diff options
author | Viet-Hoa Do <viet-hoa.do@arm.com> | 2022-06-01 11:47:14 +0100 |
---|---|---|
committer | Viet-Hoa Do <viet-hoa.do@arm.com> | 2022-11-28 16:57:42 +0000 |
commit | 03b2971ac69a86f10a1566938d1a25afee15746c (patch) | |
tree | aec7cfc047e1da278b4b71a706cda7b1b0faa158 /src/core/NEON/kernels/arm_gemm/std_transforms_sme.hpp | |
parent | 7dc0234331f2150a6b4ac5c2b49de419870f7cf5 (diff) | |
download | ComputeLibrary-03b2971ac69a86f10a1566938d1a25afee15746c.tar.gz |
Integrate SME2 kernels
* Add SME/SME2 detection.
* Integrate SME2 implementation for:
- Normal convolution
- Winograd
- Depthwise convolution
- Pooling
Resolves: COMPMID-5700
Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Change-Id: I2f1ca1d05f8cfeee9309ed1c0a36096a4a6aad5c
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8692
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/std_transforms_sme.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/std_transforms_sme.hpp | 75 |
1 files changed, 75 insertions, 0 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/std_transforms_sme.hpp b/src/core/NEON/kernels/arm_gemm/std_transforms_sme.hpp new file mode 100644 index 0000000000..054088e0b5 --- /dev/null +++ b/src/core/NEON/kernels/arm_gemm/std_transforms_sme.hpp @@ -0,0 +1,75 @@ +/* + * Copyright (c) 2022 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. + */ +#pragma once + +#include "interleave_indirect.hpp" +#include "transform.hpp" + +namespace arm_gemm { + +/* + * Define "standard" transforms for the blocked GEMMs for SVE. + * + * This assumes that A is interleaved 'height' ways, B is interleaved + * 'width'xVL ways and transposed, and that the merge needs to work in + * 'height' x 'width'xVL blocks. + * + * The optional 'block' parameter is for kernels using dot-product type + * instructions like UDOT and SDOT. + */ +template<typename TOperand, typename TResult, unsigned int height_vectors, unsigned int width_vectors, unsigned int block=1, bool integrate_sums=false> +class StdTransformsSME +{ +public: + template<typename TIn> + void PrepareA(TOperand *out, const TIn *in, const int stride, const int y0, + const int ymax, const int k0, const int kmax, int32_t row_sum_multiplier) { + Interleave<height_vectors, block, VLType::SME>(out, in, stride, y0, ymax, k0, kmax, integrate_sums, row_sum_multiplier); + } + + template<typename TIn> + void PrepareA_indirect(TOperand *out, const TIn * const * const *ptr, size_t stringlen, size_t rounded_stringlen, const int y0, + const int ymax, const int k0, const int kmax, int32_t row_sum_multiplier) { + IndirectInterleave<height_vectors, block, VLType::SME>(out, ptr, stringlen, rounded_stringlen, y0, ymax, k0, kmax, integrate_sums, row_sum_multiplier); + } + + template<typename TIn> + void PrepareA_convolution(TOperand *out, const TIn *ptr, size_t stride, const convolver<TIn> &conv, size_t rounded_stringlen, + const int y0, const int ymax, const int k0, const int kmax, int32_t row_sum_multiplier) { + ConvolutionInterleave<height_vectors, block, VLType::SME>(out, ptr, stride, conv, rounded_stringlen, y0, ymax, k0, kmax, integrate_sums, row_sum_multiplier); + } + + template<typename TIn> + void PrepareB(TOperand *out, const TIn *in, const int stride, const int x0, + const int xmax, const int k0, const int kmax) { + Transform<width_vectors, block, true, VLType::SME>(out, in, stride, x0, xmax, k0, kmax); + } + + template<typename TOut> + void Merge(TOut *out, const TResult *in, int stride, int y0, int ymax, int x0, int xmax, const TOut *bias, const Activation act, bool accumulate) { + // Separate merge not supported for SME. + } +}; + +} // namespace arm_gemm |