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/NEGEMMLowpMatrixMultiplyCore.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/NEGEMMLowpMatrixMultiplyCore.cpp')
| -rw-r--r-- | src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp index cbec73fc31..30dd289326 100644 --- a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp +++ b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp @@ -42,8 +42,8 @@ using namespace arm_compute::misc::shape_calculator; NEGEMMLowpMatrixMultiplyCore::NEGEMMLowpMatrixMultiplyCore(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), _mtx_a_reduction_kernel(), - _mtx_b_reduction_kernel(), _offset_contribution_kernel(), _vector_sum_col(), _vector_sum_row(), _tmp_a(), _tmp_b(), _workspace(), _a_offset(0), _b_offset(0), _run_vector_matrix_multiplication(false), - _dot_product_path(false), _is_first_run(true), _reshape_b_only_on_first_run(false) + _mtx_b_reduction_kernel(), _offset_contribution_kernel(), _vector_sum_col(), _vector_sum_row(), _tmp_a(), _tmp_b(), _workspace(), _B_pretranspose(), _a_offset(0), _b_offset(0), + _run_vector_matrix_multiplication(false), _dot_product_path(false), _is_first_run(true), _reshape_b_only_on_first_run(false) { } @@ -62,13 +62,13 @@ void NEGEMMLowpMatrixMultiplyCore::configure(const ITensor *a, const ITensor *b, { case DataType::S8: { - _dot_product_path = setup_assembly_kernel(a, b, output, 1.f, 1.f, _workspace, _memory_group, _asm_glue_signed); + _dot_product_path = setup_assembly_kernel(a, b, output, 1.f, 1.f, true, _workspace, _B_pretranspose, _memory_group, _asm_glue_signed); break; } case DataType::QASYMM8: case DataType::U8: { - _dot_product_path = setup_assembly_kernel(a, b, output, 1.f, 1.f, _workspace, _memory_group, _asm_glue_unsigned); + _dot_product_path = setup_assembly_kernel(a, b, output, 1.f, 1.f, true, _workspace, _B_pretranspose, _memory_group, _asm_glue_unsigned); break; } default: |