diff options
Diffstat (limited to 'src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp')
| -rw-r--r-- | src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp | 7 |
1 files changed, 4 insertions, 3 deletions
diff --git a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp index 842ee73397..c2e18a760a 100644 --- a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp +++ b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp @@ -205,16 +205,17 @@ Status CLGEMMLowpMatrixMultiplyCore::validate(const ITensorInfo *a, const ITenso const int k = a->dimension(0); constexpr int mult_transpose1xW_width = 1; constexpr int mult_interleave4x4_height = 1; - const GEMMReshapeInfo reshape_info(m, n, k, mult_transpose1xW_width, mult_interleave4x4_height); + const int depth_output_gemm3d = gemm_info.depth_output_gemm3d(); + const GEMMReshapeInfo reshape_info(m, n, k, mult_transpose1xW_width, mult_interleave4x4_height, depth_output_gemm3d); bool reshape_matrices = is_interleaved_transposed(m, n, k, gemm_info.reshape_b_only_on_first_run(), CLScheduler::get().target()); if(reshape_matrices) { - TensorInfo info_a(compute_interleaved_shape(*a, mult_interleave4x4_height), 1, a->data_type()); + TensorInfo info_a(compute_interleaved_shape(*a, mult_interleave4x4_height, gemm_info.reinterpret_input_as_3d()), 1, a->data_type()); TensorInfo info_b(compute_transpose1xW_with_element_size_shape(*b, mult_transpose1xW_width), 1, b->data_type()); - ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMInterleave4x4Kernel::validate(a, &info_a, mult_interleave4x4_height)); + ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMInterleave4x4Kernel::validate(a, &info_a, mult_interleave4x4_height, gemm_info.reinterpret_input_as_3d())); ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMTranspose1xWKernel::validate(b, &info_b, mult_transpose1xW_width)); ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixMultiplyKernel::validate(&info_a, &info_b, output, reshape_matrices, reshape_info)); } |