COMPMID-1277 - Optimizing CLIm2ColKernel for NHWC.
This patch includes:
- Im2Col optimizations for NHWC using a new data layout
- Refactoring of CLIm2ColKernel adding validation method and auto-init
- Removed im2col_reduced from CLIm2ColKernel and created a new kernel CLFlattenLayerKernel
Change-Id: I1620640b6796baa268324b33ae92cdd8de53e27c
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/141241
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
diff --git a/src/runtime/CL/functions/CLFlattenLayer.cpp b/src/runtime/CL/functions/CLFlattenLayer.cpp
index f5809a2..b372c35 100644
--- a/src/runtime/CL/functions/CLFlattenLayer.cpp
+++ b/src/runtime/CL/functions/CLFlattenLayer.cpp
@@ -23,8 +23,7 @@
*/
#include "arm_compute/runtime/CL/functions/CLFlattenLayer.h"
-#include "arm_compute/core/CL/kernels/CLIm2ColKernel.h"
-#include "arm_compute/core/Size2D.h"
+#include "arm_compute/core/CL/kernels/CLFlattenLayerKernel.h"
#include "arm_compute/runtime/CL/CLScheduler.h"
#include "support/ToolchainSupport.h"
@@ -32,8 +31,13 @@
void CLFlattenLayer::configure(const ICLTensor *input, ICLTensor *output)
{
- auto k = arm_compute::support::cpp14::make_unique<CLIm2ColKernel>();
- k->configure(input, output, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false);
+ auto k = arm_compute::support::cpp14::make_unique<CLFlattenLayerKernel>();
+ k->configure(input, output);
_kernel = std::move(k);
CLScheduler::get().tune_kernel_static(*_kernel);
}
+
+Status CLFlattenLayer::validate(const ITensorInfo *input, const ITensorInfo *output)
+{
+ return CLFlattenLayerKernel::validate(input, output);
+}
\ No newline at end of file
diff --git a/src/runtime/CL/functions/CLFullyConnectedLayer.cpp b/src/runtime/CL/functions/CLFullyConnectedLayer.cpp
index 6fd78a3..60c28a0 100644
--- a/src/runtime/CL/functions/CLFullyConnectedLayer.cpp
+++ b/src/runtime/CL/functions/CLFullyConnectedLayer.cpp
@@ -73,12 +73,11 @@
}
CLFullyConnectedLayer::CLFullyConnectedLayer(std::shared_ptr<IMemoryManager> memory_manager)
- : _memory_group(memory_manager), _im2col_kernel(), _convert_weights(), _reshape_weights_kernel(), _mm_gemm(memory_manager), _mm_gemmlowp(memory_manager), _gemmlowp_output_stage(),
- _accumulate_biases_kernel(), _im2col_output(), _gemmlowp_output(), _converted_weights_output(), _reshape_weights_output(), _are_weights_converted(true), _are_weights_reshaped(true),
+ : _memory_group(memory_manager), _convert_weights(), _flatten_layer(), _reshape_weights_kernel(), _mm_gemm(memory_manager), _mm_gemmlowp(memory_manager), _gemmlowp_output_stage(),
+ _accumulate_biases_kernel(), _flatten_output(), _gemmlowp_output(), _converted_weights_output(), _reshape_weights_output(), _are_weights_converted(true), _are_weights_reshaped(true),
_is_fc_after_conv(true), _accumulate_biases(false), _is_quantized(false), _is_prepared(false), _original_weights(nullptr)
{
}
-
void CLFullyConnectedLayer::configure_mm(const ICLTensor *input, const ICLTensor *weights, ICLTensor *output)
{
if(_is_quantized)
@@ -111,20 +110,19 @@
// If the fully connected layer is called after a convolution layer, the input tensor must be linearized
- // Initialize output tensor for im2col
- TensorShape shape_im2col = compute_im2col_fc_shape(input->info());
- _im2col_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(shape_im2col).set_data_layout(DataLayout::NCHW));
+ // Initialize output tensor for flatten
+ TensorShape shape_flatten = compute_flatten_shape(input->info());
+ _flatten_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(shape_flatten).set_data_layout(DataLayout::NCHW));
- // Configure im2col kernel
- _memory_group.manage(&_im2col_output);
- _im2col_kernel.configure(input, &_im2col_output, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false);
- CLScheduler::get().tune_kernel_static(_im2col_kernel);
+ // Configure flatten kernel
+ _memory_group.manage(&_flatten_output);
+ _flatten_layer.configure(input, &_flatten_output);
// Configure matrix multiply kernel
- configure_mm(&_im2col_output, weights, output);
+ configure_mm(&_flatten_output, weights, output);
- // Allocate the output tensor for im2col once all the configure methods have been called
- _im2col_output.allocator()->allocate();
+ // Allocate the output tensor for flatten once all the configure methods have been called
+ _flatten_output.allocator()->allocate();
}
void CLFullyConnectedLayer::configure_fc_fc(const ICLTensor *input, const ICLTensor *weights, ICLTensor *output)
@@ -254,7 +252,7 @@
bool is_quantized = is_data_type_quantized_asymmetric(input->data_type());
const GPUTarget gpu_target = CLScheduler::get().target();
- const ITensorInfo &im2col_input = TensorInfo(input->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_im2col_fc_shape(input)).set_data_layout(DataLayout::NCHW));
+ const ITensorInfo &flatten_input = TensorInfo(input->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_flatten_shape(input)).set_data_layout(DataLayout::NCHW));
const ITensorInfo &reshaped_weights = TensorInfo(weights->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_transposed_shape(*weights)));
const ITensorInfo &converted_weights = weights_reshaped ? TensorInfo(weights->clone()->set_is_resizable(true).reset_padding()) : TensorInfo(*reshaped_weights.clone());
const ITensorInfo &gemmlowp_output = TensorInfo(output->clone()->set_is_resizable(true).reset_padding().set_data_type(DataType::S32));
@@ -311,9 +309,9 @@
// Fully Connected layer after a Convolution Layer without batches
ARM_COMPUTE_RETURN_ERROR_ON((weights_to_use->dimension(1) != (input->dimension(0) * input->dimension(1) * input->dimension(2))));
- // Validate im2col kernel
- ARM_COMPUTE_RETURN_ON_ERROR(CLIm2ColKernel::validate(input, &im2col_input, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false));
- input_to_use = &im2col_input;
+ // Validate flatten kernel
+ ARM_COMPUTE_RETURN_ON_ERROR(CLFlattenLayer::validate(input, &flatten_input));
+ input_to_use = &flatten_input;
}
else
{
@@ -341,7 +339,7 @@
// Linearize input if it comes from a convolutional layer
if(_is_fc_after_conv)
{
- CLScheduler::get().enqueue(_im2col_kernel, false);
+ _flatten_layer.run();
}
// Run matrix multiply
diff --git a/src/runtime/CL/functions/CLGEMM.cpp b/src/runtime/CL/functions/CLGEMM.cpp
index 1d1b17b..a8d7058 100644
--- a/src/runtime/CL/functions/CLGEMM.cpp
+++ b/src/runtime/CL/functions/CLGEMM.cpp
@@ -171,6 +171,7 @@
Status CLGEMM::validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *c, const ITensorInfo *output, float alpha, float beta, const GEMMInfo &gemm_info)
{
ARM_COMPUTE_UNUSED(alpha);
+ ARM_COMPUTE_UNUSED(output);
// Check if we need to reshape the matrix B only on the first run
const bool reshape_b_only_on_first_run = gemm_info.reshape_b_only_on_first_run();
@@ -180,7 +181,7 @@
TensorInfo tmp_a_info{};
TensorInfo tmp_b_info{};
- TensorInfo tmp_output_info = *output->clone();
+ TensorInfo tmp_output_info{};
// Get the GPU target
const GPUTarget gpu_target = CLScheduler::get().target();
diff --git a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp
index fb90415..49549a0 100644
--- a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp
+++ b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp
@@ -171,7 +171,6 @@
const DataLayout data_layout = input->info()->data_layout();
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);
- const int idx_channel = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
const int idx_kernels = get_data_layout_dimension_index(data_layout, DataLayoutDimension::BATCHES);
const unsigned int kernel_width = weights->info()->dimension(idx_width);
@@ -193,7 +192,6 @@
ICLTensor *gemm_output_to_use = output;
ICLTensor *gemm_output_staged_to_use = output;
- const unsigned bias_element = (_append_bias && !_skip_im2col) ? 1 : 0;
const ICLTensor *biases_to_use = (_append_bias && !_skip_im2col) ? biases : nullptr;
// Get parameters from conv_info
@@ -212,7 +210,6 @@
dilation);
unsigned int mat_weights_cols = weights->info()->dimension(idx_kernels);
- unsigned int mat_weights_rows = weights->info()->dimension(idx_width) * weights->info()->dimension(idx_height) * weights->info()->dimension(idx_channel) + bias_element;
// _weights_reshaped will be auto configured in the kernel.
// Just append biases and do not transpose 1xW as it will be reshaped in CLGEMM
@@ -223,25 +220,13 @@
// Create tensor to store im2col reshaped inputs
if(!_skip_im2col)
{
- // Calculate im2col shape
- // For OpenCL the batch size is on the third dimension
- // TODO (giaiod01): Use auto-init COMPMID-1277
- TensorShape shape_im2col = input->info()->tensor_shape();
- if(shape_im2col.num_dimensions() >= 3)
- {
- shape_im2col.remove_dimension(2);
- }
- shape_im2col.set(0, mat_weights_rows);
- shape_im2col.set(1, conv_w * conv_h);
-
- // FIXME: input->clone() doesn't work with subtensors for grouped convolutions.
- TensorInfo im2col_reshaped_info(shape_im2col, 1, data_type);
- im2col_reshaped_info.set_quantization_info(input->info()->quantization_info());
- _im2col_output.allocator()->init(im2col_reshaped_info);
_memory_group.manage(&_im2col_output);
- // Configure and tune im2col
+ // Configure and tune im2col. im2col output shape is auto-initialized
_im2col_kernel.configure(input, &_im2col_output, Size2D(kernel_width, kernel_height), conv_info, _append_bias, dilation);
+
+ // Set quantization info
+ _im2col_output.info()->set_quantization_info(input->info()->quantization_info());
CLScheduler::get().tune_kernel_static(_im2col_kernel);
// Update GEMM input
@@ -350,11 +335,10 @@
const ITensorInfo *gemm_output_staged_to_use = output;
const ITensorInfo *weights_to_use = weights;
- const bool is_nhwc = data_layout == DataLayout::NHWC;
- const bool is_quantized = is_data_type_quantized_asymmetric(data_type);
- const bool skip_im2col = (data_layout == DataLayout::NHWC && kernel_width == 1 && kernel_height == 1 && conv_info.stride().first == 1 && conv_info.stride().second == 1) && !is_quantized;
- const bool append_bias = (biases != nullptr) && (!is_quantized);
- const unsigned bias_element = (append_bias && !skip_im2col) ? 1 : 0;
+ const bool is_nhwc = data_layout == DataLayout::NHWC;
+ const bool is_quantized = is_data_type_quantized_asymmetric(data_type);
+ const bool skip_im2col = (data_layout == DataLayout::NHWC && kernel_width == 1 && kernel_height == 1 && conv_info.stride().first == 1 && conv_info.stride().second == 1) && !is_quantized;
+ const bool append_bias = (biases != nullptr) && (!is_quantized);
ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(idx_channel) != input->dimension(idx_channel));
ARM_COMPUTE_RETURN_ERROR_ON(weights->num_dimensions() > 4);
@@ -391,7 +375,6 @@
dilation);
unsigned int mat_weights_cols = weights->dimension(idx_kernels);
- unsigned int mat_weights_rows = weights->dimension(idx_width) * weights->dimension(idx_height) * weights->dimension(idx_channel) + bias_element;
// Output tensor auto inizialitation if not yet initialized
ARM_COMPUTE_RETURN_ON_ERROR(CLConvolutionLayerReshapeWeights::validate(weights, is_quantized ? nullptr : biases, nullptr));
@@ -400,19 +383,14 @@
if(!skip_im2col)
{
- // Create tensor info for im2col reshaped inputs
- // For OpenCL the batch size is on the third dimension
- // TODO (giaiod01): Use auto-init COMPMID-1277
- TensorShape shape_im2col = input->tensor_shape();
- if(input->tensor_shape().num_dimensions() >= 3)
- {
- shape_im2col.remove_dimension(2);
- }
- shape_im2col.set(0, mat_weights_rows);
- shape_im2col.set(1, conv_w * conv_h);
- im2col_reshaped_info = TensorInfo(shape_im2col, 1, data_type);
- im2col_reshaped_info.set_quantization_info(input->quantization_info());
- ARM_COMPUTE_RETURN_ON_ERROR(CLIm2ColKernel::validate(input, &im2col_reshaped_info, Size2D(kernel_width, kernel_height), conv_info, append_bias, dilation));
+ const Size2D kernel_dims(kernel_width, kernel_height);
+
+ // Output tensor auto initialization if not yet initialized
+ TensorShape expected_output_shape = compute_im2col_conv_shape(input, kernel_dims, conv_info, append_bias, dilation, true);
+
+ auto_init_if_empty(im2col_reshaped_info, input->clone()->set_tensor_shape(expected_output_shape));
+
+ ARM_COMPUTE_RETURN_ON_ERROR(CLIm2ColKernel::validate(input, &im2col_reshaped_info, kernel_dims, conv_info, append_bias, dilation));
gemm_input_to_use = &im2col_reshaped_info;
}
else if(append_bias)
diff --git a/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp b/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
index 25b8adc..c2f0283 100644
--- a/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
+++ b/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
@@ -113,7 +113,7 @@
// If the fully connected layer is called after a convolution layer, the input tensor must be linearized
// Initialize output tensor for im2col
- TensorShape shape_im2col = compute_im2col_fc_shape(input->info());
+ TensorShape shape_im2col = compute_flatten_shape(input->info());
_im2col_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(shape_im2col));
// Configure im2col kernel
@@ -249,7 +249,7 @@
bool is_fc_after_conv = true;
bool is_quantized = is_data_type_quantized_asymmetric(input->data_type());
- const ITensorInfo &im2col_input = TensorInfo(input->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_im2col_fc_shape(input)));
+ const ITensorInfo &im2col_input = TensorInfo(input->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_flatten_shape(input)));
const ITensorInfo &reshaped_weights = TensorInfo(weights->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_transposed_shape(*weights)));
const ITensorInfo &converted_weights = weights_reshaped ? TensorInfo(weights->clone()->set_is_resizable(true).reset_padding()) : TensorInfo(*reshaped_weights.clone());
const ITensorInfo &gemmlowp_output = TensorInfo(output->clone()->set_is_resizable(true).reset_padding().set_data_type(DataType::S32));
@@ -420,4 +420,4 @@
_is_prepared = true;
}
-}
+}
\ No newline at end of file
diff --git a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
index c0a5d0a..df4a040 100644
--- a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
+++ b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
@@ -223,7 +223,7 @@
{
// Calculate im2col shape
// For NEON the batch size is on the fourth dimension
- // TODO (giaiod01): Use auto-init COMPMID-1277
+ // TODO (giaiod01): Auto-initialize the output shape of im2col COMPMID-1482
TensorShape shape_im2col = input->info()->tensor_shape();
shape_im2col.set(0, mat_weights_rows);
shape_im2col.set(1, conv_w * conv_h);
@@ -232,7 +232,7 @@
_im2col_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(shape_im2col));
_memory_group.manage(&_im2col_output);
- // Configure and tune im2col
+ // Configure
_im2col_kernel.configure(input, &_im2col_output, Size2D(kernel_width, kernel_height), conv_info, _append_bias, false, false, dilation);
// Update GEMM input