON(act != ActivationLayerInfo::ActivationLayerInfo::ActivationFunction::RELU
- && act != ActivationLayerInfo::ActivationLayerInfo::ActivationFunction::BOUNDED_RELU
- && act != ActivationLayerInfo::ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU);
+ ARM_COMPUTE_RETURN_ERROR_ON(act != ActivationLayerInfo::ActivationLayerInfo::ActivationFunction::RELU &&
+ act != ActivationLayerInfo::ActivationLayerInfo::ActivationFunction::BOUNDED_RELU &&
+ act !=
+ ActivationLayerInfo::ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU);
ARM_COMPUTE_RETURN_ERROR_ON(act_info.b() > act_info.a());
}
- if(output != nullptr && output->total_size() != 0)
+ if (output != nullptr && output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
@@ -86,14 +92,15 @@
std::pair<Status, Window> validate_and_configure_window_nchw(ITensorInfo *input, ITensorInfo *output)
{
- const unsigned int num_elems_processed_per_iteration = adjust_vec_size(16 / input->element_size(), input->dimension(0));
+ const unsigned int num_elems_processed_per_iteration =
+ adjust_vec_size(16 / input->element_size(), input->dimension(0));
// Configure kernel window
Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration));
AccessWindowHorizontal input_access(input, 0, num_elems_processed_per_iteration);
bool window_changed = false;
- if(output != nullptr)
+ if (output != nullptr)
{
AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration);
window_changed = update_window_and_padding(win, input_access, output_access);
@@ -104,30 +111,50 @@
window_changed = update_window_and_padding(win, input_access);
}
- Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+ Status err =
+ (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
return std::make_pair(err, win);
}
} // namespace
CLBatchNormalizationLayerKernel::CLBatchNormalizationLayerKernel()
- : _input(nullptr), _output(nullptr), _mean(nullptr), _var(nullptr), _beta(nullptr), _gamma(nullptr), _epsilon(0), _run_in_place(false)
+ : _input(nullptr),
+ _output(nullptr),
+ _mean(nullptr),
+ _var(nullptr),
+ _beta(nullptr),
+ _gamma(nullptr),
+ _epsilon(0),
+ _run_in_place(false)
{
_type = CLKernelType::ELEMENTWISE;
}
-void CLBatchNormalizationLayerKernel::configure(ICLTensor *input, ICLTensor *output, const ICLTensor *mean, const ICLTensor *var, const ICLTensor *beta, const ICLTensor *gamma,
- float epsilon, ActivationLayerInfo act_info)
+void CLBatchNormalizationLayerKernel::configure(ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *mean,
+ const ICLTensor *var,
+ const ICLTensor *beta,
+ const ICLTensor *gamma,
+ float epsilon,
+ ActivationLayerInfo act_info)
{
configure(CLKernelLibrary::get().get_compile_context(), input, output, mean, var, beta, gamma, epsilon, act_info);
}
-void CLBatchNormalizationLayerKernel::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, const ICLTensor *mean, const ICLTensor *var, const ICLTensor *beta,
- const ICLTensor *gamma,
- float epsilon, ActivationLayerInfo act_info)
+void CLBatchNormalizationLayerKernel::configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *mean,
+ const ICLTensor *var,
+ const ICLTensor *beta,
+ const ICLTensor *gamma,
+ float epsilon,
+ ActivationLayerInfo act_info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, mean, var);
- auto padding_info = get_padding_info({ input, output, mean, var, beta, gamma });
+ auto padding_info = get_padding_info({input, output, mean, var, beta, gamma});
_input = input;
_output = output;
_mean = mean;
@@ -142,13 +169,15 @@
mean->info(), var->info(), (beta != nullptr) ? beta->info() : nullptr,
(gamma != nullptr) ? gamma->info() : nullptr, epsilon, act_info));
- unsigned int num_elems_processed_per_iteration = adjust_vec_size(16 / input->info()->element_size(), input->info()->dimension(0));
+ unsigned int num_elems_processed_per_iteration =
+ adjust_vec_size(16 / input->info()->element_size(), input->info()->dimension(0));
// Set build options
CLBuildOptions build_opts;
build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
- build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(input->info()->dimension(0) % num_elems_processed_per_iteration));
+ build_opts.add_option("-DVEC_SIZE_LEFTOVER=" +
+ support::cpp11::to_string(input->info()->dimension(0) % num_elems_processed_per_iteration));
build_opts.add_option("-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(act_info.activation())));
build_opts.add_option_if(act_info.enabled(), "-DA_VAL=" + float_to_string_with_full_precision(act_info.a()));
build_opts.add_option_if(act_info.enabled(), "-DB_VAL=" + float_to_string_with_full_precision(act_info.b()));
@@ -157,29 +186,33 @@
build_opts.add_option_if(gamma == nullptr, "-DUSE_DEFAULT_GAMMA");
// Create kernel
- _kernel = create_kernel(compile_context, "batchnormalization_layer_" + lower_string(string_from_data_layout(input->info()->data_layout())), build_opts.options());
+ _kernel =
+ create_kernel(compile_context,
+ "batchnormalization_layer_" + lower_string(string_from_data_layout(input->info()->data_layout())),
+ build_opts.options());
// Set kernel static arguments
unsigned int include_output = (!_run_in_place) ? 1 : 0;
- unsigned int idx = (1 + include_output) * num_arguments_per_3D_tensor() + 2 * num_arguments_per_1D_tensor(); // Skip the input and output parameters
- if(_beta != nullptr)
+ unsigned int idx = (1 + include_output) * num_arguments_per_3D_tensor() +
+ 2 * num_arguments_per_1D_tensor(); // Skip the input and output parameters
+ if (_beta != nullptr)
{
idx += num_arguments_per_1D_tensor(); // Skip beta parameter
}
- if(_gamma != nullptr)
+ if (_gamma != nullptr)
{
idx += num_arguments_per_1D_tensor(); // Skip gamma parameter
}
_kernel.setArg<cl_float>(idx++, _epsilon);
- if(output != nullptr)
+ if (output != nullptr)
{
// Output tensor auto initialization if not yet initialized
auto_init_if_empty(*output->info(), *input->info()->clone());
}
// Configure kernel window
- if(input->info()->data_layout() == DataLayout::NHWC)
+ if (input->info()->data_layout() == DataLayout::NHWC)
{
Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration));
ICLKernel::configure_internal(win);
@@ -205,18 +238,23 @@
_config_id += lower_string(string_from_data_layout(input->info()->data_layout()));
}
-Status CLBatchNormalizationLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output,
- const ITensorInfo *mean, const ITensorInfo *var,
- const ITensorInfo *beta, const ITensorInfo *gamma,
- float epsilon, ActivationLayerInfo act_info)
+Status CLBatchNormalizationLayerKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const ITensorInfo *mean,
+ const ITensorInfo *var,
+ const ITensorInfo *beta,
+ const ITensorInfo *gamma,
+ float epsilon,
+ ActivationLayerInfo act_info)
{
const bool run_in_place = (output == nullptr) || (output == input);
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, mean, var, beta, gamma, epsilon, act_info));
- if(input->data_layout() != DataLayout::NHWC)
+ if (input->data_layout() != DataLayout::NHWC)
{
- ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_nchw(input->clone().get(), (run_in_place) ? nullptr : output->clone().get())
- .first);
+ ARM_COMPUTE_RETURN_ON_ERROR(
+ validate_and_configure_window_nchw(input->clone().get(), (run_in_place) ? nullptr : output->clone().get())
+ .first);
}
return Status{};
@@ -236,11 +274,11 @@
unsigned int idx = (1 + include_output) * num_arguments_per_3D_tensor();
add_1D_tensor_argument(idx, _mean, vector_slice);
add_1D_tensor_argument(idx, _var, vector_slice);
- if(_beta != nullptr)
+ if (_beta != nullptr)
{
add_1D_tensor_argument(idx, _beta, vector_slice);
}
- if(_gamma != nullptr)
+ if (_gamma != nullptr)
{
add_1D_tensor_argument(idx, _gamma, vector_slice);
}
@@ -249,11 +287,10 @@
{
idx = 0;
add_3D_tensor_argument(idx, _input, slice);
- if(!_run_in_place)
+ if (!_run_in_place)
{
add_3D_tensor_argument(idx, _output, slice);
}
enqueue(queue, *this, slice, lws_hint());
- }
- while(window.slide_window_slice_3D(slice));
+ } while (window.slide_window_slice_3D(slice));
}
diff --git a/src/core/CL/kernels/CLBatchNormalizationLayerKernel.h b/src/core/CL/kernels/CLBatchNormalizationLayerKernel.h
index acbe0f2..1a88d2a 100644
--- a/src/core/CL/kernels/CLBatchNormalizationLayerKernel.h
+++ b/src/core/CL/kernels/CLBatchNormalizationLayerKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLBATCHNORMALIZATIONLAYERKERNEL_H
#include "arm_compute/function_info/ActivationLayerInfo.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -64,7 +65,13 @@
* @param[in] epsilon (Optional) Small value to avoid division with zero. Default value is 0.001f.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU supported.
*/
- void configure(ICLTensor *input, ICLTensor *output, const ICLTensor *mean, const ICLTensor *var, const ICLTensor *beta = nullptr, const ICLTensor *gamma = nullptr, float epsilon = 0.001f,
+ void configure(ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *mean,
+ const ICLTensor *var,
+ const ICLTensor *beta = nullptr,
+ const ICLTensor *gamma = nullptr,
+ float epsilon = 0.001f,
ActivationLayerInfo act_info = ActivationLayerInfo());
/** Set the input and output tensors.
*
@@ -82,8 +89,15 @@
* @param[in] epsilon (Optional) Small value to avoid division with zero. Default value is 0.001f.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU supported.
*/
- void configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, const ICLTensor *mean, const ICLTensor *var, const ICLTensor *beta = nullptr,
- const ICLTensor *gamma = nullptr, float epsilon = 0.001f, ActivationLayerInfo act_info = ActivationLayerInfo());
+ void configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *mean,
+ const ICLTensor *var,
+ const ICLTensor *beta = nullptr,
+ const ICLTensor *gamma = nullptr,
+ float epsilon = 0.001f,
+ ActivationLayerInfo act_info = ActivationLayerInfo());
/** Static function to check if given info will lead to a valid configuration of @ref CLBatchNormalizationLayerKernel
*
* @param[in] input Source tensor info. In case of @p output tensor info = nullptr, this tensor will store the result.
@@ -99,10 +113,14 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *output,
- const ITensorInfo *mean, const ITensorInfo *var,
- const ITensorInfo *beta = nullptr, const ITensorInfo *gamma = nullptr,
- float epsilon = 0.001f, ActivationLayerInfo act_info = ActivationLayerInfo());
+ static Status validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const ITensorInfo *mean,
+ const ITensorInfo *var,
+ const ITensorInfo *beta = nullptr,
+ const ITensorInfo *gamma = nullptr,
+ float epsilon = 0.001f,
+ ActivationLayerInfo act_info = ActivationLayerInfo());
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLBatchToSpaceLayerKernel.cpp b/src/core/CL/kernels/CLBatchToSpaceLayerKernel.cpp
index 143a842..c640b5a 100644
--- a/src/core/CL/kernels/CLBatchToSpaceLayerKernel.cpp
+++ b/src/core/CL/kernels/CLBatchToSpaceLayerKernel.cpp
@@ -25,13 +25,14 @@
#include "arm_compute/core/CL/CLHelpers.h"
#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
#include "support/StringSupport.h"
-#include "arm_compute/core/TensorInfo.h"
using namespace arm_compute::misc::shape_calculator;
namespace arm_compute
@@ -46,7 +47,7 @@
ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
// Validate output if initialized
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON(output->num_dimensions() > 4);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
@@ -54,7 +55,11 @@
return Status{};
}
-Status validate_arguments_static(const ITensorInfo *input, const int block_shape_x, const int block_shape_y, const ITensorInfo *output, const CropInfo &crop_info)
+Status validate_arguments_static(const ITensorInfo *input,
+ const int block_shape_x,
+ const int block_shape_y,
+ const ITensorInfo *output,
+ const CropInfo &crop_info)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
@@ -66,10 +71,11 @@
ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[idx_batch] % (block_shape_x * block_shape_y) != 0);
// Validate output if initialized
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
- const TensorShape expected_output_shape = compute_batch_to_space_shape(input->data_layout(), input->tensor_shape(), block_shape_x, block_shape_y, crop_info);
- const TensorInfo expected_output = output->clone()->set_tensor_shape(expected_output_shape);
+ const TensorShape expected_output_shape = compute_batch_to_space_shape(
+ input->data_layout(), input->tensor_shape(), block_shape_x, block_shape_y, crop_info);
+ const TensorInfo expected_output = output->clone()->set_tensor_shape(expected_output_shape);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, &expected_output);
ARM_COMPUTE_RETURN_ERROR_ON(output->num_dimensions() > 4);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
@@ -79,8 +85,7 @@
}
} // namespace
-CLBatchToSpaceLayerKernel::CLBatchToSpaceLayerKernel()
- : _input(nullptr), _block_shape(nullptr), _output(nullptr)
+CLBatchToSpaceLayerKernel::CLBatchToSpaceLayerKernel() : _input(nullptr), _block_shape(nullptr), _output(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
@@ -90,11 +95,14 @@
configure(CLKernelLibrary::get().get_compile_context(), input, block_shape, output);
}
-void CLBatchToSpaceLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *block_shape, ICLTensor *output)
+void CLBatchToSpaceLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *block_shape,
+ ICLTensor *output)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- auto padding_info = get_padding_info({ input, block_shape, output });
+ auto padding_info = get_padding_info({input, block_shape, output});
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), block_shape->info(), output->info()));
@@ -106,8 +114,9 @@
CLBuildOptions build_opts;
build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(output->info()->data_type()));
build_opts.add_option("-DBATCH_SIZE=" + support::cpp11::to_string(output->info()->dimension(3)));
- _kernel = create_kernel(compile_context, "batch_to_space_" + lower_string(string_from_data_layout(input->info()->data_layout())), build_opts.options());
-
+ _kernel = create_kernel(compile_context,
+ "batch_to_space_" + lower_string(string_from_data_layout(input->info()->data_layout())),
+ build_opts.options());
// Configure kernel window
Window win = calculate_max_window(*output->info(), Steps());
@@ -116,47 +125,65 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-void CLBatchToSpaceLayerKernel::configure(const ICLTensor *input, const int32_t block_shape_x, const int32_t block_shape_y, ICLTensor *output, const CropInfo &crop_info)
+void CLBatchToSpaceLayerKernel::configure(const ICLTensor *input,
+ const int32_t block_shape_x,
+ const int32_t block_shape_y,
+ ICLTensor *output,
+ const CropInfo &crop_info)
{
configure(CLKernelLibrary::get().get_compile_context(), input, block_shape_x, block_shape_y, output, crop_info);
}
-void CLBatchToSpaceLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const int32_t block_shape_x, const int32_t block_shape_y, ICLTensor *output,
- const CropInfo &crop_info)
+void CLBatchToSpaceLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const int32_t block_shape_x,
+ const int32_t block_shape_y,
+ ICLTensor *output,
+ const CropInfo &crop_info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- const TensorShape output_shape = compute_batch_to_space_shape(input->info()->data_layout(), input->info()->tensor_shape(), block_shape_x, block_shape_y);
+ const TensorShape output_shape = compute_batch_to_space_shape(
+ input->info()->data_layout(), input->info()->tensor_shape(), block_shape_x, block_shape_y);
auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape));
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments_static(input->info(), block_shape_x, block_shape_y, output->info(), crop_info));
+ ARM_COMPUTE_ERROR_THROW_ON(
+ validate_arguments_static(input->info(), block_shape_x, block_shape_y, output->info(), crop_info));
_input = input;
_output = output;
// Create kernel
CLBuildOptions build_opts;
- build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(data_size_from_type(input->info()->data_type())));
+ build_opts.add_option("-DDATA_TYPE=" +
+ get_cl_unsigned_type_from_element_size(data_size_from_type(input->info()->data_type())));
build_opts.add_option("-DBATCH_SIZE=" + support::cpp11::to_string(output->info()->dimension(3)));
build_opts.add_option("-DBLOCK_SHAPE_X=" + support::cpp11::to_string(block_shape_x));
build_opts.add_option("-DBLOCK_SHAPE_Y=" + support::cpp11::to_string(block_shape_y));
build_opts.add_option("-DCROP_LEFT=" + support::cpp11::to_string(crop_info.left));
build_opts.add_option("-DCROP_TOP=" + support::cpp11::to_string(crop_info.top));
- _kernel = create_kernel(compile_context, "batch_to_space_static_" + lower_string(string_from_data_layout(input->info()->data_layout())), build_opts.options());
+ _kernel = create_kernel(
+ compile_context, "batch_to_space_static_" + lower_string(string_from_data_layout(input->info()->data_layout())),
+ build_opts.options());
// Configure kernel window
Window win = calculate_max_window(*output->info(), Steps());
ICLKernel::configure_internal(win);
}
-Status CLBatchToSpaceLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *block_shape, const ITensorInfo *output)
+Status
+CLBatchToSpaceLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *block_shape, const ITensorInfo *output)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, block_shape, output);
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, block_shape, output));
return Status{};
}
-Status CLBatchToSpaceLayerKernel::validate(const ITensorInfo *input, const int32_t block_shape_x, const int32_t block_shape_y, const ITensorInfo *output, const CropInfo &crop_info)
+Status CLBatchToSpaceLayerKernel::validate(const ITensorInfo *input,
+ const int32_t block_shape_x,
+ const int32_t block_shape_y,
+ const ITensorInfo *output,
+ const CropInfo &crop_info)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_static(input, block_shape_x, block_shape_y, output, crop_info));
@@ -185,7 +212,7 @@
unsigned int idx = 0;
add_4D_tensor_argument(idx, _input, slice_in);
add_argument(idx, batch_id);
- if(_block_shape != nullptr)
+ if (_block_shape != nullptr)
{
add_1D_tensor_argument(idx, _block_shape, vector_slice);
}
@@ -193,7 +220,6 @@
enqueue(queue, *this, slice_out, lws_hint());
++batch_id;
- }
- while(window.slide_window_slice_3D(slice_out));
+ } while (window.slide_window_slice_3D(slice_out));
}
-} // namespace arm_compute
\ No newline at end of file
+} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLBatchToSpaceLayerKernel.h b/src/core/CL/kernels/CLBatchToSpaceLayerKernel.h
index a05184c..b9d3e66 100644
--- a/src/core/CL/kernels/CLBatchToSpaceLayerKernel.h
+++ b/src/core/CL/kernels/CLBatchToSpaceLayerKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLBATCHTOSPACELAYERKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -65,7 +66,10 @@
*
* @deprecated This method for dynamic block shape is not fully mature and will be removed in 23.08 release
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *block_shape, ICLTensor *output);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *block_shape,
+ ICLTensor *output);
/** Initialise the kernel's inputs and output (Static block shape).
*
* @param[in] input Tensor input. Supported tensor rank: 4. Data types supported: All.
@@ -74,7 +78,11 @@
* @param[out] output Tensor output. Data types supported: same as @p input
* @param[in] crop_info Specifies how the output shape is cropped after batch to space is performed
*/
- void configure(const ICLTensor *input, const int32_t block_shape_x, const int32_t block_shape_y, ICLTensor *output, const CropInfo &crop_info);
+ void configure(const ICLTensor *input,
+ const int32_t block_shape_x,
+ const int32_t block_shape_y,
+ ICLTensor *output,
+ const CropInfo &crop_info);
/** Initialise the kernel's inputs and output (Static block shape).
*
* @param[in] compile_context The compile context to be used.
@@ -84,7 +92,12 @@
* @param[out] output Tensor output. Data types supported: same as @p input
* @param[in] crop_info Specifies how the output shape is cropped after batch to space is performed
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const int32_t block_shape_x, const int32_t block_shape_y, ICLTensor *output, const CropInfo &crop_info);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const int32_t block_shape_x,
+ const int32_t block_shape_y,
+ ICLTensor *output,
+ const CropInfo &crop_info);
/** Static function to check if given info will lead to a valid configuration of @ref CLBatchToSpaceLayerKernel
*
* @param[in] input Tensor input. Supported tensor rank: 4. Data types supported: All.
@@ -106,7 +119,11 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const int32_t block_shape_x, const int32_t block_shape_y, const ITensorInfo *output, const CropInfo &crop_info);
+ static Status validate(const ITensorInfo *input,
+ const int32_t block_shape_x,
+ const int32_t block_shape_y,
+ const ITensorInfo *output,
+ const CropInfo &crop_info);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLBitwiseKernel.cpp b/src/core/CL/kernels/CLBitwiseKernel.cpp
index 11e6d02..de3fb43 100644
--- a/src/core/CL/kernels/CLBitwiseKernel.cpp
+++ b/src/core/CL/kernels/CLBitwiseKernel.cpp
@@ -28,25 +28,29 @@
#include "arm_compute/core/CL/OpenCL.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/Utils.h"
-#include "arm_compute/core/Validate.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
+#include "arm_compute/core/Validate.h"
+
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
#include "support/StringSupport.h"
namespace arm_compute
{
-CLBitwiseKernel::CLBitwiseKernel()
- : _input1(nullptr), _input2(nullptr), _output(nullptr)
+CLBitwiseKernel::CLBitwiseKernel() : _input1(nullptr), _input2(nullptr), _output(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
-void CLBitwiseKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, BitwiseOperation op)
+void CLBitwiseKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input1,
+ const ICLTensor *input2,
+ ICLTensor *output,
+ BitwiseOperation op)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input1);
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::U8);
- if(op != BitwiseOperation::NOT)
+ if (op != BitwiseOperation::NOT)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input2);
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input2, 1, DataType::U8);
@@ -56,7 +60,7 @@
// Output auto inizialitation if not yet initialized
auto_init_if_empty(*(output->info()), *(input1->info()));
- auto padding_info = get_padding_info({ input1, input2, output });
+ auto padding_info = get_padding_info({input1, input2, output});
// Configure kernel window
const unsigned int vec_size_x = adjust_vec_size(16 / output->info()->element_size(), output->info()->dimension(0));
@@ -68,7 +72,7 @@
// Create kernel
std::string kernel_name = "";
- switch(op)
+ switch (op)
{
case BitwiseOperation::AND:
kernel_name = "bitwise_and";
@@ -107,13 +111,12 @@
{
unsigned int idx = 0;
add_2D_tensor_argument(idx, _input1, slice);
- if(_input2 != nullptr)
+ if (_input2 != nullptr)
{
add_2D_tensor_argument(idx, _input2, slice);
}
add_2D_tensor_argument(idx, _output, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(window.slide_window_slice_2D(slice));
+ } while (window.slide_window_slice_2D(slice));
}
-} // namespace arm_compute
\ No newline at end of file
+} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLBitwiseKernel.h b/src/core/CL/kernels/CLBitwiseKernel.h
index c5a9996..2c74955 100644
--- a/src/core/CL/kernels/CLBitwiseKernel.h
+++ b/src/core/CL/kernels/CLBitwiseKernel.h
@@ -59,7 +59,11 @@
* @param[out] output Destination tensor. Data types supported: U8.
* @param[in] op Bitwise operation to perform. Supported: AND, OR, NOT, XOR.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, BitwiseOperation op);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input1,
+ const ICLTensor *input2,
+ ICLTensor *output,
+ BitwiseOperation op);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLBoundingBoxTransformKernel.cpp b/src/core/CL/kernels/CLBoundingBoxTransformKernel.cpp
index 72de854..f32c518 100644
--- a/src/core/CL/kernels/CLBoundingBoxTransformKernel.cpp
+++ b/src/core/CL/kernels/CLBoundingBoxTransformKernel.cpp
@@ -31,6 +31,7 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -40,7 +41,10 @@
{
namespace
{
-Status validate_arguments(const ITensorInfo *boxes, const ITensorInfo *pred_boxes, const ITensorInfo *deltas, const BoundingBoxTransformInfo &info)
+Status validate_arguments(const ITensorInfo *boxes,
+ const ITensorInfo *pred_boxes,
+ const ITensorInfo *deltas,
+ const BoundingBoxTransformInfo &info)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(boxes, pred_boxes, deltas);
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(boxes);
@@ -53,7 +57,7 @@
ARM_COMPUTE_RETURN_ERROR_ON(boxes->num_dimensions() > 2);
const bool is_qasymm16 = boxes->data_type() == DataType::QASYMM16;
- if(is_qasymm16)
+ if (is_qasymm16)
{
const UniformQuantizationInfo boxes_qinfo = boxes->quantization_info().uniform();
ARM_COMPUTE_RETURN_ERROR_ON(boxes_qinfo.scale != 0.125f);
@@ -65,12 +69,12 @@
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(boxes, deltas);
}
- if(pred_boxes->total_size() > 0)
+ if (pred_boxes->total_size() > 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(pred_boxes->tensor_shape(), deltas->tensor_shape());
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(pred_boxes, boxes);
ARM_COMPUTE_RETURN_ERROR_ON(pred_boxes->num_dimensions() > 2);
- if(is_qasymm16)
+ if (is_qasymm16)
{
const UniformQuantizationInfo pred_boxes_qinfo = pred_boxes->quantization_info().uniform();
ARM_COMPUTE_RETURN_ERROR_ON(pred_boxes_qinfo.scale != 0.125f);
@@ -83,22 +87,31 @@
}
} // namespace
-CLBoundingBoxTransformKernel::CLBoundingBoxTransformKernel()
- : _boxes(nullptr), _pred_boxes(nullptr), _deltas(nullptr)
+CLBoundingBoxTransformKernel::CLBoundingBoxTransformKernel() : _boxes(nullptr), _pred_boxes(nullptr), _deltas(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
-void CLBoundingBoxTransformKernel::configure(const ICLTensor *boxes, ICLTensor *pred_boxes, const ICLTensor *deltas, const BoundingBoxTransformInfo &info)
+void CLBoundingBoxTransformKernel::configure(const ICLTensor *boxes,
+ ICLTensor *pred_boxes,
+ const ICLTensor *deltas,
+ const BoundingBoxTransformInfo &info)
{
configure(CLKernelLibrary::get().get_compile_context(), boxes, pred_boxes, deltas, info);
}
-void CLBoundingBoxTransformKernel::configure(const CLCompileContext &compile_context, const ICLTensor *boxes, ICLTensor *pred_boxes, const ICLTensor *deltas, const BoundingBoxTransformInfo &info)
+void CLBoundingBoxTransformKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *boxes,
+ ICLTensor *pred_boxes,
+ const ICLTensor *deltas,
+ const BoundingBoxTransformInfo &info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(boxes, pred_boxes, deltas);
- auto padding_info = get_padding_info({ boxes, pred_boxes, deltas });
- auto_init_if_empty(*pred_boxes->info(), deltas->info()->clone()->set_data_type(boxes->info()->data_type()).set_quantization_info(boxes->info()->quantization_info()));
+ auto padding_info = get_padding_info({boxes, pred_boxes, deltas});
+ auto_init_if_empty(*pred_boxes->info(), deltas->info()
+ ->clone()
+ ->set_data_type(boxes->info()->data_type())
+ .set_quantization_info(boxes->info()->quantization_info()));
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(boxes->info(), pred_boxes->info(), deltas->info(), info));
@@ -128,7 +141,7 @@
build_opts.add_option_if(info.apply_scale(), "-DSCALE_AFTER=" + float_to_string_with_full_precision(info.scale()));
build_opts.add_option_if(info.correct_transform_coords(), "-DOFFSET=1");
- if(is_quantized)
+ if (is_quantized)
{
build_opts.add_option("-DDATA_TYPE_DELTAS=" + get_cl_type_from_data_type(deltas->info()->data_type()));
const UniformQuantizationInfo boxes_qinfo = boxes->info()->quantization_info().uniform();
@@ -148,12 +161,15 @@
// Since the number of columns is a multiple of 4 by definition, we don't need to pad the tensor
const unsigned int num_elems_processed_per_iteration = 4;
- Window win = calculate_max_window(*deltas->info(), Steps(num_elems_processed_per_iteration));
+ Window win = calculate_max_window(*deltas->info(), Steps(num_elems_processed_per_iteration));
ICLKernel::configure_internal(win);
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLBoundingBoxTransformKernel::validate(const ITensorInfo *boxes, const ITensorInfo *pred_boxes, const ITensorInfo *deltas, const BoundingBoxTransformInfo &info)
+Status CLBoundingBoxTransformKernel::validate(const ITensorInfo *boxes,
+ const ITensorInfo *pred_boxes,
+ const ITensorInfo *deltas,
+ const BoundingBoxTransformInfo &info)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(boxes, pred_boxes, deltas, info));
return Status{};
diff --git a/src/core/CL/kernels/CLBoundingBoxTransformKernel.h b/src/core/CL/kernels/CLBoundingBoxTransformKernel.h
index 08f350e..9a1bb49 100644
--- a/src/core/CL/kernels/CLBoundingBoxTransformKernel.h
+++ b/src/core/CL/kernels/CLBoundingBoxTransformKernel.h
@@ -58,7 +58,10 @@
* @note Only single image prediction is supported. Height and Width (and scale) of the image will be contained in the BoundingBoxTransformInfo struct.
*
*/
- void configure(const ICLTensor *boxes, ICLTensor *pred_boxes, const ICLTensor *deltas, const BoundingBoxTransformInfo &info);
+ void configure(const ICLTensor *boxes,
+ ICLTensor *pred_boxes,
+ const ICLTensor *deltas,
+ const BoundingBoxTransformInfo &info);
/** Set the input and output tensors.
*
* @param[in] compile_context The compile context to be used.
@@ -71,7 +74,11 @@
* @note Only single image prediction is supported. Height and Width (and scale) of the image will be contained in the BoundingBoxTransformInfo struct.
*
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *boxes, ICLTensor *pred_boxes, const ICLTensor *deltas, const BoundingBoxTransformInfo &info);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *boxes,
+ ICLTensor *pred_boxes,
+ const ICLTensor *deltas,
+ const BoundingBoxTransformInfo &info);
/** Static function to check if given info will lead to a valid configuration of @ref CLBoundingBoxTransform
*
@@ -85,7 +92,10 @@
*
* @return a Status
*/
- static Status validate(const ITensorInfo *boxes, const ITensorInfo *pred_boxes, const ITensorInfo *deltas, const BoundingBoxTransformInfo &info);
+ static Status validate(const ITensorInfo *boxes,
+ const ITensorInfo *pred_boxes,
+ const ITensorInfo *deltas,
+ const BoundingBoxTransformInfo &info);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLChannelShuffleLayerKernel.cpp b/src/core/CL/kernels/CLChannelShuffleLayerKernel.cpp
index a2a0bc4..ec58bf9 100644
--- a/src/core/CL/kernels/CLChannelShuffleLayerKernel.cpp
+++ b/src/core/CL/kernels/CLChannelShuffleLayerKernel.cpp
@@ -31,6 +31,7 @@
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -46,15 +47,19 @@
ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
ARM_COMPUTE_RETURN_ERROR_ON_MSG(num_groups < 2, "Channel shuffling with less than 2 groups would be inefficient");
- const unsigned int channels = input->dimension(get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::CHANNEL));
+ const unsigned int channels =
+ input->dimension(get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::CHANNEL));
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(num_groups == channels, "Channel shuffling with same number of groups as number of channels would be inefficient");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(
+ num_groups == channels,
+ "Channel shuffling with same number of groups as number of channels would be inefficient");
// There cannot be more groups than channels
ARM_COMPUTE_RETURN_ERROR_ON(num_groups > channels);
- ARM_COMPUTE_RETURN_ERROR_ON_MSG((channels % num_groups) != 0, "The number of channels must be a multiple of the number of groups");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG((channels % num_groups) != 0,
+ "The number of channels must be a multiple of the number of groups");
// Checks performed when output is configured
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
@@ -70,11 +75,12 @@
auto_init_if_empty(*output, *input->clone());
const bool is_nhwc = input->data_layout() == DataLayout::NHWC;
- if(is_nhwc)
+ if (is_nhwc)
{
- unsigned int num_elems_processed_per_iteration_x = adjust_vec_size(max_cl_vector_width / input->element_size(), input->dimension(0));
- Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration_x));
- Window win_collapsed = win.collapse(win, Window::DimZ);
+ unsigned int num_elems_processed_per_iteration_x =
+ adjust_vec_size(max_cl_vector_width / input->element_size(), input->dimension(0));
+ Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration_x));
+ Window win_collapsed = win.collapse(win, Window::DimZ);
return std::make_pair(Status{}, win_collapsed);
}
else
@@ -83,22 +89,25 @@
constexpr unsigned int num_elems_processed_per_iteration_y = 2;
// Configure kernel window
- Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
- AccessWindowRectangle input_access(input, 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
- AccessWindowRectangle output_access(output, 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
+ Window win = calculate_max_window(
+ *input, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
+ AccessWindowRectangle input_access(input, 0, 0, num_elems_processed_per_iteration_x,
+ num_elems_processed_per_iteration_y);
+ AccessWindowRectangle output_access(output, 0, 0, num_elems_processed_per_iteration_x,
+ num_elems_processed_per_iteration_y);
const bool window_changed = update_window_and_padding(win, input_access, output_access);
Window win_collapsed = win.collapse(win, Window::DimZ);
- Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+ Status err =
+ (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
return std::make_pair(err, win_collapsed);
}
}
} // namespace
-CLChannelShuffleLayerKernel::CLChannelShuffleLayerKernel()
- : _input(nullptr), _output(nullptr)
+CLChannelShuffleLayerKernel::CLChannelShuffleLayerKernel() : _input(nullptr), _output(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
@@ -108,23 +117,27 @@
configure(CLKernelLibrary::get().get_compile_context(), input, output, num_groups);
}
-void CLChannelShuffleLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, unsigned int num_groups)
+void CLChannelShuffleLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ unsigned int num_groups)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), num_groups));
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
_input = input;
_output = output;
- const DataLayout data_layout = input->info()->data_layout();
- const bool is_nhwc = data_layout == DataLayout::NHWC;
- const unsigned int channels = input->info()->dimension(get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL));
- unsigned int vec_size_x = 0;
- unsigned int vec_size_x_leftovers = 0;
- if(is_nhwc)
+ const DataLayout data_layout = input->info()->data_layout();
+ const bool is_nhwc = data_layout == DataLayout::NHWC;
+ const unsigned int channels =
+ input->info()->dimension(get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL));
+ unsigned int vec_size_x = 0;
+ unsigned int vec_size_x_leftovers = 0;
+ if (is_nhwc)
{
- vec_size_x = adjust_vec_size(max_cl_vector_width / input->info()->element_size(), input->info()->dimension(0));
+ vec_size_x = adjust_vec_size(max_cl_vector_width / input->info()->element_size(), input->info()->dimension(0));
vec_size_x_leftovers = input->info()->dimension(0) % vec_size_x;
}
else
@@ -170,13 +183,14 @@
_config_id += support::cpp11::to_string(output->info()->dimension(1));
_config_id += "_";
_config_id += support::cpp11::to_string(output->info()->dimension(2));
- if(data_layout == DataLayout::NHWC)
+ if (data_layout == DataLayout::NHWC)
{
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
}
-Status CLChannelShuffleLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int num_groups)
+Status
+CLChannelShuffleLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int num_groups)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, num_groups));
ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first);
diff --git a/src/core/CL/kernels/CLChannelShuffleLayerKernel.h b/src/core/CL/kernels/CLChannelShuffleLayerKernel.h
index 31c007f..43c939e 100644
--- a/src/core/CL/kernels/CLChannelShuffleLayerKernel.h
+++ b/src/core/CL/kernels/CLChannelShuffleLayerKernel.h
@@ -60,7 +60,10 @@
* @param[out] output Output tensor. Data type supported: Same as @p input
* @param[in] num_groups Number of groups. Must be greater than 1 and the number of channels of the tensors must be a multiple of the number of groups.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, unsigned int num_groups);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ unsigned int num_groups);
/** Static function to check if given info will lead to a valid configuration of @ref CLChannelShuffleLayerKernel
*
* @param[in] input Input tensor info. Data types supported: All.
diff --git a/src/core/CL/kernels/CLComparisonKernel.cpp b/src/core/CL/kernels/CLComparisonKernel.cpp
index f4d6316..f272707 100644
--- a/src/core/CL/kernels/CLComparisonKernel.cpp
+++ b/src/core/CL/kernels/CLComparisonKernel.cpp
@@ -26,6 +26,7 @@
#include "arm_compute/core/CL/CLHelpers.h"
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -38,14 +39,10 @@
namespace
{
// Create supported comparisons map
-const std::map<ComparisonOperation, std::string> supported_comparison_ops =
-{
- { ComparisonOperation::Equal, "EQUAL" },
- { ComparisonOperation::NotEqual, "NOTEQUAL" },
- { ComparisonOperation::Greater, "GREATER" },
- { ComparisonOperation::GreaterEqual, "GREATEREQUAL" },
- { ComparisonOperation::Less, "LESS" },
- { ComparisonOperation::LessEqual, "LESSEQUAL" },
+const std::map<ComparisonOperation, std::string> supported_comparison_ops = {
+ {ComparisonOperation::Equal, "EQUAL"}, {ComparisonOperation::NotEqual, "NOTEQUAL"},
+ {ComparisonOperation::Greater, "GREATER"}, {ComparisonOperation::GreaterEqual, "GREATEREQUAL"},
+ {ComparisonOperation::Less, "LESS"}, {ComparisonOperation::LessEqual, "LESSEQUAL"},
};
int calculate_num_elems_processed_per_iteration(const ITensorInfo &input)
@@ -53,7 +50,10 @@
return 16 / input.element_size();
}
-Status validate_arguments(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output, ComparisonOperation operation)
+Status validate_arguments(const ITensorInfo &input1,
+ const ITensorInfo &input2,
+ const ITensorInfo &output,
+ ComparisonOperation operation)
{
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(&input1);
ARM_COMPUTE_RETURN_ERROR_ON(input1.data_type() == DataType::UNKNOWN);
@@ -64,7 +64,7 @@
ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible");
// Validate in case of configured output
- if(output.total_size() > 0)
+ if (output.total_size() > 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&output, 1, DataType::U8);
ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(out_shape, output.tensor_shape(), 0),
@@ -76,7 +76,7 @@
std::pair<Status, Window> validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
{
- const TensorShape &out_shape = TensorShape::broadcast_shape(input1.tensor_shape(), input2.tensor_shape());
+ const TensorShape &out_shape = TensorShape::broadcast_shape(input1.tensor_shape(), input2.tensor_shape());
const unsigned int num_elems_processed_per_iteration = calculate_num_elems_processed_per_iteration(input1);
// Auto initialize output if not initialized
@@ -90,27 +90,34 @@
AccessWindowHorizontal input2_access(&input2, 0, num_elems_processed_per_iteration);
AccessWindowHorizontal output_access(&output, 0, num_elems_processed_per_iteration);
- bool window_changed = update_window_and_padding(win_input1, input1_access)
- || update_window_and_padding(win_input2, input2_access)
- || update_window_and_padding(win, output_access);
+ bool window_changed = update_window_and_padding(win_input1, input1_access) ||
+ update_window_and_padding(win_input2, input2_access) ||
+ update_window_and_padding(win, output_access);
- Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+ Status err =
+ (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
return std::make_pair(err, win);
}
} // namespace
-CLComparisonKernel::CLComparisonKernel()
- : _input1(nullptr), _input2(nullptr), _output(nullptr)
+CLComparisonKernel::CLComparisonKernel() : _input1(nullptr), _input2(nullptr), _output(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
-void CLComparisonKernel::configure(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, ComparisonOperation operation)
+void CLComparisonKernel::configure(const ICLTensor *input1,
+ const ICLTensor *input2,
+ ICLTensor *output,
+ ComparisonOperation operation)
{
configure(CLKernelLibrary::get().get_compile_context(), input1, input2, output, operation);
}
-void CLComparisonKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, ComparisonOperation operation)
+void CLComparisonKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input1,
+ const ICLTensor *input2,
+ ICLTensor *output,
+ ComparisonOperation operation)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(*input1->info(), *input2->info(), *output->info(), operation));
@@ -129,10 +136,11 @@
// Set kernel build options
std::set<std::string> build_opts;
build_opts.emplace("-DDATA_TYPE=" + get_cl_type_from_data_type(input1->info()->data_type()));
- build_opts.emplace("-DVEC_SIZE=" + support::cpp11::to_string(calculate_num_elems_processed_per_iteration(*input1->info())));
+ build_opts.emplace("-DVEC_SIZE=" +
+ support::cpp11::to_string(calculate_num_elems_processed_per_iteration(*input1->info())));
build_opts.emplace("-DOP=" + operation_name);
build_opts.emplace("-DOP_NAME=" + lower_string(operation_name));
- if(is_data_type_quantized(input1->info()->data_type()))
+ if (is_data_type_quantized(input1->info()->data_type()))
{
const UniformQuantizationInfo iq1_info = input1->info()->quantization_info().uniform();
const UniformQuantizationInfo iq2_info = input2->info()->quantization_info().uniform();
@@ -160,12 +168,16 @@
_config_id += lower_string(string_from_data_layout(input1->info()->data_layout()));
}
-Status CLComparisonKernel::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ComparisonOperation operation)
+Status CLComparisonKernel::validate(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ const ITensorInfo *output,
+ ComparisonOperation operation)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input1, input2, output);
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(*input1, *input2, *output, operation));
- ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(*input1->clone(), *input2->clone(), *output->clone()).first);
+ ARM_COMPUTE_RETURN_ON_ERROR(
+ validate_and_configure_window(*input1->clone(), *input2->clone(), *output->clone()).first);
return Status{};
}
@@ -181,17 +193,18 @@
bool can_collapse = true;
const bool is_vector = in_shape1.num_dimensions() == 1 || in_shape2.num_dimensions() == 1;
- if(std::min(in_shape1.total_size(), in_shape2.total_size()) > 1 && !is_vector)
+ if (std::min(in_shape1.total_size(), in_shape2.total_size()) > 1 && !is_vector)
{
can_collapse = (std::min(in_shape1.num_dimensions(), in_shape2.num_dimensions()) > Window::DimZ);
- for(size_t d = Window::DimZ; can_collapse && (d < out_shape.num_dimensions()); d++)
+ for (size_t d = Window::DimZ; can_collapse && (d < out_shape.num_dimensions()); d++)
{
can_collapse = (in_shape1[d] == in_shape2[d]);
}
}
bool has_collapsed = false;
- Window collapsed = can_collapse ? window.collapse_if_possible(ICLKernel::window(), Window::DimZ, &has_collapsed) : window;
+ Window collapsed =
+ can_collapse ? window.collapse_if_possible(ICLKernel::window(), Window::DimZ, &has_collapsed) : window;
const TensorShape &in_shape1_collapsed = has_collapsed ? in_shape1.collapsed_from(Window::DimZ) : in_shape1;
const TensorShape &in_shape2_collapsed = has_collapsed ? in_shape2.collapsed_from(Window::DimZ) : in_shape2;
@@ -212,16 +225,16 @@
ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_input1));
ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_input2));
- }
- while(collapsed.slide_window_slice_3D(slice));
+ } while (collapsed.slide_window_slice_3D(slice));
}
BorderSize CLComparisonKernel::border_size() const
{
const int num_elems_processed_per_iteration = calculate_num_elems_processed_per_iteration(*_input1->info());
- const unsigned int replicateSize = _output->info()->dimension(0) - std::min(_input1->info()->dimension(0), _input2->info()->dimension(0));
- const unsigned int border = std::min<unsigned int>(num_elems_processed_per_iteration - 1U, replicateSize);
- return BorderSize{ 0, border, 0, 0 };
+ const unsigned int replicateSize =
+ _output->info()->dimension(0) - std::min(_input1->info()->dimension(0), _input2->info()->dimension(0));
+ const unsigned int border = std::min<unsigned int>(num_elems_processed_per_iteration - 1U, replicateSize);
+ return BorderSize{0, border, 0, 0};
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLComparisonKernel.h b/src/core/CL/kernels/CLComparisonKernel.h
index 0b94190..174a6c9 100644
--- a/src/core/CL/kernels/CLComparisonKernel.h
+++ b/src/core/CL/kernels/CLComparisonKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLCOMPARISONKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -64,7 +65,11 @@
* @param[out] output Destination tensor. Data types supported: U8.
* @param[in] operation Comparison operation to use.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, ComparisonOperation operation);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input1,
+ const ICLTensor *input2,
+ ICLTensor *output,
+ ComparisonOperation operation);
/** Static function to check if given info will lead to a valid configuration of @ref CLComparisonKernel
*
* @param[in] input1 Source tensor. Data types supported: All.
@@ -74,10 +79,13 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ComparisonOperation operation);
+ static Status validate(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ const ITensorInfo *output,
+ ComparisonOperation operation);
// Inherited methods overridden:
- void run(const Window &window, cl::CommandQueue &queue) override;
+ void run(const Window &window, cl::CommandQueue &queue) override;
BorderSize border_size() const override;
private:
diff --git a/src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.cpp b/src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.cpp
index 76af5d5..f8ecc4c 100644
--- a/src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.cpp
+++ b/src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.cpp
@@ -29,6 +29,7 @@
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/Validate.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -40,7 +41,8 @@
_type = CLKernelType::ELEMENTWISE;
}
-Status CLDeconvolutionLayerUpsampleKernel::validate(const ITensorInfo *input, const ITensorInfo *output,
+Status CLDeconvolutionLayerUpsampleKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
const PadStrideInfo &info)
{
ARM_COMPUTE_UNUSED(info);
@@ -60,7 +62,7 @@
ARM_COMPUTE_RETURN_ERROR_ON(output->dimension(idx_h) == 0);
ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(idx_c) != output->dimension(idx_c));
- for(size_t i = 3; i < Coordinates::num_max_dimensions; ++i)
+ for (size_t i = 3; i < Coordinates::num_max_dimensions; ++i)
{
ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(i) != output->dimension(i));
}
@@ -68,20 +70,21 @@
return Status{};
}
-void CLDeconvolutionLayerUpsampleKernel::configure(const ICLTensor *input, ICLTensor *output,
- const PadStrideInfo &info)
+void CLDeconvolutionLayerUpsampleKernel::configure(const ICLTensor *input, ICLTensor *output, const PadStrideInfo &info)
{
configure(CLKernelLibrary::get().get_compile_context(), input, output, info);
}
-void CLDeconvolutionLayerUpsampleKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output,
- const PadStrideInfo &info)
+void CLDeconvolutionLayerUpsampleKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const PadStrideInfo &info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
// Perform validation step
ARM_COMPUTE_ERROR_THROW_ON(CLDeconvolutionLayerUpsampleKernel::validate(input->info(), output->info(), info));
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
_input = input;
_output = output;
@@ -119,7 +122,7 @@
const int out_end_y = _output->info()->dimension(idx_h) - _info.pad_bottom() + _info.stride().second - 1;
const int out_step_y = _info.stride().second;
- switch(_data_layout)
+ switch (_data_layout)
{
case DataLayout::NCHW:
{
@@ -137,8 +140,7 @@
add_3D_tensor_argument(idx, _input, slice_in);
add_3D_tensor_argument(idx, _output, slice_out);
enqueue(queue, *this, slice_out, lws_hint());
- }
- while(collapsed.slide_window_slice_3D(slice_in) && collapsed.slide_window_slice_3D(slice_out));
+ } while (collapsed.slide_window_slice_3D(slice_in) && collapsed.slide_window_slice_3D(slice_out));
break;
}
case DataLayout::NHWC:
@@ -156,8 +158,7 @@
add_3D_tensor_argument(idx, _input, slice_in);
add_3D_tensor_argument(idx, _output, slice_out);
enqueue(queue, *this, slice_out, lws_hint());
- }
- while(window.slide_window_slice_3D(slice_in) && window.slide_window_slice_3D(slice_out));
+ } while (window.slide_window_slice_3D(slice_in) && window.slide_window_slice_3D(slice_out));
break;
}
default:
diff --git a/src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h b/src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h
index e0d1322..762989a 100644
--- a/src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h
+++ b/src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h
@@ -62,7 +62,10 @@
* @param[out] output Destination tensor. Data types supported: same as @p input. All but the lowest two dimensions must be the same size as in the input tensor, i.e. scaling is only performed within the XY-plane.
* @param[in] info Contains padding and stride information described in @ref PadStrideInfo.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const PadStrideInfo &info);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const PadStrideInfo &info);
/** Static function to check if given info will lead to a valid configuration of @ref CLDeconvolutionLayerUpsample
*
* @param[in] input Source tensor info. Data types supported: All.
diff --git a/src/core/CL/kernels/CLDeconvolutionReshapeOutputKernel.cpp b/src/core/CL/kernels/CLDeconvolutionReshapeOutputKernel.cpp
index 0fc0ff8..b33e0a8 100644
--- a/src/core/CL/kernels/CLDeconvolutionReshapeOutputKernel.cpp
+++ b/src/core/CL/kernels/CLDeconvolutionReshapeOutputKernel.cpp
@@ -27,9 +27,10 @@
#include "arm_compute/core/CL/CLKernelLibrary.h"
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/Validate.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/core/utils/StringUtils.h"
+#include "arm_compute/core/Validate.h"
+
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
#include "support/StringSupport.h"
@@ -38,7 +39,11 @@
{
namespace
{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const ITensorInfo *input_info, const ITensorInfo *weights_info,
+Status validate_arguments(const ITensorInfo *input,
+ const ITensorInfo *bias,
+ const ITensorInfo *output,
+ const ITensorInfo *input_info,
+ const ITensorInfo *weights_info,
const PadStrideInfo &deconv_info)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output, input_info, weights_info);
@@ -53,19 +58,21 @@
ARM_COMPUTE_RETURN_ERROR_ON(weights_info->dimension(idx_w) != deconv_info.stride().first);
ARM_COMPUTE_RETURN_ERROR_ON(weights_info->dimension(idx_h) != deconv_info.stride().second);
- ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32, DataType::F16, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::S32);
- if(!is_qasymm)
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32, DataType::F16, DataType::QASYMM8,
+ DataType::QASYMM8_SIGNED, DataType::S32);
+ if (!is_qasymm)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, input_info, weights_info);
}
- ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(0) != weights_info->dimension(idx_w) * weights_info->dimension(idx_h) * weights_info->dimension(idx_b));
+ ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(0) != weights_info->dimension(idx_w) * weights_info->dimension(idx_h) *
+ weights_info->dimension(idx_b));
ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(1) != input_info->dimension(idx_w));
ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(2) != input_info->dimension(idx_h));
ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(3) != input_info->dimension(idx_b));
- if(bias != nullptr)
+ if (bias != nullptr)
{
- if(is_qasymm)
+ if (is_qasymm)
{
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(bias, 1, DataType::S32);
}
@@ -76,19 +83,26 @@
ARM_COMPUTE_RETURN_ERROR_ON(bias->dimension(0) != weights_info->dimension(idx_b));
}
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
const PadStrideInfo stride_info(deconv_info.stride().first, deconv_info.stride().second);
- auto out_dims = deconvolution_output_dimensions(input_info->dimension(idx_w), input_info->dimension(idx_h), weights_info->dimension(idx_w), weights_info->dimension(idx_h), stride_info);
+ auto out_dims = deconvolution_output_dimensions(input_info->dimension(idx_w), input_info->dimension(idx_h),
+ weights_info->dimension(idx_w), weights_info->dimension(idx_h),
+ stride_info);
- const TensorShape output_shape = misc::shape_calculator::compute_deconvolution_output_shape(out_dims, *input_info, *weights_info);
+ const TensorShape output_shape =
+ misc::shape_calculator::compute_deconvolution_output_shape(out_dims, *input_info, *weights_info);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), output_shape);
}
return Status{};
}
-std::pair<Status, Window> validate_and_configure_window(const ITensorInfo *input, ITensorInfo *output, const ITensorInfo *input_info, const ITensorInfo *weights_info, const PadStrideInfo &deconv_info)
+std::pair<Status, Window> validate_and_configure_window(const ITensorInfo *input,
+ ITensorInfo *output,
+ const ITensorInfo *input_info,
+ const ITensorInfo *weights_info,
+ const PadStrideInfo &deconv_info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
@@ -97,11 +111,17 @@
const size_t idx_h = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
const PadStrideInfo stride_info(deconv_info.stride().first, deconv_info.stride().second);
- auto out_dims = deconvolution_output_dimensions(input_info->dimension(idx_w), input_info->dimension(idx_h), weights_info->dimension(idx_w), weights_info->dimension(idx_h), stride_info);
+ auto out_dims =
+ deconvolution_output_dimensions(input_info->dimension(idx_w), input_info->dimension(idx_h),
+ weights_info->dimension(idx_w), weights_info->dimension(idx_h), stride_info);
- const TensorShape output_shape = misc::shape_calculator::compute_deconvolution_output_shape(out_dims, *input_info, *weights_info);
+ const TensorShape output_shape =
+ misc::shape_calculator::compute_deconvolution_output_shape(out_dims, *input_info, *weights_info);
- auto_init_if_empty(*output, input->clone()->set_tensor_shape(output_shape).set_data_layout(data_layout).set_quantization_info(input->quantization_info()));
+ auto_init_if_empty(*output, input->clone()
+ ->set_tensor_shape(output_shape)
+ .set_data_layout(data_layout)
+ .set_quantization_info(input->quantization_info()));
Window win = calculate_max_window(*input);
@@ -109,29 +129,37 @@
}
} // namespace
-CLDeconvolutionReshapeOutputKernel::CLDeconvolutionReshapeOutputKernel()
- : _add_bias(false),
- _bias(nullptr)
+CLDeconvolutionReshapeOutputKernel::CLDeconvolutionReshapeOutputKernel() : _add_bias(false), _bias(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
-void CLDeconvolutionReshapeOutputKernel::configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const ITensorInfo *input_info, const ITensorInfo *weights_info,
+void CLDeconvolutionReshapeOutputKernel::configure(const ICLTensor *input,
+ const ICLTensor *bias,
+ ICLTensor *output,
+ const ITensorInfo *input_info,
+ const ITensorInfo *weights_info,
const PadStrideInfo &deconv_info)
{
configure(CLKernelLibrary::get().get_compile_context(), input, bias, output, input_info, weights_info, deconv_info);
}
-void CLDeconvolutionReshapeOutputKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const ITensorInfo *input_info,
- const ITensorInfo *weights_info,
- const PadStrideInfo &deconv_info)
+void CLDeconvolutionReshapeOutputKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *bias,
+ ICLTensor *output,
+ const ITensorInfo *input_info,
+ const ITensorInfo *weights_info,
+ const PadStrideInfo &deconv_info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, input_info, weights_info);
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (bias != nullptr ? bias->info() : nullptr), output->info(), input_info, weights_info, deconv_info));
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (bias != nullptr ? bias->info() : nullptr),
+ output->info(), input_info, weights_info, deconv_info));
- auto padding_info = get_padding_info({ input, bias, output });
+ auto padding_info = get_padding_info({input, bias, output});
// Configure kernel window
- auto win_config = validate_and_configure_window(input->info(), output->info(), input_info, weights_info, deconv_info);
+ auto win_config =
+ validate_and_configure_window(input->info(), output->info(), input_info, weights_info, deconv_info);
ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
const DataLayout data_layout = input_info->data_layout();
@@ -178,7 +206,11 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLDeconvolutionReshapeOutputKernel::validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const ITensorInfo *input_info, const ITensorInfo *weights_info,
+Status CLDeconvolutionReshapeOutputKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *bias,
+ const ITensorInfo *output,
+ const ITensorInfo *input_info,
+ const ITensorInfo *weights_info,
const PadStrideInfo &deconv_info)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, bias, output, input_info, weights_info, deconv_info));
@@ -194,7 +226,7 @@
unsigned int idx = 0;
add_3D_tensor_argument(idx, _input, collapsed);
add_3D_tensor_argument(idx, _output, collapsed);
- if(_add_bias)
+ if (_add_bias)
{
add_1D_tensor_argument(idx, _bias, collapsed);
}
diff --git a/src/core/CL/kernels/CLDeconvolutionReshapeOutputKernel.h b/src/core/CL/kernels/CLDeconvolutionReshapeOutputKernel.h
index ce354fa..8f436b0 100644
--- a/src/core/CL/kernels/CLDeconvolutionReshapeOutputKernel.h
+++ b/src/core/CL/kernels/CLDeconvolutionReshapeOutputKernel.h
@@ -67,7 +67,12 @@
* @param[in] weights_info Deconvolution weights tensor info. Supported data types: same as @p input. Supported data layouts: same as @p input.
* @param[in] deconv_info Contains padding and policies to be used in the deconvolution, this is described in @ref PadStrideInfo. This kernel supports only stride_x = weights.width && stride_y = weights.height. Moreover, padding is not supported.
*/
- void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const ITensorInfo *input_info, const ITensorInfo *weights_info, const PadStrideInfo &deconv_info);
+ void configure(const ICLTensor *input,
+ const ICLTensor *bias,
+ ICLTensor *output,
+ const ITensorInfo *input_info,
+ const ITensorInfo *weights_info,
+ const PadStrideInfo &deconv_info);
/** Initialise the kernel's source and destination.
*
* @param[in] compile_context The compile context to be used.
@@ -79,8 +84,13 @@
* @param[in] weights_info Deconvolution weights tensor info. Supported data types: same as @p input. Supported data layouts: same as @p input.
* @param[in] deconv_info Contains padding and policies to be used in the deconvolution, this is described in @ref PadStrideInfo. This kernel supports only stride_x = weights.width && stride_y = weights.height. Moreover, padding is not supported.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const ITensorInfo *input_info, const ITensorInfo *weights_info,
- const PadStrideInfo &deconv_info);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *bias,
+ ICLTensor *output,
+ const ITensorInfo *input_info,
+ const ITensorInfo *weights_info,
+ const PadStrideInfo &deconv_info);
/** Static function to check if given info will lead to a valid configuration of @ref CLDeconvolutionReshapeOutputKernel.
*
@@ -93,7 +103,12 @@
*
* @return a Status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const ITensorInfo *input_info, const ITensorInfo *weights_info, const PadStrideInfo &deconv_info);
+ static Status validate(const ITensorInfo *input,
+ const ITensorInfo *bias,
+ const ITensorInfo *output,
+ const ITensorInfo *input_info,
+ const ITensorInfo *weights_info,
+ const PadStrideInfo &deconv_info);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLDepthToSpaceLayerKernel.cpp b/src/core/CL/kernels/CLDepthToSpaceLayerKernel.cpp
index 5c1dc4f..cdf19ab 100644
--- a/src/core/CL/kernels/CLDepthToSpaceLayerKernel.cpp
+++ b/src/core/CL/kernels/CLDepthToSpaceLayerKernel.cpp
@@ -27,6 +27,7 @@
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -49,12 +50,14 @@
ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[idx_channel] % (block_shape * block_shape) != 0);
// Validate output if initialized
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
const int idx_width = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
const int idx_height = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
- ARM_COMPUTE_RETURN_ERROR_ON(output->tensor_shape()[idx_width] != (block_shape * input->tensor_shape()[idx_width]));
- ARM_COMPUTE_RETURN_ERROR_ON(output->tensor_shape()[idx_height] != (block_shape * input->tensor_shape()[idx_height]));
+ ARM_COMPUTE_RETURN_ERROR_ON(output->tensor_shape()[idx_width] !=
+ (block_shape * input->tensor_shape()[idx_width]));
+ ARM_COMPUTE_RETURN_ERROR_ON(output->tensor_shape()[idx_height] !=
+ (block_shape * input->tensor_shape()[idx_height]));
ARM_COMPUTE_RETURN_ERROR_ON(output->num_dimensions() > 4);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
}
@@ -63,8 +66,7 @@
}
} // namespace
-CLDepthToSpaceLayerKernel::CLDepthToSpaceLayerKernel()
- : _input(nullptr), _output(nullptr), _block_shape()
+CLDepthToSpaceLayerKernel::CLDepthToSpaceLayerKernel() : _input(nullptr), _output(nullptr), _block_shape()
{
_type = CLKernelType::ELEMENTWISE;
}
@@ -74,14 +76,18 @@
configure(CLKernelLibrary::get().get_compile_context(), input, output, block_shape);
}
-void CLDepthToSpaceLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, int32_t block_shape)
+void CLDepthToSpaceLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ int32_t block_shape)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- TensorShape output_shape = compute_depth_to_space_shape(input->info()->tensor_shape(), input->info()->data_layout(), block_shape);
+ TensorShape output_shape =
+ compute_depth_to_space_shape(input->info()->tensor_shape(), input->info()->data_layout(), block_shape);
auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type());
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), block_shape));
@@ -98,7 +104,9 @@
build_opts.add_option("-DCHANNEL_SIZE=" + support::cpp11::to_string(input->info()->dimension(idx_channel)));
build_opts.add_option("-DBLOCK_SHAPE=" + support::cpp11::to_string(block_shape));
build_opts.add_option("-DWIDTH_IN=" + support::cpp11::to_string(input->info()->dimension(idx_width)));
- _kernel = create_kernel(compile_context, "depth_to_space_" + lower_string(string_from_data_layout(input->info()->data_layout())), build_opts.options());
+ _kernel = create_kernel(compile_context,
+ "depth_to_space_" + lower_string(string_from_data_layout(input->info()->data_layout())),
+ build_opts.options());
// Configure kernel window
Window win = calculate_max_window(*input->info(), Steps());
@@ -137,7 +145,6 @@
enqueue(queue, *this, slice_in, lws_hint());
++batch_id;
- }
- while(window.slide_window_slice_3D(slice_in));
+ } while (window.slide_window_slice_3D(slice_in));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLDepthToSpaceLayerKernel.h b/src/core/CL/kernels/CLDepthToSpaceLayerKernel.h
index 1f7f77b..cef70c4 100644
--- a/src/core/CL/kernels/CLDepthToSpaceLayerKernel.h
+++ b/src/core/CL/kernels/CLDepthToSpaceLayerKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLDEPTHTOSPACELAYERKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -61,7 +62,8 @@
* @param[out] output Tensor output. Data types supported: same as @p input
* @param[in] block_shape Block shape value.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, int32_t block_shape);
+ void
+ configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, int32_t block_shape);
/** Static function to check if given info will lead to a valid configuration of @ref CLDepthToSpaceLayerKernel.
*
* @param[in] input Tensor input info. Supported tensor rank: 4. Data types supported: All.
diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp
index e34b692..b95abe7 100644
--- a/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp
+++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp
@@ -23,16 +23,17 @@
*/
#include "src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.h"
-#include "arm_compute/core/utils/ActivationFunctionUtils.h"
#include "arm_compute/core/CL/CLHelpers.h"
#include "arm_compute/core/CL/CLKernelLibrary.h"
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/utils/ActivationFunctionUtils.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/core/utils/quantization/AsymmHelpers.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLUtils.h"
#include "src/core/CL/CLValidate.h"
#include "src/core/CL/ICLKernel.h"
@@ -45,12 +46,18 @@
{
namespace
{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const DWCComputeKernelInfo &dwc_info,
- const ConvolutionInfo &conv_info, const ITensorInfo *output_multipliers, const ITensorInfo *output_shifts)
+Status validate_arguments(const ITensorInfo *input,
+ const ITensorInfo *weights,
+ const ITensorInfo *biases,
+ const ITensorInfo *output,
+ const DWCComputeKernelInfo &dwc_info,
+ const ConvolutionInfo &conv_info,
+ const ITensorInfo *output_multipliers,
+ const ITensorInfo *output_shifts)
{
ARM_COMPUTE_UNUSED(dwc_info);
bool in_place = false;
- if(output == nullptr || output == input)
+ if (output == nullptr || output == input)
{
in_place = true;
output = input;
@@ -58,11 +65,14 @@
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights);
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(input, DataLayout::NHWC);
- ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
+ DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON(conv_info.pad_stride_info.stride().first > 1 && dwc_info.m0 != 1);
ARM_COMPUTE_RETURN_ERROR_ON(conv_info.dilation.x() > 1 && dwc_info.m0 != 1);
ARM_COMPUTE_RETURN_ERROR_ON((dwc_info.export_input_to_cl_image == true));
- ARM_COMPUTE_RETURN_ERROR_ON_MSG((dwc_info.export_weights_to_cl_image == true) && (export_to_cl_image(weights) == false), "Weights cannot be exported to cl_image!");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG((dwc_info.export_weights_to_cl_image == true) &&
+ (export_to_cl_image(weights) == false),
+ "Weights cannot be exported to cl_image!");
ARM_COMPUTE_RETURN_ERROR_ON((dwc_info.export_weights_to_cl_image == true) && ((dwc_info.n0 % 4) != 0));
ARM_COMPUTE_RETURN_ERROR_ON(conv_info.pad_stride_info.stride().first < 1);
ARM_COMPUTE_RETURN_ERROR_ON(conv_info.pad_stride_info.stride().second < 1);
@@ -72,33 +82,40 @@
ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(idx_c) != (input->dimension(idx_c) * conv_info.depth_multiplier));
// In place restrictions
- if(in_place)
+ if (in_place)
{
- const int weights_width_idx = get_data_layout_dimension_index(weights->data_layout(), DataLayoutDimension::WIDTH);
- const int weights_height_idx = get_data_layout_dimension_index(weights->data_layout(), DataLayoutDimension::HEIGHT);
- ARM_COMPUTE_RETURN_ERROR_ON(weights->tensor_shape()[weights_width_idx] != 1U || weights->tensor_shape()[weights_height_idx] != 1U);
+ const int weights_width_idx =
+ get_data_layout_dimension_index(weights->data_layout(), DataLayoutDimension::WIDTH);
+ const int weights_height_idx =
+ get_data_layout_dimension_index(weights->data_layout(), DataLayoutDimension::HEIGHT);
+ ARM_COMPUTE_RETURN_ERROR_ON(weights->tensor_shape()[weights_width_idx] != 1U ||
+ weights->tensor_shape()[weights_height_idx] != 1U);
ARM_COMPUTE_RETURN_ERROR_ON(conv_info.depth_multiplier != 1U);
ARM_COMPUTE_RETURN_ERROR_ON(conv_info.pad_stride_info.stride() != std::make_pair(1U, 1U));
ARM_COMPUTE_RETURN_ERROR_ON(conv_info.dilation != Size2D(1U, 1U));
- ARM_COMPUTE_RETURN_ERROR_ON(conv_info.pad_stride_info.has_padding()); // Note that in princple padding can be supported with in_place but we choose not to support it
+ ARM_COMPUTE_RETURN_ERROR_ON(
+ conv_info.pad_stride_info
+ .has_padding()); // Note that in princple padding can be supported with in_place but we choose not to support it
}
- const ConvolutionInfo info{ conv_info.pad_stride_info, conv_info.depth_multiplier, ActivationLayerInfo(), conv_info.dilation };
- const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_depthwise_convolution_shape(*input, *weights, conv_info);
+ const ConvolutionInfo info{conv_info.pad_stride_info, conv_info.depth_multiplier, ActivationLayerInfo(),
+ conv_info.dilation};
+ const TensorShape output_shape =
+ arm_compute::misc::shape_calculator::compute_depthwise_convolution_shape(*input, *weights, conv_info);
- if(conv_info.depth_multiplier > 1 && dwc_info.n0 > 1)
+ if (conv_info.depth_multiplier > 1 && dwc_info.n0 > 1)
{
ARM_COMPUTE_RETURN_ERROR_ON((conv_info.depth_multiplier % dwc_info.n0) != 0);
}
const bool is_quantized = is_data_type_quantized(input->data_type());
- if(biases != nullptr)
+ if (biases != nullptr)
{
ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != output_shape[idx_c]);
ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1);
- if(is_quantized)
+ if (is_quantized)
{
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(biases, 1, DataType::S32);
}
@@ -108,7 +125,7 @@
}
}
- if(is_quantized)
+ if (is_quantized)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output_multipliers, output_shifts);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output_multipliers, 1, DataType::S32);
@@ -116,7 +133,7 @@
ARM_COMPUTE_RETURN_ERROR_ON(output_multipliers->num_dimensions() > 1);
ARM_COMPUTE_RETURN_ERROR_ON(output_shifts->num_dimensions() > 1);
- if(is_data_type_quantized_per_channel(weights->data_type()))
+ if (is_data_type_quantized_per_channel(weights->data_type()))
{
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(weights, 1, DataType::QSYMM8_PER_CHANNEL);
ARM_COMPUTE_RETURN_ERROR_ON(output_shape[idx_c] != output_multipliers->dimension(0));
@@ -134,22 +151,24 @@
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights);
}
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), output_shape);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
}
- if(is_data_type_quantized(input->data_type()))
+ if (is_data_type_quantized(input->data_type()))
{
const UniformQuantizationInfo iq_info = input->quantization_info().uniform();
const UniformQuantizationInfo wq_info = weights->quantization_info().uniform();
- const UniformQuantizationInfo oq_info = (output->total_size() != 0) ? output->quantization_info().uniform() : iq_info;
+ const UniformQuantizationInfo oq_info =
+ (output->total_size() != 0) ? output->quantization_info().uniform() : iq_info;
float multiplier = iq_info.scale * wq_info.scale / oq_info.scale;
int output_multiplier = 0;
int output_shift = 0;
- ARM_COMPUTE_RETURN_ON_ERROR(quantization::calculate_quantized_multiplier(multiplier, &output_multiplier, &output_shift));
+ ARM_COMPUTE_RETURN_ON_ERROR(
+ quantization::calculate_quantized_multiplier(multiplier, &output_multiplier, &output_shift));
}
return Status{};
@@ -171,30 +190,48 @@
_type = CLKernelType::DEPTHWISE;
}
-void CLDepthwiseConvolutionLayerNativeKernel::configure(ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output,
- const DWCComputeKernelInfo &dwc_info, const ConvolutionInfo &conv_info,
- const ICLTensor *output_multipliers, const ICLTensor *output_shifts)
+void CLDepthwiseConvolutionLayerNativeKernel::configure(ICLTensor *input,
+ const ICLTensor *weights,
+ const ICLTensor *biases,
+ ICLTensor *output,
+ const DWCComputeKernelInfo &dwc_info,
+ const ConvolutionInfo &conv_info,
+ const ICLTensor *output_multipliers,
+ const ICLTensor *output_shifts)
{
- configure(CLKernelLibrary::get().get_compile_context(), input, weights, biases, output, dwc_info, conv_info, output_multipliers, output_shifts);
+ configure(CLKernelLibrary::get().get_compile_context(), input, weights, biases, output, dwc_info, conv_info,
+ output_multipliers, output_shifts);
}
-void CLDepthwiseConvolutionLayerNativeKernel::configure(const CLCompileContext &compile_context, ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output,
- const DWCComputeKernelInfo &dwc_info, const ConvolutionInfo &conv_info,
- const ICLTensor *output_multipliers, const ICLTensor *output_shifts)
+void CLDepthwiseConvolutionLayerNativeKernel::configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ const ICLTensor *weights,
+ const ICLTensor *biases,
+ ICLTensor *output,
+ const DWCComputeKernelInfo &dwc_info,
+ const ConvolutionInfo &conv_info,
+ const ICLTensor *output_multipliers,
+ const ICLTensor *output_shifts)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights);
- if(output == nullptr)
+ if (output == nullptr)
{
// In-place
output = input;
}
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), weights->info(), (biases != nullptr) ? biases->info() : nullptr, output->info(),
- dwc_info, conv_info, (output_multipliers != nullptr) ? output_multipliers->info() : nullptr, (output_shifts != nullptr) ? output_shifts->info() : nullptr));
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(
+ input->info(), weights->info(), (biases != nullptr) ? biases->info() : nullptr, output->info(), dwc_info,
+ conv_info, (output_multipliers != nullptr) ? output_multipliers->info() : nullptr,
+ (output_shifts != nullptr) ? output_shifts->info() : nullptr));
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
- const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_depthwise_convolution_shape(*(input->info()), *(weights->info()), conv_info);
- auto_init_if_empty(*(output->info()), input->info()->clone()->set_tensor_shape(output_shape).set_quantization_info(output->info()->quantization_info()));
+ const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_depthwise_convolution_shape(
+ *(input->info()), *(weights->info()), conv_info);
+ auto_init_if_empty(*(output->info()), input->info()
+ ->clone()
+ ->set_tensor_shape(output_shape)
+ .set_quantization_info(output->info()->quantization_info()));
_input = input;
_output = output;
@@ -214,12 +251,12 @@
CLBuildOptions build_opts;
// Update the padding for the input/weights tensor if we can export to cl_image
- if(_export_input_to_cl_image)
+ if (_export_input_to_cl_image)
{
arm_compute::opencl::kernels::gemm::update_padding_for_cl_image(input->info());
}
- if(_export_weights_to_cl_image)
+ if (_export_weights_to_cl_image)
{
arm_compute::opencl::kernels::gemm::update_padding_for_cl_image(weights->info());
}
@@ -229,9 +266,10 @@
const auto act_function = conv_info.act_info.activation();
const auto dst_data_type = _output->info()->data_type();
- if((gpu_target != GPUTarget::G71 && (gpu_target & GPUTarget::GPU_ARCH_MASK) == GPUTarget::BIFROST)
- && (act_function == ActivationLayerInfo::ActivationFunction::BOUNDED_RELU || act_function == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
- && (dst_data_type == DataType::F32 || dst_data_type == DataType::F16))
+ if ((gpu_target != GPUTarget::G71 && (gpu_target & GPUTarget::GPU_ARCH_MASK) == GPUTarget::BIFROST) &&
+ (act_function == ActivationLayerInfo::ActivationFunction::BOUNDED_RELU ||
+ act_function == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU) &&
+ (dst_data_type == DataType::F32 || dst_data_type == DataType::F16))
{
// -cl-fast-relaxed-math also sets -cl-finite-math-only and -cl-unsafe-math-optimizations
// to disable -cl-finite-math-only, we only include -cl-unsafe-math-optimizations
@@ -268,23 +306,24 @@
build_opts.add_option("-DN0=" + support::cpp11::to_string(n0));
build_opts.add_option("-DM0=" + support::cpp11::to_string(m0));
build_opts.add_option("-DM0_A=" + support::cpp11::to_string(_weights->info()->dimension(1) + m0 - 1));
- build_opts.add_option_if_else(conv_info.depth_multiplier > 1, "-DN0_A=1", "-DN0_A=" + support::cpp11::to_string(n0));
+ build_opts.add_option_if_else(conv_info.depth_multiplier > 1, "-DN0_A=1",
+ "-DN0_A=" + support::cpp11::to_string(n0));
build_opts.add_option("-DPARTIAL_N0=" + support::cpp11::to_string(_output->info()->dimension(0) % n0));
build_opts.add_option_if(_input->info()->num_dimensions() > 3, "-DBATCHED_EXECUTION");
// Force unroll with pragma when any of the following values exceed the maximum number of manual unroll
- set_unroll_with_pragma(build_opts, { static_cast<int>(_weights->info()->dimension(1) + m0 - 1),
- static_cast<int>(_weights->info()->dimension(1)),
- static_cast<int>(_weights->info()->dimension(2))
- });
+ set_unroll_with_pragma(build_opts, {static_cast<int>(_weights->info()->dimension(1) + m0 - 1),
+ static_cast<int>(_weights->info()->dimension(1)),
+ static_cast<int>(_weights->info()->dimension(2))});
- if(biases != nullptr)
+ if (biases != nullptr)
{
build_opts.add_option(std::string("-DHAS_BIAS"));
- build_opts.add_option(std::string("-DBIA_DATA_TYPE=" + get_cl_type_from_data_type(biases->info()->data_type())));
+ build_opts.add_option(
+ std::string("-DBIA_DATA_TYPE=" + get_cl_type_from_data_type(biases->info()->data_type())));
}
- if(_is_quantized)
+ if (_is_quantized)
{
kernel_name = "dwc_native_quantized_nhwc";
const UniformQuantizationInfo iqinfo = input->info()->quantization_info().uniform();
@@ -306,13 +345,17 @@
build_opts.add_option("-DDST_OFFSET=" + support::cpp11::to_string(oqinfo.offset));
build_opts.add_option("-DZERO_VALUE=" + support::cpp11::to_string(zero_value_s32));
build_opts.add_option("-DACC_DATA_TYPE=" + get_cl_type_from_data_type(DataType::S32));
- build_opts.add_option("-DDST_MULTIPLIERS_DATA_TYPE=" + get_cl_type_from_data_type(_output_multipliers->info()->data_type()));
- build_opts.add_option("-DDST_SHIFTS_DATA_TYPE=" + get_cl_type_from_data_type(_output_shifts->info()->data_type()));
- build_opts.add_option_if_else(weights->info()->data_type() == DataType::QSYMM8_PER_CHANNEL, "-DQUANTIZATION_TYPE=PER_CHANNEL", "-DQUANTIZATION_TYPE=PER_TENSOR");
+ build_opts.add_option("-DDST_MULTIPLIERS_DATA_TYPE=" +
+ get_cl_type_from_data_type(_output_multipliers->info()->data_type()));
+ build_opts.add_option("-DDST_SHIFTS_DATA_TYPE=" +
+ get_cl_type_from_data_type(_output_shifts->info()->data_type()));
+ build_opts.add_option_if_else(weights->info()->data_type() == DataType::QSYMM8_PER_CHANNEL,
+ "-DQUANTIZATION_TYPE=PER_CHANNEL", "-DQUANTIZATION_TYPE=PER_TENSOR");
// Note: We expect the input and output tensors to always adopt a per-tensor quantization approach
int a_val{};
int b_val{};
- std::tie(b_val, a_val) = get_quantized_activation_min_max(conv_info.act_info, input->info()->data_type(), oqinfo);
+ std::tie(b_val, a_val) =
+ get_quantized_activation_min_max(conv_info.act_info, input->info()->data_type(), oqinfo);
build_opts.add_option_if(conv_info.act_info.enabled(), "-DA_VAL=" + support::cpp11::to_string(a_val));
build_opts.add_option_if(conv_info.act_info.enabled(), "-DB_VAL=" + support::cpp11::to_string(b_val));
@@ -321,8 +364,10 @@
{
kernel_name = "dwc_native_fp_nhwc";
build_opts.add_option("-DACC_DATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
- build_opts.add_option_if(conv_info.act_info.enabled(), "-DA_VAL=" + float_to_string_with_full_precision(conv_info.act_info.a()));
- build_opts.add_option_if(conv_info.act_info.enabled(), "-DB_VAL=" + float_to_string_with_full_precision(conv_info.act_info.b()));
+ build_opts.add_option_if(conv_info.act_info.enabled(),
+ "-DA_VAL=" + float_to_string_with_full_precision(conv_info.act_info.a()));
+ build_opts.add_option_if(conv_info.act_info.enabled(),
+ "-DB_VAL=" + float_to_string_with_full_precision(conv_info.act_info.b()));
}
Window win = calculate_max_window(*(output->info()), Steps(n0, m0));
@@ -350,10 +395,17 @@
_config_id += string_from_data_type(input->info()->data_type());
}
-Status CLDepthwiseConvolutionLayerNativeKernel::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output,
- const DWCComputeKernelInfo &dwc_info, const ConvolutionInfo &conv_info, const ITensorInfo *output_multipliers, const ITensorInfo *output_shifts)
+Status CLDepthwiseConvolutionLayerNativeKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *weights,
+ const ITensorInfo *biases,
+ const ITensorInfo *output,
+ const DWCComputeKernelInfo &dwc_info,
+ const ConvolutionInfo &conv_info,
+ const ITensorInfo *output_multipliers,
+ const ITensorInfo *output_shifts)
{
- ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, weights, biases, output, dwc_info, conv_info, output_multipliers, output_shifts));
+ ARM_COMPUTE_RETURN_ON_ERROR(
+ validate_arguments(input, weights, biases, output, dwc_info, conv_info, output_multipliers, output_shifts));
return Status{};
}
@@ -370,47 +422,52 @@
cl::Image2D input_cl_image;
cl::Image2D weights_cl_image;
- if(_export_input_to_cl_image || _export_weights_to_cl_image)
+ if (_export_input_to_cl_image || _export_weights_to_cl_image)
{
// Export cl_buffer to cl_image
- if(_export_input_to_cl_image)
+ if (_export_input_to_cl_image)
{
- const size_t image_w = _input->info()->dimension(0) / 4;
- const size_t image_h = _input->info()->dimension(1) * _input->info()->dimension(2) * _input->info()->dimension(3);
+ const size_t image_w = _input->info()->dimension(0) / 4;
+ const size_t image_h =
+ _input->info()->dimension(1) * _input->info()->dimension(2) * _input->info()->dimension(3);
const TensorShape shape2d(image_w, image_h);
const size_t image_row_pitch = _input->info()->strides_in_bytes()[1];
- input_cl_image = create_image2d_from_buffer(CLKernelLibrary::get().context(), _input->cl_buffer(), shape2d, _input->info()->data_type(), image_row_pitch, CLImage2DType::ReadOnly);
+ input_cl_image =
+ create_image2d_from_buffer(CLKernelLibrary::get().context(), _input->cl_buffer(), shape2d,
+ _input->info()->data_type(), image_row_pitch, CLImage2DType::ReadOnly);
}
- if(_export_weights_to_cl_image)
+ if (_export_weights_to_cl_image)
{
- const size_t image_w = _weights->info()->dimension(0) / 4;
- const size_t image_h = _weights->info()->dimension(1) * _weights->info()->dimension(2) * _weights->info()->dimension(3);
+ const size_t image_w = _weights->info()->dimension(0) / 4;
+ const size_t image_h =
+ _weights->info()->dimension(1) * _weights->info()->dimension(2) * _weights->info()->dimension(3);
const TensorShape shape2d(image_w, image_h);
const size_t image_row_pitch = _weights->info()->strides_in_bytes()[1];
- weights_cl_image = create_image2d_from_buffer(CLKernelLibrary::get().context(), _weights->cl_buffer(), shape2d, _weights->info()->data_type(), image_row_pitch,
- CLImage2DType::ReadOnly);
+ weights_cl_image =
+ create_image2d_from_buffer(CLKernelLibrary::get().context(), _weights->cl_buffer(), shape2d,
+ _weights->info()->data_type(), image_row_pitch, CLImage2DType::ReadOnly);
}
}
unsigned int idx = 0;
- if(_export_input_to_cl_image)
+ if (_export_input_to_cl_image)
{
_kernel.setArg(idx++, input_cl_image);
}
add_4d_tensor_nhwc_argument(idx, _input);
add_4d_tensor_nhwc_argument(idx, _output);
- if(_export_weights_to_cl_image)
+ if (_export_weights_to_cl_image)
{
_kernel.setArg(idx++, weights_cl_image);
}
add_4d_tensor_nhwc_argument(idx, _weights);
- if(_is_quantized)
+ if (_is_quantized)
{
add_1D_tensor_argument(idx, _output_multipliers, slice);
add_1D_tensor_argument(idx, _output_shifts, slice);
}
- if(_biases != nullptr)
+ if (_biases != nullptr)
{
add_1D_tensor_argument(idx, _biases, slice);
}
diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.h b/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.h
index 8eee7b2..d34a662 100644
--- a/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.h
+++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.h
@@ -24,11 +24,11 @@
#ifndef ARM_COMPUTE_CLDEPTHWISECONVOLUTIONLAYERNATIVEKERNEL_H
#define ARM_COMPUTE_CLDEPTHWISECONVOLUTIONLAYERNATIVEKERNEL_H
-#include "src/core/CL/ICLKernel.h"
-
#include "arm_compute/core/KernelDescriptors.h"
#include "arm_compute/function_info/ConvolutionInfo.h"
+#include "src/core/CL/ICLKernel.h"
+
namespace arm_compute
{
class ICLTensor;
@@ -74,15 +74,28 @@
* * no padding
* * no change of data layout after configure
*/
- void configure(const CLCompileContext &compile_context, ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const DWCComputeKernelInfo &dwc_info,
- const ConvolutionInfo &conv_info, const ICLTensor *output_multipliers = nullptr, const ICLTensor *output_shifts = nullptr);
+ void configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ const ICLTensor *weights,
+ const ICLTensor *biases,
+ ICLTensor *output,
+ const DWCComputeKernelInfo &dwc_info,
+ const ConvolutionInfo &conv_info,
+ const ICLTensor *output_multipliers = nullptr,
+ const ICLTensor *output_shifts = nullptr);
/** Static function to check if given info will lead to a valid configuration of @ref CLDepthwiseConvolutionLayerNativeKernel
*
* Similar to @ref CLDepthwiseConvolutionLayerNativeKernel::configure()
*/
- void configure(ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const DWCComputeKernelInfo &dwc_info,
- const ConvolutionInfo &conv_info, const ICLTensor *output_multipliers = nullptr, const ICLTensor *output_shifts = nullptr);
+ void configure(ICLTensor *input,
+ const ICLTensor *weights,
+ const ICLTensor *biases,
+ ICLTensor *output,
+ const DWCComputeKernelInfo &dwc_info,
+ const ConvolutionInfo &conv_info,
+ const ICLTensor *output_multipliers = nullptr,
+ const ICLTensor *output_shifts = nullptr);
/** Static function to check if given info will lead to a valid configuration of @ref CLDepthwiseConvolutionLayerNativeKernel
*
@@ -90,23 +103,29 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const DWCComputeKernelInfo &dwc_info,
- const ConvolutionInfo &conv_info, const ITensorInfo *output_multipliers = nullptr, const ITensorInfo *output_shifts = nullptr);
+ static Status validate(const ITensorInfo *input,
+ const ITensorInfo *weights,
+ const ITensorInfo *biases,
+ const ITensorInfo *output,
+ const DWCComputeKernelInfo &dwc_info,
+ const ConvolutionInfo &conv_info,
+ const ITensorInfo *output_multipliers = nullptr,
+ const ITensorInfo *output_shifts = nullptr);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
private:
- const ICLTensor *_input {};
+ const ICLTensor *_input{};
const ICLTensor *_weights{};
const ICLTensor *_biases{};
ICLTensor *_output{};
- unsigned int _depth_multiplier{ 0 };
+ unsigned int _depth_multiplier{0};
const ICLTensor *_output_multipliers{};
const ICLTensor *_output_shifts{};
- bool _export_input_to_cl_image{ false };
- bool _export_weights_to_cl_image{ true };
- bool _is_quantized{ false };
+ bool _export_input_to_cl_image{false};
+ bool _export_weights_to_cl_image{true};
+ bool _is_quantized{false};
};
} // namespace arm_compute
#endif /*ARM_COMPUTE_CLDEPTHWISECONVOLUTIONLAYERNATIVEKERNEL_H */
diff --git a/src/core/CL/kernels/CLFFTDigitReverseKernel.cpp b/src/core/CL/kernels/CLFFTDigitReverseKernel.cpp
index 9b514ed..3d8f875 100644
--- a/src/core/CL/kernels/CLFFTDigitReverseKernel.cpp
+++ b/src/core/CL/kernels/CLFFTDigitReverseKernel.cpp
@@ -28,6 +28,7 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -37,17 +38,20 @@
{
namespace
{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *idx, const FFTDigitReverseKernelInfo &config)
+Status validate_arguments(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const ITensorInfo *idx,
+ const FFTDigitReverseKernelInfo &config)
{
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(input, DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON(input->num_channels() != 1 && input->num_channels() != 2);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(idx, 1, DataType::U32);
- ARM_COMPUTE_RETURN_ERROR_ON(std::set<unsigned int>({ 0, 1 }).count(config.axis) == 0);
+ ARM_COMPUTE_RETURN_ERROR_ON(std::set<unsigned int>({0, 1}).count(config.axis) == 0);
ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[config.axis] != idx->tensor_shape().x());
// Checks performed when output is configured
- if((output != nullptr) && (output->total_size() != 0))
+ if ((output != nullptr) && (output->total_size() != 0))
{
ARM_COMPUTE_RETURN_ERROR_ON(output->num_channels() != 2);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
@@ -57,7 +61,10 @@
return Status{};
}
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, ITensorInfo *idx, const FFTDigitReverseKernelInfo &config)
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input,
+ ITensorInfo *output,
+ ITensorInfo *idx,
+ const FFTDigitReverseKernelInfo &config)
{
ARM_COMPUTE_UNUSED(idx, config);
@@ -69,21 +76,27 @@
}
} // namespace
-CLFFTDigitReverseKernel::CLFFTDigitReverseKernel()
- : _input(nullptr), _output(nullptr), _idx(nullptr)
+CLFFTDigitReverseKernel::CLFFTDigitReverseKernel() : _input(nullptr), _output(nullptr), _idx(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
-void CLFFTDigitReverseKernel::configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *idx, const FFTDigitReverseKernelInfo &config)
+void CLFFTDigitReverseKernel::configure(const ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *idx,
+ const FFTDigitReverseKernelInfo &config)
{
configure(CLKernelLibrary::get().get_compile_context(), input, output, idx, config);
}
-void CLFFTDigitReverseKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const ICLTensor *idx, const FFTDigitReverseKernelInfo &config)
+void CLFFTDigitReverseKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *idx,
+ const FFTDigitReverseKernelInfo &config)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, idx);
- auto padding_info = get_padding_info({ input, output, idx });
+ auto padding_info = get_padding_info({input, output, idx});
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), idx->info(), config));
_input = input;
@@ -114,10 +127,14 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLFFTDigitReverseKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *idx, const FFTDigitReverseKernelInfo &config)
+Status CLFFTDigitReverseKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const ITensorInfo *idx,
+ const FFTDigitReverseKernelInfo &config)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, idx, config));
- ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get(), idx->clone().get(), config).first);
+ ARM_COMPUTE_RETURN_ON_ERROR(
+ validate_and_configure_window(input->clone().get(), output->clone().get(), idx->clone().get(), config).first);
return Status{};
}
@@ -137,7 +154,6 @@
add_3D_tensor_argument(idx, _output, slice);
add_1D_tensor_argument(idx, _idx, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(collapsed.slide_window_slice_3D(slice));
+ } while (collapsed.slide_window_slice_3D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLFFTDigitReverseKernel.h b/src/core/CL/kernels/CLFFTDigitReverseKernel.h
index e5583a4..fdd1bcc 100644
--- a/src/core/CL/kernels/CLFFTDigitReverseKernel.h
+++ b/src/core/CL/kernels/CLFFTDigitReverseKernel.h
@@ -24,10 +24,10 @@
#ifndef ARM_COMPUTE_CLFFTDIGITREVERSEKERNEL_H
#define ARM_COMPUTE_CLFFTDIGITREVERSEKERNEL_H
-#include "src/core/CL/ICLKernel.h"
-
#include "arm_compute/core/KernelDescriptors.h"
+#include "src/core/CL/ICLKernel.h"
+
namespace arm_compute
{
// Forward declarations
@@ -56,7 +56,8 @@
* @param[in] idx Digit reverse index tensor. Data type supported: U32
* @param[in] config Kernel configuration.
*/
- void configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *idx, const FFTDigitReverseKernelInfo &config);
+ void
+ configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *idx, const FFTDigitReverseKernelInfo &config);
/** Set the input and output tensors.
*
* @param[in] compile_context The compile context to be used.
@@ -65,7 +66,11 @@
* @param[in] idx Digit reverse index tensor. Data type supported: U32
* @param[in] config Kernel configuration.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const ICLTensor *idx, const FFTDigitReverseKernelInfo &config);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *idx,
+ const FFTDigitReverseKernelInfo &config);
/** Static function to check if given info will lead to a valid configuration of @ref CLFFTDigitReverseKernel
*
* @param[in] input Source tensor info. Data types supported: F16/F32.
@@ -75,7 +80,10 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *idx, const FFTDigitReverseKernelInfo &config);
+ static Status validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const ITensorInfo *idx,
+ const FFTDigitReverseKernelInfo &config);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLFFTRadixStageKernel.cpp b/src/core/CL/kernels/CLFFTRadixStageKernel.cpp
index 95f4b64..3729e6b 100644
--- a/src/core/CL/kernels/CLFFTRadixStageKernel.cpp
+++ b/src/core/CL/kernels/CLFFTRadixStageKernel.cpp
@@ -29,6 +29,7 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -46,11 +47,11 @@
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 2, DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON(CLFFTRadixStageKernel::supported_radix().count(config.radix) == 0);
- ARM_COMPUTE_RETURN_ERROR_ON(std::set<unsigned int>({ 0, 1 }).count(config.axis) == 0);
+ ARM_COMPUTE_RETURN_ERROR_ON(std::set<unsigned int>({0, 1}).count(config.axis) == 0);
ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[config.axis] % config.radix);
// Checks performed when output is configured
- if((output != nullptr) && (output->total_size() != 0))
+ if ((output != nullptr) && (output->total_size() != 0))
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
@@ -59,9 +60,10 @@
return Status{};
}
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, const FFTRadixStageKernelInfo &config)
+std::pair<Status, Window>
+validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, const FFTRadixStageKernelInfo &config)
{
- if(output != nullptr)
+ if (output != nullptr)
{
auto_init_if_empty(*output, *input);
}
@@ -76,8 +78,7 @@
}
} // namespace
-CLFFTRadixStageKernel::CLFFTRadixStageKernel()
- : _input(nullptr), _output(nullptr), _run_in_place(false)
+CLFFTRadixStageKernel::CLFFTRadixStageKernel() : _input(nullptr), _output(nullptr), _run_in_place(false)
{
_type = CLKernelType::ELEMENTWISE;
}
@@ -87,11 +88,15 @@
configure(CLKernelLibrary::get().get_compile_context(), input, output, config);
}
-void CLFFTRadixStageKernel::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, const FFTRadixStageKernelInfo &config)
+void CLFFTRadixStageKernel::configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ ICLTensor *output,
+ const FFTRadixStageKernelInfo &config)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input);
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (output != nullptr) ? output->info() : nullptr, config));
- auto padding_info = get_padding_info({ input, output });
+ ARM_COMPUTE_ERROR_THROW_ON(
+ validate_arguments(input->info(), (output != nullptr) ? output->info() : nullptr, config));
+ auto padding_info = get_padding_info({input, output});
_input = input;
_output = output;
@@ -110,11 +115,12 @@
_kernel = create_kernel(compile_context, kernel_name, build_opts.options());
// Set static arguments if not the first stage
- if(!config.is_first_stage)
+ if (!config.is_first_stage)
{
const unsigned int Ni = config.Nx * config.radix;
const float exp_const = (-2.0 * M_PI) / static_cast<float>(Ni);
- unsigned int idx = (1 + (_run_in_place ? 0 : 1)) * num_arguments_per_3D_tensor(); // Skip the input and output parameters
+ unsigned int idx =
+ (1 + (_run_in_place ? 0 : 1)) * num_arguments_per_3D_tensor(); // Skip the input and output parameters
_kernel.setArg<cl_uint>(idx++, config.Nx);
_kernel.setArg<cl_uint>(idx++, Ni);
_kernel.setArg<cl_float>(idx, exp_const);
@@ -136,21 +142,22 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLFFTRadixStageKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const FFTRadixStageKernelInfo &config)
+Status CLFFTRadixStageKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const FFTRadixStageKernelInfo &config)
{
const bool run_in_place = (output == nullptr) || (output == input);
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, config));
- ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(),
- (run_in_place) ? nullptr : output->clone().get(),
- config)
- .first);
+ ARM_COMPUTE_RETURN_ON_ERROR(
+ validate_and_configure_window(input->clone().get(), (run_in_place) ? nullptr : output->clone().get(), config)
+ .first);
return Status{};
}
std::set<unsigned int> CLFFTRadixStageKernel::supported_radix()
{
- return std::set<unsigned int> { 2, 3, 4, 5, 7, 8 };
+ return std::set<unsigned int>{2, 3, 4, 5, 7, 8};
}
void CLFFTRadixStageKernel::run(const Window &window, cl::CommandQueue &queue)
@@ -165,12 +172,11 @@
{
unsigned int idx = 0;
add_3D_tensor_argument(idx, _input, slice);
- if(!_run_in_place)
+ if (!_run_in_place)
{
add_3D_tensor_argument(idx, _output, slice);
}
enqueue(queue, *this, slice, lws_hint());
- }
- while(collapsed.slide_window_slice_3D(slice));
+ } while (collapsed.slide_window_slice_3D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLFFTRadixStageKernel.h b/src/core/CL/kernels/CLFFTRadixStageKernel.h
index 9bb310d..de80bfc 100644
--- a/src/core/CL/kernels/CLFFTRadixStageKernel.h
+++ b/src/core/CL/kernels/CLFFTRadixStageKernel.h
@@ -24,10 +24,10 @@
#ifndef ARM_COMPUTE_CLFFTRADIXSTAGEKERNEL_H
#define ARM_COMPUTE_CLFFTRADIXSTAGEKERNEL_H
-#include "src/core/CL/ICLKernel.h"
-
#include "arm_compute/core/KernelDescriptors.h"
+#include "src/core/CL/ICLKernel.h"
+
#include <set>
namespace arm_compute
@@ -69,7 +69,10 @@
* @param[out] output Destination tensor. Can be nullptr. Data type supported: same as @p input
* @param[in] config FFT descriptor metadata.
*/
- void configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, const FFTRadixStageKernelInfo &config);
+ void configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ ICLTensor *output,
+ const FFTRadixStageKernelInfo &config);
/** Static function to check if given info will lead to a valid configuration of @ref CLFFTRadixStageKernel
*
* @param[in] input Source tensor info. Data types supported: F16/F32.
diff --git a/src/core/CL/kernels/CLFFTScaleKernel.cpp b/src/core/CL/kernels/CLFFTScaleKernel.cpp
index 8a714d7..be6e16b 100644
--- a/src/core/CL/kernels/CLFFTScaleKernel.cpp
+++ b/src/core/CL/kernels/CLFFTScaleKernel.cpp
@@ -28,6 +28,7 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -43,7 +44,7 @@
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 2, DataType::F16, DataType::F32);
// Checks performed when output is configured
- if((output != nullptr) && (output->total_size() != 0))
+ if ((output != nullptr) && (output->total_size() != 0))
{
ARM_COMPUTE_RETURN_ERROR_ON(output->num_channels() != 1 && output->num_channels() != 2);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
@@ -54,8 +55,7 @@
}
} // namespace
-CLFFTScaleKernel::CLFFTScaleKernel()
- : _input(nullptr), _output(nullptr), _run_in_place(false)
+CLFFTScaleKernel::CLFFTScaleKernel() : _input(nullptr), _output(nullptr), _run_in_place(false)
{
_type = CLKernelType::ELEMENTWISE;
}
@@ -65,11 +65,14 @@
configure(CLKernelLibrary::get().get_compile_context(), input, output, config);
}
-void CLFFTScaleKernel::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, const FFTScaleKernelInfo &config)
+void CLFFTScaleKernel::configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ ICLTensor *output,
+ const FFTScaleKernelInfo &config)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (output != nullptr) ? output->info() : nullptr));
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
_input = input;
_output = output;
@@ -78,20 +81,22 @@
// Create kernel
CLBuildOptions build_opts;
build_opts.add_option_if(_run_in_place, "-DIN_PLACE");
- build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(output != nullptr ? output->info()->num_channels() : input->info()->num_channels()));
+ build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(output != nullptr ? output->info()->num_channels()
+ : input->info()->num_channels()));
build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
build_opts.add_option_if(config.conjugate, "-DCONJ");
std::string kernel_name = "fft_scale_conj";
_kernel = create_kernel(compile_context, kernel_name, build_opts.options());
// Set static arguments
- unsigned int idx = (1 + (_run_in_place ? 0 : 1)) * num_arguments_per_3D_tensor(); // Skip the input and output parameters
+ unsigned int idx =
+ (1 + (_run_in_place ? 0 : 1)) * num_arguments_per_3D_tensor(); // Skip the input and output parameters
_kernel.setArg<cl_float>(idx, config.scale);
// Configure kernel window
Window win = calculate_max_window(*input->info(), Steps());
- if(output != nullptr)
+ if (output != nullptr)
{
// Output auto inizialitation if not yet initialized
auto_init_if_empty(*output->info(), *input->info()->clone());
@@ -130,12 +135,11 @@
{
unsigned int idx = 0;
add_3D_tensor_argument(idx, _input, slice);
- if(!_run_in_place)
+ if (!_run_in_place)
{
add_3D_tensor_argument(idx, _output, slice);
}
enqueue(queue, *this, slice, lws_hint());
- }
- while(collapsed.slide_window_slice_3D(slice));
+ } while (collapsed.slide_window_slice_3D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLFFTScaleKernel.h b/src/core/CL/kernels/CLFFTScaleKernel.h
index cc518be..b995282 100644
--- a/src/core/CL/kernels/CLFFTScaleKernel.h
+++ b/src/core/CL/kernels/CLFFTScaleKernel.h
@@ -24,10 +24,10 @@
#ifndef ARM_COMPUTE_CLFFTSCALEKERNEL_H
#define ARM_COMPUTE_CLFFTSCALEKERNEL_H
-#include "src/core/CL/ICLKernel.h"
-
#include "arm_compute/core/KernelDescriptors.h"
+#include "src/core/CL/ICLKernel.h"
+
namespace arm_compute
{
// Forward declarations
@@ -63,7 +63,10 @@
* @param[out] output Destination tensor. Data type supported: same as @p input
* @param[in] config Kernel configuration
*/
- void configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, const FFTScaleKernelInfo &config);
+ void configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ ICLTensor *output,
+ const FFTScaleKernelInfo &config);
/** Static function to check if given info will lead to a valid configuration of @ref CLFFTScaleKernel
*
* @param[in] input Source tensor info. Data types supported: F16/F32.
diff --git a/src/core/CL/kernels/CLFillBorderKernel.cpp b/src/core/CL/kernels/CLFillBorderKernel.cpp
index fcd99a4..86bb502 100644
--- a/src/core/CL/kernels/CLFillBorderKernel.cpp
+++ b/src/core/CL/kernels/CLFillBorderKernel.cpp
@@ -31,14 +31,14 @@
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/StringUtils.h"
#include "arm_compute/core/Validate.h"
+
#include "src/core/helpers/WindowHelpers.h"
#include "support/Cast.h"
#include "support/StringSupport.h"
namespace arm_compute
{
-CLFillBorderKernel::CLFillBorderKernel()
- : ICLKernel(), _tensor(nullptr)
+CLFillBorderKernel::CLFillBorderKernel() : ICLKernel(), _tensor(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
@@ -56,27 +56,38 @@
ICLKernel::add_argument<T>(idx, static_cast<T>(value));
}
-void CLFillBorderKernel::configure(ICLTensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value)
+void CLFillBorderKernel::configure(ICLTensor *tensor,
+ BorderSize border_size,
+ BorderMode border_mode,
+ const PixelValue &constant_border_value)
{
configure(CLKernelLibrary::get().get_compile_context(), tensor, border_size, border_mode, constant_border_value);
}
-void CLFillBorderKernel::configure(const CLCompileContext &compile_context, ICLTensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value)
+void CLFillBorderKernel::configure(const CLCompileContext &compile_context,
+ ICLTensor *tensor,
+ BorderSize border_size,
+ BorderMode border_mode,
+ const PixelValue &constant_border_value)
{
_tensor = tensor;
configure(compile_context, tensor->info(), border_size, border_mode, constant_border_value);
}
-void CLFillBorderKernel::configure(const CLCompileContext &compile_context, ITensorInfo *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value)
+void CLFillBorderKernel::configure(const CLCompileContext &compile_context,
+ ITensorInfo *tensor,
+ BorderSize border_size,
+ BorderMode border_mode,
+ const PixelValue &constant_border_value)
{
ARM_COMPUTE_ERROR_ON(tensor == nullptr);
ARM_COMPUTE_ERROR_ON(tensor->num_channels() != 1);
- auto padding_info = get_padding_info({ tensor });
+ auto padding_info = get_padding_info({tensor});
border_size.limit(tensor->padding());
// If there is no border: early exit
- if(border_size.empty() || border_mode == BorderMode::UNDEFINED)
+ if (border_size.empty() || border_mode == BorderMode::UNDEFINED)
{
return;
}
@@ -98,25 +109,22 @@
_kernel = create_kernel(compile_context, kernel_name, build_opts.options());
// Create static kernel arguments
- const unsigned int valid_width = tensor->valid_region().shape[0];
- const unsigned int valid_height = tensor->valid_region().shape[1];
- const cl_int2 valid_region_coords =
- {
- {
- static_cast<cl_int>(tensor->valid_region().anchor[0]),
- static_cast<cl_int>(tensor->valid_region().anchor[1]),
- }
- };
- const unsigned int total_valid_width = border_size.left + valid_width + border_size.right;
+ const unsigned int valid_width = tensor->valid_region().shape[0];
+ const unsigned int valid_height = tensor->valid_region().shape[1];
+ const cl_int2 valid_region_coords = {{
+ static_cast<cl_int>(tensor->valid_region().anchor[0]),
+ static_cast<cl_int>(tensor->valid_region().anchor[1]),
+ }};
+ const unsigned int total_valid_width = border_size.left + valid_width + border_size.right;
// Set static kernel arguments
unsigned int idx = num_arguments_per_3D_tensor(); //Skip the tensor parameters
ICLKernel::add_argument<cl_uint>(idx, valid_width);
ICLKernel::add_argument<cl_uint>(idx, valid_height);
ICLKernel::add_argument<cl_int2>(idx, valid_region_coords);
- if(BorderMode::CONSTANT == border_mode)
+ if (BorderMode::CONSTANT == border_mode)
{
- switch(dt)
+ switch (dt)
{
case DataType::U8:
case DataType::QASYMM8:
@@ -175,12 +183,13 @@
void CLFillBorderKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
{
// Border mode undefined or border width == 0
- if(_kernel() == nullptr)
+ if (_kernel() == nullptr)
{
return;
}
- const auto tensor = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
+ const auto tensor =
+ utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window);
@@ -193,14 +202,13 @@
unsigned int idx = 0;
add_3D_tensor_argument(idx, tensor, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(collapsed.slide_window_slice_3D(slice));
+ } while (collapsed.slide_window_slice_3D(slice));
}
void CLFillBorderKernel::run(const Window &window, cl::CommandQueue &queue)
{
// Border mode undefined or border width == 0
- if(_kernel() == nullptr)
+ if (_kernel() == nullptr)
{
return;
}
@@ -216,7 +224,6 @@
unsigned int idx = 0;
add_3D_tensor_argument(idx, _tensor, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(collapsed.slide_window_slice_3D(slice));
+ } while (collapsed.slide_window_slice_3D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLFillBorderKernel.h b/src/core/CL/kernels/CLFillBorderKernel.h
index 7951f48..5782143 100644
--- a/src/core/CL/kernels/CLFillBorderKernel.h
+++ b/src/core/CL/kernels/CLFillBorderKernel.h
@@ -26,6 +26,7 @@
#include "arm_compute/core/PixelValue.h"
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -57,7 +58,11 @@
* @param[in] border_mode Border mode to use for the convolution.
* @param[in] constant_border_value (Optional) Constant value to use for borders if border_mode is set to CONSTANT.
*/
- void configure(const CLCompileContext &compile_context, ICLTensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value = PixelValue());
+ void configure(const CLCompileContext &compile_context,
+ ICLTensor *tensor,
+ BorderSize border_size,
+ BorderMode border_mode,
+ const PixelValue &constant_border_value = PixelValue());
/** Initialise the kernel's input, output and border mode.
*
* @param[in,out] tensor Tensor to process Data types supported: U8/QASYMM8/S8/QASYMM8_SIGNED/U16/S16/U32/S32/F16/F32.
@@ -65,7 +70,10 @@
* @param[in] border_mode Border mode to use for the convolution.
* @param[in] constant_border_value (Optional) Constant value to use for borders if border_mode is set to CONSTANT.
*/
- void configure(ICLTensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value = PixelValue());
+ void configure(ICLTensor *tensor,
+ BorderSize border_size,
+ BorderMode border_mode,
+ const PixelValue &constant_border_value = PixelValue());
/** Initialise the kernel's input, output and border mode.
*
* @param[in] compile_context The compile context to be used.
@@ -74,7 +82,11 @@
* @param[in] border_mode Border mode to use for the convolution.
* @param[in] constant_border_value (Optional) Constant value to use for borders if border_mode is set to CONSTANT.
*/
- void configure(const CLCompileContext &compile_context, ITensorInfo *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value = PixelValue());
+ void configure(const CLCompileContext &compile_context,
+ ITensorInfo *tensor,
+ BorderSize border_size,
+ BorderMode border_mode,
+ const PixelValue &constant_border_value = PixelValue());
/** Function to set the constant value on fill border kernel depending on type.
*
diff --git a/src/core/CL/kernels/CLFuseBatchNormalizationKernel.cpp b/src/core/CL/kernels/CLFuseBatchNormalizationKernel.cpp
index 68fe324..7da0679 100644
--- a/src/core/CL/kernels/CLFuseBatchNormalizationKernel.cpp
+++ b/src/core/CL/kernels/CLFuseBatchNormalizationKernel.cpp
@@ -30,20 +30,26 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
-
#include "support/StringSupport.h"
namespace arm_compute
{
namespace
{
-Status validate_arguments(const ITensorInfo *input_weights, const ITensorInfo *bn_mean, const ITensorInfo *bn_var,
- const ITensorInfo *fused_weights, const ITensorInfo *fused_bias,
- const ITensorInfo *input_bias, const ITensorInfo *bn_beta, const ITensorInfo *bn_gamma,
- float epsilon, FuseBatchNormalizationType fbn_type)
+Status validate_arguments(const ITensorInfo *input_weights,
+ const ITensorInfo *bn_mean,
+ const ITensorInfo *bn_var,
+ const ITensorInfo *fused_weights,
+ const ITensorInfo *fused_bias,
+ const ITensorInfo *input_bias,
+ const ITensorInfo *bn_beta,
+ const ITensorInfo *bn_gamma,
+ float epsilon,
+ FuseBatchNormalizationType fbn_type)
{
ARM_COMPUTE_UNUSED(epsilon);
ARM_COMPUTE_ERROR_ON_NULLPTR(input_weights, bn_mean, bn_var);
@@ -54,43 +60,44 @@
ARM_COMPUTE_RETURN_ERROR_ON(input_bias == nullptr && fused_bias == nullptr);
ARM_COMPUTE_RETURN_ERROR_ON(bn_mean->num_dimensions() > 1);
- if(fbn_type == FuseBatchNormalizationType::CONVOLUTION)
+ if (fbn_type == FuseBatchNormalizationType::CONVOLUTION)
{
ARM_COMPUTE_RETURN_ERROR_ON(input_weights->dimension(3) != bn_mean->dimension(0));
}
else
{
- const size_t channel_idx = get_data_layout_dimension_index(input_weights->data_layout(), DataLayoutDimension::CHANNEL);
+ const size_t channel_idx =
+ get_data_layout_dimension_index(input_weights->data_layout(), DataLayoutDimension::CHANNEL);
ARM_COMPUTE_RETURN_ERROR_ON(input_weights->dimension(channel_idx) != bn_mean->dimension(0));
}
// Validate bias
- if(input_bias != nullptr)
+ if (input_bias != nullptr)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(bn_mean, input_bias);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input_weights, input_bias);
}
// Validate beta
- if(bn_beta != nullptr)
+ if (bn_beta != nullptr)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(bn_mean, bn_beta);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input_weights, bn_beta);
}
// Validate gamma
- if(bn_gamma != nullptr)
+ if (bn_gamma != nullptr)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(bn_mean, bn_gamma);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input_weights, bn_gamma);
}
// Validate output weights
- if(fused_weights != nullptr && fused_weights->total_size() != 0)
+ if (fused_weights != nullptr && fused_weights->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input_weights, fused_weights);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input_weights, fused_weights);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input_weights, fused_weights);
}
// Validate output bias
- if(fused_bias != nullptr && fused_bias->total_size() != 0)
+ if (fused_bias != nullptr && fused_bias->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(bn_mean, fused_bias);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input_weights, fused_bias);
@@ -101,28 +108,52 @@
} // namespace
CLFuseBatchNormalizationKernel::CLFuseBatchNormalizationKernel()
- : _input_weights(nullptr), _input_bias(nullptr), _bn_mean(nullptr), _bn_var(nullptr), _bn_gamma(nullptr), _bn_beta(nullptr), _fused_weights(nullptr), _fused_bias(nullptr), _epsilon(),
- _run_in_place_weights(false), _run_in_place_bias(false)
+ : _input_weights(nullptr),
+ _input_bias(nullptr),
+ _bn_mean(nullptr),
+ _bn_var(nullptr),
+ _bn_gamma(nullptr),
+ _bn_beta(nullptr),
+ _fused_weights(nullptr),
+ _fused_bias(nullptr),
+ _epsilon(),
+ _run_in_place_weights(false),
+ _run_in_place_bias(false)
{
_type = CLKernelType::ELEMENTWISE;
}
-void CLFuseBatchNormalizationKernel::configure(const ICLTensor *input_weights, const ICLTensor *bn_mean, const ICLTensor *bn_var,
- ICLTensor *fused_weights, ICLTensor *fused_bias,
- const ICLTensor *input_bias, const ICLTensor *bn_beta, const ICLTensor *bn_gamma,
- float epsilon, FuseBatchNormalizationType fbn_type)
+void CLFuseBatchNormalizationKernel::configure(const ICLTensor *input_weights,
+ const ICLTensor *bn_mean,
+ const ICLTensor *bn_var,
+ ICLTensor *fused_weights,
+ ICLTensor *fused_bias,
+ const ICLTensor *input_bias,
+ const ICLTensor *bn_beta,
+ const ICLTensor *bn_gamma,
+ float epsilon,
+ FuseBatchNormalizationType fbn_type)
{
- configure(CLKernelLibrary::get().get_compile_context(), input_weights, bn_mean, bn_var, fused_weights, fused_bias, input_bias, bn_beta, bn_gamma, epsilon, fbn_type);
+ configure(CLKernelLibrary::get().get_compile_context(), input_weights, bn_mean, bn_var, fused_weights, fused_bias,
+ input_bias, bn_beta, bn_gamma, epsilon, fbn_type);
}
-void CLFuseBatchNormalizationKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input_weights, const ICLTensor *bn_mean, const ICLTensor *bn_var,
- ICLTensor *fused_weights, ICLTensor *fused_bias,
- const ICLTensor *input_bias, const ICLTensor *bn_beta, const ICLTensor *bn_gamma,
- float epsilon, FuseBatchNormalizationType fbn_type)
+void CLFuseBatchNormalizationKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input_weights,
+ const ICLTensor *bn_mean,
+ const ICLTensor *bn_var,
+ ICLTensor *fused_weights,
+ ICLTensor *fused_bias,
+ const ICLTensor *input_bias,
+ const ICLTensor *bn_beta,
+ const ICLTensor *bn_gamma,
+ float epsilon,
+ FuseBatchNormalizationType fbn_type)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input_weights, bn_mean, bn_var);
- auto padding_info = get_padding_info({ input_weights, bn_mean, bn_var, fused_weights, fused_bias, input_bias, bn_beta, bn_gamma });
+ auto padding_info =
+ get_padding_info({input_weights, bn_mean, bn_var, fused_weights, fused_bias, input_bias, bn_beta, bn_gamma});
_input_weights = input_weights;
_input_bias = input_bias;
@@ -135,28 +166,28 @@
_epsilon = epsilon;
_run_in_place_weights = (fused_weights == nullptr) || (fused_weights == input_weights);
- _run_in_place_bias = (input_bias != nullptr && fused_bias == nullptr) || (input_bias != nullptr && fused_bias == input_bias);
+ _run_in_place_bias =
+ (input_bias != nullptr && fused_bias == nullptr) || (input_bias != nullptr && fused_bias == input_bias);
// Auto initialize outputs
- if(_fused_weights != nullptr)
+ if (_fused_weights != nullptr)
{
// Output tensor auto initialization if not yet initialized
auto_init_if_empty(*_fused_weights->info(), *_input_weights->info()->clone());
}
- if(_fused_bias != nullptr)
+ if (_fused_bias != nullptr)
{
// Output tensor auto initialization if not yet initialized
auto_init_if_empty(*_fused_bias->info(), *_bn_mean->info()->clone());
}
// Validate arguments
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input_weights->info(), bn_mean->info(), bn_var->info(),
- (fused_weights != nullptr) ? fused_weights->info() : nullptr,
- (fused_bias != nullptr) ? fused_bias->info() : nullptr,
- (input_bias != nullptr) ? input_bias->info() : nullptr,
- (bn_beta != nullptr) ? bn_beta->info() : nullptr,
- (bn_gamma != nullptr) ? bn_gamma->info() : nullptr,
- epsilon, fbn_type));
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(
+ input_weights->info(), bn_mean->info(), bn_var->info(),
+ (fused_weights != nullptr) ? fused_weights->info() : nullptr,
+ (fused_bias != nullptr) ? fused_bias->info() : nullptr, (input_bias != nullptr) ? input_bias->info() : nullptr,
+ (bn_beta != nullptr) ? bn_beta->info() : nullptr, (bn_gamma != nullptr) ? bn_gamma->info() : nullptr, epsilon,
+ fbn_type));
// Configure kernel window
Window win = calculate_max_window(*input_weights->info());
@@ -165,7 +196,8 @@
// Set build options
CLBuildOptions build_opts;
build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input_weights->info()->data_type()));
- build_opts.add_option_if(fbn_type == FuseBatchNormalizationType::CONVOLUTION, "-DDIM2=" + support::cpp11::to_string(input_weights->info()->dimension(2)));
+ build_opts.add_option_if(fbn_type == FuseBatchNormalizationType::CONVOLUTION,
+ "-DDIM2=" + support::cpp11::to_string(input_weights->info()->dimension(2)));
build_opts.add_option("-DEPSILON=" + float_to_string_with_full_precision(epsilon));
build_opts.add_option_if(_input_weights->info()->data_layout() == DataLayout::NHWC, "-DNHWC");
build_opts.add_option_if(_run_in_place_weights, "-DIN_PLACE_W");
@@ -180,12 +212,19 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLFuseBatchNormalizationKernel::validate(const ITensorInfo *input_weights, const ITensorInfo *bn_mean, const ITensorInfo *bn_var,
- const ITensorInfo *fused_weights, const ITensorInfo *fused_bias,
- const ITensorInfo *input_bias, const ITensorInfo *bn_beta, const ITensorInfo *bn_gamma,
- float epsilon, FuseBatchNormalizationType fbn_type)
+Status CLFuseBatchNormalizationKernel::validate(const ITensorInfo *input_weights,
+ const ITensorInfo *bn_mean,
+ const ITensorInfo *bn_var,
+ const ITensorInfo *fused_weights,
+ const ITensorInfo *fused_bias,
+ const ITensorInfo *input_bias,
+ const ITensorInfo *bn_beta,
+ const ITensorInfo *bn_gamma,
+ float epsilon,
+ FuseBatchNormalizationType fbn_type)
{
- ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input_weights, bn_mean, bn_var, fused_weights, fused_bias, input_bias, bn_beta, bn_gamma, epsilon, fbn_type));
+ ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input_weights, bn_mean, bn_var, fused_weights, fused_bias,
+ input_bias, bn_beta, bn_gamma, epsilon, fbn_type));
return Status{};
}
@@ -202,25 +241,25 @@
// Add kernel arguments
unsigned int idx = 0;
add_3D_tensor_argument(idx, _input_weights, slice_3d);
- if(_input_bias != nullptr)
+ if (_input_bias != nullptr)
{
add_1D_tensor_argument(idx, _input_bias, slice_1d);
}
add_1D_tensor_argument(idx, _bn_mean, slice_1d);
add_1D_tensor_argument(idx, _bn_var, slice_1d);
- if(!_run_in_place_weights)
+ if (!_run_in_place_weights)
{
add_3D_tensor_argument(idx, _fused_weights, slice_3d);
}
- if(!_run_in_place_bias)
+ if (!_run_in_place_bias)
{
add_1D_tensor_argument(idx, _fused_bias, slice_1d);
}
- if(_bn_beta != nullptr)
+ if (_bn_beta != nullptr)
{
add_1D_tensor_argument(idx, _bn_beta, slice_1d);
}
- if(_bn_gamma != nullptr)
+ if (_bn_gamma != nullptr)
{
add_1D_tensor_argument(idx, _bn_gamma, slice_1d);
}
diff --git a/src/core/CL/kernels/CLFuseBatchNormalizationKernel.h b/src/core/CL/kernels/CLFuseBatchNormalizationKernel.h
index 78b1e74..76ec7a7 100644
--- a/src/core/CL/kernels/CLFuseBatchNormalizationKernel.h
+++ b/src/core/CL/kernels/CLFuseBatchNormalizationKernel.h
@@ -62,9 +62,16 @@
* @param[in] epsilon (Optional) Batch normalization layer epsilon parameter. Defaults to 0.001f.
* @param[in] fbn_type (Optional) Fused batch normalization type. Defaults to CONVOLUTION.
*/
- void configure(const ICLTensor *input_weights, const ICLTensor *bn_mean, const ICLTensor *bn_var, ICLTensor *fused_weights, ICLTensor *fused_bias,
- const ICLTensor *input_bias = nullptr, const ICLTensor *bn_beta = nullptr, const ICLTensor *bn_gamma = nullptr,
- float epsilon = 0.001f, FuseBatchNormalizationType fbn_type = FuseBatchNormalizationType::CONVOLUTION);
+ void configure(const ICLTensor *input_weights,
+ const ICLTensor *bn_mean,
+ const ICLTensor *bn_var,
+ ICLTensor *fused_weights,
+ ICLTensor *fused_bias,
+ const ICLTensor *input_bias = nullptr,
+ const ICLTensor *bn_beta = nullptr,
+ const ICLTensor *bn_gamma = nullptr,
+ float epsilon = 0.001f,
+ FuseBatchNormalizationType fbn_type = FuseBatchNormalizationType::CONVOLUTION);
/** Set the source, destination of the kernel
*
* @param[in] compile_context The compile context to be used.
@@ -81,9 +88,17 @@
* @param[in] epsilon (Optional) Batch normalization layer epsilon parameter. Defaults to 0.001f.
* @param[in] fbn_type (Optional) Fused batch normalization type. Defaults to CONVOLUTION.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input_weights, const ICLTensor *bn_mean, const ICLTensor *bn_var, ICLTensor *fused_weights, ICLTensor *fused_bias,
- const ICLTensor *input_bias = nullptr, const ICLTensor *bn_beta = nullptr, const ICLTensor *bn_gamma = nullptr,
- float epsilon = 0.001f, FuseBatchNormalizationType fbn_type = FuseBatchNormalizationType::CONVOLUTION);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input_weights,
+ const ICLTensor *bn_mean,
+ const ICLTensor *bn_var,
+ ICLTensor *fused_weights,
+ ICLTensor *fused_bias,
+ const ICLTensor *input_bias = nullptr,
+ const ICLTensor *bn_beta = nullptr,
+ const ICLTensor *bn_gamma = nullptr,
+ float epsilon = 0.001f,
+ FuseBatchNormalizationType fbn_type = FuseBatchNormalizationType::CONVOLUTION);
/** Static function to check if given info will lead to a valid configuration of @ref CLFuseBatchNormalizationKernel
*
* @param[in] input_weights Input weights tensor info for convolution or depthwise convolution layer. Data type supported: F16/F32. Data layout supported: NCHW, NHWC
@@ -101,10 +116,16 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input_weights, const ITensorInfo *bn_mean, const ITensorInfo *bn_var,
- const ITensorInfo *fused_weights, const ITensorInfo *fused_bias,
- const ITensorInfo *input_bias = nullptr, const ITensorInfo *bn_beta = nullptr, const ITensorInfo *bn_gamma = nullptr,
- float epsilon = 0.001f, FuseBatchNormalizationType fbn_type = FuseBatchNormalizationType::CONVOLUTION);
+ static Status validate(const ITensorInfo *input_weights,
+ const ITensorInfo *bn_mean,
+ const ITensorInfo *bn_var,
+ const ITensorInfo *fused_weights,
+ const ITensorInfo *fused_bias,
+ const ITensorInfo *input_bias = nullptr,
+ const ITensorInfo *bn_beta = nullptr,
+ const ITensorInfo *bn_gamma = nullptr,
+ float epsilon = 0.001f,
+ FuseBatchNormalizationType fbn_type = FuseBatchNormalizationType::CONVOLUTION);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLGatherKernel.cpp b/src/core/CL/kernels/CLGatherKernel.cpp
index 5495023..c11a189 100644
--- a/src/core/CL/kernels/CLGatherKernel.cpp
+++ b/src/core/CL/kernels/CLGatherKernel.cpp
@@ -22,8 +22,10 @@
* SOFTWARE.
*/
#include "src/core/CL/kernels/CLGatherKernel.h"
+
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
#include "support/StringSupport.h"
@@ -34,7 +36,8 @@
{
namespace
{
-inline Status validate_arguments(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, int axis)
+inline Status
+validate_arguments(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, int axis)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, indices, output);
const uint32_t actual_axis = wrap_around(axis, static_cast<int>(input->num_dimensions()));
@@ -43,11 +46,12 @@
ARM_COMPUTE_RETURN_ERROR_ON(actual_axis >= input->num_dimensions());
ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
- TensorShape output_shape = arm_compute::misc::shape_calculator::compute_gather_shape(input->tensor_shape(), indices->tensor_shape(), actual_axis);
+ TensorShape output_shape = arm_compute::misc::shape_calculator::compute_gather_shape(
+ input->tensor_shape(), indices->tensor_shape(), actual_axis);
ARM_COMPUTE_RETURN_ERROR_ON(output_shape.total_size() != output->tensor_shape().total_size());
}
@@ -56,12 +60,14 @@
return Status{};
}
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *indices, ITensorInfo *output, int axis)
+std::pair<Status, Window>
+validate_and_configure_window(ITensorInfo *input, ITensorInfo *indices, ITensorInfo *output, int axis)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, indices);
const uint32_t actual_axis = wrap_around(axis, static_cast<int>(input->num_dimensions()));
// Output auto initialization if not yet initialized
- TensorShape output_shape = arm_compute::misc::shape_calculator::compute_gather_shape(input->tensor_shape(), indices->tensor_shape(), actual_axis);
+ TensorShape output_shape = arm_compute::misc::shape_calculator::compute_gather_shape(
+ input->tensor_shape(), indices->tensor_shape(), actual_axis);
auto_init_if_empty((*output), output_shape, 1, input->data_type());
// Create window
@@ -72,8 +78,7 @@
} // namespace
-CLGatherKernel::CLGatherKernel()
- : _input(nullptr), _indices(nullptr), _output(nullptr), _axis(0)
+CLGatherKernel::CLGatherKernel() : _input(nullptr), _indices(nullptr), _output(nullptr), _axis(0)
{
_type = CLKernelType::ELEMENTWISE;
}
@@ -83,10 +88,14 @@
configure(CLKernelLibrary::get().get_compile_context(), input, indices, output, axis);
}
-void CLGatherKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *indices, ICLTensor *output, int axis)
+void CLGatherKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *indices,
+ ICLTensor *output,
+ int axis)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, indices);
- auto padding_info = get_padding_info({ input, output, indices });
+ auto padding_info = get_padding_info({input, output, indices});
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), indices->info(), output->info(), axis));
// Configure kernel window
@@ -100,7 +109,8 @@
// Set build options
CLBuildOptions build_opts;
- build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(data_size_from_type(input->info()->data_type())));
+ build_opts.add_option("-DDATA_TYPE=" +
+ get_cl_unsigned_type_from_element_size(data_size_from_type(input->info()->data_type())));
build_opts.add_option("-DOUTPUT_DIM_Z=" + support::cpp11::to_string(output->info()->dimension(2)));
build_opts.add_option("-DINDICES_DIM_Z=" + support::cpp11::to_string(indices->info()->dimension(2)));
build_opts.add_option("-DINPUT_DIM_Z=" + support::cpp11::to_string(input->info()->dimension(2)));
@@ -114,10 +124,12 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLGatherKernel::validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, int axis)
+Status
+CLGatherKernel::validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, int axis)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, indices, output, axis));
- ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), indices->clone().get(), output->clone().get(), axis).first);
+ ARM_COMPUTE_RETURN_ON_ERROR(
+ validate_and_configure_window(input->clone().get(), indices->clone().get(), output->clone().get(), axis).first);
return Status{};
}
diff --git a/src/core/CL/kernels/CLGatherKernel.h b/src/core/CL/kernels/CLGatherKernel.h
index 8f472a4..db4b49d 100644
--- a/src/core/CL/kernels/CLGatherKernel.h
+++ b/src/core/CL/kernels/CLGatherKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLGATHERKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -63,7 +64,11 @@
* @param[out] output Destination tensor. Data type supported: Same as @p input
* @param[in] axis (Optional) The axis in @p input to gather @p indices from. Negative values wrap around. Defaults to 0
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *indices, ICLTensor *output, int axis = 0);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *indices,
+ ICLTensor *output,
+ int axis = 0);
/** Static function to check if given info will lead to a valid configuration of @ref CLGatherKernel
*
@@ -74,7 +79,8 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, int axis = 0);
+ static Status
+ validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, int axis = 0);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLGenerateProposalsLayerKernel.cpp b/src/core/CL/kernels/CLGenerateProposalsLayerKernel.cpp
index 088c454..b9ff72b 100644
--- a/src/core/CL/kernels/CLGenerateProposalsLayerKernel.cpp
+++ b/src/core/CL/kernels/CLGenerateProposalsLayerKernel.cpp
@@ -31,6 +31,7 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -47,7 +48,7 @@
ARM_COMPUTE_RETURN_ERROR_ON(anchors->dimension(0) != info.values_per_roi());
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(anchors, DataType::QSYMM16, DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON(anchors->num_dimensions() > 2);
- if(all_anchors->total_size() > 0)
+ if (all_anchors->total_size() > 0)
{
size_t feature_height = info.feat_height();
size_t feature_width = info.feat_width();
@@ -57,7 +58,7 @@
ARM_COMPUTE_RETURN_ERROR_ON(all_anchors->dimension(0) != info.values_per_roi());
ARM_COMPUTE_RETURN_ERROR_ON(all_anchors->dimension(1) != feature_height * feature_width * num_anchors);
- if(is_data_type_quantized(anchors->data_type()))
+ if (is_data_type_quantized(anchors->data_type()))
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(anchors, all_anchors);
}
@@ -66,21 +67,25 @@
}
} // namespace
-CLComputeAllAnchorsKernel::CLComputeAllAnchorsKernel()
- : _anchors(nullptr), _all_anchors(nullptr)
+CLComputeAllAnchorsKernel::CLComputeAllAnchorsKernel() : _anchors(nullptr), _all_anchors(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
-void CLComputeAllAnchorsKernel::configure(const ICLTensor *anchors, ICLTensor *all_anchors, const ComputeAnchorsInfo &info)
+void CLComputeAllAnchorsKernel::configure(const ICLTensor *anchors,
+ ICLTensor *all_anchors,
+ const ComputeAnchorsInfo &info)
{
configure(CLKernelLibrary::get().get_compile_context(), anchors, all_anchors, info);
}
-void CLComputeAllAnchorsKernel::configure(const CLCompileContext &compile_context, const ICLTensor *anchors, ICLTensor *all_anchors, const ComputeAnchorsInfo &info)
+void CLComputeAllAnchorsKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *anchors,
+ ICLTensor *all_anchors,
+ const ComputeAnchorsInfo &info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(anchors, all_anchors);
- auto padding_info = get_padding_info({ anchors, all_anchors });
+ auto padding_info = get_padding_info({anchors, all_anchors});
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(anchors->info(), all_anchors->info(), info));
// Metadata
@@ -91,7 +96,8 @@
// Initialize the output if empty
const TensorShape output_shape(info.values_per_roi(), width * height * num_anchors);
- auto_init_if_empty(*all_anchors->info(), TensorInfo(output_shape, 1, data_type, anchors->info()->quantization_info()));
+ auto_init_if_empty(*all_anchors->info(),
+ TensorInfo(output_shape, 1, data_type, anchors->info()->quantization_info()));
// Set instance variables
_anchors = anchors;
@@ -108,7 +114,7 @@
build_opts.add_option("-DNUM_ANCHORS=" + support::cpp11::to_string(num_anchors));
build_opts.add_option("-DNUM_ROI_FIELDS=" + support::cpp11::to_string(info.values_per_roi()));
- if(is_quantized)
+ if (is_quantized)
{
const UniformQuantizationInfo qinfo = anchors->info()->quantization_info().uniform();
build_opts.add_option("-DSCALE=" + float_to_string_with_full_precision(qinfo.scale));
@@ -116,8 +122,9 @@
}
// Create kernel
- const std::string kernel_name = (is_quantized) ? "generate_proposals_compute_all_anchors_quantized" : "generate_proposals_compute_all_anchors";
- _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
+ const std::string kernel_name =
+ (is_quantized) ? "generate_proposals_compute_all_anchors_quantized" : "generate_proposals_compute_all_anchors";
+ _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
// The tensor all_anchors can be interpreted as an array of structs (each structs has values_per_roi fields).
// This means we don't need to pad on the X dimension, as we know in advance how many fields
@@ -127,7 +134,9 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLComputeAllAnchorsKernel::validate(const ITensorInfo *anchors, const ITensorInfo *all_anchors, const ComputeAnchorsInfo &info)
+Status CLComputeAllAnchorsKernel::validate(const ITensorInfo *anchors,
+ const ITensorInfo *all_anchors,
+ const ComputeAnchorsInfo &info)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(anchors, all_anchors, info));
return Status{};
diff --git a/src/core/CL/kernels/CLGenerateProposalsLayerKernel.h b/src/core/CL/kernels/CLGenerateProposalsLayerKernel.h
index d26795a..e08f281 100644
--- a/src/core/CL/kernels/CLGenerateProposalsLayerKernel.h
+++ b/src/core/CL/kernels/CLGenerateProposalsLayerKernel.h
@@ -62,7 +62,10 @@
* @param[in] info Contains Compute Anchors operation information described in @ref ComputeAnchorsInfo
*
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *anchors, ICLTensor *all_anchors, const ComputeAnchorsInfo &info);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *anchors,
+ ICLTensor *all_anchors,
+ const ComputeAnchorsInfo &info);
/** Static function to check if given info will lead to a valid configuration of @ref CLComputeAllAnchorsKernel
*
@@ -81,5 +84,5 @@
const ICLTensor *_anchors;
ICLTensor *_all_anchors;
};
-} // arm_compute
+} // namespace arm_compute
#endif // ARM_COMPUTE_CLGENERATEPROSPOSALSLAYERKERNEL_H
diff --git a/src/core/CL/kernels/CLInstanceNormalizationLayerKernel.cpp b/src/core/CL/kernels/CLInstanceNormalizationLayerKernel.cpp
index 7ed323c..b13eb16 100644
--- a/src/core/CL/kernels/CLInstanceNormalizationLayerKernel.cpp
+++ b/src/core/CL/kernels/CLInstanceNormalizationLayerKernel.cpp
@@ -30,6 +30,7 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -39,17 +40,20 @@
{
namespace
{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const InstanceNormalizationLayerKernelInfo &info)
+Status validate_arguments(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const InstanceNormalizationLayerKernelInfo &info)
{
ARM_COMPUTE_RETURN_ERROR_ON_MSG(info.epsilon == 0.f, "Epsilon must be different than 0");
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(input, DataType::F16, DataType::F32);
- if(output != nullptr && output->total_size() != 0)
+ if (output != nullptr && output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->num_channels() != output->num_channels(), "Input and output have different number of channels");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->num_channels() != output->num_channels(),
+ "Input and output have different number of channels");
}
return Status{};
@@ -59,27 +63,30 @@
{
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(input, DataType::F16, DataType::F32);
- if(output != nullptr && output->total_size() != 0)
+ if (output != nullptr && output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->num_channels() != output->num_channels(), "Input and output have different number of channels");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->num_channels() != output->num_channels(),
+ "Input and output have different number of channels");
}
return Status{};
}
} // namespace
-CLComputeMeanVariance::CLComputeMeanVariance()
- : _input(nullptr), _output(nullptr)
+CLComputeMeanVariance::CLComputeMeanVariance() : _input(nullptr), _output(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
-void CLComputeMeanVariance::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, bool use_mixed_precision)
+void CLComputeMeanVariance::configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ ICLTensor *output,
+ bool use_mixed_precision)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input);
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
_input = input;
_output = output == nullptr ? input : output;
@@ -88,7 +95,8 @@
const unsigned int num_elems_processed_per_iteration = 16 / input->info()->element_size();
CLBuildOptions build_opts;
- build_opts.add_option("-DINTERNAL_DATA_TYPE=" + (use_mixed_precision ? "float" : get_cl_type_from_data_type(input->info()->data_type())));
+ build_opts.add_option("-DINTERNAL_DATA_TYPE=" +
+ (use_mixed_precision ? "float" : get_cl_type_from_data_type(input->info()->data_type())));
build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
build_opts.add_option("-DDIM_X=" + support::cpp11::to_string(input->info()->dimension(0)));
@@ -108,7 +116,7 @@
const TensorShape out_shape(input_channel, 2u, input_batches);
// Output auto initialization if not yet initialized
- if(use_mixed_precision)
+ if (use_mixed_precision)
{
auto_init_if_empty(*_output->info(), out_shape, 1, DataType::F32);
}
@@ -134,7 +142,7 @@
Window collapsed_window = window.collapse(window, Window::DimZ);
// We will process the planes together
- if(_input->info()->data_layout() == DataLayout::NCHW)
+ if (_input->info()->data_layout() == DataLayout::NCHW)
{
collapsed_window.set(Window::DimX, Window::Dimension(0, 1, 1));
collapsed_window.set(Window::DimY, Window::Dimension(0, 1, 1));
@@ -157,10 +165,14 @@
_type = CLKernelType::ELEMENTWISE;
}
-void CLInstanceNormalizationLayerKernel::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *mean_var, ICLTensor *output, const InstanceNormalizationLayerKernelInfo &info)
+void CLInstanceNormalizationLayerKernel::configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ ICLTensor *mean_var,
+ ICLTensor *output,
+ const InstanceNormalizationLayerKernelInfo &info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input);
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
_input = input;
_output = output == nullptr ? input : output;
@@ -172,7 +184,9 @@
CLBuildOptions build_opts;
build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
- build_opts.add_option("-DINTERNAL_DATA_TYPE=" + (info.use_mixed_precision ? "float" : get_cl_type_from_data_type(input->info()->data_type())));
+ build_opts.add_option("-DINTERNAL_DATA_TYPE=" + (info.use_mixed_precision
+ ? "float"
+ : get_cl_type_from_data_type(input->info()->data_type())));
build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
build_opts.add_option("-DDIM_X=" + support::cpp11::to_string(input->info()->dimension(0)));
build_opts.add_option("-DDIM_Y=" + support::cpp11::to_string(input->info()->dimension(1)));
@@ -188,7 +202,7 @@
// Configure kernel window
Window win = calculate_max_window(*input->info(), Steps(1));
- if(output != nullptr)
+ if (output != nullptr)
{
auto_init_if_empty(*output->info(), input->info()->tensor_shape(), 1, input->info()->data_type());
}
@@ -197,7 +211,9 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLInstanceNormalizationLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const InstanceNormalizationLayerKernelInfo &info)
+Status CLInstanceNormalizationLayerKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const InstanceNormalizationLayerKernelInfo &info)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, info));
return Status{};
@@ -211,7 +227,7 @@
Window collapsed_window = window.collapse(window, Window::DimZ);
// We will process the planes together
- if(_input->info()->data_layout() == DataLayout::NCHW)
+ if (_input->info()->data_layout() == DataLayout::NCHW)
{
collapsed_window.set(Window::DimX, Window::Dimension(0, 1, 1));
collapsed_window.set(Window::DimY, Window::Dimension(0, 1, 1));
@@ -226,7 +242,7 @@
add_4D_tensor_argument(idx, _input, collapsed_window);
add_3D_tensor_argument(idx, _mean, collapsed_window);
- if(!_run_in_place)
+ if (!_run_in_place)
{
add_4D_tensor_argument(idx, _output, collapsed_window);
}
diff --git a/src/core/CL/kernels/CLInstanceNormalizationLayerKernel.h b/src/core/CL/kernels/CLInstanceNormalizationLayerKernel.h
index 2f9014a..9f436da 100644
--- a/src/core/CL/kernels/CLInstanceNormalizationLayerKernel.h
+++ b/src/core/CL/kernels/CLInstanceNormalizationLayerKernel.h
@@ -24,10 +24,10 @@
#ifndef ARM_COMPUTE_CLINSTANCENORMALIZATIONLAYERKERNEL_H
#define ARM_COMPUTE_CLINSTANCENORMALIZATIONLAYERKERNEL_H
-#include "src/core/CL/ICLKernel.h"
-
#include "arm_compute/core/KernelDescriptors.h"
+#include "src/core/CL/ICLKernel.h"
+
namespace arm_compute
{
// Forward declarations
@@ -59,7 +59,11 @@
* @param[out] output Destination tensor. Data types and data layouts supported: same as @p input.
* @param[in] info Kernel meta-data descriptor
*/
- void configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *mean_var, ICLTensor *output, const InstanceNormalizationLayerKernelInfo &info);
+ void configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ ICLTensor *mean_var,
+ ICLTensor *output,
+ const InstanceNormalizationLayerKernelInfo &info);
/** Static function to check if given info will lead to a valid configuration of @ref CLInstanceNormalizationLayer.
*
@@ -69,7 +73,8 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *output, const InstanceNormalizationLayerKernelInfo &info);
+ static Status
+ validate(const ITensorInfo *input, const ITensorInfo *output, const InstanceNormalizationLayerKernelInfo &info);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
@@ -106,7 +111,8 @@
* @param[out] output Destination tensor. Data types and data layouts supported: same as @p input.
* @param[in] use_mixed_precision Use mixed precision in case of FP16 execution
*/
- void configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, bool use_mixed_precision);
+ void
+ configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, bool use_mixed_precision);
/** Static function to check if given info will lead to a valid configuration of @ref CLInstanceNormalizationLayer.
*
diff --git a/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp b/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp
index 542d380..9ed9d7c 100644
--- a/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp
+++ b/src/core/CL/kernels/CLL2NormalizeLayerKernel.cpp
@@ -31,10 +31,10 @@
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
#include "arm_compute/core/Validate.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
-
#include "support/StringSupport.h"
namespace arm_compute
@@ -43,7 +43,8 @@
{
constexpr int max_input_tensor_dim = 3;
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *sum, const ITensorInfo *output, int axis, float epsilon)
+Status
+validate_arguments(const ITensorInfo *input, const ITensorInfo *sum, const ITensorInfo *output, int axis, float epsilon)
{
ARM_COMPUTE_UNUSED(epsilon);
@@ -53,14 +54,15 @@
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON_MSG(actual_axis > 2, "Actual axis greater than 2 is not supported");
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(actual_axis >= TensorShape::num_max_dimensions, "Actual normalization axis greater than max number of dimensions");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(actual_axis >= TensorShape::num_max_dimensions,
+ "Actual normalization axis greater than max number of dimensions");
// Reduce shape on axis
TensorShape sum_shape = input->tensor_shape();
sum_shape.set(actual_axis, 1);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(sum->tensor_shape(), sum_shape);
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
@@ -78,16 +80,22 @@
_type = CLKernelType::ELEMENTWISE;
}
-void CLL2NormalizeLayerKernel::configure(const ICLTensor *input, const ICLTensor *sum, ICLTensor *output, int axis, float epsilon)
+void CLL2NormalizeLayerKernel::configure(
+ const ICLTensor *input, const ICLTensor *sum, ICLTensor *output, int axis, float epsilon)
{
configure(CLKernelLibrary::get().get_compile_context(), input, sum, output, axis, epsilon);
}
-void CLL2NormalizeLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *sum, ICLTensor *output, int axis, float epsilon)
+void CLL2NormalizeLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *sum,
+ ICLTensor *output,
+ int axis,
+ float epsilon)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, sum, output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), sum->info(), output->info(), axis, epsilon));
- auto padding_info = get_padding_info({ input, sum, output });
+ auto padding_info = get_padding_info({input, sum, output});
_input = input;
_sum = sum;
@@ -95,8 +103,9 @@
_actual_axis = wrap_around(axis, max_input_tensor_dim);
_epsilon = epsilon;
- const unsigned int vec_size_x = adjust_vec_size(max_cl_vector_width / input->info()->element_size(), input->info()->dimension(0));
- const int vec_size_x_leftovers = input->info()->dimension(0) % vec_size_x;
+ const unsigned int vec_size_x =
+ adjust_vec_size(max_cl_vector_width / input->info()->element_size(), input->info()->dimension(0));
+ const int vec_size_x_leftovers = input->info()->dimension(0) % vec_size_x;
// Set build options
CLBuildOptions build_opts;
@@ -107,7 +116,7 @@
// Create kernel
std::string kernel_name;
unsigned int idx = 0;
- switch(_actual_axis)
+ switch (_actual_axis)
{
case 0:
kernel_name = "l2_normalize_x";
@@ -127,7 +136,7 @@
_kernel = create_kernel(compile_context, kernel_name, build_opts.options());
// Set epsilon argument
- if(input->info()->data_type() == DataType::F32)
+ if (input->info()->data_type() == DataType::F32)
{
_kernel.setArg<cl_float>(idx, _epsilon);
}
@@ -146,7 +155,8 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLL2NormalizeLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *sum, const ITensorInfo *output, int axis, float epsilon)
+Status CLL2NormalizeLayerKernel::validate(
+ const ITensorInfo *input, const ITensorInfo *sum, const ITensorInfo *output, int axis, float epsilon)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, sum, output, axis, epsilon));
return Status{};
@@ -159,7 +169,7 @@
Window window_sum(window);
- switch(_actual_axis)
+ switch (_actual_axis)
{
case 0:
{
@@ -173,8 +183,7 @@
add_2D_tensor_argument(idx, _sum, sum_slice);
add_2D_tensor_argument(idx, _output, in_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(window.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(sum_slice));
+ } while (window.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(sum_slice));
}
break;
case 1:
@@ -189,8 +198,7 @@
add_2D_tensor_argument(idx, _sum, sum_slice);
add_2D_tensor_argument(idx, _output, in_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(window.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(sum_slice));
+ } while (window.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(sum_slice));
}
break;
case 2:
@@ -205,8 +213,7 @@
add_3D_tensor_argument(idx, _sum, sum_slice);
add_3D_tensor_argument(idx, _output, in_slice);
enqueue(queue, *this, in_slice, lws_hint());
- }
- while(window.slide_window_slice_3D(in_slice) && window.slide_window_slice_3D(sum_slice));
+ } while (window.slide_window_slice_3D(in_slice) && window.slide_window_slice_3D(sum_slice));
}
break;
default:
diff --git a/src/core/CL/kernels/CLL2NormalizeLayerKernel.h b/src/core/CL/kernels/CLL2NormalizeLayerKernel.h
index edc0585..5c9ab94 100644
--- a/src/core/CL/kernels/CLL2NormalizeLayerKernel.h
+++ b/src/core/CL/kernels/CLL2NormalizeLayerKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLL2NORMALIZELAYERKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -70,7 +71,12 @@
* @param[in] axis Axis along which to reduce. Negative values wrap around. Maximum supported actual reduction axis : 2
* @param[in] epsilon Lower bound value for the normalization.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *sum, ICLTensor *output, int axis, float epsilon);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *sum,
+ ICLTensor *output,
+ int axis,
+ float epsilon);
/** Static function to check if given info will lead to a valid configuration of @ref CLL2NormalizeLayerKernel.
*
@@ -84,7 +90,8 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *sum, const ITensorInfo *output, int axis, float epsilon);
+ static Status
+ validate(const ITensorInfo *input, const ITensorInfo *sum, const ITensorInfo *output, int axis, float epsilon);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLMaxUnpoolingLayerKernel.cpp b/src/core/CL/kernels/CLMaxUnpoolingLayerKernel.cpp
index dc9d686..e560f1d 100644
--- a/src/core/CL/kernels/CLMaxUnpoolingLayerKernel.cpp
+++ b/src/core/CL/kernels/CLMaxUnpoolingLayerKernel.cpp
@@ -31,6 +31,7 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -42,26 +43,31 @@
namespace
{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const PoolingLayerInfo &pool_info, const ITensorInfo *indices)
+Status validate_arguments(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const PoolingLayerInfo &pool_info,
+ const ITensorInfo *indices)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output, indices);
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
- ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
+ DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(indices, 1, DataType::U32);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, indices);
- int pool_stride_x = 0;
- int pool_stride_y = 0;
- PoolingType pool_type = pool_info.pool_type;
- const PadStrideInfo pad_stride_info = pool_info.pad_stride_info;
+ int pool_stride_x = 0;
+ int pool_stride_y = 0;
+ PoolingType pool_type = pool_info.pool_type;
+ const PadStrideInfo pad_stride_info = pool_info.pad_stride_info;
std::tie(pool_stride_x, pool_stride_y) = pad_stride_info.stride();
- const int pool_size_x = pool_info.pool_size.width;
- const int pool_size_y = pool_info.pool_size.height;
+ const int pool_size_x = pool_info.pool_size.width;
+ const int pool_size_y = pool_info.pool_size.height;
const Size2D pool_size(pool_size_x, pool_size_y);
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(pool_type != PoolingType::MAX, "Pooling indices only supported for MAX pooling method");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(pool_type != PoolingType::MAX,
+ "Pooling indices only supported for MAX pooling method");
ARM_COMPUTE_RETURN_ERROR_ON_MSG((pool_size != Size2D(2, 2)), "Pooling indices only supported for pool size 2x2");
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
@@ -71,17 +77,20 @@
}
} // namespace
-CLMaxUnpoolingLayerKernel::CLMaxUnpoolingLayerKernel()
- : _input(nullptr), _output(nullptr), _indices(nullptr)
+CLMaxUnpoolingLayerKernel::CLMaxUnpoolingLayerKernel() : _input(nullptr), _output(nullptr), _indices(nullptr)
{
_type = CLKernelType::POOL;
}
-void CLMaxUnpoolingLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *indices, ICLTensor *output, const PoolingLayerInfo &pool_info)
+void CLMaxUnpoolingLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *indices,
+ ICLTensor *output,
+ const PoolingLayerInfo &pool_info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), pool_info, indices->info()));
- auto padding_info = get_padding_info({ input, indices, output });
+ auto padding_info = get_padding_info({input, indices, output});
_input = input;
_output = output;
@@ -119,7 +128,10 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLMaxUnpoolingLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, const PoolingLayerInfo &pool_info)
+Status CLMaxUnpoolingLayerKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *indices,
+ const ITensorInfo *output,
+ const PoolingLayerInfo &pool_info)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, indices, output);
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, pool_info, indices));
@@ -140,7 +152,6 @@
add_3D_tensor_argument(idx, _output, slice);
add_3D_tensor_argument(idx, _indices, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(window.slide_window_slice_3D(slice));
+ } while (window.slide_window_slice_3D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLMaxUnpoolingLayerKernel.h b/src/core/CL/kernels/CLMaxUnpoolingLayerKernel.h
index 45481d0..eb18a46 100644
--- a/src/core/CL/kernels/CLMaxUnpoolingLayerKernel.h
+++ b/src/core/CL/kernels/CLMaxUnpoolingLayerKernel.h
@@ -59,7 +59,11 @@
* @param[out] output Destination tensor. Data types supported: Same as @p input.
* @param[in] pool_info Contains pooling operation information described in @ref PoolingLayerInfo.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *indices, ICLTensor *output, const PoolingLayerInfo &pool_info);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *indices,
+ ICLTensor *output,
+ const PoolingLayerInfo &pool_info);
/** Static function to check if given info will lead to a valid configuration of @ref CLMaxUnpoolingLayerKernel
*
* @param[in] input Source tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32.
@@ -72,7 +76,10 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *indices, const ITensorInfo *output, const PoolingLayerInfo &pool_info);
+ static Status validate(const ITensorInfo *input,
+ const ITensorInfo *indices,
+ const ITensorInfo *output,
+ const PoolingLayerInfo &pool_info);
// Inherited methods overridden
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLMeanStdDevNormalizationKernel.cpp b/src/core/CL/kernels/CLMeanStdDevNormalizationKernel.cpp
index ac33468..8632bdf 100644
--- a/src/core/CL/kernels/CLMeanStdDevNormalizationKernel.cpp
+++ b/src/core/CL/kernels/CLMeanStdDevNormalizationKernel.cpp
@@ -31,6 +31,7 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -49,7 +50,7 @@
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32);
// Checks performed when output is configured
- if((output != nullptr) && (output->total_size() != 0))
+ if ((output != nullptr) && (output->total_size() != 0))
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
@@ -69,15 +70,19 @@
configure(CLKernelLibrary::get().get_compile_context(), input, output, epsilon);
}
-void CLMeanStdDevNormalizationKernel::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, float epsilon)
+void CLMeanStdDevNormalizationKernel::configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ ICLTensor *output,
+ float epsilon)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input);
_run_in_place = (output == nullptr) || (output == input);
- ARM_COMPUTE_ERROR_THROW_ON(CLMeanStdDevNormalizationKernel::validate(input->info(), (output != nullptr) ? output->info() : nullptr, epsilon));
+ ARM_COMPUTE_ERROR_THROW_ON(CLMeanStdDevNormalizationKernel::validate(
+ input->info(), (output != nullptr) ? output->info() : nullptr, epsilon));
- if(output != nullptr)
+ if (output != nullptr)
{
auto_init_if_empty(*output->info(), *input->info());
}
@@ -85,7 +90,8 @@
_input = input;
_output = output;
- const unsigned int num_elems_processed_per_iteration = adjust_vec_size(16 / input->info()->element_size(), input->info()->dimension(0));
+ const unsigned int num_elems_processed_per_iteration =
+ adjust_vec_size(16 / input->info()->element_size(), input->info()->dimension(0));
// Set build options
CLBuildOptions build_opts;
@@ -134,7 +140,6 @@
add_2D_tensor_argument_if((!_run_in_place), idx, _output, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(window.slide_window_slice_2D(slice));
+ } while (window.slide_window_slice_2D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLMeanStdDevNormalizationKernel.h b/src/core/CL/kernels/CLMeanStdDevNormalizationKernel.h
index a1ba2b9..e02a3c5 100644
--- a/src/core/CL/kernels/CLMeanStdDevNormalizationKernel.h
+++ b/src/core/CL/kernels/CLMeanStdDevNormalizationKernel.h
@@ -66,7 +66,10 @@
* @param[out] output (Optional) Destination tensor. It can be nullptr in case of in-place computation. Data type supported: same as @p input
* @param[in] epsilon (Optional) Small float to avoid division by zero in case of zero standard deviation. Defaults to 1e-8.
*/
- void configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output = nullptr, float epsilon = 1e-8f);
+ void configure(const CLCompileContext &compile_context,
+ ICLTensor *input,
+ ICLTensor *output = nullptr,
+ float epsilon = 1e-8f);
/** Static function to check if given info will lead to a valid configuration of @ref CLMeanStdDevNormalizationKernel
*
* @param[in] input Source tensor info with 2 dimensions. In case of @p output tensor info = nullptr,
diff --git a/src/core/CL/kernels/CLNormalizationLayerKernel.cpp b/src/core/CL/kernels/CLNormalizationLayerKernel.cpp
index c6c4229..b636c48 100644
--- a/src/core/CL/kernels/CLNormalizationLayerKernel.cpp
+++ b/src/core/CL/kernels/CLNormalizationLayerKernel.cpp
@@ -32,6 +32,7 @@
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
#include "arm_compute/core/utils/StringUtils.h"
#include "arm_compute/core/Window.h"
+
#include "src/core/AccessWindowStatic.h"
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
@@ -53,7 +54,7 @@
ARM_COMPUTE_RETURN_ERROR_ON_MSG(!(norm_info.norm_size() % 2), "Normalization size should be odd");
// Checks performed when output is configured
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
@@ -63,7 +64,8 @@
return Status{};
}
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, NormalizationLayerInfo norm_info)
+std::pair<Status, Window>
+validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, NormalizationLayerInfo norm_info)
{
// Output tensor auto initialization if not yet initialized
auto_init_if_empty(*output, *input->clone());
@@ -71,9 +73,10 @@
bool window_changed = false;
Window win;
const DataLayout data_layout = input->data_layout();
- if(data_layout == DataLayout::NCHW)
+ if (data_layout == DataLayout::NCHW)
{
- const unsigned int vec_size_x = adjust_vec_size(max_cl_vector_width / input->element_size(), input->dimension(0));
+ const unsigned int vec_size_x =
+ adjust_vec_size(max_cl_vector_width / input->element_size(), input->dimension(0));
const unsigned int norm_idx = get_normalization_dimension_index(input->data_layout(), norm_info);
const bool is_norm_across_width = norm_idx == 0;
@@ -87,15 +90,16 @@
// The output has 1 right padding because of the vec_size_x.
// The input has 1 left padding because radius = 1.
// The input has 2 right padding because of radius = 1 AND because of the extra output padding
- const unsigned int border_width_left = is_norm_across_width ? norm_radius : 0;
- const unsigned int border_width_right = is_norm_across_width ? norm_radius + (vec_size_x - input->dimension(0) % vec_size_x) : 0;
- const BorderSize border_size = BorderSize(0, border_width_right, 0, border_width_left);
+ const unsigned int border_width_left = is_norm_across_width ? norm_radius : 0;
+ const unsigned int border_width_right =
+ is_norm_across_width ? norm_radius + (vec_size_x - input->dimension(0) % vec_size_x) : 0;
+ const BorderSize border_size = BorderSize(0, border_width_right, 0, border_width_left);
win = calculate_max_window(*input, Steps(vec_size_x));
// We do not use a Rectangle window for IN_MAP_2D as we clamp the top and bottom accesses inside the kernel, avoiding padding
// Reads can occur within the valid region of the input
- if(is_norm_across_width)
+ if (is_norm_across_width)
{
AccessWindowStatic input_access(input, -border_size.left, 0, input->dimension(0) + border_size.right, 0);
window_changed = window_changed || update_window_and_padding(win, input_access);
@@ -112,13 +116,14 @@
else
{
unsigned int vec_size_x = adjust_vec_size(max_cl_vector_width / input->element_size(), input->dimension(0));
- if(norm_info.is_cross_map())
+ if (norm_info.is_cross_map())
{
vec_size_x = 1;
}
win = calculate_max_window(*input, Steps(vec_size_x));
}
- Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+ Status err =
+ (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
return std::make_pair(err, win);
}
} // namespace
@@ -139,10 +144,13 @@
configure(CLKernelLibrary::get().get_compile_context(), input, output, norm_info);
}
-void CLNormalizationLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, NormalizationLayerInfo norm_info)
+void CLNormalizationLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ NormalizationLayerInfo norm_info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
// Perform validation step
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), norm_info));
@@ -152,16 +160,17 @@
_input = input;
_output = output;
- const DataLayout data_layout = input->info()->data_layout();
- unsigned int vec_size_x = adjust_vec_size(max_cl_vector_width / input->info()->element_size(), input->info()->dimension(0));
- int vec_size_x_leftovers = input->info()->dimension(0) % vec_size_x;
- if(norm_info.is_cross_map() && data_layout == DataLayout::NHWC)
+ const DataLayout data_layout = input->info()->data_layout();
+ unsigned int vec_size_x =
+ adjust_vec_size(max_cl_vector_width / input->info()->element_size(), input->info()->dimension(0));
+ int vec_size_x_leftovers = input->info()->dimension(0) % vec_size_x;
+ if (norm_info.is_cross_map() && data_layout == DataLayout::NHWC)
{
vec_size_x = 1;
vec_size_x_leftovers = 0;
}
- if(data_layout == DataLayout::NCHW)
+ if (data_layout == DataLayout::NCHW)
{
const unsigned int norm_idx = get_normalization_dimension_index(data_layout, norm_info);
_is_norm_across_width = norm_idx == 0;
@@ -175,9 +184,10 @@
// The output has 1 right padding because of the vec_size_x.
// The input has 1 left padding because radius = 1.
// The input has 2 right padding because of radius = 1 AND the extra output padding
- const unsigned int border_width_left = _is_norm_across_width ? norm_radius : 0;
- const unsigned int border_width_right = _is_norm_across_width ? norm_radius + (vec_size_x - input->info()->dimension(0) % vec_size_x) : 0;
- _border_size = BorderSize(0, border_width_right, 0, border_width_left);
+ const unsigned int border_width_left = _is_norm_across_width ? norm_radius : 0;
+ const unsigned int border_width_right =
+ _is_norm_across_width ? norm_radius + (vec_size_x - input->info()->dimension(0) % vec_size_x) : 0;
+ _border_size = BorderSize(0, border_width_right, 0, border_width_left);
}
const bool is_in_map_2D = (norm_info.type() == NormType::IN_MAP_2D);
@@ -193,12 +203,14 @@
build_opts.add_option(("-DRADIUS=" + support::cpp11::to_string(norm_info.norm_size() / 2)));
build_opts.add_option(("-DNUM_SLICES=" + support::cpp11::to_string(input->info()->dimension(2))));
build_opts.add_option_if(is_in_map_2D, "-DIN_MAP_2D");
- build_opts.add_option_if(norm_info.is_in_map() || (data_layout == DataLayout::NHWC && norm_info.is_cross_map()), "-DWIDTH_SIZE=" + support::cpp11::to_string(input->info()->dimension(0)));
- build_opts.add_option_if(norm_info.is_in_map() && data_layout == DataLayout::NHWC, "-DDIM1_SIZE=" + support::cpp11::to_string(input->info()->dimension(1)));
+ build_opts.add_option_if(norm_info.is_in_map() || (data_layout == DataLayout::NHWC && norm_info.is_cross_map()),
+ "-DWIDTH_SIZE=" + support::cpp11::to_string(input->info()->dimension(0)));
+ build_opts.add_option_if(norm_info.is_in_map() && data_layout == DataLayout::NHWC,
+ "-DDIM1_SIZE=" + support::cpp11::to_string(input->info()->dimension(1)));
// Create kernel
std::string kernel_name;
- if(norm_info.is_in_map())
+ if (norm_info.is_in_map())
{
kernel_name = "normalization_layer_in_map_" + lower_string(string_from_data_layout(data_layout));
}
@@ -222,16 +234,19 @@
_config_id += support::cpp11::to_string(input->info()->dimension(0));
_config_id += "_";
_config_id += support::cpp11::to_string(input->info()->dimension(1));
- if(data_layout == DataLayout::NHWC)
+ if (data_layout == DataLayout::NHWC)
{
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
}
-Status CLNormalizationLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, NormalizationLayerInfo norm_info)
+Status CLNormalizationLayerKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ NormalizationLayerInfo norm_info)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, norm_info));
- ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get(), norm_info).first);
+ ARM_COMPUTE_RETURN_ON_ERROR(
+ validate_and_configure_window(input->clone().get(), output->clone().get(), norm_info).first);
return Status{};
}
@@ -251,7 +266,6 @@
add_3D_tensor_argument(idx, _input, slice);
add_3D_tensor_argument(idx, _output, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(window_collapsed.slide_window_slice_3D(slice));
+ } while (window_collapsed.slide_window_slice_3D(slice));
}
-} // namespace arm_compute
\ No newline at end of file
+} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLNormalizationLayerKernel.h b/src/core/CL/kernels/CLNormalizationLayerKernel.h
index 739a2ae..5517ba6 100644
--- a/src/core/CL/kernels/CLNormalizationLayerKernel.h
+++ b/src/core/CL/kernels/CLNormalizationLayerKernel.h
@@ -63,7 +63,10 @@
* Data layouts supported: same as @p input.
* @param[in] norm_info Normalization layer information like the normalization type, normalization size and other parameters.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, NormalizationLayerInfo norm_info);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ NormalizationLayerInfo norm_info);
/** Static function to check if given info will lead to a valid configuration of @ref CLNormalizationLayerKernel
*
* @param[in] input Source tensor. 3 lower dims represent a single input with dimensions [width, height, IFM],
@@ -77,7 +80,7 @@
static Status validate(const ITensorInfo *input, const ITensorInfo *output, NormalizationLayerInfo norm_info);
// Inherited methods overridden:
- void run(const Window &window, cl::CommandQueue &queue) override;
+ void run(const Window &window, cl::CommandQueue &queue) override;
BorderSize border_size() const override;
private:
diff --git a/src/core/CL/kernels/CLNormalizePlanarYUVLayerKernel.cpp b/src/core/CL/kernels/CLNormalizePlanarYUVLayerKernel.cpp
index 6b0400d..59352a8 100644
--- a/src/core/CL/kernels/CLNormalizePlanarYUVLayerKernel.cpp
+++ b/src/core/CL/kernels/CLNormalizePlanarYUVLayerKernel.cpp
@@ -31,32 +31,35 @@
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/AccessWindowStatic.h"
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
-
#include "support/StringSupport.h"
namespace arm_compute
{
namespace
{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *mean, const ITensorInfo *std)
+Status
+validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *mean, const ITensorInfo *std)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
- ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
+ DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, mean, std);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(mean, std);
ARM_COMPUTE_RETURN_ERROR_ON_MSG(mean->num_dimensions() > 1, "mean and std must be vectors");
- const unsigned int channel_idx = get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::CHANNEL);
+ const unsigned int channel_idx =
+ get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::CHANNEL);
ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(channel_idx) != mean->dimension(0));
// Checks performed when output is configured
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
@@ -77,7 +80,8 @@
bool window_changed = update_window_and_padding(win, input_access, output_access);
- Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+ Status err =
+ (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
return std::make_pair(err, win);
}
} // namespace
@@ -88,12 +92,19 @@
_type = CLKernelType::ELEMENTWISE;
}
-void CLNormalizePlanarYUVLayerKernel::configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *mean, const ICLTensor *std)
+void CLNormalizePlanarYUVLayerKernel::configure(const ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *mean,
+ const ICLTensor *std)
{
configure(CLKernelLibrary::get().get_compile_context(), input, output, mean, std);
}
-void CLNormalizePlanarYUVLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const ICLTensor *mean, const ICLTensor *std)
+void CLNormalizePlanarYUVLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *mean,
+ const ICLTensor *std)
{
// Perform validation step
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, mean, std);
@@ -102,7 +113,7 @@
// Output tensor auto initialization if not yet initialized
auto_init_if_empty(*output->info(), *input->info()->clone());
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
_input = input;
_output = output;
@@ -112,9 +123,10 @@
const DataLayout data_layout = input->info()->data_layout();
// Get number of elements to process per iterations
- const unsigned int num_elems_processed_per_iteration = (data_layout == DataLayout::NHWC) ? adjust_vec_size(16 / input->info()->element_size(),
- input->info()->dimension(0)) :
- (16 / input->info()->element_size());
+ const unsigned int num_elems_processed_per_iteration =
+ (data_layout == DataLayout::NHWC)
+ ? adjust_vec_size(16 / input->info()->element_size(), input->info()->dimension(0))
+ : (16 / input->info()->element_size());
const unsigned int channel_idx = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
const DataType dt = input->info()->data_type();
@@ -122,11 +134,12 @@
CLBuildOptions build_opts;
build_opts.add_option(("-DDATA_TYPE=" + get_cl_type_from_data_type(dt)));
build_opts.add_option(("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration)));
- build_opts.add_option(("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(input->info()->dimension(0) % num_elems_processed_per_iteration)));
+ build_opts.add_option(("-DVEC_SIZE_LEFTOVER=" +
+ support::cpp11::to_string(input->info()->dimension(0) % num_elems_processed_per_iteration)));
build_opts.add_option(("-DNUM_CHANNELS=" + support::cpp11::to_string(input->info()->dimension(channel_idx))));
std::string kernel_name = "normalize_planar_yuv_layer_";
- if(is_data_type_quantized(dt))
+ if (is_data_type_quantized(dt))
{
const UniformQuantizationInfo qinfo = input->info()->quantization_info().uniform();
build_opts.add_option(("-DOFFSET=" + support::cpp11::to_string(qinfo.offset)));
@@ -139,7 +152,7 @@
_kernel = create_kernel(compile_context, kernel_name, build_opts.options());
// Configure kernel window
- if(data_layout == DataLayout::NHWC)
+ if (data_layout == DataLayout::NHWC)
{
Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration));
ICLKernel::configure_internal(win);
@@ -165,12 +178,16 @@
_config_id += support::cpp11::to_string(input->info()->dimension(2));
}
-Status CLNormalizePlanarYUVLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *mean, const ITensorInfo *std)
+Status CLNormalizePlanarYUVLayerKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const ITensorInfo *mean,
+ const ITensorInfo *std)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, mean, std));
- if(input->data_layout() == DataLayout::NCHW)
+ if (input->data_layout() == DataLayout::NCHW)
{
- ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_nchw(input->clone().get(), output->clone().get()).first);
+ ARM_COMPUTE_RETURN_ON_ERROR(
+ validate_and_configure_window_nchw(input->clone().get(), output->clone().get()).first);
}
return Status{};
}
@@ -196,7 +213,6 @@
add_3D_tensor_argument(idx, _input, slice);
add_3D_tensor_argument(idx, _output, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(collapsed.slide_window_slice_3D(slice));
+ } while (collapsed.slide_window_slice_3D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLNormalizePlanarYUVLayerKernel.h b/src/core/CL/kernels/CLNormalizePlanarYUVLayerKernel.h
index 6db4433..341b404 100644
--- a/src/core/CL/kernels/CLNormalizePlanarYUVLayerKernel.h
+++ b/src/core/CL/kernels/CLNormalizePlanarYUVLayerKernel.h
@@ -67,7 +67,11 @@
* @param[in] std Standard deviation values tensor. 1 dimension with size equal to the number of input channels.
* Data types supported: same as @p input
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const ICLTensor *mean, const ICLTensor *std);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *mean,
+ const ICLTensor *std);
/** Static function to check if given info will lead to a valid configuration of @ref CLNormalizePlanarYUVLayerKernel
*
* @param[in] input Source tensor info. 3 lower dimensions represent a single input with dimensions [width, height, channels].
@@ -79,7 +83,8 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *mean, const ITensorInfo *std);
+ static Status
+ validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *mean, const ITensorInfo *std);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLPadLayerKernel.cpp b/src/core/CL/kernels/CLPadLayerKernel.cpp
index 53f313c..0ac2850 100644
--- a/src/core/CL/kernels/CLPadLayerKernel.cpp
+++ b/src/core/CL/kernels/CLPadLayerKernel.cpp
@@ -27,6 +27,7 @@
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
#include "support/StringSupport.h"
@@ -35,25 +36,29 @@
{
namespace
{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const PaddingList &padding, PixelValue constant_value, PaddingMode mode)
+Status validate_arguments(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const PaddingList &padding,
+ PixelValue constant_value,
+ PaddingMode mode)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_UNUSED(constant_value);
ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
ARM_COMPUTE_RETURN_ERROR_ON((padding.size() < 1) || (padding.size() > input->num_dimensions()));
- if(mode == PaddingMode::REFLECT || mode == PaddingMode::SYMMETRIC)
+ if (mode == PaddingMode::REFLECT || mode == PaddingMode::SYMMETRIC)
{
ARM_COMPUTE_RETURN_ERROR_ON(padding.size() > 3);
const auto is_reflect = static_cast<unsigned int>(mode == PaddingMode::REFLECT);
- for(size_t i = 0; i < padding.size(); ++i)
+ for (size_t i = 0; i < padding.size(); ++i)
{
ARM_COMPUTE_RETURN_ERROR_ON(padding.at(i).first > (input->dimension(i) - is_reflect));
ARM_COMPUTE_RETURN_ERROR_ON(padding.at(i).second > (input->dimension(i) - is_reflect));
}
}
- if(output->total_size() > 0)
+ if (output->total_size() > 0)
{
TensorShape padded_shape = misc::shape_calculator::compute_padded_shape(input->tensor_shape(), padding);
@@ -65,41 +70,51 @@
}
} // namespace
-CLPadLayerKernel::CLPadLayerKernel()
- : _input(nullptr), _output(nullptr), _4d_enabled(false)
+CLPadLayerKernel::CLPadLayerKernel() : _input(nullptr), _output(nullptr), _4d_enabled(false)
{
_type = CLKernelType::ELEMENTWISE;
}
-void CLPadLayerKernel::configure(const ICLTensor *input, ICLTensor *output, const PaddingList &padding, PixelValue constant_value, PaddingMode mode)
+void CLPadLayerKernel::configure(
+ const ICLTensor *input, ICLTensor *output, const PaddingList &padding, PixelValue constant_value, PaddingMode mode)
{
configure(CLKernelLibrary::get().get_compile_context(), input, output, padding, constant_value, mode);
}
-void CLPadLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const PaddingList &padding, PixelValue constant_value, PaddingMode mode)
+void CLPadLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const PaddingList &padding,
+ PixelValue constant_value,
+ PaddingMode mode)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(misc::shape_calculator::compute_padded_shape(input->info()->tensor_shape(), padding)));
+ auto_init_if_empty(*output->info(),
+ input->info()->clone()->set_tensor_shape(
+ misc::shape_calculator::compute_padded_shape(input->info()->tensor_shape(), padding)));
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), padding, constant_value, mode));
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
_input = input;
_output = output;
_4d_enabled = (mode == PaddingMode::CONSTANT) && (padding.size() > 3);
// Set build options
- const DataType &data_type = input->info()->data_type();
- const unsigned int input_width = input->info()->dimension(0);
- const unsigned int input_height = input->info()->dimension(1);
- const unsigned int input_depth = input->info()->dimension(2);
- const unsigned int pad_x_before = padding.at(0).first;
- const unsigned int pad_y_before = padding.size() > 1 ? padding.at(1).first : 0;
- const unsigned int pad_z_before = padding.size() > 2 ? padding.at(2).first : 0;
- const unsigned int vec_size = adjust_vec_size(std::min(16U, 32U / static_cast<unsigned int>(element_size_from_data_type(input->info()->data_type()))), input_width);
- const unsigned int pad_right_start = input_width + pad_x_before;
- const unsigned int pad_x_before_remainder = pad_x_before % vec_size;
- const unsigned int vec_size_leftover_write = vec_size - (ceil_to_multiple(output->info()->dimension(0), vec_size) - output->info()->dimension(0));
+ const DataType &data_type = input->info()->data_type();
+ const unsigned int input_width = input->info()->dimension(0);
+ const unsigned int input_height = input->info()->dimension(1);
+ const unsigned int input_depth = input->info()->dimension(2);
+ const unsigned int pad_x_before = padding.at(0).first;
+ const unsigned int pad_y_before = padding.size() > 1 ? padding.at(1).first : 0;
+ const unsigned int pad_z_before = padding.size() > 2 ? padding.at(2).first : 0;
+ const unsigned int vec_size = adjust_vec_size(
+ std::min(16U, 32U / static_cast<unsigned int>(element_size_from_data_type(input->info()->data_type()))),
+ input_width);
+ const unsigned int pad_right_start = input_width + pad_x_before;
+ const unsigned int pad_x_before_remainder = pad_x_before % vec_size;
+ const unsigned int vec_size_leftover_write =
+ vec_size - (ceil_to_multiple(output->info()->dimension(0), vec_size) - output->info()->dimension(0));
CLBuildOptions build_opts;
build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(data_type));
@@ -108,12 +123,12 @@
build_opts.add_option("-DSRC_WIDTH=" + support::cpp11::to_string(input_width));
build_opts.add_option("-DPAD_X_BEFORE_REMAINDER=" + support::cpp11::to_string(pad_x_before_remainder));
build_opts.add_option("-DVEC_SIZE_LEFTOVER_WRITE=" + support::cpp11::to_string(vec_size_leftover_write));
- if(padding.size() > 1)
+ if (padding.size() > 1)
{
build_opts.add_option("-DPAD_Y_BEFORE=" + support::cpp11::to_string(pad_y_before));
build_opts.add_option("-DSRC_HEIGHT=" + support::cpp11::to_string(input_height));
- if(padding.size() > 2)
+ if (padding.size() > 2)
{
build_opts.add_option("-DPAD_Z_BEFORE=" + support::cpp11::to_string(pad_z_before));
build_opts.add_option("-DSRC_DEPTH=" + support::cpp11::to_string(input_depth));
@@ -121,23 +136,25 @@
}
std::string kernel_name = "pad_layer_";
- switch(mode)
+ switch (mode)
{
case PaddingMode::CONSTANT:
{
kernel_name += "constant";
- const unsigned int vec_size_leftover_read = vec_size - (ceil_to_multiple(pad_right_start, vec_size) - pad_right_start);
+ const unsigned int vec_size_leftover_read =
+ vec_size - (ceil_to_multiple(pad_right_start, vec_size) - pad_right_start);
build_opts.add_option("-DCONST_VAL=" + string_from_pixel_value(constant_value, data_type));
build_opts.add_option("-DVEC_SIZE_LEFTOVER_READ=" + support::cpp11::to_string(vec_size_leftover_read));
- if(pad_x_before >= vec_size)
+ if (pad_x_before >= vec_size)
{
build_opts.add_option("-DTHREADS_TO_SKIP_BEFORE=" + support::cpp11::to_string(pad_x_before / vec_size));
- build_opts.add_option("-DTHREADS_TO_SKIP_AFTER=" + support::cpp11::to_string(pad_right_start / vec_size));
+ build_opts.add_option("-DTHREADS_TO_SKIP_AFTER=" +
+ support::cpp11::to_string(pad_right_start / vec_size));
}
- if(_4d_enabled)
+ if (_4d_enabled)
{
build_opts.add_option("-DPAD_W_BEFORE=" + support::cpp11::to_string(padding.at(3).first));
build_opts.add_option("-DSRC_BATCH=" + support::cpp11::to_string(input->info()->dimension(3)));
@@ -154,14 +171,17 @@
const unsigned int pad_x_after_remainder = pad_right_start % vec_size;
const unsigned int after_pad_fact_x = (2 * input_width + pad_x_before) - is_reflect;
- const unsigned int output_last_x = ceil_to_multiple(pad_right_start + padding.at(0).second, vec_size);
+ const unsigned int output_last_x = ceil_to_multiple(pad_right_start + padding.at(0).second, vec_size);
build_opts.add_option("-DIS_REFLECT=" + support::cpp11::to_string(is_reflect));
build_opts.add_option("-DPAD_X_AFTER_REMAINDER=" + support::cpp11::to_string(pad_x_after_remainder));
- build_opts.add_option("-DPAD_X_BEFORE_REMAINDER_REFL=" + support::cpp11::to_string((pad_x_before_remainder + is_reflect) % vec_size));
- build_opts.add_option("-DPAD_X_AFTER_REMAINDER_REFL=" + support::cpp11::to_string((pad_x_after_remainder - is_reflect) % vec_size));
+ build_opts.add_option("-DPAD_X_BEFORE_REMAINDER_REFL=" +
+ support::cpp11::to_string((pad_x_before_remainder + is_reflect) % vec_size));
+ build_opts.add_option("-DPAD_X_AFTER_REMAINDER_REFL=" +
+ support::cpp11::to_string((pad_x_after_remainder - is_reflect) % vec_size));
build_opts.add_option("-DAFTER_PAD_FACT_X=" + support::cpp11::to_string(after_pad_fact_x));
- build_opts.add_option_if(after_pad_fact_x < output_last_x, "-DAFTER_PAD_REM=" + support::cpp11::to_string(after_pad_fact_x % vec_size));
+ build_opts.add_option_if(after_pad_fact_x < output_last_x,
+ "-DAFTER_PAD_REM=" + support::cpp11::to_string(after_pad_fact_x % vec_size));
break;
}
@@ -179,7 +199,11 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLPadLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const PaddingList &padding, PixelValue constant_value, PaddingMode mode)
+Status CLPadLayerKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const PaddingList &padding,
+ PixelValue constant_value,
+ PaddingMode mode)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, padding, constant_value, mode));
return Status{};
@@ -197,13 +221,12 @@
unsigned int idx = 0;
add_3D_tensor_argument(idx, _input, slice);
add_3D_tensor_argument(idx, _output, slice);
- if(_4d_enabled)
+ if (_4d_enabled)
{
add_argument<unsigned int>(idx, batch++);
}
enqueue(queue, *this, slice, lws_hint());
- }
- while(window.slide_window_slice_3D(slice));
+ } while (window.slide_window_slice_3D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLPadLayerKernel.h b/src/core/CL/kernels/CLPadLayerKernel.h
index 90af337..dca121b 100644
--- a/src/core/CL/kernels/CLPadLayerKernel.h
+++ b/src/core/CL/kernels/CLPadLayerKernel.h
@@ -56,7 +56,11 @@
* @param[in] mode (Optional) Controls whether the padding should be filled with @p constant_value using CONSTANT,
* or reflect the input, either including the border values (SYMMETRIC) or not (REFLECT).
*/
- void configure(const ICLTensor *input, ICLTensor *output, const PaddingList &padding, PixelValue constant_value = PixelValue(), PaddingMode mode = PaddingMode::CONSTANT);
+ void configure(const ICLTensor *input,
+ ICLTensor *output,
+ const PaddingList &padding,
+ PixelValue constant_value = PixelValue(),
+ PaddingMode mode = PaddingMode::CONSTANT);
/** Set the input and output tensor.
*
* @param[in] compile_context The compile context to be used.
@@ -68,8 +72,12 @@
* @param[in] mode (Optional) Controls whether the padding should be filled with @p constant_value using CONSTANT,
* or reflect the input, either including the border values (SYMMETRIC) or not (REFLECT).
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const PaddingList &padding, PixelValue constant_value = PixelValue(),
- PaddingMode mode = PaddingMode::CONSTANT);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const PaddingList &padding,
+ PixelValue constant_value = PixelValue(),
+ PaddingMode mode = PaddingMode::CONSTANT);
/** Static function to check if given info will lead to a valid configuration of @ref CLPadLayerKernel
*
* @param[in] input Source tensor info. Data types supported: All.
@@ -80,7 +88,11 @@
* @param[in] mode (Optional) Controls whether the padding should be filled with @p constant_value using CONSTANT,
* or reflect the input, either including the border values (SYMMETRIC) or not (REFLECT).
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *output, const PaddingList &padding, PixelValue constant_value = PixelValue(), PaddingMode mode = PaddingMode::CONSTANT);
+ static Status validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const PaddingList &padding,
+ PixelValue constant_value = PixelValue(),
+ PaddingMode mode = PaddingMode::CONSTANT);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLPriorBoxLayerKernel.cpp b/src/core/CL/kernels/CLPriorBoxLayerKernel.cpp
index bf1b874..7dcdf1d 100644
--- a/src/core/CL/kernels/CLPriorBoxLayerKernel.cpp
+++ b/src/core/CL/kernels/CLPriorBoxLayerKernel.cpp
@@ -30,10 +30,10 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
-
#include "support/StringSupport.h"
using namespace arm_compute::misc::shape_calculator;
@@ -42,7 +42,10 @@
{
namespace
{
-Status validate_arguments(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const PriorBoxLayerInfo &info)
+Status validate_arguments(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ const ITensorInfo *output,
+ const PriorBoxLayerInfo &info)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input1, input2, output);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::F32);
@@ -51,10 +54,10 @@
// Check variances
const int var_size = info.variances().size();
- if(var_size > 1)
+ if (var_size > 1)
{
ARM_COMPUTE_RETURN_ERROR_ON_MSG(var_size != 4, "Must provide 4 variance values");
- for(int i = 0; i < var_size; ++i)
+ for (int i = 0; i < var_size; ++i)
{
ARM_COMPUTE_RETURN_ERROR_ON_MSG(var_size <= 0, "Must be greater than 0");
}
@@ -62,17 +65,19 @@
ARM_COMPUTE_RETURN_ERROR_ON_MSG(info.steps()[0] < 0.f, "Step x should be greater or equal to 0");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(info.steps()[1] < 0.f, "Step y should be greater or equal to 0");
- if(!info.max_sizes().empty())
+ if (!info.max_sizes().empty())
{
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(info.max_sizes().size() != info.min_sizes().size(), "Max and min sizes dimensions should match");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(info.max_sizes().size() != info.min_sizes().size(),
+ "Max and min sizes dimensions should match");
}
- for(unsigned int i = 0; i < info.max_sizes().size(); ++i)
+ for (unsigned int i = 0; i < info.max_sizes().size(); ++i)
{
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(info.max_sizes()[i] < info.min_sizes()[i], "Max size should be greater than min size");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(info.max_sizes()[i] < info.min_sizes()[i],
+ "Max size should be greater than min size");
}
- if(output != nullptr && output->total_size() != 0)
+ if (output != nullptr && output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON(output->dimension(1) != 2);
}
@@ -80,7 +85,11 @@
return Status{};
}
-std::pair<Status, Window> validate_and_configure_window(const ITensorInfo *input1, const ITensorInfo *input2, ITensorInfo *output, const PriorBoxLayerInfo &info, int num_priors)
+std::pair<Status, Window> validate_and_configure_window(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ ITensorInfo *output,
+ const PriorBoxLayerInfo &info,
+ int num_priors)
{
ARM_COMPUTE_UNUSED(input2);
// Output tensor auto initialization if not yet initialized
@@ -88,10 +97,11 @@
auto_init_if_empty(*output, output_shape, 1, input1->data_type());
const unsigned int num_elems_processed_per_iteration = 4 * num_priors;
- Window win = calculate_max_window(*output, Steps(num_elems_processed_per_iteration));
+ Window win = calculate_max_window(*output, Steps(num_elems_processed_per_iteration));
AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration);
bool window_changed = update_window_and_padding(win, output_access);
- Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+ Status err =
+ (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
return std::make_pair(err, win);
}
} // namespace
@@ -102,13 +112,25 @@
_type = CLKernelType::ELEMENTWISE;
}
-void CLPriorBoxLayerKernel::configure(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, const PriorBoxLayerInfo &info, cl::Buffer *min, cl::Buffer *max, cl::Buffer *aspect_ratios)
+void CLPriorBoxLayerKernel::configure(const ICLTensor *input1,
+ const ICLTensor *input2,
+ ICLTensor *output,
+ const PriorBoxLayerInfo &info,
+ cl::Buffer *min,
+ cl::Buffer *max,
+ cl::Buffer *aspect_ratios)
{
configure(CLKernelLibrary::get().get_compile_context(), input1, input2, output, info, min, max, aspect_ratios);
}
-void CLPriorBoxLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, const PriorBoxLayerInfo &info, cl::Buffer *min,
- cl::Buffer *max, cl::Buffer *aspect_ratios)
+void CLPriorBoxLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input1,
+ const ICLTensor *input2,
+ ICLTensor *output,
+ const PriorBoxLayerInfo &info,
+ cl::Buffer *min,
+ cl::Buffer *max,
+ cl::Buffer *aspect_ratios)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
@@ -135,7 +157,7 @@
int img_width = info.img_size().x;
int img_height = info.img_size().y;
- if(img_width == 0 || img_height == 0)
+ if (img_width == 0 || img_height == 0)
{
img_width = input2->info()->dimension(width_idx);
img_height = input2->info()->dimension(height_idx);
@@ -143,7 +165,7 @@
float step_x = info.steps()[0];
float step_y = info.steps()[0];
- if(step_x == 0.f || step_y == 0.f)
+ if (step_x == 0.f || step_y == 0.f)
{
step_x = static_cast<float>(img_width) / layer_width;
step_y = static_cast<float>(img_height) / layer_height;
@@ -162,18 +184,20 @@
build_opts.add_option("-DOFFSET=" + support::cpp11::to_string(info.offset()));
build_opts.add_option_if(info.clip(), "-DIN_PLACE");
- if(info.variances().size() > 1)
+ if (info.variances().size() > 1)
{
- for(unsigned int i = 0; i < info.variances().size(); ++i)
+ for (unsigned int i = 0; i < info.variances().size(); ++i)
{
- build_opts.add_option("-DVARIANCE_" + support::cpp11::to_string(i) + "=" + support::cpp11::to_string(info.variances().at(i)));
+ build_opts.add_option("-DVARIANCE_" + support::cpp11::to_string(i) + "=" +
+ support::cpp11::to_string(info.variances().at(i)));
}
}
else
{
- for(unsigned int i = 0; i < 4; ++i)
+ for (unsigned int i = 0; i < 4; ++i)
{
- build_opts.add_option("-DVARIANCE_" + support::cpp11::to_string(i) + "=" + support::cpp11::to_string(info.variances().at(0)));
+ build_opts.add_option("-DVARIANCE_" + support::cpp11::to_string(i) + "=" +
+ support::cpp11::to_string(info.variances().at(0)));
}
}
@@ -194,13 +218,17 @@
ICLKernel::configure_internal(win_config.second);
}
-Status CLPriorBoxLayerKernel::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const PriorBoxLayerInfo &info)
+Status CLPriorBoxLayerKernel::validate(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ const ITensorInfo *output,
+ const PriorBoxLayerInfo &info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input1, input2, output, info));
const int num_priors = info.aspect_ratios().size() * info.min_sizes().size() + info.max_sizes().size();
- ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input1->clone().get(), input2->clone().get(), output->clone().get(), info, num_priors)
- .first);
+ ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input1->clone().get(), input2->clone().get(),
+ output->clone().get(), info, num_priors)
+ .first);
return Status{};
}
@@ -211,8 +239,9 @@
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window);
queue.enqueueWriteBuffer(*_min, CL_TRUE, 0, _info.min_sizes().size() * sizeof(float), _info.min_sizes().data());
- queue.enqueueWriteBuffer(*_aspect_ratios, CL_TRUE, 0, _info.aspect_ratios().size() * sizeof(float), _info.aspect_ratios().data());
- if(!_info.max_sizes().empty())
+ queue.enqueueWriteBuffer(*_aspect_ratios, CL_TRUE, 0, _info.aspect_ratios().size() * sizeof(float),
+ _info.aspect_ratios().data());
+ if (!_info.max_sizes().empty())
{
queue.enqueueWriteBuffer(*_max, CL_TRUE, 0, _info.max_sizes().size() * sizeof(float), _info.max_sizes().data());
}
diff --git a/src/core/CL/kernels/CLPriorBoxLayerKernel.h b/src/core/CL/kernels/CLPriorBoxLayerKernel.h
index 6c369a7..a50e0c5 100644
--- a/src/core/CL/kernels/CLPriorBoxLayerKernel.h
+++ b/src/core/CL/kernels/CLPriorBoxLayerKernel.h
@@ -57,7 +57,13 @@
* @param[in] max Maximum prior box values
* @param[in] aspect_ratios Aspect ratio values
*/
- void configure(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, const PriorBoxLayerInfo &info, cl::Buffer *min, cl::Buffer *max, cl::Buffer *aspect_ratios);
+ void configure(const ICLTensor *input1,
+ const ICLTensor *input2,
+ ICLTensor *output,
+ const PriorBoxLayerInfo &info,
+ cl::Buffer *min,
+ cl::Buffer *max,
+ cl::Buffer *aspect_ratios);
/** Set the input and output tensors.
*
* @param[in] compile_context The compile context to be used.
@@ -69,8 +75,14 @@
* @param[in] max Maximum prior box values
* @param[in] aspect_ratios Aspect ratio values
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, const PriorBoxLayerInfo &info, cl::Buffer *min, cl::Buffer *max,
- cl::Buffer *aspect_ratios);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input1,
+ const ICLTensor *input2,
+ ICLTensor *output,
+ const PriorBoxLayerInfo &info,
+ cl::Buffer *min,
+ cl::Buffer *max,
+ cl::Buffer *aspect_ratios);
/** Static function to check if given info will lead to a valid configuration of @ref CLPriorBoxLayerKernel
*
* @param[in] input1 First source tensor info. Data types supported: F32. Data layouts supported: NCHW/NHWC.
@@ -80,14 +92,17 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const PriorBoxLayerInfo &info);
+ static Status validate(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ const ITensorInfo *output,
+ const PriorBoxLayerInfo &info);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
private:
- const ICLTensor *_input1;
- const ICLTensor *_input2;
+ const ICLTensor *_input1;
+ const ICLTensor *_input2;
ICLTensor *_output;
PriorBoxLayerInfo _info;
int _num_priors;
diff --git a/src/core/CL/kernels/CLQLSTMLayerNormalizationKernel.cpp b/src/core/CL/kernels/CLQLSTMLayerNormalizationKernel.cpp
index bd573e5..731fcb8 100644
--- a/src/core/CL/kernels/CLQLSTMLayerNormalizationKernel.cpp
+++ b/src/core/CL/kernels/CLQLSTMLayerNormalizationKernel.cpp
@@ -22,10 +22,12 @@
* SOFTWARE.
*/
#include "src/core/CL/kernels/CLQLSTMLayerNormalizationKernel.h"
+
#include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/utils/quantization/AsymmHelpers.h"
#include "arm_compute/core/Utils.h"
+#include "arm_compute/core/utils/quantization/AsymmHelpers.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
#include "support/StringSupport.h"
@@ -49,14 +51,19 @@
const uint32_t temp_num_elems_processed_per_iteration = max_cl_vector_width / input->element_size();
/* If width is less then step, then make step same as width to avoid global size being step instead of actual width. */
/* Or we should fix in arm_compute::enqueue() or arm_compute::calculate_max_window(). */
- const uint32_t num_elems_processed_per_iteration = (input->dimension(0) < temp_num_elems_processed_per_iteration) ? input->dimension(0) : temp_num_elems_processed_per_iteration;
+ const uint32_t num_elems_processed_per_iteration = (input->dimension(0) < temp_num_elems_processed_per_iteration)
+ ? input->dimension(0)
+ : temp_num_elems_processed_per_iteration;
// This kernel doesn't need padding
Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration));
return std::make_pair(Status{}, win);
}
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *weight, const ITensorInfo *bias)
+Status validate_arguments(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const ITensorInfo *weight,
+ const ITensorInfo *bias)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weight, bias, output);
@@ -72,7 +79,7 @@
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(weight, bias);
// Checks performed when output is configured
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
@@ -87,10 +94,14 @@
_type = CLKernelType::ELEMENTWISE;
}
-void CLQLSTMLayerNormalizationKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const ICLTensor *weight, const ICLTensor *bias)
+void CLQLSTMLayerNormalizationKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *weight,
+ const ICLTensor *bias)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, weight, bias, output);
- auto padding_info = get_padding_info({ input, weight, bias, output });
+ auto padding_info = get_padding_info({input, weight, bias, output});
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), weight->info(), bias->info()));
@@ -104,7 +115,8 @@
int32_t output_multiplier{};
int32_t output_shift{};
const UniformQuantizationInfo quan_info = _weight->info()->quantization_info().uniform();
- const Status status = quantization::calculate_quantized_multiplier(quan_info.scale, &output_multiplier, &output_shift);
+ const Status status =
+ quantization::calculate_quantized_multiplier(quan_info.scale, &output_multiplier, &output_shift);
output_shift *= -1;
// Set build options
@@ -114,8 +126,12 @@
build_opts.add_option("-DWIDTH=" + support::cpp11::to_string(input->info()->dimension(0)));
build_opts.add_option("-DOUTPUT_MULTIPLIER=" + support::cpp11::to_string(output_multiplier));
build_opts.add_option("-DOUTPUT_SHIFT=" + support::cpp11::to_string(output_shift));
- build_opts.add_option("-DMIN_BOUND=" + support::cpp11::to_string(std::get<0>(quantization::get_min_max_values_from_quantized_data_type(input->info()->data_type()))));
- build_opts.add_option("-DMAX_BOUND=" + support::cpp11::to_string(std::get<1>(quantization::get_min_max_values_from_quantized_data_type(input->info()->data_type()))));
+ build_opts.add_option("-DMIN_BOUND=" +
+ support::cpp11::to_string(std::get<0>(
+ quantization::get_min_max_values_from_quantized_data_type(input->info()->data_type()))));
+ build_opts.add_option("-DMAX_BOUND=" +
+ support::cpp11::to_string(std::get<1>(
+ quantization::get_min_max_values_from_quantized_data_type(input->info()->data_type()))));
// Create kernel
_kernel = create_kernel(compile_context, "qlstm_layer_normalization", build_opts.options());
@@ -135,12 +151,18 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-void CLQLSTMLayerNormalizationKernel::configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *weight, const ICLTensor *bias)
+void CLQLSTMLayerNormalizationKernel::configure(const ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *weight,
+ const ICLTensor *bias)
{
configure(CLKernelLibrary::get().get_compile_context(), input, output, weight, bias);
}
-Status CLQLSTMLayerNormalizationKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *weight, const ITensorInfo *bias)
+Status CLQLSTMLayerNormalizationKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const ITensorInfo *weight,
+ const ITensorInfo *bias)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, weight, bias));
ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first);
@@ -171,7 +193,6 @@
add_2D_tensor_argument(idx, _output, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(window.slide_window_slice_2D(slice));
+ } while (window.slide_window_slice_2D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLQLSTMLayerNormalizationKernel.h b/src/core/CL/kernels/CLQLSTMLayerNormalizationKernel.h
index 31085c3..ba912e1 100644
--- a/src/core/CL/kernels/CLQLSTMLayerNormalizationKernel.h
+++ b/src/core/CL/kernels/CLQLSTMLayerNormalizationKernel.h
@@ -63,7 +63,11 @@
* @param[in] weight Weight tensor. Data types supported: Same as @p input.
* @param[in] bias Bias tensor. Data types supported: S32.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const ICLTensor *weight, const ICLTensor *bias);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *weight,
+ const ICLTensor *bias);
/** Static function to check if given info will lead to a valid configuration of @ref CLQLSTMLayerNormalizationKernel
*
* @param[in] input Source tensor info with 2 dimensions. Data types supported: QSYMM16.
@@ -73,7 +77,8 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *weight, const ITensorInfo *bias);
+ static Status
+ validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *weight, const ITensorInfo *bias);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLROIAlignLayerKernel.cpp b/src/core/CL/kernels/CLROIAlignLayerKernel.cpp
index 69a6fa5..c97910e 100644
--- a/src/core/CL/kernels/CLROIAlignLayerKernel.cpp
+++ b/src/core/CL/kernels/CLROIAlignLayerKernel.cpp
@@ -31,6 +31,7 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -42,24 +43,29 @@
{
namespace
{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *rois, ITensorInfo *output, const ROIPoolingLayerInfo &pool_info)
+Status validate_arguments(const ITensorInfo *input,
+ const ITensorInfo *rois,
+ ITensorInfo *output,
+ const ROIPoolingLayerInfo &pool_info)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, rois, output);
ARM_COMPUTE_RETURN_ERROR_ON(rois->dimension(0) != 5);
ARM_COMPUTE_RETURN_ERROR_ON(rois->num_dimensions() > 2);
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
- ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F32, DataType::F16);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
+ DataType::F32, DataType::F16);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(input, DataLayout::NHWC, DataLayout::NCHW);
ARM_COMPUTE_RETURN_ERROR_ON((pool_info.pooled_width() == 0) || (pool_info.pooled_height() == 0));
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
- ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(compute_roi_align_shape(*input, *rois, pool_info), output->tensor_shape());
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(compute_roi_align_shape(*input, *rois, pool_info),
+ output->tensor_shape());
}
- if(is_data_type_quantized_asymmetric(input->data_type()))
+ if (is_data_type_quantized_asymmetric(input->data_type()))
{
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(rois, 1, DataType::QASYMM16);
@@ -82,12 +88,19 @@
_type = CLKernelType::ELEMENTWISE;
}
-void CLROIAlignLayerKernel::configure(const ICLTensor *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info)
+void CLROIAlignLayerKernel::configure(const ICLTensor *input,
+ const ICLTensor *rois,
+ ICLTensor *output,
+ const ROIPoolingLayerInfo &pool_info)
{
configure(CLKernelLibrary::get().get_compile_context(), input, rois, output, pool_info);
}
-void CLROIAlignLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info)
+void CLROIAlignLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *rois,
+ ICLTensor *output,
+ const ROIPoolingLayerInfo &pool_info)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, rois);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), rois->info(), output->info(), pool_info));
@@ -97,7 +110,7 @@
auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type());
output->info()->set_data_layout(input->info()->data_layout());
- auto padding_info = get_padding_info({ input, rois, output });
+ auto padding_info = get_padding_info({input, rois, output});
_input = input;
_output = output;
@@ -111,16 +124,23 @@
CLBuildOptions build_opts;
build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(data_type));
build_opts.add_option("-DDATA_SIZE=" + get_data_size_from_data_type(input->info()->data_type()));
- build_opts.add_option("-DMAX_DIM_X=" + support::cpp11::to_string(_input->info()->dimension(get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::WIDTH))));
- build_opts.add_option("-DMAX_DIM_Y=" + support::cpp11::to_string(_input->info()->dimension(get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::HEIGHT))));
- build_opts.add_option("-DMAX_DIM_Z=" + support::cpp11::to_string(_input->info()->dimension(get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::CHANNEL))));
+ build_opts.add_option("-DMAX_DIM_X=" +
+ support::cpp11::to_string(_input->info()->dimension(get_data_layout_dimension_index(
+ input->info()->data_layout(), DataLayoutDimension::WIDTH))));
+ build_opts.add_option("-DMAX_DIM_Y=" +
+ support::cpp11::to_string(_input->info()->dimension(get_data_layout_dimension_index(
+ input->info()->data_layout(), DataLayoutDimension::HEIGHT))));
+ build_opts.add_option("-DMAX_DIM_Z=" +
+ support::cpp11::to_string(_input->info()->dimension(get_data_layout_dimension_index(
+ input->info()->data_layout(), DataLayoutDimension::CHANNEL))));
build_opts.add_option("-DPOOLED_DIM_X=" + support::cpp11::to_string(pool_info.pooled_width()));
build_opts.add_option("-DPOOLED_DIM_Y=" + support::cpp11::to_string(pool_info.pooled_height()));
build_opts.add_option("-DSPATIAL_SCALE=" + float_to_string_with_full_precision(pool_info.spatial_scale()));
build_opts.add_option_if(input->info()->data_layout() == DataLayout::NHWC, "-DNHWC");
- build_opts.add_option_if(pool_info.sampling_ratio() > 0, "-DSAMPLING_RATIO=" + support::cpp11::to_string(pool_info.sampling_ratio()));
+ build_opts.add_option_if(pool_info.sampling_ratio() > 0,
+ "-DSAMPLING_RATIO=" + support::cpp11::to_string(pool_info.sampling_ratio()));
- if(is_qasymm)
+ if (is_qasymm)
{
const UniformQuantizationInfo iq_info = input->info()->quantization_info().uniform();
const UniformQuantizationInfo roisq_info = rois->info()->quantization_info().uniform();
@@ -144,7 +164,10 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLROIAlignLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *rois, ITensorInfo *output, const ROIPoolingLayerInfo &pool_info)
+Status CLROIAlignLayerKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *rois,
+ ITensorInfo *output,
+ const ROIPoolingLayerInfo &pool_info)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, rois, output, pool_info));
return Status{};
diff --git a/src/core/CL/kernels/CLROIAlignLayerKernel.h b/src/core/CL/kernels/CLROIAlignLayerKernel.h
index 5284a59..2e84e5d 100644
--- a/src/core/CL/kernels/CLROIAlignLayerKernel.h
+++ b/src/core/CL/kernels/CLROIAlignLayerKernel.h
@@ -61,7 +61,8 @@
* @note The z dimensions of @p output tensor and @p input tensor must be the same.
* @note The fourth dimension of @p output tensor must be the same as the number of elements in @p rois array.
*/
- void configure(const ICLTensor *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info);
+ void
+ configure(const ICLTensor *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info);
/** Set the input and output tensors.
*
* @param[in] compile_context The compile context to be used.
@@ -77,7 +78,11 @@
* @note The z dimensions of @p output tensor and @p input tensor must be the same.
* @note The fourth dimension of @p output tensor must be the same as the number of elements in @p rois array.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *rois,
+ ICLTensor *output,
+ const ROIPoolingLayerInfo &pool_info);
/** Static function to check if given info will lead to a valid configuration of @ref CLROIAlignLayerKernel
*
* @param[in] input Source tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32.
@@ -93,7 +98,10 @@
*
* @return a Status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *rois, ITensorInfo *output, const ROIPoolingLayerInfo &pool_info);
+ static Status validate(const ITensorInfo *input,
+ const ITensorInfo *rois,
+ ITensorInfo *output,
+ const ROIPoolingLayerInfo &pool_info);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue);
diff --git a/src/core/CL/kernels/CLROIPoolingLayerKernel.cpp b/src/core/CL/kernels/CLROIPoolingLayerKernel.cpp
index f6933c6..1b2c414 100644
--- a/src/core/CL/kernels/CLROIPoolingLayerKernel.cpp
+++ b/src/core/CL/kernels/CLROIPoolingLayerKernel.cpp
@@ -31,6 +31,7 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -48,7 +49,10 @@
_type = CLKernelType::ELEMENTWISE;
}
-Status CLROIPoolingLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *rois, const ITensorInfo *output, const ROIPoolingLayerInfo &pool_info)
+Status CLROIPoolingLayerKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *rois,
+ const ITensorInfo *output,
+ const ROIPoolingLayerInfo &pool_info)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, rois, output);
@@ -61,10 +65,11 @@
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32, DataType::F16, DataType::QASYMM8);
ARM_COMPUTE_RETURN_ERROR_ON((pool_info.pooled_width() == 0) || (pool_info.pooled_height() == 0));
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
- ARM_COMPUTE_RETURN_ERROR_ON((output->dimension(0) != pool_info.pooled_width()) || (output->dimension(1) != pool_info.pooled_height()));
+ ARM_COMPUTE_RETURN_ERROR_ON((output->dimension(0) != pool_info.pooled_width()) ||
+ (output->dimension(1) != pool_info.pooled_height()));
ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(2) != output->dimension(2));
ARM_COMPUTE_RETURN_ERROR_ON(rois->dimension(1) != output->dimension(3));
}
@@ -72,20 +77,30 @@
return Status{};
}
-void CLROIPoolingLayerKernel::configure(const ICLTensor *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info)
+void CLROIPoolingLayerKernel::configure(const ICLTensor *input,
+ const ICLTensor *rois,
+ ICLTensor *output,
+ const ROIPoolingLayerInfo &pool_info)
{
configure(CLKernelLibrary::get().get_compile_context(), input, rois, output, pool_info);
}
-void CLROIPoolingLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *rois, const ICLTensor *output, const ROIPoolingLayerInfo &pool_info)
+void CLROIPoolingLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *rois,
+ const ICLTensor *output,
+ const ROIPoolingLayerInfo &pool_info)
{
- ARM_COMPUTE_ERROR_THROW_ON(CLROIPoolingLayerKernel::validate(input->info(), rois->info(), output->info(), pool_info));
+ ARM_COMPUTE_ERROR_THROW_ON(
+ CLROIPoolingLayerKernel::validate(input->info(), rois->info(), output->info(), pool_info));
- auto padding_info = get_padding_info({ input, rois, output });
+ auto padding_info = get_padding_info({input, rois, output});
// Output auto initialization if not yet initialized
- TensorShape output_shape(pool_info.pooled_width(), pool_info.pooled_height(), input->info()->dimension(2), rois->info()->dimension(1));
- auto_init_if_empty(*(output->info()), output_shape, 1, input->info()->data_type(), output->info()->quantization_info());
+ TensorShape output_shape(pool_info.pooled_width(), pool_info.pooled_height(), input->info()->dimension(2),
+ rois->info()->dimension(1));
+ auto_init_if_empty(*(output->info()), output_shape, 1, input->info()->data_type(),
+ output->info()->quantization_info());
// Set instance variables
_input = input;
@@ -107,11 +122,12 @@
build_opts.add_option("-DPOOLED_DIM_Y=" + support::cpp11::to_string(pool_info.pooled_height()));
build_opts.add_option("-DSPATIAL_SCALE=" + support::cpp11::to_string(pool_info.spatial_scale()));
- if(is_qasymm)
+ if (is_qasymm)
{
// Determine quantization info scale, offset
UniformQuantizationInfo uqinfo = UniformQuantizationInfo();
- uqinfo = compute_requantization_scale_offset(_input->info()->quantization_info().uniform(), _output->info()->quantization_info().uniform());
+ uqinfo = compute_requantization_scale_offset(_input->info()->quantization_info().uniform(),
+ _output->info()->quantization_info().uniform());
build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(uqinfo.offset));
build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(uqinfo.scale));
diff --git a/src/core/CL/kernels/CLROIPoolingLayerKernel.h b/src/core/CL/kernels/CLROIPoolingLayerKernel.h
index 7b7b457..80bfb63 100644
--- a/src/core/CL/kernels/CLROIPoolingLayerKernel.h
+++ b/src/core/CL/kernels/CLROIPoolingLayerKernel.h
@@ -59,7 +59,8 @@
* @note The z dimensions of @p output tensor and @p input tensor must be the same.
* @note The fourth dimension of @p output tensor must be the same as the number of elements in @p rois array.
*/
- void configure(const ICLTensor *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info);
+ void
+ configure(const ICLTensor *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info);
/** Set the input and output tensors.
*
* @param[in] compile_context The compile context to be used.
@@ -74,7 +75,11 @@
* @note The z dimensions of @p output tensor and @p input tensor must be the same.
* @note The fourth dimension of @p output tensor must be the same as the number of elements in @p rois array.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *rois, const ICLTensor *output, const ROIPoolingLayerInfo &pool_info);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *rois,
+ const ICLTensor *output,
+ const ROIPoolingLayerInfo &pool_info);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
@@ -92,7 +97,10 @@
* @note The z dimensions of @p output tensor and @p input tensor must be the same.
* @note The fourth dimension of @p output tensor must be the same as the number of elements in @p rois array.
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *rois, const ITensorInfo *output, const ROIPoolingLayerInfo &pool_info);
+ static Status validate(const ITensorInfo *input,
+ const ITensorInfo *rois,
+ const ITensorInfo *output,
+ const ROIPoolingLayerInfo &pool_info);
private:
const ICLTensor *_input;
diff --git a/src/core/CL/kernels/CLRangeKernel.cpp b/src/core/CL/kernels/CLRangeKernel.cpp
index a06c2ee..622f621 100644
--- a/src/core/CL/kernels/CLRangeKernel.cpp
+++ b/src/core/CL/kernels/CLRangeKernel.cpp
@@ -28,6 +28,7 @@
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -42,11 +43,8 @@
Status validate_arguments(const ITensorInfo *output, const float start, const float end, const float step)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
- ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output,
- 1,
- DataType::U8, DataType::S8, DataType::QASYMM8,
- DataType::U16, DataType::S16,
- DataType::U32, DataType::S32,
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U8, DataType::S8, DataType::QASYMM8,
+ DataType::U16, DataType::S16, DataType::U32, DataType::S32,
DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(output);
@@ -56,19 +54,22 @@
ARM_COMPUTE_RETURN_ERROR_ON_MSG((start == end), "start of the requested sequence must not be equal to the end");
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(start, output->data_type(), output->quantization_info()), "start value is outside the range of the data type");
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(end, output->data_type(), output->quantization_info()), "end value is outside the range of the data type");
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(step, output->data_type(), output->quantization_info()), "step value is outside the range of the data type");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(start, output->data_type(), output->quantization_info()),
+ "start value is outside the range of the data type");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(end, output->data_type(), output->quantization_info()),
+ "end value is outside the range of the data type");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(step, output->data_type(), output->quantization_info()),
+ "step value is outside the range of the data type");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->num_dimensions() != 1, "Output has to be a 1-D tensor");
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->tensor_shape().total_size() < num_of_elements_in_range(start, end, step), "Output tensor size is incorrect");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->tensor_shape().total_size() < num_of_elements_in_range(start, end, step),
+ "Output tensor size is incorrect");
return Status{};
}
} // namespace
-CLRangeKernel::CLRangeKernel()
- : _start(0), _end(1), _step(1), _output(nullptr)
+CLRangeKernel::CLRangeKernel() : _start(0), _end(1), _step(1), _output(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
@@ -78,16 +79,18 @@
configure(CLKernelLibrary::get().get_compile_context(), output, start, end, step);
}
-void CLRangeKernel::configure(const CLCompileContext &compile_context, ICLTensor *output, const float start, const float end, const float step)
+void CLRangeKernel::configure(
+ const CLCompileContext &compile_context, ICLTensor *output, const float start, const float end, const float step)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(output->info(), start, end, step));
// Configure kernel window
- unsigned int num_elems_processed_per_iteration = adjust_vec_size(vector_size_byte_opencl / output->info()->element_size(), output->info()->dimension(0));
- Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration));
+ unsigned int num_elems_processed_per_iteration =
+ adjust_vec_size(vector_size_byte_opencl / output->info()->element_size(), output->info()->dimension(0));
+ Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration));
- auto padding_info = get_padding_info({ output });
+ auto padding_info = get_padding_info({output});
_start = start;
_end = end;
@@ -100,10 +103,11 @@
CLBuildOptions build_opts;
build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(output->info()->data_type()));
build_opts.add_option("-DVECTOR_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
- build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(output->info()->dimension(0) % num_elems_processed_per_iteration));
+ build_opts.add_option("-DVEC_SIZE_LEFTOVER=" +
+ support::cpp11::to_string(output->info()->dimension(0) % num_elems_processed_per_iteration));
build_opts.add_option("-DSTART=" + support::cpp11::to_string(start));
build_opts.add_option("-DSTEP=" + support::cpp11::to_string(step));
- if(is_data_type_quantized_asymmetric(output->info()->data_type()))
+ if (is_data_type_quantized_asymmetric(output->info()->data_type()))
{
const UniformQuantizationInfo qinfo = output->info()->quantization_info().uniform();
build_opts.add_option("-DOFFSET_OUT=" + support::cpp11::to_string(qinfo.offset));
diff --git a/src/core/CL/kernels/CLRangeKernel.h b/src/core/CL/kernels/CLRangeKernel.h
index 1b94a09..65251a1 100644
--- a/src/core/CL/kernels/CLRangeKernel.h
+++ b/src/core/CL/kernels/CLRangeKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLRANGEKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
diff --git a/src/core/CL/kernels/CLReductionOperationKernel.cpp b/src/core/CL/kernels/CLReductionOperationKernel.cpp
index e5cfb99..70875a2 100644
--- a/src/core/CL/kernels/CLReductionOperationKernel.cpp
+++ b/src/core/CL/kernels/CLReductionOperationKernel.cpp
@@ -28,15 +28,15 @@
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/TensorInfo.h"
-#include "arm_compute/core/Validate.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/core/utils/StringUtils.h"
+#include "arm_compute/core/Validate.h"
+
#include "src/core/AccessWindowStatic.h"
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
-
#include "support/StringSupport.h"
namespace arm_compute
@@ -47,23 +47,28 @@
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
- if(input->num_channels() == 1)
+ if (input->num_channels() == 1)
{
- ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::S32, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
+ DataType::S32, DataType::F16, DataType::F32);
}
else
{
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 2, DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON(axis == 0);
}
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(op == ReductionOperation::SUM_SQUARE && input->data_type() == DataType::QASYMM8, "Not supported reduction operation for QASYMM8");
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis >= TensorShape::num_max_dimensions, "Reduction axis greater than max number of dimensions");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(op == ReductionOperation::SUM_SQUARE && input->data_type() == DataType::QASYMM8,
+ "Not supported reduction operation for QASYMM8");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis >= TensorShape::num_max_dimensions,
+ "Reduction axis greater than max number of dimensions");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis > 3, "Unsupported reduction axis");
- ARM_COMPUTE_RETURN_ERROR_ON((op == ReductionOperation::MEAN_SUM) && (axis == 0) && (input->dimension(0) == 0) && (input->data_type() != DataType::QASYMM8)
- && (input->data_type() != DataType::QASYMM8_SIGNED));
- ARM_COMPUTE_RETURN_ERROR_ON_MSG((op == ReductionOperation::ARG_IDX_MAX) || (op == ReductionOperation::ARG_IDX_MIN), "Not supported reduction operation, use CLArgMinMaxLayer");
+ ARM_COMPUTE_RETURN_ERROR_ON((op == ReductionOperation::MEAN_SUM) && (axis == 0) && (input->dimension(0) == 0) &&
+ (input->data_type() != DataType::QASYMM8) &&
+ (input->data_type() != DataType::QASYMM8_SIGNED));
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG((op == ReductionOperation::ARG_IDX_MAX) || (op == ReductionOperation::ARG_IDX_MIN),
+ "Not supported reduction operation, use CLArgMinMaxLayer");
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
@@ -79,33 +84,42 @@
_type = CLKernelType::ELEMENTWISE;
}
-void CLReductionOperationKernel::configure(const ICLTensor *input, ICLTensor *output, unsigned int axis, ReductionOperation op)
+void CLReductionOperationKernel::configure(const ICLTensor *input,
+ ICLTensor *output,
+ unsigned int axis,
+ ReductionOperation op)
{
configure(CLKernelLibrary::get().get_compile_context(), input, output, axis, op);
}
-void CLReductionOperationKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, unsigned int axis, ReductionOperation op)
+void CLReductionOperationKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ unsigned int axis,
+ ReductionOperation op)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), axis, op));
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
_input = input;
_output = output;
_reduction_axis = axis;
_op = op;
- const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_reduced_shape(input->info()->tensor_shape(), axis, true);
- auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape).reset_padding().set_is_resizable(true));
+ const TensorShape output_shape =
+ arm_compute::misc::shape_calculator::compute_reduced_shape(input->info()->tensor_shape(), axis, true);
+ auto_init_if_empty(*output->info(),
+ input->info()->clone()->set_tensor_shape(output_shape).reset_padding().set_is_resizable(true));
// Set build options
CLBuildOptions build_opts;
DataType data_type = input->info()->data_type();
std::string data_type_promoted{};
- if(is_data_type_quantized(data_type))
+ if (is_data_type_quantized(data_type))
{
data_type_promoted = "int";
}
@@ -130,10 +144,14 @@
build_opts.add_option_if(op == ReductionOperation::PROD, "-DPROD");
build_opts.add_option_if(op == ReductionOperation::MIN, "-DMIN");
build_opts.add_option_if(op == ReductionOperation::MAX, "-DMAX");
- build_opts.add_option_if(is_data_type_quantized(data_type), "-DOFFSET=" + support::cpp11::to_string(input->info()->quantization_info().uniform().offset));
- build_opts.add_option_if(is_data_type_quantized(data_type), "-DSCALE=" + float_to_string_with_full_precision(input->info()->quantization_info().uniform().scale));
+ build_opts.add_option_if(is_data_type_quantized(data_type),
+ "-DOFFSET=" +
+ support::cpp11::to_string(input->info()->quantization_info().uniform().offset));
+ build_opts.add_option_if(
+ is_data_type_quantized(data_type),
+ "-DSCALE=" + float_to_string_with_full_precision(input->info()->quantization_info().uniform().scale));
- switch(op)
+ switch (op)
{
case ReductionOperation::SUM_SQUARE:
build_opts.add_option(("-DOPERATION=square_sum"));
@@ -159,7 +177,7 @@
std::string kernel_axis_name;
const bool is_serial_op = needs_serialized_reduction(_op, _input->info()->data_type(), _reduction_axis);
- switch(axis)
+ switch (axis)
{
case 0:
{
@@ -187,13 +205,17 @@
// Configure kernel window
Window win = calculate_max_window(*input->info(), Steps(vec_size));
- win.set(Window::DimX, Window::Dimension(win.x().start(), win.x().end() * _input->info()->num_channels(), win.x().step()));
+ win.set(Window::DimX,
+ Window::Dimension(win.x().start(), win.x().end() * _input->info()->num_channels(), win.x().step()));
ICLKernel::configure_internal(win);
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLReductionOperationKernel::validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int axis, ReductionOperation op)
+Status CLReductionOperationKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ unsigned int axis,
+ ReductionOperation op)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, axis, op));
return Status{};
@@ -205,18 +227,19 @@
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window);
const bool is_serial_op = needs_serialized_reduction(_op, _input->info()->data_type(), _reduction_axis);
- switch(_reduction_axis)
+ switch (_reduction_axis)
{
case 0:
{
// We use parallel reduction only in non quantized types
- if(is_serial_op)
+ if (is_serial_op)
{
// Get first input and output slices
- Window window_in{ window };
- window_in.set(Window::DimX, Window::Dimension(0, _input->info()->dimension(0), _input->info()->dimension(0)));
+ Window window_in{window};
+ window_in.set(Window::DimX,
+ Window::Dimension(0, _input->info()->dimension(0), _input->info()->dimension(0)));
- Window out_window{ window };
+ Window out_window{window};
out_window.set(Window::DimX, Window::Dimension(0, 0, 0));
Window in_slice = window_in.first_slice_window_1D();
@@ -228,8 +251,7 @@
add_1D_tensor_argument(idx, _input, in_slice);
add_1D_tensor_argument(idx, _output, out_slice);
enqueue(queue, *this, in_slice);
- }
- while(window_in.slide_window_slice_1D(in_slice) && out_window.slide_window_slice_1D(out_slice));
+ } while (window_in.slide_window_slice_1D(in_slice) && out_window.slide_window_slice_1D(out_slice));
}
else
{
@@ -251,8 +273,9 @@
case 1:
{
// Get first input and output slices
- Window window_in{ window };
- window_in.set(Window::DimY, Window::Dimension(0, _input->info()->dimension(1), _input->info()->dimension(1)));
+ Window window_in{window};
+ window_in.set(Window::DimY,
+ Window::Dimension(0, _input->info()->dimension(1), _input->info()->dimension(1)));
Window in_slice = window_in.first_slice_window_2D();
Window out_slice = window.first_slice_window_2D();
@@ -262,15 +285,15 @@
add_2D_tensor_argument(idx, _input, in_slice);
add_2D_tensor_argument(idx, _output, out_slice);
enqueue(queue, *this, in_slice);
- }
- while(window_in.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(out_slice));
+ } while (window_in.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(out_slice));
}
break;
case 2:
{
// Get first input and output slices
- Window window_in{ window };
- window_in.set(Window::DimZ, Window::Dimension(0, _input->info()->dimension(2), _input->info()->dimension(2)));
+ Window window_in{window};
+ window_in.set(Window::DimZ,
+ Window::Dimension(0, _input->info()->dimension(2), _input->info()->dimension(2)));
Window in_slice = window_in.first_slice_window_3D();
Window out_slice = window.first_slice_window_3D();
@@ -280,14 +303,13 @@
add_3D_tensor_argument(idx, _input, in_slice);
add_3D_tensor_argument(idx, _output, out_slice);
enqueue(queue, *this, in_slice);
- }
- while(window_in.slide_window_slice_3D(in_slice) && window.slide_window_slice_3D(out_slice));
+ } while (window_in.slide_window_slice_3D(in_slice) && window.slide_window_slice_3D(out_slice));
}
break;
case 3:
{
// Get first input and output slices
- Window window_in{ window };
+ Window window_in{window};
window_in.set(3, Window::Dimension(0, 1, 1));
Window in_slice = window_in.first_slice_window_4D();
Window out_slice = window.first_slice_window_4D();
@@ -298,8 +320,7 @@
add_4D_tensor_argument(idx, _input, in_slice);
add_4D_tensor_argument(idx, _output, out_slice);
enqueue(queue, *this, in_slice);
- }
- while(window_in.slide_window_slice_4D(in_slice) && window.slide_window_slice_4D(out_slice));
+ } while (window_in.slide_window_slice_4D(in_slice) && window.slide_window_slice_4D(out_slice));
}
break;
default:
diff --git a/src/core/CL/kernels/CLReductionOperationKernel.h b/src/core/CL/kernels/CLReductionOperationKernel.h
index b456378..2f94b2a 100644
--- a/src/core/CL/kernels/CLReductionOperationKernel.h
+++ b/src/core/CL/kernels/CLReductionOperationKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLREDUCTIONOPERATIONKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -67,7 +68,11 @@
* @param[in] axis Axis along which to reduce. Supported reduction axis : 0,1,2,3
* @param[in] op Reduction operation to perform. Operations supported: MEAN_SUM, PROD, SUM_SQUARE, SUM, MIN, MAX
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, unsigned int axis, ReductionOperation op);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ unsigned int axis,
+ ReductionOperation op);
/** Static function to check if given info will lead to a valid configuration of @ref CLReductionOperationKernel.
*
@@ -79,7 +84,8 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int axis, ReductionOperation op);
+ static Status
+ validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int axis, ReductionOperation op);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLReorgLayerKernel.cpp b/src/core/CL/kernels/CLReorgLayerKernel.cpp
index 3c74e80..9fd2194 100644
--- a/src/core/CL/kernels/CLReorgLayerKernel.cpp
+++ b/src/core/CL/kernels/CLReorgLayerKernel.cpp
@@ -28,9 +28,10 @@
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/TensorInfo.h"
-#include "arm_compute/core/Validate.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/core/utils/StringUtils.h"
+#include "arm_compute/core/Validate.h"
+
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
#include "support/StringSupport.h"
@@ -51,13 +52,16 @@
const size_t idx_height = get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::HEIGHT);
ARM_COMPUTE_RETURN_ERROR_ON(stride <= 0);
- ARM_COMPUTE_RETURN_ERROR_ON_MSG((input->tensor_shape()[idx_width] % stride) != 0, "The width of the input tensor must be a multiple of stride");
- ARM_COMPUTE_RETURN_ERROR_ON_MSG((input->tensor_shape()[idx_height] % stride) != 0, "The height of the input tensor must be a multiple of stride");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG((input->tensor_shape()[idx_width] % stride) != 0,
+ "The width of the input tensor must be a multiple of stride");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG((input->tensor_shape()[idx_height] % stride) != 0,
+ "The height of the input tensor must be a multiple of stride");
// Validate output if initialized
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
- const TensorInfo tensor_info_output = output->clone()->set_tensor_shape(misc::shape_calculator::compute_reorg_output_shape(*input, stride));
+ const TensorInfo tensor_info_output =
+ output->clone()->set_tensor_shape(misc::shape_calculator::compute_reorg_output_shape(*input, stride));
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, &tensor_info_output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
}
@@ -66,8 +70,7 @@
}
} // namespace
-CLReorgLayerKernel::CLReorgLayerKernel()
- : _input(nullptr), _output(nullptr)
+CLReorgLayerKernel::CLReorgLayerKernel() : _input(nullptr), _output(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
@@ -77,17 +80,22 @@
configure(CLKernelLibrary::get().get_compile_context(), input, output, stride);
}
-void CLReorgLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, int32_t stride)
+void CLReorgLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ int32_t stride)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), stride));
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
_input = input;
_output = output;
- std::string kernel_name = std::string("reorg_layer_") + lower_string(string_from_data_layout(input->info()->data_layout()));
- const size_t idx_channel = get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::CHANNEL);
+ std::string kernel_name =
+ std::string("reorg_layer_") + lower_string(string_from_data_layout(input->info()->data_layout()));
+ const size_t idx_channel =
+ get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::CHANNEL);
// Create kernel
CLBuildOptions build_opts;
@@ -98,7 +106,9 @@
// Configure window
// auto inizialize the output tensor if not yet initialized
- auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(misc::shape_calculator::compute_reorg_output_shape(*input->info(), stride)));
+ auto_init_if_empty(*output->info(),
+ input->info()->clone()->set_tensor_shape(
+ misc::shape_calculator::compute_reorg_output_shape(*input->info(), stride)));
Window win = calculate_max_window(*output->info(), Steps());
@@ -119,7 +129,9 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLReorgLayerKernel::validate(const arm_compute::ITensorInfo *input, const arm_compute::ITensorInfo *output, int32_t stride)
+Status CLReorgLayerKernel::validate(const arm_compute::ITensorInfo *input,
+ const arm_compute::ITensorInfo *output,
+ int32_t stride)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, stride));
@@ -139,7 +151,6 @@
add_3D_tensor_argument(idx, _input, slice);
add_3D_tensor_argument(idx, _output, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(window.slide_window_slice_3D(slice));
+ } while (window.slide_window_slice_3D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLReorgLayerKernel.h b/src/core/CL/kernels/CLReorgLayerKernel.h
index 455a617..f335071 100644
--- a/src/core/CL/kernels/CLReorgLayerKernel.h
+++ b/src/core/CL/kernels/CLReorgLayerKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLREORGLAYERKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
diff --git a/src/core/CL/kernels/CLReverseKernel.cpp b/src/core/CL/kernels/CLReverseKernel.cpp
index 0d70ff4..79a0f03 100644
--- a/src/core/CL/kernels/CLReverseKernel.cpp
+++ b/src/core/CL/kernels/CLReverseKernel.cpp
@@ -30,6 +30,7 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -49,7 +50,7 @@
ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis->dimension(0) > 4, "Only up to 4 dimensions can be reversed");
// Checks performed when output is configured
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
@@ -60,8 +61,7 @@
}
} // namespace
-CLReverseKernel::CLReverseKernel()
- : _input(nullptr), _output(nullptr), _axis(nullptr)
+CLReverseKernel::CLReverseKernel() : _input(nullptr), _output(nullptr), _axis(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
@@ -71,10 +71,13 @@
configure(CLKernelLibrary::get().get_compile_context(), input, output, axis);
}
-void CLReverseKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const ICLTensor *axis)
+void CLReverseKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *axis)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, axis);
- auto padding_info = get_padding_info({ input, output, axis });
+ auto padding_info = get_padding_info({input, output, axis});
_input = input;
_output = output;
@@ -138,7 +141,6 @@
add_1D_tensor_argument(idx, _axis, axis_slice);
add_4D_tensor_argument(idx, _output, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(collapsed.slide_window_slice_4D(slice));
+ } while (collapsed.slide_window_slice_4D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLReverseKernel.h b/src/core/CL/kernels/CLReverseKernel.h
index 4a21e4f..fbd99dc 100644
--- a/src/core/CL/kernels/CLReverseKernel.h
+++ b/src/core/CL/kernels/CLReverseKernel.h
@@ -60,7 +60,10 @@
* @param[out] output Output tensor. Data type supported: Same as @p input
* @param[in] axis Axis tensor. Contains the indices of the dimensions to reverse. Data type supported: U32
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const ICLTensor *axis);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const ICLTensor *axis);
/** Static function to check if given info will lead to a valid configuration of @ref CLReverseKernel
*
diff --git a/src/core/CL/kernels/CLSelectKernel.cpp b/src/core/CL/kernels/CLSelectKernel.cpp
index c0e014e..703c64d 100644
--- a/src/core/CL/kernels/CLSelectKernel.cpp
+++ b/src/core/CL/kernels/CLSelectKernel.cpp
@@ -30,10 +30,10 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/helpers/AdjustVecSize.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
-
#include "support/StringSupport.h"
namespace arm_compute
@@ -51,9 +51,11 @@
const bool is_same_rank = (c->tensor_shape().num_dimensions() == x->tensor_shape().num_dimensions());
ARM_COMPUTE_RETURN_ERROR_ON(is_same_rank && (x->tensor_shape() != c->tensor_shape()));
- ARM_COMPUTE_RETURN_ERROR_ON(!is_same_rank && ((c->tensor_shape().num_dimensions() > 1) || (c->tensor_shape().x() != x->tensor_shape()[x->tensor_shape().num_dimensions() - 1])));
+ ARM_COMPUTE_RETURN_ERROR_ON(!is_same_rank &&
+ ((c->tensor_shape().num_dimensions() > 1) ||
+ (c->tensor_shape().x() != x->tensor_shape()[x->tensor_shape().num_dimensions() - 1])));
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(x, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(x, output);
@@ -63,13 +65,16 @@
}
} // namespace
-CLSelectKernel::CLSelectKernel()
- : _c(nullptr), _x(nullptr), _y(nullptr), _output(nullptr), _has_same_rank(false)
+CLSelectKernel::CLSelectKernel() : _c(nullptr), _x(nullptr), _y(nullptr), _output(nullptr), _has_same_rank(false)
{
_type = CLKernelType::ELEMENTWISE;
}
-void CLSelectKernel::configure(const CLCompileContext &compile_context, const ICLTensor *c, const ICLTensor *x, const ICLTensor *y, ICLTensor *output)
+void CLSelectKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *c,
+ const ICLTensor *x,
+ const ICLTensor *y,
+ ICLTensor *output)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(c, x, y, output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(c->info(), x->info(), y->info(), output->info()));
@@ -80,7 +85,7 @@
_output = output;
_has_same_rank = (c->info()->tensor_shape().num_dimensions() == x->info()->tensor_shape().num_dimensions());
- auto padding_info = get_padding_info({ c, x, y, output });
+ auto padding_info = get_padding_info({c, x, y, output});
const unsigned int vec_size_x = adjust_vec_size(16 / x->info()->element_size(), x->info()->dimension(0));
const int vec_size_x_leftovers = output->info()->dimension(0) % vec_size_x;
@@ -92,14 +97,14 @@
// Create kernel
std::string kernel_name = "select";
- if(_has_same_rank)
+ if (_has_same_rank)
{
kernel_name += "_same_rank";
}
else
{
const bool is_input_rank_greater_than_two = x->info()->tensor_shape().num_dimensions() > 2;
- if(is_input_rank_greater_than_two)
+ if (is_input_rank_greater_than_two)
{
const size_t width = x->info()->tensor_shape().x();
const size_t height = x->info()->tensor_shape().y();
@@ -128,7 +133,8 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLSelectKernel::validate(const ITensorInfo *c, const ITensorInfo *x, const ITensorInfo *y, const ITensorInfo *output)
+Status
+CLSelectKernel::validate(const ITensorInfo *c, const ITensorInfo *x, const ITensorInfo *y, const ITensorInfo *output)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(c, x, y, output));
return Status{};
@@ -142,7 +148,7 @@
Window collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
Window slice = collapsed.first_slice_window_3D();
- if(!_has_same_rank)
+ if (!_has_same_rank)
{
Window vector_slice = window.first_slice_window_1D();
vector_slice.set(Window::DimX, Window::Dimension(0, 0, 0));
@@ -153,7 +159,7 @@
do
{
unsigned int idx = _has_same_rank ? 0 : num_arguments_per_1D_tensor();
- if(_has_same_rank)
+ if (_has_same_rank)
{
add_3D_tensor_argument(idx, _c, slice);
}
@@ -162,7 +168,6 @@
add_3D_tensor_argument(idx, _output, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(collapsed.slide_window_slice_3D(slice));
+ } while (collapsed.slide_window_slice_3D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLSelectKernel.h b/src/core/CL/kernels/CLSelectKernel.h
index b8c10cd..c4256fd 100644
--- a/src/core/CL/kernels/CLSelectKernel.h
+++ b/src/core/CL/kernels/CLSelectKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLSELECTKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -60,7 +61,11 @@
* @param[out] y Second input tensor. Data types supported: Same as @p x
* @param[in] output Output tensor. Data types supported: Same as @p x.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *c, const ICLTensor *x, const ICLTensor *y, ICLTensor *output);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *c,
+ const ICLTensor *x,
+ const ICLTensor *y,
+ ICLTensor *output);
/** Static function to check if given info will lead to a valid configuration of @ref CLSelectKernel
*
* @param[in] c Condition input tensor. Data types supported: U8.
diff --git a/src/core/CL/kernels/CLSpaceToBatchLayerKernel.cpp b/src/core/CL/kernels/CLSpaceToBatchLayerKernel.cpp
index 3632ae2..f4c0839 100644
--- a/src/core/CL/kernels/CLSpaceToBatchLayerKernel.cpp
+++ b/src/core/CL/kernels/CLSpaceToBatchLayerKernel.cpp
@@ -27,6 +27,7 @@
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -38,19 +39,22 @@
{
namespace
{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *block_info, const ITensorInfo *paddings, const ITensorInfo *output)
+Status validate_arguments(const ITensorInfo *input,
+ const ITensorInfo *block_info,
+ const ITensorInfo *paddings,
+ const ITensorInfo *output)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, block_info, paddings, output);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(block_info, 1, DataType::S32);
ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
ARM_COMPUTE_RETURN_ERROR_ON(block_info->num_dimensions() > 1);
- ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(block_info->tensor_shape(), TensorShape{ 2 });
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(block_info->tensor_shape(), TensorShape{2});
ARM_COMPUTE_RETURN_ERROR_ON(paddings->num_dimensions() > 2);
- ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(paddings->tensor_shape(), TensorShape{ 2, 2 });
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(paddings->tensor_shape(), TensorShape{2, 2});
// Validate output if initialized
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
const DataLayout data_layout = input->data_layout();
const int idx_channel = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
@@ -61,7 +65,11 @@
return Status{};
}
-Status validate_arguments_static(const ITensorInfo *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right,
+Status validate_arguments_static(const ITensorInfo *input,
+ const int block_shape_x,
+ const int block_shape_y,
+ const Size2D &padding_left,
+ const Size2D &padding_right,
const ITensorInfo *output)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
@@ -70,9 +78,10 @@
ARM_COMPUTE_RETURN_ERROR_ON(block_shape_x < 1 || block_shape_y < 1);
// Validate output if initialized
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
- TensorShape expected_output_shape = misc::shape_calculator::compute_space_to_batch_shape(input, block_shape_x, block_shape_y, padding_left, padding_right);
+ TensorShape expected_output_shape = misc::shape_calculator::compute_space_to_batch_shape(
+ input, block_shape_x, block_shape_y, padding_left, padding_right);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), expected_output_shape);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
@@ -88,16 +97,24 @@
_type = CLKernelType::ELEMENTWISE;
}
-void CLSpaceToBatchLayerKernel::configure(const ICLTensor *input, const ICLTensor *block_shape, const ICLTensor *paddings, ICLTensor *output)
+void CLSpaceToBatchLayerKernel::configure(const ICLTensor *input,
+ const ICLTensor *block_shape,
+ const ICLTensor *paddings,
+ ICLTensor *output)
{
configure(CLKernelLibrary::get().get_compile_context(), input, block_shape, paddings, output);
}
-void CLSpaceToBatchLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *block_shape, const ICLTensor *paddings, ICLTensor *output)
+void CLSpaceToBatchLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *block_shape,
+ const ICLTensor *paddings,
+ ICLTensor *output)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, block_shape, paddings, output);
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), block_shape->info(), paddings->info(), output->info()));
- auto padding_info = get_padding_info({ input, block_shape, paddings, output });
+ ARM_COMPUTE_ERROR_THROW_ON(
+ validate_arguments(input->info(), block_shape->info(), paddings->info(), output->info()));
+ auto padding_info = get_padding_info({input, block_shape, paddings, output});
_input = input;
_block_shape = block_shape;
@@ -111,14 +128,17 @@
// Create kernel
CLBuildOptions build_opts;
- build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(data_size_from_type(input->info()->data_type())));
+ build_opts.add_option("-DDATA_TYPE=" +
+ get_cl_unsigned_type_from_element_size(data_size_from_type(input->info()->data_type())));
build_opts.add_option("-DWIDTH_OUT=" + support::cpp11::to_string(output->info()->dimension(idx_width)));
build_opts.add_option("-DHEIGHT_OUT=" + support::cpp11::to_string(output->info()->dimension(idx_height)));
build_opts.add_option("-DBATCH_SIZE=" + support::cpp11::to_string(output->info()->dimension(idx_batch)));
build_opts.add_option("-DWIDTH_IN=" + support::cpp11::to_string(input->info()->dimension(idx_width)));
build_opts.add_option("-DHEIGHT_IN=" + support::cpp11::to_string(input->info()->dimension(idx_height)));
build_opts.add_option("-DBATCH_IN=" + support::cpp11::to_string(input->info()->dimension(idx_batch)));
- _kernel = create_kernel(compile_context, "space_to_batch_" + lower_string(string_from_data_layout(input->info()->data_layout())), build_opts.options());
+ _kernel = create_kernel(compile_context,
+ "space_to_batch_" + lower_string(string_from_data_layout(input->info()->data_layout())),
+ build_opts.options());
// Configure kernel window
Window win = calculate_max_window(*output->info(), Steps());
@@ -126,22 +146,34 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-void CLSpaceToBatchLayerKernel::configure(const ICLTensor *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right,
- ICLTensor *output)
+void CLSpaceToBatchLayerKernel::configure(const ICLTensor *input,
+ const int block_shape_x,
+ const int block_shape_y,
+ const Size2D &padding_left,
+ const Size2D &padding_right,
+ ICLTensor *output)
{
- configure(CLKernelLibrary::get().get_compile_context(), input, block_shape_x, block_shape_y, padding_left, padding_right, output);
+ configure(CLKernelLibrary::get().get_compile_context(), input, block_shape_x, block_shape_y, padding_left,
+ padding_right, output);
}
-void CLSpaceToBatchLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left,
- const Size2D &padding_right,
- ICLTensor *output)
+void CLSpaceToBatchLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const int block_shape_x,
+ const int block_shape_y,
+ const Size2D &padding_left,
+ const Size2D &padding_right,
+ ICLTensor *output)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- TensorShape output_shape = misc::shape_calculator::compute_space_to_batch_shape(input->info(), block_shape_x, block_shape_y, padding_left, padding_right);
- auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type(), input->info()->quantization_info());
+ TensorShape output_shape = misc::shape_calculator::compute_space_to_batch_shape(
+ input->info(), block_shape_x, block_shape_y, padding_left, padding_right);
+ auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type(),
+ input->info()->quantization_info());
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments_static(input->info(), block_shape_x, block_shape_y, padding_left, padding_right, output->info()));
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments_static(input->info(), block_shape_x, block_shape_y, padding_left,
+ padding_right, output->info()));
_input = input;
_output = output;
@@ -153,7 +185,8 @@
// Create kernel
CLBuildOptions build_opts;
- build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(data_size_from_type(input->info()->data_type())));
+ build_opts.add_option("-DDATA_TYPE=" +
+ get_cl_unsigned_type_from_element_size(data_size_from_type(input->info()->data_type())));
build_opts.add_option("-DWIDTH_OUT=" + support::cpp11::to_string(output->info()->dimension(idx_width)));
build_opts.add_option("-DHEIGHT_OUT=" + support::cpp11::to_string(output->info()->dimension(idx_height)));
build_opts.add_option("-DBATCH_SIZE=" + support::cpp11::to_string(output->info()->dimension(idx_batch)));
@@ -166,22 +199,32 @@
build_opts.add_option("-DPAD_RIGHT_X=" + support::cpp11::to_string(padding_right.x()));
build_opts.add_option("-DPAD_LEFT_Y=" + support::cpp11::to_string(padding_left.y()));
build_opts.add_option("-DPAD_RIGHT_Y=" + support::cpp11::to_string(padding_right.y()));
- _kernel = create_kernel(compile_context, "space_to_batch_static_" + lower_string(string_from_data_layout(input->info()->data_layout())), build_opts.options());
+ _kernel = create_kernel(
+ compile_context, "space_to_batch_static_" + lower_string(string_from_data_layout(input->info()->data_layout())),
+ build_opts.options());
// Configure kernel window
Window win = calculate_max_window(*output->info(), Steps());
ICLKernel::configure_internal(win);
}
-Status CLSpaceToBatchLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *block_shape, const ITensorInfo *paddings, const ITensorInfo *output)
+Status CLSpaceToBatchLayerKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *block_shape,
+ const ITensorInfo *paddings,
+ const ITensorInfo *output)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, block_shape, paddings, output));
return Status{};
}
-Status CLSpaceToBatchLayerKernel::validate(const ITensorInfo *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right,
+Status CLSpaceToBatchLayerKernel::validate(const ITensorInfo *input,
+ const int block_shape_x,
+ const int block_shape_y,
+ const Size2D &padding_left,
+ const Size2D &padding_right,
const ITensorInfo *output)
{
- ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_static(input, block_shape_x, block_shape_y, padding_left, padding_right, output));
+ ARM_COMPUTE_RETURN_ON_ERROR(
+ validate_arguments_static(input, block_shape_x, block_shape_y, padding_left, padding_right, output));
return Status{};
}
@@ -218,7 +261,6 @@
add_3D_tensor_argument(idx, _output, slice_out);
enqueue(queue, *this, slice_out, lws_hint());
++batch_id;
- }
- while(window.slide_window_slice_3D(slice_out));
+ } while (window.slide_window_slice_3D(slice_out));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLSpaceToBatchLayerKernel.h b/src/core/CL/kernels/CLSpaceToBatchLayerKernel.h
index 4817cfe..f9dce9d 100644
--- a/src/core/CL/kernels/CLSpaceToBatchLayerKernel.h
+++ b/src/core/CL/kernels/CLSpaceToBatchLayerKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLSPACETOBATCHLAYERKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -63,7 +64,11 @@
* @param[in] paddings 2-D tensor with shape [2, M] (First dimension is the fastest-changing dimension). Supported M: 2. Data types supported: S32
* @param[out] output Tensor output. Data types supported: same as @p input
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *block_shape, const ICLTensor *paddings, ICLTensor *output);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const ICLTensor *block_shape,
+ const ICLTensor *paddings,
+ ICLTensor *output);
/** Initialise the kernel's input and output. (Static block shape and paddings)
*
* @param[in] input Tensor input. Supported tensor rank: 4. Data types supported: All.
@@ -73,7 +78,12 @@
* @param[in] padding_right The padding at the end of every dimension of the output tensor.
* @param[out] output Tensor output. Data types supported: same as @p input
*/
- void configure(const ICLTensor *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right, ICLTensor *output);
+ void configure(const ICLTensor *input,
+ const int block_shape_x,
+ const int block_shape_y,
+ const Size2D &padding_left,
+ const Size2D &padding_right,
+ ICLTensor *output);
/** Initialise the kernel's input and output. (Static block shape and paddings)
*
* @param[in] compile_context The compile context to be used.
@@ -84,8 +94,13 @@
* @param[in] padding_right The padding at the end of every dimension of the output tensor.
* @param[out] output Tensor output. Data types supported: same as @p input
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right,
- ICLTensor *output);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ const int block_shape_x,
+ const int block_shape_y,
+ const Size2D &padding_left,
+ const Size2D &padding_right,
+ ICLTensor *output);
/** Static function to check if given info will lead to a valid configuration of @ref CLSpaceToBatchLayerKernel
*
* @param[in] input Tensor input. Supported tensor rank: 4. Data types supported: All.
@@ -95,7 +110,10 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *block_shape, const ITensorInfo *paddings, const ITensorInfo *output);
+ static Status validate(const ITensorInfo *input,
+ const ITensorInfo *block_shape,
+ const ITensorInfo *paddings,
+ const ITensorInfo *output);
/** Static function to check if given info will lead to a valid configuration of @ref CLSpaceToBatchLayerKernel (Static block shape and paddings)
*
* @param[in] input Tensor input. Supported tensor rank: 4. Data types supported: All.
@@ -107,7 +125,12 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right, const ITensorInfo *output);
+ static Status validate(const ITensorInfo *input,
+ const int block_shape_x,
+ const int block_shape_y,
+ const Size2D &padding_left,
+ const Size2D &padding_right,
+ const ITensorInfo *output);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLSpaceToDepthLayerKernel.cpp b/src/core/CL/kernels/CLSpaceToDepthLayerKernel.cpp
index c5ffdb5..25662b5 100644
--- a/src/core/CL/kernels/CLSpaceToDepthLayerKernel.cpp
+++ b/src/core/CL/kernels/CLSpaceToDepthLayerKernel.cpp
@@ -27,6 +27,7 @@
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
@@ -45,7 +46,7 @@
ARM_COMPUTE_RETURN_ERROR_ON(block_shape < 1);
// Validate output if initialized
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
const DataLayout data_layout = input->data_layout();
const int idx_width = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
@@ -64,8 +65,7 @@
}
} // namespace
-CLSpaceToDepthLayerKernel::CLSpaceToDepthLayerKernel()
- : _input(nullptr), _output(nullptr), _block_shape()
+CLSpaceToDepthLayerKernel::CLSpaceToDepthLayerKernel() : _input(nullptr), _output(nullptr), _block_shape()
{
_type = CLKernelType::ELEMENTWISE;
}
@@ -75,10 +75,13 @@
configure(CLKernelLibrary::get().get_compile_context(), input, output, block_shape);
}
-void CLSpaceToDepthLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, int32_t block_shape)
+void CLSpaceToDepthLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ int32_t block_shape)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- auto padding_info = get_padding_info({ input, output });
+ auto padding_info = get_padding_info({input, output});
TensorShape output_shape = compute_space_to_depth_shape(input->info(), block_shape);
auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type());
@@ -94,11 +97,14 @@
// Create kernel
CLBuildOptions build_opts;
- build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(data_size_from_type(output->info()->data_type())));
+ build_opts.add_option("-DDATA_TYPE=" +
+ get_cl_unsigned_type_from_element_size(data_size_from_type(output->info()->data_type())));
build_opts.add_option("-DCHANNEL_SIZE=" + support::cpp11::to_string(output->info()->dimension(idx_channel)));
build_opts.add_option("-DBLOCK_SHAPE=" + support::cpp11::to_string(block_shape));
build_opts.add_option("-DWIDTH_IN=" + support::cpp11::to_string(output->info()->dimension(idx_width)));
- _kernel = create_kernel(compile_context, "space_to_depth_" + lower_string(string_from_data_layout(input->info()->data_layout())), build_opts.options());
+ _kernel = create_kernel(compile_context,
+ "space_to_depth_" + lower_string(string_from_data_layout(input->info()->data_layout())),
+ build_opts.options());
// Configure kernel window
Window win = calculate_max_window(*output->info(), Steps());
@@ -136,7 +142,6 @@
enqueue(queue, *this, slice_out, lws_hint());
++batch_id;
- }
- while(window.slide_window_slice_3D(slice_out));
+ } while (window.slide_window_slice_3D(slice_out));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLSpaceToDepthLayerKernel.h b/src/core/CL/kernels/CLSpaceToDepthLayerKernel.h
index bb1ac5f..d093291 100644
--- a/src/core/CL/kernels/CLSpaceToDepthLayerKernel.h
+++ b/src/core/CL/kernels/CLSpaceToDepthLayerKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CLSPACETODEPTHLAYERKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -61,7 +62,8 @@
* @param[out] output Tensor output. Data types supported: same as @p input
* @param[in] block_shape Block shape value.
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, int32_t block_shape);
+ void
+ configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, int32_t block_shape);
/** Static function to check if given info will lead to a valid configuration of @ref CLSpaceToDepthLayerKernel.
*
* @param[in] input Tensor input info. Supported tensor rank: 4. Data types supported: All.
diff --git a/src/core/CL/kernels/CLStackLayerKernel.cpp b/src/core/CL/kernels/CLStackLayerKernel.cpp
index 075c93a..23e2671 100644
--- a/src/core/CL/kernels/CLStackLayerKernel.cpp
+++ b/src/core/CL/kernels/CLStackLayerKernel.cpp
@@ -30,10 +30,10 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
#include "src/core/CL/CLValidate.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
-
#include "support/StringSupport.h"
using namespace arm_compute::misc::shape_calculator;
@@ -42,7 +42,11 @@
{
namespace
{
-Status validate_arguments(const ITensorInfo *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, const ITensorInfo *output)
+Status validate_arguments(const ITensorInfo *input,
+ unsigned int axis,
+ unsigned int idx_input,
+ unsigned int num_tensors,
+ const ITensorInfo *output)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
@@ -51,9 +55,10 @@
ARM_COMPUTE_RETURN_ERROR_ON(axis > input->num_dimensions());
ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
- ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), compute_stack_shape(*input, axis, num_tensors));
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(),
+ compute_stack_shape(*input, axis, num_tensors));
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
}
@@ -61,7 +66,8 @@
return Status{};
}
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, unsigned int axis, unsigned int num_tensors, ITensorInfo *output)
+std::pair<Status, Window>
+validate_and_configure_window(ITensorInfo *input, unsigned int axis, unsigned int num_tensors, ITensorInfo *output)
{
// Output auto inizialitation if not yet initialized
auto_init_if_empty(*output, input->clone()->set_tensor_shape(compute_stack_shape(*input, axis, num_tensors)));
@@ -73,18 +79,23 @@
}
} // namespace
-CLStackLayerKernel::CLStackLayerKernel()
- : _input(nullptr), _output(nullptr)
+CLStackLayerKernel::CLStackLayerKernel() : _input(nullptr), _output(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
-void CLStackLayerKernel::configure(const ICLTensor *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, ICLTensor *output)
+void CLStackLayerKernel::configure(
+ const ICLTensor *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, ICLTensor *output)
{
configure(CLKernelLibrary::get().get_compile_context(), input, axis, idx_input, num_tensors, output);
}
-void CLStackLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, ICLTensor *output)
+void CLStackLayerKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ unsigned int axis,
+ unsigned int idx_input,
+ unsigned int num_tensors,
+ ICLTensor *output)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), axis, idx_input, num_tensors, output->info()));
@@ -112,10 +123,15 @@
_kernel.setArg<cl_uint>(idx, idx_input);
}
-Status CLStackLayerKernel::validate(const ITensorInfo *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, const ITensorInfo *output)
+Status CLStackLayerKernel::validate(const ITensorInfo *input,
+ unsigned int axis,
+ unsigned int idx_input,
+ unsigned int num_tensors,
+ const ITensorInfo *output)
{
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, axis, idx_input, num_tensors, output));
- ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), axis, num_tensors, output->clone().get()).first);
+ ARM_COMPUTE_RETURN_ON_ERROR(
+ validate_and_configure_window(input->clone().get(), axis, num_tensors, output->clone().get()).first);
return Status{};
}
diff --git a/src/core/CL/kernels/CLStackLayerKernel.h b/src/core/CL/kernels/CLStackLayerKernel.h
index 2865127..d3c17f5 100644
--- a/src/core/CL/kernels/CLStackLayerKernel.h
+++ b/src/core/CL/kernels/CLStackLayerKernel.h
@@ -26,6 +26,7 @@
#define ARM_COMPUTE_CLSTACKLAYERKERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
namespace arm_compute
@@ -60,7 +61,8 @@
* @param[out] output Output tensor. Data types supported: Same as @p input.
*
*/
- void configure(const ICLTensor *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, ICLTensor *output);
+ void configure(
+ const ICLTensor *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, ICLTensor *output);
/** Initialise the kernel's inputs and output
*
* @note Supported input tensor rank: up to 4
@@ -74,7 +76,12 @@
* @param[out] output Output tensor. Data types supported: Same as @p input.
*
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, ICLTensor *output);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ unsigned int axis,
+ unsigned int idx_input,
+ unsigned int num_tensors,
+ ICLTensor *output);
/** Static function to check if given info will lead to a valid configuration of @ref CLStackLayerKernel
*
* @note Supported input tensor rank: up to 4
@@ -88,7 +95,11 @@
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, unsigned int axis, unsigned int idx_input, unsigned int num_tensors, const ITensorInfo *output);
+ static Status validate(const ITensorInfo *input,
+ unsigned int axis,
+ unsigned int idx_input,
+ unsigned int num_tensors,
+ const ITensorInfo *output);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLStridedSliceKernel.cpp b/src/core/CL/kernels/CLStridedSliceKernel.cpp
index 9acbafd..a8f6112 100644
--- a/src/core/CL/kernels/CLStridedSliceKernel.cpp
+++ b/src/core/CL/kernels/CLStridedSliceKernel.cpp
@@ -22,11 +22,13 @@
* SOFTWARE.
*/
#include "src/core/CL/kernels/CLStridedSliceKernel.h"
+
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/utils/helpers/tensor_transform.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
#include "src/core/utils/helpers/bit_ops.h"
@@ -37,9 +39,14 @@
{
namespace
{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output,
- const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
- int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask)
+Status validate_arguments(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const Coordinates &starts,
+ const Coordinates &ends,
+ const BiStrides &strides,
+ int32_t begin_mask,
+ int32_t end_mask,
+ int32_t shrink_axis_mask)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
@@ -48,19 +55,16 @@
ARM_COMPUTE_RETURN_ERROR_ON(starts.num_dimensions() > input->num_dimensions());
ARM_COMPUTE_RETURN_ERROR_ON(ends.num_dimensions() > input->num_dimensions());
ARM_COMPUTE_RETURN_ERROR_ON(strides.num_dimensions() > input->num_dimensions());
- ARM_COMPUTE_RETURN_ERROR_ON(std::any_of(strides.cbegin(), strides.cbegin() + strides.num_dimensions(), [](int i)
- {
- return i == 0;
- }));
+ ARM_COMPUTE_RETURN_ERROR_ON(
+ std::any_of(strides.cbegin(), strides.cbegin() + strides.num_dimensions(), [](int i) { return i == 0; }));
// Get expected output shape
- const TensorShape exp_output_shape = arm_compute::misc::shape_calculator::compute_strided_slice_shape(*input,
- starts, ends, strides,
- begin_mask, end_mask, shrink_axis_mask);
+ const TensorShape exp_output_shape = arm_compute::misc::shape_calculator::compute_strided_slice_shape(
+ *input, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask);
ARM_COMPUTE_RETURN_ERROR_ON(exp_output_shape.total_size() == 0);
// Checks output if configured
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
const TensorInfo exp_output_info = output->clone()->set_tensor_shape(exp_output_shape);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, &exp_output_info);
@@ -76,28 +80,33 @@
_type = CLKernelType::ELEMENTWISE;
}
-void CLStridedSliceKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *input, ITensorInfo *output,
- const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
- int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask)
+void CLStridedSliceKernel::configure(const CLCompileContext &compile_context,
+ const ITensorInfo *input,
+ ITensorInfo *output,
+ const Coordinates &starts,
+ const Coordinates &ends,
+ const BiStrides &strides,
+ int32_t begin_mask,
+ int32_t end_mask,
+ int32_t shrink_axis_mask)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
- auto padding_info = get_padding_info({ input, output });
- ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, output, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask));
+ auto padding_info = get_padding_info({input, output});
+ ARM_COMPUTE_ERROR_THROW_ON(
+ validate_arguments(input, output, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask));
const TensorShape &input_shape = input->tensor_shape();
Coordinates starts_abs;
Coordinates ends_abs;
Coordinates final_strides;
- std::tie(starts_abs, ends_abs, final_strides) = arm_compute::helpers::tensor_transform::calculate_strided_slice_coords(
- input_shape,
- starts, ends, strides,
- begin_mask, end_mask, shrink_axis_mask);
+ std::tie(starts_abs, ends_abs, final_strides) =
+ arm_compute::helpers::tensor_transform::calculate_strided_slice_coords(input_shape, starts, ends, strides,
+ begin_mask, end_mask, shrink_axis_mask);
// Configure kernel window
- const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_strided_slice_shape(*input,
- starts, ends, strides,
- begin_mask, end_mask, shrink_axis_mask);
+ const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_strided_slice_shape(
+ *input, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask);
auto_init_if_empty(*output, input->clone()->set_tensor_shape(output_shape));
Window win = calculate_max_window(*output, Steps());
@@ -108,29 +117,33 @@
const bool multi_access_x = !is_shrink_on_x && (final_strides.x() == 1) && (output_width_x / vec_size_x > 0);
// Update window if needed
- if(multi_access_x)
+ if (multi_access_x)
{
Window &updated_window = win;
updated_window.set(Window::DimX,
- Window::Dimension(updated_window.x().start(), ceil_to_multiple(updated_window.x().end(), vec_size_x), vec_size_x));
+ Window::Dimension(updated_window.x().start(),
+ ceil_to_multiple(updated_window.x().end(), vec_size_x), vec_size_x));
}
ICLKernel::configure_internal(win);
// Create build options
CLBuildOptions build_opts;
- build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(data_size_from_type(input->data_type())));
- for(unsigned int i = 0; i < input_shape.num_dimensions(); ++i)
+ build_opts.add_option("-DDATA_TYPE=" +
+ get_cl_unsigned_type_from_element_size(data_size_from_type(input->data_type())));
+ for (unsigned int i = 0; i < input_shape.num_dimensions(); ++i)
{
const bool is_shrink = arm_compute::helpers::bit_ops::is_bit_set(shrink_axis_mask, i);
- build_opts.add_option("-DSTART_" + support::cpp11::to_string(i) + "=" + support::cpp11::to_string(starts_abs[i]));
- build_opts.add_option("-DSTRIDE_" + support::cpp11::to_string(i) + "=" + support::cpp11::to_string(final_strides[i]));
+ build_opts.add_option("-DSTART_" + support::cpp11::to_string(i) + "=" +
+ support::cpp11::to_string(starts_abs[i]));
+ build_opts.add_option("-DSTRIDE_" + support::cpp11::to_string(i) + "=" +
+ support::cpp11::to_string(final_strides[i]));
build_opts.add_option_if(is_shrink, "-DSHRINK_" + support::cpp11::to_string(i));
}
- build_opts.add_option_if(multi_access_x, "-DLAST_ACCESSED_X=" + support::cpp11::to_string(std::max<int>(output_width_x - vec_size_x, 0)));
+ build_opts.add_option_if(multi_access_x, "-DLAST_ACCESSED_X=" + support::cpp11::to_string(
+ std::max<int>(output_width_x - vec_size_x, 0)));
build_opts.add_option_if(multi_access_x, "-DVEC_SIZE=" + support::cpp11::to_string(vec_size_x));
build_opts.add_option_if_else(input_shape.num_dimensions() > 2,
- "-DSRC_DEPTH=" + support::cpp11::to_string(input_shape.z()),
- "-DSRC_DEPTH=1");
+ "-DSRC_DEPTH=" + support::cpp11::to_string(input_shape.z()), "-DSRC_DEPTH=1");
build_opts.add_option_if_else(output->num_dimensions() > 2,
"-DDST_DEPTH=" + support::cpp11::to_string(output->tensor_shape().z()),
"-DDST_DEPTH=1");
@@ -142,7 +155,7 @@
_config_id = "strided_slice";
_config_id += "_";
_config_id += lower_string(string_from_data_type(input->data_type()));
- for(unsigned int i = 0; i < input_shape.num_dimensions(); ++i)
+ for (unsigned int i = 0; i < input_shape.num_dimensions(); ++i)
{
_config_id += "_";
_config_id += support::cpp11::to_string(input->dimension(i));
@@ -156,11 +169,17 @@
ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}
-Status CLStridedSliceKernel::validate(const ITensorInfo *input, const ITensorInfo *output,
- const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
- int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask)
+Status CLStridedSliceKernel::validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const Coordinates &starts,
+ const Coordinates &ends,
+ const BiStrides &strides,
+ int32_t begin_mask,
+ int32_t end_mask,
+ int32_t shrink_axis_mask)
{
- ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask));
+ ARM_COMPUTE_RETURN_ON_ERROR(
+ validate_arguments(input, output, starts, ends, strides, begin_mask, end_mask, shrink_axis_mask));
return Status{};
}
@@ -170,8 +189,9 @@
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
- const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
- auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
+ const auto src =
+ utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
+ auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
Window window_collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
Window slice = window_collapsed.first_slice_window_4D();
@@ -182,7 +202,6 @@
add_4D_tensor_argument(idx, src, slice);
add_4D_tensor_argument(idx, dst, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(window_collapsed.slide_window_slice_4D(slice));
+ } while (window_collapsed.slide_window_slice_4D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLStridedSliceKernel.h b/src/core/CL/kernels/CLStridedSliceKernel.h
index 4c20150..1cf5bca 100644
--- a/src/core/CL/kernels/CLStridedSliceKernel.h
+++ b/src/core/CL/kernels/CLStridedSliceKernel.h
@@ -25,6 +25,7 @@
#define ARM_COMPUTE_CL_STRIDED_SLICE_KERNEL_H
#include "arm_compute/core/Types.h"
+
#include "src/core/CL/ICLKernel.h"
#include <cstdint>
@@ -53,9 +54,15 @@
* @param[in] shrink_axis_mask If the ith bit of shrink_axis_mask is set, it implies that the ith specification shrinks the dimensionality by 1.
* A slice of size 1 starting from starts[i] in the dimension must be preserved.
*/
- void configure(const CLCompileContext &compile_context, const ITensorInfo *input, ITensorInfo *output,
- const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
- int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask);
+ void configure(const CLCompileContext &compile_context,
+ const ITensorInfo *input,
+ ITensorInfo *output,
+ const Coordinates &starts,
+ const Coordinates &ends,
+ const BiStrides &strides,
+ int32_t begin_mask,
+ int32_t end_mask,
+ int32_t shrink_axis_mask);
/** Static function to check if given info will lead to a valid configuration of @ref CLStridedSliceKernel
*
@@ -71,9 +78,14 @@
* @param[in] shrink_axis_mask If the ith bit of shrink_axis_mask is set, it implies that the ith specification shrinks the dimensionality by 1.
* A slice of size 1 starting from starts[i] in the dimension must be preserved.
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *output,
- const Coordinates &starts, const Coordinates &ends, const BiStrides &strides,
- int32_t begin_mask, int32_t end_mask, int32_t shrink_axis_mask);
+ static Status validate(const ITensorInfo *input,
+ const ITensorInfo *output,
+ const Coordinates &starts,
+ const Coordinates &ends,
+ const BiStrides &strides,
+ int32_t begin_mask,
+ int32_t end_mask,
+ int32_t shrink_axis_mask);
// Inherited methods overridden:
void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
diff --git a/src/core/CL/kernels/CLTileKernel.cpp b/src/core/CL/kernels/CLTileKernel.cpp
index 3e7015c..fa996c4 100644
--- a/src/core/CL/kernels/CLTileKernel.cpp
+++ b/src/core/CL/kernels/CLTileKernel.cpp
@@ -22,9 +22,11 @@
* SOFTWARE.
*/
#include "src/core/CL/kernels/CLTileKernel.h"
+
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
#include "arm_compute/core/utils/StringUtils.h"
+
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
#include "support/StringSupport.h"
@@ -39,15 +41,13 @@
ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
ARM_COMPUTE_RETURN_ERROR_ON(multiples.size() > 4);
ARM_COMPUTE_RETURN_ERROR_ON(multiples.empty());
- ARM_COMPUTE_RETURN_ERROR_ON(std::any_of(multiples.begin(), multiples.end(), [](uint32_t e)
- {
- return e == 0;
- }));
+ ARM_COMPUTE_RETURN_ERROR_ON(std::any_of(multiples.begin(), multiples.end(), [](uint32_t e) { return e == 0; }));
// Validate output if initialized
- if(output->total_size() != 0)
+ if (output->total_size() != 0)
{
- ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(misc::shape_calculator::compute_tiled_shape(input->tensor_shape(), multiples), output->tensor_shape());
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(
+ misc::shape_calculator::compute_tiled_shape(input->tensor_shape(), multiples), output->tensor_shape());
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
}
@@ -55,8 +55,7 @@
}
} // namespace
-CLTileKernel::CLTileKernel()
- : _input(nullptr), _output(nullptr)
+CLTileKernel::CLTileKernel() : _input(nullptr), _output(nullptr)
{
_type = CLKernelType::ELEMENTWISE;
}
@@ -66,7 +65,10 @@
configure(CLKernelLibrary::get().get_compile_context(), input, output, multiples);
}
-void CLTileKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const Multiples &multiples)
+void CLTileKernel::configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const Multiples &multiples)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
@@ -104,15 +106,14 @@
// Configure window without padding
Window win = calculate_max_window(*output->info());
- if(multi_access_x)
+ if (multi_access_x)
{
// If multi-access is enabled, no thread should cross the tile boundaries. This means we need
// as many threads as those to cover a single tile times multiples[0]. Note that if threads
// do not cross the boundaries of the tiles, they won't cross the boundaries of the last tile, and
// we don't need to pad the output
const unsigned int size_win_x = ceil_to_multiple(input->info()->dimension(0), vec_size_x) * multiples[0];
- win.set(Window::DimX,
- Window::Dimension(win.x().start(), size_win_x, vec_size_x));
+ win.set(Window::DimX, Window::Dimension(win.x().start(), size_win_x, vec_size_x));
}
ICLKernel::configure_internal(win);
@@ -121,7 +122,7 @@
_config_id = "tile";
_config_id += "_";
_config_id += lower_string(string_from_data_type(input->info()->data_type()));
- for(unsigned int i = 0; i < multiples.size(); ++i)
+ for (unsigned int i = 0; i < multiples.size(); ++i)
{
_config_id += "_";
_config_id += support::cpp11::to_string(input->info()->dimension(i));
@@ -150,7 +151,6 @@
add_4D_tensor_argument(idx, _input, slice);
add_4D_tensor_argument(idx, _output, slice);
enqueue(queue, *this, slice, lws_hint());
- }
- while(collapsed.slide_window_slice_4D(slice));
+ } while (collapsed.slide_window_slice_4D(slice));
}
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLTileKernel.h b/src/core/CL/kernels/CLTileKernel.h
index 41752ca..c3486ae 100644
--- a/src/core/CL/kernels/CLTileKernel.h
+++ b/src/core/CL/kernels/CLTileKernel.h
@@ -64,7 +64,10 @@
* @param[out] output Destination tensor. Same as @p input
*
*/
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const Multiples &multiples);
+ void configure(const CLCompileContext &compile_context,
+ const ICLTensor *input,
+ ICLTensor *output,
+ const Multiples &multiples);
/** Static function to check if given info will lead to a valid configuration of @ref CLTileKernel
*
* @param[in] input Source tensor info. Data type supported: All.