diff options
| author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2018-08-08 12:29:38 +0100 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:54:54 +0000 |
| commit | db9d46da3a8645d0c2cc71d035448999a36770ec (patch) | |
| tree | 79c61d6ad845e60d86fa3059056b8ac3fffc236b /src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp | |
| parent | 980002bd5848f065b02a31bb105e47a5deb7bc98 (diff) | |
| download | ComputeLibrary-db9d46da3a8645d0c2cc71d035448999a36770ec.tar.gz | |
COMPMID-1485 - Add support for NHWC when running NEGEMMConvolutionLayer with FP16/QASYMM8
When the GEMM3D check fails, now we fallback to the classic implementation with im2col
and col2im. In this manner the function can work with QASYMM8 and FP16
Change-Id: I359e9da3a63956f33b5acbc9bca4383b14af10e2
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/143372
Tested-by: Jenkins <bsgcomp@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 | 2 |
1 files changed, 2 insertions, 0 deletions
diff --git a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp index 773492d0ce..828011d019 100644 --- a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp +++ b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp @@ -195,6 +195,8 @@ Status NEGEMMLowpMatrixMultiplyCore::validate(const ITensorInfo *a, const ITenso ARM_COMPUTE_UNUSED(gemm_info); ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.is_a_reshaped(), "Matrix A already reshaped is not supported"); ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.is_b_reshaped(), "Matrix B already reshaped is not supported"); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.reinterpret_input_as_3d(), "NEGEMMLowpMatrixMultiplyCore cannot reinterpret the input tensor as 3D"); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.depth_output_gemm3d() != 1, "NEGEMMLowpMatrixMultiplyCore cannot reinterpret the output tensor as 3D"); int32_t a_offset = a->quantization_info().offset; int32_t b_offset = b->quantization_info().offset; |