diff options
author | Michele Di Giorgio <michele.digiorgio@arm.com> | 2021-06-21 12:00:43 +0100 |
---|---|---|
committer | Michele Di Giorgio <michele.digiorgio@arm.com> | 2021-06-29 13:29:01 +0000 |
commit | 93b75e0c072c3cc5654fcdf6aed1068b40012081 (patch) | |
tree | 08acbf1bcafaa326bea1d8e472ad66b955c7c17f /src/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.cpp | |
parent | 5fdde99f4271891a40c02cd1e89f1344aa84583a (diff) | |
download | ComputeLibrary-93b75e0c072c3cc5654fcdf6aed1068b40012081.tar.gz |
Port NEGEMM to memory injecting interface (Part 1)
- Start porting NEGEMM to the new API
- Port NEGEMMInterleave4x4Kernel to the new API
- Port NEGEMMMatrixAdditionKernel to the new API
- Port NEGEMMTranspose1xWKernel to the new API
- Remove padding from NEGEMMMatrixAdditionKernel
- Remove unused INESimpleKernel and ICPPSimpleKernel
Partially resolves: COMPMID-4402
Change-Id: I63edadddfe00a54586e5384d6a0211db25ae9042
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5857
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.cpp')
-rw-r--r-- | src/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.cpp | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/src/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.cpp b/src/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.cpp index b95bdd4ca5..6bcf59ee96 100644 --- a/src/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.cpp +++ b/src/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.cpp @@ -686,7 +686,7 @@ void inline matrix_multiply_s8(Iterator &ina, Iterator &inb, Iterator &out, int const auto width_out = static_cast<int>(out_info.dimension(0)); const auto height_out = static_cast<int>(out_info.dimension(1)); const size_t out_stride = out_info.strides_in_bytes()[1] / out_info.element_size(); - // The implementation assumes that the matrix A and Matrix B have been reshaped respectively with NEGEMMInterleave4x4 and NEGEMMTranspose1xW + // The implementation assumes that the matrix A and Matrix B have been reshaped respectively with CpuGemmInterleave4x4 and CpuGemmTranspose1xW // The reshaping of the matrices helps to have a cache friendly implementation and helps to avoid the data re-arrangements needed for computing 16x4 elements per iteration // All the values needed for computing a single 4x4 block will be read from consecutive memory positions execute_window_loop(window, [&](const Coordinates & id) |