diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2018-05-03 13:44:35 +0100 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:51:17 +0000 |
commit | 932b561159cd6a8c9230bbd0343790c85755846e (patch) | |
tree | f5345af71844e8f78873258bddec2cb37b39b8f8 /src/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.cpp | |
parent | 563494c2f447e201e88e6d7133a41e12971777eb (diff) | |
download | ComputeLibrary-932b561159cd6a8c9230bbd0343790c85755846e.tar.gz |
COMPMID-959: Perform pretranspose if allowed on NEON assembly
Change-Id: I281699ce7270aec1317c47b5a13799954cf6c9e8
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/130010
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Tello <pablo.tello@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Diffstat (limited to 'src/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.cpp')
-rw-r--r-- | src/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.cpp | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/src/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.cpp b/src/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.cpp index 27dd6c51d7..bd81bf202f 100644 --- a/src/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.cpp +++ b/src/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.cpp @@ -39,7 +39,7 @@ using namespace arm_compute; NEGEMMLowpAssemblyMatrixMultiplyCore::NEGEMMLowpAssemblyMatrixMultiplyCore(std::shared_ptr<IMemoryManager> memory_manager) : _memory_group(std::move(memory_manager)), _asm_glue_unsigned(), _asm_glue_signed(), _mm_kernel(nullptr), _mtx_a_reshape_kernel(nullptr), _mtx_b_reshape_kernel(nullptr), _tmp_a(), _tmp_b(), - _workspace() + _workspace(), _B_pretransposed() { } @@ -58,13 +58,13 @@ void NEGEMMLowpAssemblyMatrixMultiplyCore::configure(const ITensor *a, const ITe { case DataType::S8: { - run_optimised = setup_assembly_kernel(a, b, output, 1.f, 1.f, _workspace, _memory_group, _asm_glue_signed); + run_optimised = setup_assembly_kernel(a, b, output, 1.f, 1.f, true, _workspace, _B_pretransposed, _memory_group, _asm_glue_signed); break; } case DataType::QASYMM8: case DataType::U8: { - run_optimised = setup_assembly_kernel(a, b, output, 1.f, 1.f, _workspace, _memory_group, _asm_glue_unsigned); + run_optimised = setup_assembly_kernel(a, b, output, 1.f, 1.f, true, _workspace, _B_pretransposed, _memory_group, _asm_glue_unsigned); break; } default: |