diff options
| -rw-r--r-- | src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp | 2 | ||||
| -rw-r--r-- | src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp | 4 |
2 files changed, 3 insertions, 3 deletions
diff --git a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp index 2c072a8ba0..1a1a4b7c3d 100644 --- a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp +++ b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp @@ -89,7 +89,7 @@ Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *input1, lhs_info.k0 = 4; lhs_info.v0 = mult_interleave4x4_height; lhs_info.interleave = true; - lhs_info.transpose = unroll_block; + lhs_info.transpose = !unroll_block; TensorShape tensor_shape0{ input0->tensor_shape() }; tensor_shape0.set(0, k); diff --git a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp index edb3107173..4b72878b5f 100644 --- a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp +++ b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp @@ -132,7 +132,7 @@ void CLGEMMLowpMatrixMultiplyCore::configure(const ICLTensor *a, const ICLTensor lhs_info.k0 = 4; lhs_info.v0 = mult_interleave4x4_height; lhs_info.interleave = true; - lhs_info.transpose = unroll_block; + lhs_info.transpose = !unroll_block; // Check if we need to reshape the matrix A and matrix B _is_interleaved_transposed = is_interleaved_transposed(m, n, k, _reshape_b_only_on_first_run, gpu_target); @@ -268,7 +268,7 @@ Status CLGEMMLowpMatrixMultiplyCore::validate(const ITensorInfo *a, const ITenso lhs_info.k0 = 4; lhs_info.v0 = mult_interleave4x4_height; lhs_info.interleave = true; - lhs_info.transpose = unroll_block; + lhs_info.transpose = !unroll_block; bool reshape_matrices = is_interleaved_transposed(m, n, k, gemm_info.reshape_b_only_on_first_run(), CLScheduler::get().target()); |