diff options
author | Giorgio Arena <giorgio.arena@arm.com> | 2018-08-23 12:00:02 +0100 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:54:54 +0000 |
commit | 226e4b92b191491ffa57ede66eba1d5d6fcf3b76 (patch) | |
tree | 334705a1e743e3465400208d582cf0b25bf950fa /src/runtime | |
parent | 35aea3776449557c438e264bae7af5b1fe0e5ff6 (diff) | |
download | ComputeLibrary-226e4b92b191491ffa57ede66eba1d5d6fcf3b76.tar.gz |
COMPMID-1470 Add auto-init of the output in NECol2im
The output of NECol2Im is already auto-initialized.
This patch is about calling ShapeCalculator instead of computing the shape inside the kernel, adding validate_and_configure_window, and standardize the way convolved dims are passed (now NEON uses Size2D, while CL passes a pair of uint values: using Size2D for both implementations)
Change-Id: I795696e1b6532f57847c3186c1b532c09f5a25da
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/145345
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele DiGiorgio <michele.digiorgio@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Diffstat (limited to 'src/runtime')
-rw-r--r-- | src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp | 4 | ||||
-rw-r--r-- | src/runtime/CL/functions/CLLocallyConnectedLayer.cpp | 4 | ||||
-rw-r--r-- | src/runtime/CL/tuners/BifrostTuner.cpp | 2 |
3 files changed, 5 insertions, 5 deletions
diff --git a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp index 782fe710e7..c9daea4169 100644 --- a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp +++ b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp @@ -285,7 +285,7 @@ void CLGEMMConvolutionLayer::configure(const ICLTensor *input, const ICLTensor * if(input->info()->data_layout() == DataLayout::NCHW) { // Configure and tune Col2Im - _col2im_kernel.configure(_is_quantized ? gemm_output_staged_to_use : gemm_output_to_use, output, std::make_pair(conv_w, conv_h), num_groups); + _col2im_kernel.configure(_is_quantized ? gemm_output_staged_to_use : gemm_output_to_use, output, Size2D(conv_w, conv_h), num_groups); CLScheduler::get().tune_kernel_static(_col2im_kernel); } else @@ -443,7 +443,7 @@ Status CLGEMMConvolutionLayer::validate(const ITensorInfo *input, const ITensorI { ARM_COMPUTE_RETURN_ON_ERROR(CLCol2ImKernel::validate(is_quantized ? gemm_output_staged_to_use : gemm_output_to_use, output, - std::make_pair(conv_w, conv_h), num_groups)); + Size2D(conv_w, conv_h), num_groups)); } } diff --git a/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp b/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp index 40bf032d69..5c6bef995f 100644 --- a/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp +++ b/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp @@ -122,7 +122,7 @@ Status CLLocallyConnectedLayer::validate(const ITensorInfo *input, const ITensor 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))); + ARM_COMPUTE_RETURN_ON_ERROR(CLCol2ImKernel::validate(&gemm_output_info, output, Size2D(conv_w, conv_h))); return Status{}; } @@ -163,7 +163,7 @@ void CLLocallyConnectedLayer::configure(const ICLTensor *input, const ICLTensor _input_im2col_kernel.configure(input, &_input_im2col_reshaped, Size2D(kernel_width, kernel_height), conv_info, _has_bias); _weights_reshape_kernel.configure(weights, biases, &_weights_reshaped); _mm_kernel.configure(&_input_im2col_reshaped, &_weights_reshaped, &_gemm_output); - _output_col2im_kernel.configure(&_gemm_output, output, std::make_pair(conv_w, conv_h)); + _output_col2im_kernel.configure(&_gemm_output, output, Size2D(conv_w, conv_h)); // Allocate intermediate tensors _input_im2col_reshaped.allocator()->allocate(); diff --git a/src/runtime/CL/tuners/BifrostTuner.cpp b/src/runtime/CL/tuners/BifrostTuner.cpp index 2d52f3392e..59d73b4e79 100644 --- a/src/runtime/CL/tuners/BifrostTuner.cpp +++ b/src/runtime/CL/tuners/BifrostTuner.cpp @@ -134,7 +134,7 @@ void tune_col2im_kernel(CLCol2ImKernel &k) // via exhaustive autotuning over 30 representative tensor shapes. if(gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72, GPUTarget::G51, GPUTarget::G51BIG, GPUTarget::G51LIT, GPUTarget::G76)) { - if((k._convolved_dims.first == 7) || (k._convolved_dims.first == 14)) + if((k._convolved_dims.width == 7) || (k._convolved_dims.width == 14)) { lws_hint = cl::NDRange(1, 7, 1); } |