COMPMID-754: Add validation to LocallyConnected and NEDeconv layers
Change-Id: Ifed8713f4d7f1315af684b30d11323db2b533f10
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/121783
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
diff --git a/arm_compute/core/CL/kernels/CLLocallyConnectedMatrixMultiplyKernel.h b/arm_compute/core/CL/kernels/CLLocallyConnectedMatrixMultiplyKernel.h
index 948f47a..fee5055 100644
--- a/arm_compute/core/CL/kernels/CLLocallyConnectedMatrixMultiplyKernel.h
+++ b/arm_compute/core/CL/kernels/CLLocallyConnectedMatrixMultiplyKernel.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -55,6 +55,15 @@
* @param[out] output Output tensor to store the result. Data type supported: same as @p input0
*/
void configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output);
+ /** Static function to check if given info will lead to a valid configuration of @ref CLLocallyConnectedMatrixMultiplyKernel
+ *
+ * @param[in] input0 First input tensor info. Data types supported: F32
+ * @param[in] input1 Second input tensor info. Data type supported: same as @p input0
+ * @param[in] output Output tensor info. Data type supported: same as @p input0
+ *
+ * @return a status
+ */
+ static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/arm_compute/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.h b/arm_compute/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.h
index 5e3410f..19a92c0 100644
--- a/arm_compute/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.h
+++ b/arm_compute/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.h
@@ -55,6 +55,15 @@
* @param[out] output Output tensor to store the result of matrix multiplication. Data type supported: same as @p input0
*/
void configure(const ITensor *input0, const ITensor *input1, ITensor *output);
+ /** Static function to check if given info will lead to a valid configuration of @ref NELocallyConnectedMatrixMultiplyKernel
+ *
+ * @param[in] input0 First input tensor info. Data types supported: F16, F32
+ * @param[in] input1 Second input tensor info. Data type supported: same as @p input0
+ * @param[in] output Output tensor info. Data type supported: same as @p input0
+ *
+ * @return a status
+ */
+ static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output);
// Inherited methods overridden:
void run(const Window &window, const ThreadInfo &info) override;
diff --git a/arm_compute/runtime/CL/functions/CLLocallyConnectedLayer.h b/arm_compute/runtime/CL/functions/CLLocallyConnectedLayer.h
index f56039f..e5aafd8 100644
--- a/arm_compute/runtime/CL/functions/CLLocallyConnectedLayer.h
+++ b/arm_compute/runtime/CL/functions/CLLocallyConnectedLayer.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -65,6 +65,20 @@
* @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo.
*/
void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info);
+ /** Static function to check if given info will lead to a valid configuration of @ref CLLocallyConnectedLayer
+ *
+ * @param[in] input Input tensor info. 3 lower dimensions represent a single input [width, height, IFM],
+ * while every optional dimension from 4 and above represent a batch of inputs.
+ * Data types supported: F32.
+ * @param[in] weights Weights tensor info. Weights are 5D tensor with dimensions [kernel_x, kernel_y, IFM, OFM, num_patches]. Data type supported:Same as @p input.
+ * @param[in] biases Biases tensor info. Shared biases supported. Biases are 2D tensor with dimensions [OFM, num_patches]. Data type supported:Same as @p input.
+ * @param[in] output Output tensor info. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs.
+ * Data types supported: Same as @p input.
+ * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo.
+ *
+ * @return a status
+ */
+ static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info);
// Inherited methods overridden:
void run() override;
diff --git a/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h
index c148c3f..1d5dd3f 100644
--- a/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h
@@ -91,6 +91,20 @@
*/
void configure(ITensor *input, const ITensor *weights, const ITensor *bias, ITensor *output, const PadStrideInfo &info,
unsigned int inner_border_right, unsigned int inner_border_top);
+ /** Static function to check if given info will lead to a valid configuration of @ref NEDeconvolutionLayer
+ *
+ * @param[in] input Input tensor info. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: F32.
+ * @param[in] weights The 4d weights info with dimensions [width, height, OFM, IFM]. Data type supported: Same as @p input.
+ * @param[in] bias (Optional) The biases have one dimension. Data type supported: Same as @p input.
+ * @param[in] output Output tensor info. The output has the same number of dimensions as the @p input.
+ * @param[in] info Contains padding and policies to be used in the deconvolution, this is decribed in @ref PadStrideInfo.
+ * @param[in] inner_border_right The number of zeros added to right edge of the input.
+ * @param[in] inner_border_top The number of zeros added to top edge of the input.
+ *
+ * @return a status
+ */
+ static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, const ITensorInfo *output, const PadStrideInfo &info,
+ unsigned int inner_border_right, unsigned int inner_border_top);
// Inherited methods overridden:
void run() override;
diff --git a/arm_compute/runtime/NEON/functions/NELocallyConnectedLayer.h b/arm_compute/runtime/NEON/functions/NELocallyConnectedLayer.h
index 18d2a1d..e07f09c 100644
--- a/arm_compute/runtime/NEON/functions/NELocallyConnectedLayer.h
+++ b/arm_compute/runtime/NEON/functions/NELocallyConnectedLayer.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -65,6 +65,20 @@
* @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo.
*/
void configure(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const PadStrideInfo &conv_info);
+ /** Static function to check if given info will lead to a valid configuration of @ref NELocallyConnectedLayer
+ *
+ * @param[in] input Input tensor info. 3 lower dimensions represent a single input [width, height, IFM],
+ * while every optional dimension from 4 and above represent a batch of inputs.
+ * Data types supported: F16, F32.
+ * @param[in] weights Weights tensor info. Weights are 5D tensor with dimensions [kernel_x, kernel_y, IFM, OFM, num_patches]. Data type supported:Same as @p input.
+ * @param[in] biases Biases tensor info. Shared biases supported. Biases are 2D tensor with dimensions [OFM, num_patches]. Data type supported:Same as @p input.
+ * @param[in] output Output tensor info. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs.
+ * Data types supported: Same as @p input.
+ * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo.
+ *
+ * @return a status
+ */
+ static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info);
// Inherited methods overridden:
void run() override;
diff --git a/src/core/CL/kernels/CLLocallyConnectedMatrixMultiplyKernel.cpp b/src/core/CL/kernels/CLLocallyConnectedMatrixMultiplyKernel.cpp
index a3af5b0..84f2e0c 100644
--- a/src/core/CL/kernels/CLLocallyConnectedMatrixMultiplyKernel.cpp
+++ b/src/core/CL/kernels/CLLocallyConnectedMatrixMultiplyKernel.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017, 2018 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -46,13 +46,44 @@
{
}
+namespace
+{
+Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output)
+{
+ ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input0, input1, output);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input0, input1, output);
+ ARM_COMPUTE_RETURN_ERROR_ON(input0->dimension(0) != input1->dimension(1));
+
+ return Status{};
+}
+
+std::tuple<Status, Window> validate_and_configure_window(ITensorInfo *input0, ITensorInfo *input1, ITensorInfo *output)
+{
+ const unsigned int num_elems_processed_per_iteration_x = max_cl_vector_width / data_size_from_type(input0->data_type());
+
+ Window win = calculate_max_window(*output, Steps(num_elems_processed_per_iteration_x));
+
+ AccessWindowHorizontal input0_access(input0, 0, num_elems_processed_per_iteration_x);
+ AccessWindowHorizontal input1_access(input1, 0, num_elems_processed_per_iteration_x);
+ AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration_x);
+
+ bool window_changed = update_window_and_padding(win, input0_access, input1_access, output_access);
+
+ output_access.set_valid_region(win, ValidRegion(Coordinates(), output->tensor_shape()));
+
+ Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+
+ return std::make_tuple(err, win);
+}
+} // namespace
+
void CLLocallyConnectedMatrixMultiplyKernel::configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output)
{
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::F32);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::F32);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F32);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input0, input1, output);
- ARM_COMPUTE_ERROR_ON(input0->info()->dimension(0) != input1->info()->dimension(1));
+ ARM_COMPUTE_ERROR_ON_NULLPTR(input0, input1, output);
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input0->info(), input1->info(), output->info()));
_input0 = input0;
_input1 = input1;
@@ -77,20 +108,20 @@
std::string data_type_name = lower_string(string_from_data_type(input0->info()->data_type()));
_kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(("gemm_lc_vm_" + data_type_name), build_opts));
- // Configure window kernel
- const unsigned int num_elems_processed_per_iteration_x = max_cl_vector_width / data_size_from_type(input0->info()->data_type());
+ // Configure kernel window
+ auto win_config = validate_and_configure_window(input0->info(), input1->info(), output->info());
- Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration_x));
+ ARM_COMPUTE_ERROR_THROW_ON(std::get<0>(win_config));
- AccessWindowRectangle input0_access(input0->info(), 0, 0, num_elems_processed_per_iteration_x, 1);
- AccessWindowRectangle input1_access(input1->info(), 0, 0, num_elems_processed_per_iteration_x, 1);
- AccessWindowRectangle output_access(output->info(), 0, 0, num_elems_processed_per_iteration_x, 1);
+ ICLKernel::configure(std::get<1>(win_config));
+}
- update_window_and_padding(win, input0_access, input1_access, output_access);
+Status CLLocallyConnectedMatrixMultiplyKernel::validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output)
+{
+ ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input0, input1, output));
+ ARM_COMPUTE_RETURN_ON_ERROR(std::get<0>(validate_and_configure_window(input0->clone().get(), input1->clone().get(), output->clone().get())));
- output_access.set_valid_region(win, ValidRegion(Coordinates(), output->info()->tensor_shape()));
-
- ICLKernel::configure(win);
+ return Status{};
}
void CLLocallyConnectedMatrixMultiplyKernel::run(const Window &window, cl::CommandQueue &queue)
diff --git a/src/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.cpp b/src/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.cpp
index 35beb82..099626d 100644
--- a/src/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.cpp
+++ b/src/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.cpp
@@ -302,6 +302,37 @@
},
ina, out);
}
+
+Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output)
+{
+ ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input0, input1, output);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F16, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input0, input1, output);
+ ARM_COMPUTE_RETURN_ERROR_ON(input0->dimension(0) != input1->dimension(1));
+
+ return Status{};
+}
+
+std::tuple<Status, Window> validate_and_configure_window(ITensorInfo *input0, ITensorInfo *input1, ITensorInfo *output)
+{
+ const unsigned int num_elems_processed_per_iteration_x = 16;
+
+ Window win = calculate_max_window(*output, Steps(num_elems_processed_per_iteration_x));
+
+ AccessWindowHorizontal input0_access(input0, 0, num_elems_processed_per_iteration_x);
+ AccessWindowHorizontal input1_access(input1, 0, num_elems_processed_per_iteration_x);
+ AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration_x);
+
+ bool window_changed = update_window_and_padding(win, input0_access, input1_access, output_access);
+
+ output_access.set_valid_region(win, ValidRegion(Coordinates(), output->tensor_shape()));
+
+ Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+
+ return std::make_tuple(err, win);
+}
} // namespace
NELocallyConnectedMatrixMultiplyKernel::NELocallyConnectedMatrixMultiplyKernel()
@@ -311,31 +342,27 @@
void NELocallyConnectedMatrixMultiplyKernel::configure(const ITensor *input0, const ITensor *input1, ITensor *output)
{
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::F16, DataType::F32);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::F16, DataType::F32);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F16, DataType::F32);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F16, DataType::F32);
- ARM_COMPUTE_ERROR_ON(input0->info()->dimension(0) != input1->info()->dimension(1));
+ ARM_COMPUTE_ERROR_ON_NULLPTR(input0, input1, output);
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input0->info(), input1->info(), output->info()));
_input0 = input0;
_input1 = input1;
_output = output;
- const unsigned int num_elems_processed_per_iteration_x = 16;
-
// Configure kernel window
- Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration_x));
+ auto win_config = validate_and_configure_window(input0->info(), input1->info(), output->info());
- AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration_x);
+ ARM_COMPUTE_ERROR_THROW_ON(std::get<0>(win_config));
- update_window_and_padding(win,
- AccessWindowHorizontal(input0->info(), 0, num_elems_processed_per_iteration_x),
- AccessWindowHorizontal(input1->info(), 0, num_elems_processed_per_iteration_x),
- output_access);
+ INEKernel::configure(std::get<1>(win_config));
+}
- output_access.set_valid_region(win, ValidRegion(Coordinates(), output->info()->tensor_shape()));
+Status NELocallyConnectedMatrixMultiplyKernel::validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output)
+{
+ ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input0, input1, output));
+ ARM_COMPUTE_RETURN_ON_ERROR(std::get<0>(validate_and_configure_window(input0->clone().get(), input1->clone().get(), output->clone().get())));
- INEKernel::configure(win);
+ return Status{};
}
void NELocallyConnectedMatrixMultiplyKernel::run(const Window &window, const ThreadInfo &info)
diff --git a/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp b/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp
index d284949..a3eb501 100644
--- a/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp
+++ b/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp
@@ -33,40 +33,102 @@
using namespace arm_compute;
+namespace
+{
+void calculate_shapes(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
+ TensorShape &shape_wr, TensorShape &shape_im2col, TensorShape &shape_gemm)
+{
+ ARM_COMPUTE_UNUSED(output);
+
+ const unsigned int kernel_width = weights->dimension(0);
+ const unsigned int kernel_height = weights->dimension(1);
+
+ bool has_bias = (biases != nullptr);
+
+ // Get convolved dimensions
+ unsigned int conv_w = 0;
+ unsigned int conv_h = 0;
+ std::tie(conv_w, conv_h) = scaled_dimensions(input->dimension(0), input->dimension(1), kernel_width, kernel_height,
+ conv_info);
+
+ const size_t mat_weights_cols = weights->dimension(3);
+ const size_t mat_weights_rows = weights->dimension(0) * weights->dimension(1) * weights->dimension(2) + ((has_bias) ? 1 : 0);
+ const size_t mat_weights_num = weights->dimension(4);
+
+ shape_wr = TensorShape(mat_weights_cols, mat_weights_rows, mat_weights_num);
+
+ const size_t mat_input_cols = mat_weights_rows;
+ const size_t mat_input_rows = conv_w * conv_h;
+
+ shape_im2col = input->tensor_shape();
+ shape_im2col.set(0, mat_input_cols);
+ shape_im2col.set(1, mat_input_rows);
+ shape_im2col.set(2, 1);
+
+ shape_gemm = shape_im2col;
+ shape_gemm.set(0, mat_weights_cols);
+ shape_gemm.set(1, mat_input_rows);
+}
+} // namespace
+
CLLocallyConnectedLayer::CLLocallyConnectedLayer(std::shared_ptr<IMemoryManager> memory_manager)
: _memory_group(std::move(memory_manager)), _input_im2col_kernel(), _weights_reshape_kernel(), _mm_kernel(), _output_col2im_kernel(), _input_im2col_reshaped(), _weights_reshaped(), _gemm_output(),
_is_first_run(false)
{
}
+Status CLLocallyConnectedLayer::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info)
+{
+ ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights, output);
+ ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(2) != input->dimension(2));
+ ARM_COMPUTE_RETURN_ERROR_ON(!conv_info.padding_is_symmetric());
+
+ bool has_bias = (biases != nullptr);
+
+ if(has_bias)
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != weights->dimension(3));
+ ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 2);
+ }
+
+ const unsigned int kernel_width = weights->dimension(0);
+ const unsigned int kernel_height = weights->dimension(1);
+
+ // Get convolved dimensions
+ unsigned int conv_w = 0;
+ unsigned int conv_h = 0;
+ std::tie(conv_w, conv_h) = scaled_dimensions(input->dimension(0), input->dimension(1), kernel_width, kernel_height,
+ conv_info);
+
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG((output->dimension(0) != conv_w) || (output->dimension(1) != conv_h), "Output shape does not match the expected one");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(weights->dimension(4) != (conv_w * conv_h), "Weights shape does not match the expected one");
+
+ // Calculate intermediate buffer shapes
+ TensorShape shape_wr;
+ TensorShape shape_im2col;
+ TensorShape shape_gemm;
+ calculate_shapes(input, weights, biases, output, conv_info, shape_wr, shape_im2col, shape_gemm);
+
+ TensorInfo weights_reshaped_info(shape_wr, 1, weights->data_type());
+ TensorInfo input_im2col_reshaped_info(shape_im2col, 1, input->data_type());
+ TensorInfo gemm_output_info(shape_gemm, 1, input->data_type());
+
+ ARM_COMPUTE_RETURN_ON_ERROR(CLIm2ColKernel::validate(input, &input_im2col_reshaped_info, Size2D(kernel_width, kernel_height), conv_info, has_bias));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLWeightsReshapeKernel::validate(weights, biases, &weights_reshaped_info));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLLocallyConnectedMatrixMultiplyKernel::validate(&input_im2col_reshaped_info, &weights_reshaped_info, &gemm_output_info));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLCol2ImKernel::validate(&gemm_output_info, output, std::make_pair(conv_w, conv_h)));
+
+ return Status{};
+}
+
void CLLocallyConnectedLayer::configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info)
{
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(weights, 1, DataType::F32);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F32);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights, output);
- ARM_COMPUTE_ERROR_ON(weights->info()->dimension(2) != input->info()->dimension(2));
- ARM_COMPUTE_ERROR_ON(!conv_info.padding_is_symmetric());
-
- if(biases != nullptr)
- {
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(biases, 1, DataType::F32);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, biases);
- ARM_COMPUTE_ERROR_ON(biases->info()->dimension(0) != weights->info()->dimension(3));
- ARM_COMPUTE_ERROR_ON(biases->info()->num_dimensions() > 2);
- }
+ ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights, output);
+ ARM_COMPUTE_ERROR_THROW_ON(CLLocallyConnectedLayer::validate(input->info(), weights->info(), biases == nullptr ? nullptr : biases->info(), output->info(), conv_info));
bool _has_bias = (biases != nullptr);
_is_first_run = true;
- // Get parameters for conv_info
- unsigned int stride_x = 0;
- unsigned int stride_y = 0;
- unsigned int pad_x = 0;
- unsigned int pad_y = 0;
- std::tie(stride_x, stride_y) = conv_info.stride();
- std::tie(pad_x, pad_y) = conv_info.pad();
-
const unsigned int kernel_width = weights->info()->dimension(0);
const unsigned int kernel_height = weights->info()->dimension(1);
@@ -76,32 +138,14 @@
std::tie(conv_w, conv_h) = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1), kernel_width, kernel_height,
conv_info);
- ARM_COMPUTE_ERROR_ON_MSG((output->info()->dimension(0) != conv_w) || (output->info()->dimension(1) != conv_h), "Output shape does not match the expected one");
- ARM_COMPUTE_ERROR_ON_MSG(weights->info()->dimension(4) != (conv_w * conv_h), "Weights shape does not match the expected one");
-
- // Create tensor to store the reshaped weights
- const size_t mat_weights_cols = weights->info()->dimension(3);
- const size_t mat_weights_rows = weights->info()->dimension(0) * weights->info()->dimension(1) * weights->info()->dimension(2) + ((_has_bias) ? 1 : 0);
- const size_t mat_weights_num = weights->info()->dimension(4);
-
- const TensorShape shape_wr(mat_weights_cols, mat_weights_rows, mat_weights_num);
+ // Calculate intermediate buffer shapes
+ TensorShape shape_wr;
+ TensorShape shape_im2col;
+ TensorShape shape_gemm;
+ calculate_shapes(input->info(), weights->info(), biases == nullptr ? nullptr : biases->info(), output->info(), conv_info, shape_wr, shape_im2col, shape_gemm);
_weights_reshaped.allocator()->init(TensorInfo(shape_wr, 1, weights->info()->data_type()));
-
- // Create tensor to store im2col reshaped inputs
- const size_t mat_input_cols = mat_weights_rows;
- const size_t mat_input_rows = conv_w * conv_h;
- TensorShape shape_im2col = input->info()->tensor_shape();
- shape_im2col.set(0, mat_input_cols);
- shape_im2col.set(1, mat_input_rows);
- shape_im2col.set(2, 1);
-
_input_im2col_reshaped.allocator()->init(TensorInfo(shape_im2col, 1, input->info()->data_type()));
-
- // Create locally connected layer output tensor
- TensorShape shape_gemm = _input_im2col_reshaped.info()->tensor_shape();
- shape_gemm.set(0, mat_weights_cols);
- shape_gemm.set(1, mat_input_rows);
_gemm_output.allocator()->init(TensorInfo(shape_gemm, 1, input->info()->data_type()));
// Manage intermediate buffers
diff --git a/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp b/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp
index 693d7a4..14d54d2 100644
--- a/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp
+++ b/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp
@@ -42,13 +42,64 @@
{
}
+Status NEDeconvolutionLayer::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, const ITensorInfo *output, const PadStrideInfo &info,
+ unsigned int inner_border_right, unsigned int inner_border_top)
+{
+ ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights, output);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(weights, 1, DataType::F32);
+ ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(0) != weights->dimension(1));
+ ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(0) < 1);
+ ARM_COMPUTE_RETURN_ERROR_ON(!info.padding_is_symmetric());
+
+ const unsigned int stride_x = info.stride().first;
+ const unsigned int stride_y = info.stride().second;
+
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(inner_border_right > stride_x - 1, "inner_border_right must be smaller than stride_x");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(inner_border_top > stride_y - 1, "inner_border_top must be smaller than stride_y");
+
+ auto out_dims = deconvolution_output_dimensions(input->dimension(0), input->dimension(1), weights->dimension(0), weights->dimension(1),
+ info.pad().first, info.pad().second, inner_border_right, inner_border_top, stride_x, stride_y);
+
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights, bias);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, weights, bias);
+
+ if(bias != nullptr)
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, bias);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, bias);
+ }
+
+ if(output->tensor_shape().total_size() > 0)
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+
+ const TensorShape output_shape = deconvolution_output_shape(out_dims, input->tensor_shape(), weights->tensor_shape());
+
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->dimension(Window::DimX) != output_shape.x(), "Output's width is invalid.");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->dimension(Window::DimY) != output_shape.y(), "Output's height is invalid.");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->dimension(Window::DimZ) != output_shape.z(), "Output's depth is invalid.");
+ }
+
+ TensorInfo scale_out_info(input->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_deconvolution_shape(*input, stride_x, stride_y, inner_border_right, inner_border_top,
+ info)));
+ const PadStrideInfo conv_info(1, 1, 0, 0, 0, 0, DimensionRoundingType::CEIL);
+
+ for(size_t i = 2; i < Coordinates::num_max_dimensions; ++i)
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(i) != scale_out_info.dimension(i));
+ }
+
+ ARM_COMPUTE_RETURN_ON_ERROR(NEConvolutionLayer::validate(&scale_out_info, weights, bias, output, info, WeightsInfo()));
+
+ return Status{};
+}
+
void NEDeconvolutionLayer::configure(ITensor *input, const ITensor *weights, const ITensor *bias, ITensor *output, const PadStrideInfo &info,
unsigned int inner_border_right, unsigned int inner_border_top)
{
- ARM_COMPUTE_ERROR_ON_NULLPTR(output);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32);
- ARM_COMPUTE_ERROR_ON(weights->info()->dimension(0) != weights->info()->dimension(1));
- ARM_COMPUTE_ERROR_ON(!info.padding_is_symmetric());
+ ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights, output);
_input = input;
_info = info;
@@ -56,15 +107,9 @@
const unsigned int stride_x = info.stride().first;
const unsigned int stride_y = info.stride().second;
- auto out_dims = deconvolution_output_dimensions(input->info()->dimension(0), input->info()->dimension(1), weights->info()->dimension(0), weights->info()->dimension(1),
- info.pad().first, info.pad().second, inner_border_right, inner_border_top, stride_x, stride_y);
- const TensorShape output_shape = deconvolution_output_shape(out_dims, input->info()->tensor_shape(), weights->info()->tensor_shape());
-
- ARM_COMPUTE_UNUSED(output_shape);
- ARM_COMPUTE_ERROR_ON_MSG(output->info()->dimension(Window::DimX) != output_shape.x(), "Output's width is invalid.");
- ARM_COMPUTE_ERROR_ON_MSG(output->info()->dimension(Window::DimY) != output_shape.y(), "Output's height is invalid.");
- ARM_COMPUTE_ERROR_ON_MSG(output->info()->dimension(Window::DimZ) != output_shape.z(), "Output's depth is invalid.");
+ // Perform validation step
+ ARM_COMPUTE_ERROR_THROW_ON(NEDeconvolutionLayer::validate(input->info(), weights->info(), bias == nullptr ? nullptr : bias->info(), output->info(), info, inner_border_right, inner_border_top));
_memory_group.manage(&_scaled_output);
diff --git a/src/runtime/NEON/functions/NELocallyConnectedLayer.cpp b/src/runtime/NEON/functions/NELocallyConnectedLayer.cpp
index 45ddb70..9735594 100644
--- a/src/runtime/NEON/functions/NELocallyConnectedLayer.cpp
+++ b/src/runtime/NEON/functions/NELocallyConnectedLayer.cpp
@@ -33,40 +33,102 @@
using namespace arm_compute;
+namespace
+{
+void calculate_shapes(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
+ TensorShape &shape_wr, TensorShape &shape_im2col, TensorShape &shape_gemm)
+{
+ ARM_COMPUTE_UNUSED(output);
+
+ const unsigned int kernel_width = weights->dimension(0);
+ const unsigned int kernel_height = weights->dimension(1);
+
+ bool has_bias = (biases != nullptr);
+
+ // Get convolved dimensions
+ unsigned int conv_w = 0;
+ unsigned int conv_h = 0;
+ std::tie(conv_w, conv_h) = scaled_dimensions(input->dimension(0), input->dimension(1), kernel_width, kernel_height,
+ conv_info);
+
+ const size_t mat_weights_cols = weights->dimension(3);
+ const size_t mat_weights_rows = weights->dimension(0) * weights->dimension(1) * weights->dimension(2) + ((has_bias) ? 1 : 0);
+ const size_t mat_weights_num = weights->dimension(4);
+
+ shape_wr = TensorShape(mat_weights_cols, mat_weights_rows, mat_weights_num);
+
+ const size_t mat_input_cols = mat_weights_rows;
+ const size_t mat_input_rows = conv_w * conv_h;
+
+ shape_im2col = input->tensor_shape();
+ shape_im2col.set(0, mat_input_cols);
+ shape_im2col.set(1, mat_input_rows);
+ shape_im2col.set(2, 1);
+
+ shape_gemm = shape_im2col;
+ shape_gemm.set(0, mat_weights_cols);
+ shape_gemm.set(1, mat_input_rows);
+}
+} // namespace
+
NELocallyConnectedLayer::NELocallyConnectedLayer(std::shared_ptr<IMemoryManager> memory_manager)
: _memory_group(std::move(memory_manager)), _input_im2col_kernel(), _weights_reshape_kernel(), _mm_kernel(), _output_col2im_kernel(), _input_im2col_reshaped(), _weights_reshaped(), _gemm_output(),
_is_first_run(false)
{
}
+Status NELocallyConnectedLayer::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info)
+{
+ ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights, output);
+ ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(2) != input->dimension(2));
+ ARM_COMPUTE_RETURN_ERROR_ON(!conv_info.padding_is_symmetric());
+
+ bool has_bias = (biases != nullptr);
+
+ if(has_bias)
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != weights->dimension(3));
+ ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 2);
+ }
+
+ const unsigned int kernel_width = weights->dimension(0);
+ const unsigned int kernel_height = weights->dimension(1);
+
+ // Get convolved dimensions
+ unsigned int conv_w = 0;
+ unsigned int conv_h = 0;
+ std::tie(conv_w, conv_h) = scaled_dimensions(input->dimension(0), input->dimension(1), kernel_width, kernel_height,
+ conv_info);
+
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG((output->dimension(0) != conv_w) || (output->dimension(1) != conv_h), "Output shape does not match the expected one");
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(weights->dimension(4) != (conv_w * conv_h), "Weights shape does not match the expected one");
+
+ // Calculate intermediate buffer shapes
+ TensorShape shape_wr;
+ TensorShape shape_im2col;
+ TensorShape shape_gemm;
+ calculate_shapes(input, weights, biases, output, conv_info, shape_wr, shape_im2col, shape_gemm);
+
+ TensorInfo weights_reshaped_info(shape_wr, 1, weights->data_type());
+ TensorInfo input_im2col_reshaped_info(shape_im2col, 1, input->data_type());
+ TensorInfo gemm_output_info(shape_gemm, 1, input->data_type());
+
+ ARM_COMPUTE_RETURN_ON_ERROR(NEIm2ColKernel::validate(input, &input_im2col_reshaped_info, Size2D(kernel_width, kernel_height), conv_info, has_bias, false));
+ ARM_COMPUTE_RETURN_ON_ERROR(NEWeightsReshapeKernel::validate(weights, biases, &weights_reshaped_info));
+ ARM_COMPUTE_RETURN_ON_ERROR(NELocallyConnectedMatrixMultiplyKernel::validate(&input_im2col_reshaped_info, &weights_reshaped_info, &gemm_output_info));
+ ARM_COMPUTE_RETURN_ON_ERROR(NECol2ImKernel::validate(&gemm_output_info, output, Size2D(conv_w, conv_h)));
+
+ return Status{};
+}
+
void NELocallyConnectedLayer::configure(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const PadStrideInfo &conv_info)
{
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(weights, 1, DataType::F32);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F32);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights, output);
- ARM_COMPUTE_ERROR_ON(weights->info()->dimension(2) != input->info()->dimension(2));
- ARM_COMPUTE_ERROR_ON(!conv_info.padding_is_symmetric());
-
- if(biases != nullptr)
- {
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(biases, 1, DataType::F32);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, biases);
- ARM_COMPUTE_ERROR_ON(biases->info()->dimension(0) != weights->info()->dimension(3));
- ARM_COMPUTE_ERROR_ON(biases->info()->num_dimensions() > 2);
- }
+ ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights, output);
+ ARM_COMPUTE_ERROR_THROW_ON(NELocallyConnectedLayer::validate(input->info(), weights->info(), biases == nullptr ? nullptr : biases->info(), output->info(), conv_info));
bool _has_bias = (biases != nullptr);
_is_first_run = true;
- // Get parameters for conv_info
- unsigned int stride_x = 0;
- unsigned int stride_y = 0;
- unsigned int pad_x = 0;
- unsigned int pad_y = 0;
- std::tie(stride_x, stride_y) = conv_info.stride();
- std::tie(pad_x, pad_y) = conv_info.pad();
-
const unsigned int kernel_width = weights->info()->dimension(0);
const unsigned int kernel_height = weights->info()->dimension(1);
@@ -76,32 +138,14 @@
std::tie(conv_w, conv_h) = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1), kernel_width, kernel_height,
conv_info);
- ARM_COMPUTE_ERROR_ON_MSG((output->info()->dimension(0) != conv_w) || (output->info()->dimension(1) != conv_h), "Output shape does not match the expected one");
- ARM_COMPUTE_ERROR_ON_MSG(weights->info()->dimension(4) != (conv_w * conv_h), "Weights shape does not match the expected one");
-
- // Create tensor to store the reshaped weights
- const size_t mat_weights_cols = weights->info()->dimension(3);
- const size_t mat_weights_rows = weights->info()->dimension(0) * weights->info()->dimension(1) * weights->info()->dimension(2) + ((_has_bias) ? 1 : 0);
- const size_t mat_weights_num = weights->info()->dimension(4);
-
- const TensorShape shape_wr(mat_weights_cols, mat_weights_rows, mat_weights_num);
+ // Calculate intermediate buffer shapes
+ TensorShape shape_wr;
+ TensorShape shape_im2col;
+ TensorShape shape_gemm;
+ calculate_shapes(input->info(), weights->info(), biases == nullptr ? nullptr : biases->info(), output->info(), conv_info, shape_wr, shape_im2col, shape_gemm);
_weights_reshaped.allocator()->init(TensorInfo(shape_wr, 1, weights->info()->data_type()));
-
- // Create tensor to store im2col reshaped inputs
- const size_t mat_input_cols = mat_weights_rows;
- const size_t mat_input_rows = conv_w * conv_h;
- TensorShape shape_im2col = input->info()->tensor_shape();
- shape_im2col.set(0, mat_input_cols);
- shape_im2col.set(1, mat_input_rows);
- shape_im2col.set(2, 1);
-
_input_im2col_reshaped.allocator()->init(TensorInfo(shape_im2col, 1, input->info()->data_type()));
-
- // Create locally connected layer output tensor
- TensorShape shape_gemm = _input_im2col_reshaped.info()->tensor_shape();
- shape_gemm.set(0, mat_weights_cols);
- shape_gemm.set(1, mat_input_rows);
_gemm_output.allocator()->init(TensorInfo(shape_gemm, 1, input->info()->data_type()));
// Manage intermediate buffers
diff --git a/tests/validation/CL/LocallyConnected.cpp b/tests/validation/CL/LocallyConnected.cpp
index 05cab29..de79d60 100644
--- a/tests/validation/CL/LocallyConnected.cpp
+++ b/tests/validation/CL/LocallyConnected.cpp
@@ -47,6 +47,67 @@
TEST_SUITE(CL)
TEST_SUITE(LocallyConnected)
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(
+ framework::dataset::make("InputInfo", { TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Mismatching data type input/weights
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Mismatching data type input/bias
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Mismatching data type input/output
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Mismatching shape input/weights
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Mismatching shape input/bias
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Mismatching shape input/output
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Asymmetric padding
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0)
+ }),
+ framework::dataset::make("WeightsInfo",{ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F16, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 274U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F32, 0)
+ })),
+ framework::dataset::make("BiasInfo", { TensorInfo(TensorShape(21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(21U, 275U), 1, DataType::F16, 0),
+ TensorInfo(TensorShape(21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(21U, 274U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(21U, 275U), 1, DataType::F32, 0)
+ })),
+ framework::dataset::make("OutputInfo", { TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F16, 0),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 25U, 22U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0)
+ })),
+ framework::dataset::make("PadStride", { PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(2, 1, 1, 0, 0, 0, DimensionRoundingType::FLOOR),
+ PadStrideInfo(2, 1, 0, 0)
+ })),
+ framework::dataset::make("Expected", { false, false, false, false, false, false, false, true })),
+ input_info, weights_info, bias_info, output_info, conv_info, expected)
+{
+ bool is_valid = bool(CLLocallyConnectedLayer::validate(&input_info.clone()->set_is_resizable(false),
+ &weights_info.clone()->set_is_resizable(false),
+ &bias_info.clone()->set_is_resizable(false),
+ &output_info.clone()->set_is_resizable(false),
+ conv_info));
+ ARM_COMPUTE_EXPECT(is_valid == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(framework::dataset::concat(datasets::SmallLocallyConnectedDataset(), datasets::LargeLocallyConnectedDataset()),
framework::dataset::make("DataType", DataType::F32)),
src_shape, weights_shape, bias_shape, dst_shape, info, data_type)
diff --git a/tests/validation/NEON/DeconvolutionLayer.cpp b/tests/validation/NEON/DeconvolutionLayer.cpp
index 566b75a..3bb6d6f 100644
--- a/tests/validation/NEON/DeconvolutionLayer.cpp
+++ b/tests/validation/NEON/DeconvolutionLayer.cpp
@@ -58,6 +58,106 @@
TEST_SUITE(NEON)
TEST_SUITE(DeconvolutionLayer)
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, (combine(datasets::SmallDeconvolutionShapes(), framework::dataset::make("DataType", DataType::F32))),
+ input_shape, data_type)
+{
+ // Create shapes
+ const unsigned int kernel_size_x = 3;
+ const unsigned int kernel_size_y = 3;
+ const unsigned int num_kernels = 1;
+ const TensorShape weights_shape(kernel_size_x, kernel_size_y, input_shape.z(), num_kernels);
+ const TensorShape bias_shape(num_kernels);
+ auto out_dim = deconvolution_output_dimensions(input_shape.x(), input_shape.y(), kernel_size_x, kernel_size_y, 1, 1, 0, 0, 1, 1);
+ TensorShape output_shape = deconvolution_output_shape(out_dim, input_shape, weights_shape);
+
+ // Create tensors
+ Tensor src = create_tensor<Tensor>(input_shape, data_type, 1);
+ Tensor weights = create_tensor<Tensor>(weights_shape, data_type, 1);
+ Tensor bias = create_tensor<Tensor>(bias_shape, data_type, 1);
+ Tensor dst = create_tensor<Tensor>(output_shape, data_type, 1);
+
+ ARM_COMPUTE_EXPECT(src.info()->is_resizable(), framework::LogLevel::ERRORS);
+ ARM_COMPUTE_EXPECT(weights.info()->is_resizable(), framework::LogLevel::ERRORS);
+ ARM_COMPUTE_EXPECT(bias.info()->is_resizable(), framework::LogLevel::ERRORS);
+ ARM_COMPUTE_EXPECT(dst.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+ // Create and configure function
+ NEDeconvolutionLayer deconv;
+ deconv.configure(&src, &weights, &bias, &dst, PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL), 0, 0);
+
+ // Validate valid region
+ const ValidRegion src_valid_region = shape_to_valid_region(input_shape);
+ const ValidRegion weights_valid_region = shape_to_valid_region(weights_shape);
+ const ValidRegion bias_valid_region = shape_to_valid_region(bias_shape);
+ const ValidRegion dst_valid_region = shape_to_valid_region(output_shape);
+
+ validate(src.info()->valid_region(), src_valid_region);
+ validate(weights.info()->valid_region(), weights_valid_region);
+ validate(bias.info()->valid_region(), bias_valid_region);
+ validate(dst.info()->valid_region(), dst_valid_region);
+}
+
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zip(
+ framework::dataset::make("InputInfo", { TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32, 0), // Mismatching data type
+ TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32, 0), // Invalid weights shape
+ TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::QS8, 4), // Non supported data type
+ TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32, 11), // Invalid bias shape
+ TensorInfo(TensorShape(13U, 11U, 4U, 3U), 1, DataType::F32, 0), // Window shrink
+ TensorInfo(TensorShape(32U, 16U, 2U), 1, DataType::F32, 0),
+ }),
+ framework::dataset::make("WeightsInfo", { TensorInfo(TensorShape(3U, 3U, 2U, 2U), 1, DataType::F16, 0),
+ TensorInfo(TensorShape(3U, 3U, 2U, 4U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 2U, 2U), 1, DataType::QS8, 5),
+ TensorInfo(TensorShape(3U, 2U, 2U, 2U), 1, DataType::F32, 11),
+ TensorInfo(TensorShape(3U, 3U, 4U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(1U, 1U, 2U, 4U), 1, DataType::F32, 0),
+ })),
+ framework::dataset::make("BiasInfo", { TensorInfo(TensorShape(1U), 1, DataType::F16, 0),
+ TensorInfo(TensorShape(1U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(1U), 1, DataType::F32, 5),
+ TensorInfo(TensorShape(25U, 11U), 1, DataType::F32, 11),
+ TensorInfo(TensorShape(1U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(4U), 1, DataType::F32, 0),
+ })),
+ framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(25U, 11U, 2U), 1, DataType::F16, 0),
+ TensorInfo(TensorShape(25U, 10U, 2U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(25U, 11U, 2U), 1, DataType::F32, 5),
+ TensorInfo(TensorShape(13U, 13U, 2U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 9U, 1U, 3U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(32U, 16U, 4U), 1, DataType::F32, 0),
+ })),
+ framework::dataset::make("PadStrideInfo", { PadStrideInfo(1, 1, 0, 0),
+ PadStrideInfo(1, 1, 0, 0),
+ PadStrideInfo(1, 1, 0, 0),
+ PadStrideInfo(1, 1, 0, 0),
+ PadStrideInfo(1, 1, 1, 1),
+ PadStrideInfo(1, 1, 0, 0),
+ })),
+ framework::dataset::make("ax", { 1U,
+ 1U,
+ 1U,
+ 1U,
+ 0U,
+ 0U,
+ })),
+ framework::dataset::make("ay", { 1U,
+ 1U,
+ 1U,
+ 1U,
+ 0U,
+ 0U,
+ })),
+ framework::dataset::make("Expected", { false, false, false, false, false, true })),
+ input_info, weights_info, bias_info, output_info, pad_info, ax, ay, expected)
+{
+ bool is_valid = bool(NEDeconvolutionLayer::validate(&input_info.clone()->set_is_resizable(false), &weights_info.clone()->set_is_resizable(false), &bias_info.clone()->set_is_resizable(false), &output_info.clone()->set_is_resizable(false), pad_info, ax, ay));
+ ARM_COMPUTE_EXPECT(is_valid == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
template <typename T>
using NEDeconvolutionLayerFixture4x4 = DeconvolutionValidationFixture<Tensor, Accessor, NEDeconvolutionLayer, T, 4, 4>;
diff --git a/tests/validation/NEON/LocallyConnected.cpp b/tests/validation/NEON/LocallyConnected.cpp
index 56430d9..b00f274 100644
--- a/tests/validation/NEON/LocallyConnected.cpp
+++ b/tests/validation/NEON/LocallyConnected.cpp
@@ -48,6 +48,67 @@
TEST_SUITE(NEON)
TEST_SUITE(LocallyConnected)
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(
+ framework::dataset::make("InputInfo", { TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Mismatching data type input/weights
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Mismatching data type input/bias
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Mismatching data type input/output
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Mismatching shape input/weights
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Mismatching shape input/bias
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Mismatching shape input/output
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0), // Asymmetric padding
+ TensorInfo(TensorShape(23U, 27U, 5U), 1, DataType::F32, 0)
+ }),
+ framework::dataset::make("WeightsInfo",{ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F16, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 274U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(3U, 3U, 5U, 21U, 275U), 1, DataType::F32, 0)
+ })),
+ framework::dataset::make("BiasInfo", { TensorInfo(TensorShape(21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(21U, 275U), 1, DataType::F16, 0),
+ TensorInfo(TensorShape(21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(21U, 274U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(21U, 275U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(21U, 275U), 1, DataType::F32, 0)
+ })),
+ framework::dataset::make("OutputInfo", { TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F16, 0),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 25U, 22U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0)
+ })),
+ framework::dataset::make("PadStride", { PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(2, 1, 1, 0, 0, 0, DimensionRoundingType::FLOOR),
+ PadStrideInfo(2, 1, 0, 0)
+ })),
+ framework::dataset::make("Expected", { false, false, false, false, false, false, false, true })),
+ input_info, weights_info, bias_info, output_info, conv_info, expected)
+{
+ bool is_valid = bool(NELocallyConnectedLayer::validate(&input_info.clone()->set_is_resizable(false),
+ &weights_info.clone()->set_is_resizable(false),
+ &bias_info.clone()->set_is_resizable(false),
+ &output_info.clone()->set_is_resizable(false),
+ conv_info));
+ ARM_COMPUTE_EXPECT(is_valid == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(framework::dataset::concat(datasets::SmallLocallyConnectedDataset(), datasets::LargeLocallyConnectedDataset()),
framework::dataset::make("DataType", DataType::F32)),
src_shape, weights_shape, bias_shape, dst_shape, info, data_type)