diff options
author | Sang-Hoon Park <sang-hoon.park@arm.com> | 2021-05-17 17:04:50 +0100 |
---|---|---|
committer | Sang-Hoon Park <sang-hoon.park@arm.com> | 2021-05-26 10:16:05 +0000 |
commit | d89e2faa60d148f3c04e57032a28f1065a1be0e8 (patch) | |
tree | c95eb97f9c79198cb5db1232b497491df10614f2 /arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h | |
parent | 8b83d4684249bb96e27f95e11cf8f38a1c33b82b (diff) | |
download | ComputeLibrary-d89e2faa60d148f3c04e57032a28f1065a1be0e8.tar.gz |
Create CpuGemmDirectConv2d
As the first phase of making NEGEMMConv2d stateless,
CpuGemmDirectConv2d operator is created. Kernels and
operators used by the operator use TensorInfo pointers
instead of Tensor pointers.
The CpuGemmDirectConv2d isn't completely stateless
because it manages one intermediate tensor internally.
This will be resolved by implementing memory injection
mechanism with the following patches.
Also, weight manager of CpuGemmAssemblyDispatch is disabled
to enable this work.
Implements: COMPMID-4506
Change-Id: Iec3ca6de29d98bef7ea95e8f4473d6dc0024a140
Signed-off-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5672
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h')
-rw-r--r-- | arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h | 3 |
1 files changed, 3 insertions, 0 deletions
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h index a292712bd7..dc9783f9eb 100644 --- a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h +++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h @@ -25,6 +25,7 @@ #define ARM_COMPUTE_NEGEMMLOWPMATRIXMULTIPLYCORE_H #include "NEActivationLayer.h" +#include "arm_compute/core/ITensorPack.h" #include "arm_compute/runtime/IFunction.h" #include "arm_compute/runtime/IMemoryManager.h" #include "arm_compute/runtime/IWeightsManager.h" @@ -169,6 +170,8 @@ private: bool _fuse_output_stage; bool _run_activation; bool _flip_signedness; + + ITensorPack _asm_glue_tensors{}; }; } // namespace arm_compute #endif /*ARM_COMPUTE_NEGEMMLOWPMATRIXMULTIPLYCORE_H */ |