diff options
author | Gunes Bayir <gunes.bayir@arm.com> | 2023-07-06 15:51:12 +0100 |
---|---|---|
committer | Gunes Bayir <gunes.bayir@arm.com> | 2023-07-07 11:19:01 +0000 |
commit | 5153fe4b989768907ed588ecdb859d6a81b72992 (patch) | |
tree | 879bc23c32cc2f5190d8c3e051ea1c7a17e06048 /src/gpu/cl/operators/ClFullyConnected.h | |
parent | acea4071a7f457bab696dc3c895ba47d60345541 (diff) | |
download | ComputeLibrary-5153fe4b989768907ed588ecdb859d6a81b72992.tar.gz |
Fix unsupported configuration in CLFullyConnected validation
When the weights to CLFullyConnected layer are not constant and the weights need reshaping, we prefer MatMul kernels instead of Gemm-based ones. The bias addition is currently unsupported in MatMull, but the validate() function does not account for this properly. This patch fixes the validation of this function.
Resolves: COMPMID-6338
Change-Id: I5c240191ae8e369753691c43ab4a30d4ae1776b0
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9882
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/gpu/cl/operators/ClFullyConnected.h')
-rw-r--r-- | src/gpu/cl/operators/ClFullyConnected.h | 4 |
1 files changed, 0 insertions, 4 deletions
diff --git a/src/gpu/cl/operators/ClFullyConnected.h b/src/gpu/cl/operators/ClFullyConnected.h index 5dc68c1bbe..9a5ba40510 100644 --- a/src/gpu/cl/operators/ClFullyConnected.h +++ b/src/gpu/cl/operators/ClFullyConnected.h @@ -132,11 +132,7 @@ private: TensorInfo _flattened_src{}; TensorInfo _converted_weights{}; TensorInfo _reshaped_weights{}; - - // Saved tensor shapes for reshaping when using matmul - TensorShape _lhs_shape_original{}; TensorInfo _lhs_to_use{}; - TensorInfo _weights_to_use{}; int _weights_to_use_idx{ ACL_SRC_1 }; |