diff options
author | Michele Di Giorgio <michele.digiorgio@arm.com> | 2021-01-22 09:47:04 +0000 |
---|---|---|
committer | Michele Di Giorgio <michele.digiorgio@arm.com> | 2021-06-18 10:33:48 +0000 |
commit | d02d5edfa15ba6c04a9986a8a362a945cb38ac31 (patch) | |
tree | ced4f49691d6c7038e347a8709b315bff59c64cf /src/runtime/CL/functions | |
parent | b014c27ba6db9840e4a72519760d51a87a2af7e7 (diff) | |
download | ComputeLibrary-d02d5edfa15ba6c04a9986a8a362a945cb38ac31.tar.gz |
Integrate improved CPU depthwise convolution kernels
* Replace assembly kernels for depthwise convolution with more optimized
ones.
* Add int8 assembly kernels.
* Fix implicit padding on optimized kernels
Resolves: COMPMID-3867, COMPMID-4361
Change-Id: I0b0867e05f61be4f368f62190d55e14d0ab3ebf2
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5622
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Diffstat (limited to 'src/runtime/CL/functions')
-rw-r--r-- | src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp | 4 | ||||
-rw-r--r-- | src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp | 2 |
2 files changed, 0 insertions, 6 deletions
diff --git a/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp b/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp index 6467caffef..c7520cd087 100644 --- a/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp +++ b/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp @@ -307,11 +307,9 @@ void CLDepthwiseConvolutionLayer::CLDepthwiseConvolutionLayerGeneric::prepare() { _output_multipliers.map(); _output_shifts.map(); - const unsigned int idx_ofms = _needs_permute ? 2 : 0; quantization::compute_quantized_multipliers_and_shifts(_input->info(), _original_weights->info(), _output->info(), - idx_ofms, reinterpret_cast<int32_t *>(_output_multipliers.ptr_to_element(Coordinates(0))), reinterpret_cast<int32_t *>(_output_shifts.ptr_to_element(Coordinates(0)))); _output_multipliers.unmap(); @@ -513,11 +511,9 @@ void CLDepthwiseConvolutionLayer::CLDepthwiseConvolutionLayerInternal3x3::prepar { _output_multipliers.map(); _output_shifts.map(); - const unsigned int idx_ofms = _is_nhwc ? 0 : 2; quantization::compute_quantized_multipliers_and_shifts(_input->info(), _original_weights->info(), _output->info(), - idx_ofms, reinterpret_cast<int32_t *>(_output_multipliers.ptr_to_element(Coordinates(0))), reinterpret_cast<int32_t *>(_output_shifts.ptr_to_element(Coordinates(0)))); _output_multipliers.unmap(); diff --git a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp index 3184d5dfe0..188f3b8819 100644 --- a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp +++ b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp @@ -352,7 +352,6 @@ void CLGEMMConvolutionLayer::configure(const CLCompileContext &compile_context, quantization::compute_quantized_multipliers_and_shifts(input->info(), weights->info(), output->info(), - idx_kernels, gemmlowp_output_stage.gemmlowp_multipliers.data(), gemmlowp_output_stage.gemmlowp_shifts.data()); gemmlowp_output_stage.gemmlowp_multiplier = gemmlowp_output_stage.gemmlowp_multipliers[0]; @@ -562,7 +561,6 @@ Status CLGEMMConvolutionLayer::validate(const ITensorInfo *input, const ITensorI quantization::compute_quantized_multipliers_and_shifts(input, weights, output, - idx_kernels, gemmlowp_output_stage.gemmlowp_multipliers.data(), gemmlowp_output_stage.gemmlowp_shifts.data()); gemmlowp_output_stage.gemmlowp_multiplier = gemmlowp_output_stage.gemmlowp_multipliers[0]; |