diff options
author | Manuel Bottini <manuel.bottini@arm.com> | 2021-04-13 13:09:30 +0100 |
---|---|---|
committer | Manuel Bottini <manuel.bottini@arm.com> | 2021-04-14 14:21:50 +0000 |
commit | 327225d3b2f716d5c62d801a7fafc7d377521f34 (patch) | |
tree | c19125b74a5ddf9a63e165cbffa7a85b01c7aff1 /src/core/cpu/kernels/CpuDirectConvolutionOutputStageKernel.h | |
parent | 21c28957f9c6fe1a28ef934e711bb7474b8d65ee (diff) | |
download | ComputeLibrary-327225d3b2f716d5c62d801a7fafc7d377521f34.tar.gz |
Port NEDirectConvolutionLayer to new API
Partially resolves: COMPMID-4009
Change-Id: I19ffb61c5c4541134a5028677d2d81228740e454
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5419
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Diffstat (limited to 'src/core/cpu/kernels/CpuDirectConvolutionOutputStageKernel.h')
-rw-r--r-- | src/core/cpu/kernels/CpuDirectConvolutionOutputStageKernel.h | 93 |
1 files changed, 93 insertions, 0 deletions
diff --git a/src/core/cpu/kernels/CpuDirectConvolutionOutputStageKernel.h b/src/core/cpu/kernels/CpuDirectConvolutionOutputStageKernel.h new file mode 100644 index 0000000000..9eeab194cb --- /dev/null +++ b/src/core/cpu/kernels/CpuDirectConvolutionOutputStageKernel.h @@ -0,0 +1,93 @@ +/* + * Copyright (c) 2017-2021 Arm Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef ARM_COMPUTE_CPU_DIRECTCONVOLUTION_OUTPUTSTAGE_KERNEL_H +#define ARM_COMPUTE_CPU_DIRECTCONVOLUTION_OUTPUTSTAGE_KERNEL_H + +#include "arm_compute/core/KernelDescriptors.h" +#include "src/core/common/Macros.h" +#include "src/core/cpu/ICpuKernel.h" + +namespace arm_compute +{ +class ITensor; +namespace cpu +{ +namespace kernels +{ +/** Kernel to accumulate the biases, if provided, or downscale in case of quantized input. + * + * @note We assume bias to be shared + * @note For quantized computations (i.e. @p src of S32 type) the output data type for auto-initialization must be passed as part + * of the @ref DirectConvolutionLayerOutputStageKernelInfo. + */ +class CpuDirectConvolutionOutputStageKernel : public ICpuKernel +{ +public: + /** Default constructor */ + CpuDirectConvolutionOutputStageKernel() = default; + ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(CpuDirectConvolutionOutputStageKernel); + /** Set the accumulate buffer and the biases of the kernel. + * + * @param[in, out] src Input to add the bias to. If @p output is not specified then accumulation is done in-place. + * Data type supported: F16/F32/S32 + * @param[in] bias (Optional) The shared bias tensor to add. It must be 1D Tensor. Data type supported: Same as @p src + * @param[out] dst (Optional) If the output tensor is specified the accumulation is done out-of-place. (Defaults to nullptr) + * Note that in-place computation is only supported for F16/F32. For S32 this must not be nullptr. + * Data type supported: F16/F32 or QASYMM8/QASYMM8_SIGNED if @p src is S32 + * @param[in] info (Optional) DirectConvolutionLayerOutputStageKernel descriptor metadata + */ + void configure(ITensorInfo *src, const ITensorInfo *bias = nullptr, ITensorInfo *dst = nullptr, + const DirectConvolutionLayerOutputStageKernelInfo &info = DirectConvolutionLayerOutputStageKernelInfo()); + /** Static function to check if given info will lead to a valid configuration of @ref CpuDirectConvolutionOutputStageKernel + * + * @param[in] src Input to add the bias to. If @p output is not specified then accumulation is done in-place. + * Data type supported: F16/F32/S32 + * @param[in] bias (Optional) The shared bias tensor to add. It must be 1D Tensor. Data type supported: Same as @p src + * @param[in] dst (Optional) If the output tensor is specified the accumulation is done out-of-place. (Defaults to nullptr) + * Note that in-place computation is only supported for F16/F32. For S32 this must not be nullptr. + * Data type supported: F16/F32 or QASYMM8/QASYMM8_SIGNED if @p src is S32 + * @param[in] info (Optional) DirectConvolutionLayerOutputStageKernel descriptor metadata + * + * @return a status + */ + static Status validate(const ITensorInfo *src, const ITensorInfo *bias = nullptr, const ITensorInfo *dst = nullptr, + const DirectConvolutionLayerOutputStageKernelInfo &info = DirectConvolutionLayerOutputStageKernelInfo()); + + // Inherited methods overridden: + void run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info) override; + const char *name() const override; + +private: + using OutputStageKernel = void(ITensor *src, const ITensor *bias, const Window &window, ITensor *dst, + int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift); + + OutputStageKernel *_func{ nullptr }; + int _result_fixedpoint_multiplier{ 0 }; + int _result_shift{ 0 }; + int _result_offset_after_shift{ 0 }; +}; +} // namespace kernels +} // namespace cpu +} // namespace arm_compute +#endif /*ARM_COMPUTE_CPU_DIRECTCONVOLUTION_OUTPUTSTAGE_KERNEL_H */ |