diff options
| author | Isabella Gottardi <isabella.gottardi@arm.com> | 2018-03-01 16:42:00 +0000 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:53:09 +0000 |
| commit | 8e74f4488daf1b628ca718396d5fc72fea95a83d (patch) | |
| tree | f372c61aab423799f82ea7a98aa6a157a4887bdc /src/runtime/CL/functions/CLGEMM.cpp | |
| parent | 0a887922c73bbe7c5d42b1eb3ae55730f0d9a139 (diff) | |
| download | ComputeLibrary-8e74f4488daf1b628ca718396d5fc72fea95a83d.tar.gz | |
COMPMID-911: Allow GEMM to work with 3D tensors
Change-Id: I8c4823a0d909e19e9ef548f00b9ae98c66de61dd
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/123569
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Diffstat (limited to 'src/runtime/CL/functions/CLGEMM.cpp')
| -rw-r--r-- | src/runtime/CL/functions/CLGEMM.cpp | 9 |
1 files changed, 3 insertions, 6 deletions
diff --git a/src/runtime/CL/functions/CLGEMM.cpp b/src/runtime/CL/functions/CLGEMM.cpp index a0ec66f804..f9713bb586 100644 --- a/src/runtime/CL/functions/CLGEMM.cpp +++ b/src/runtime/CL/functions/CLGEMM.cpp @@ -24,10 +24,6 @@ #include "arm_compute/runtime/CL/functions/CLGEMM.h" #include "arm_compute/core/CL/ICLTensor.h" -#include "arm_compute/core/CL/kernels/CLGEMMInterleave4x4Kernel.h" -#include "arm_compute/core/CL/kernels/CLGEMMMatrixAdditionKernel.h" -#include "arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h" -#include "arm_compute/core/CL/kernels/CLGEMMTranspose1xWKernel.h" #include "arm_compute/core/Error.h" #include "arm_compute/core/GPUTarget.h" #include "arm_compute/core/Helpers.h" @@ -111,6 +107,7 @@ void CLGEMM::configure(const ICLTensor *a, const ICLTensor *b, const ICLTensor * const int m = a->info()->dimension(1); const int n = b->info()->dimension(0); const int k = a->info()->dimension(0); + const int depth_output_gemm3d = gemm_info.depth_output_gemm3d(); int mult_transpose1xW_width = 1; int mult_interleave4x4_height = 1; @@ -144,7 +141,7 @@ void CLGEMM::configure(const ICLTensor *a, const ICLTensor *b, const ICLTensor * } // Configure and tune matrix multiply kernel - _mm_kernel.configure(matrix_a, matrix_b, output, alpha, _is_interleaved_transposed, GEMMReshapeInfo(m, n, k, mult_transpose1xW_width, mult_interleave4x4_height)); + _mm_kernel.configure(matrix_a, matrix_b, output, alpha, _is_interleaved_transposed, GEMMReshapeInfo(m, n, k, mult_transpose1xW_width, mult_interleave4x4_height, depth_output_gemm3d)); CLScheduler::get().tune_kernel_static(_mm_kernel); if(_is_interleaved_transposed) @@ -197,7 +194,7 @@ Status CLGEMM::validate(const ITensorInfo *a, const ITensorInfo *b, const ITenso mult_interleave4x4_height = 2; } - const GEMMReshapeInfo reshape_info = GEMMReshapeInfo(m, n, k, mult_transpose1xW_width, mult_interleave4x4_height); + const GEMMReshapeInfo reshape_info = GEMMReshapeInfo(m, n, k, mult_transpose1xW_width, mult_interleave4x4_height, gemm_info.depth_output_gemm3d()); // Check if we need to reshape the matrix A and matrix B const bool run_interleave_transpose = is_interleaved_transposed(m, n, k, a->data_type(), reshape_b_only_on_first_run, gpu_target); |