diff options
author | Mohammed Suhail Munshi <MohammedSuhail.Munshi@arm.com> | 2022-07-04 13:36:14 +0100 |
---|---|---|
committer | Mohmun02 <MohammedSuhail.Munshi@arm.com> | 2022-08-03 16:57:56 +0000 |
commit | f67903b8ab8205b47f0ee2c27aeca8bed405c58e (patch) | |
tree | f8773d534657b062b70059ee5aab623aa190c767 /src/core/experimental/dynamic_fusion/ClKernelBuildingImpl/components/ClStoreKernelComponents.cpp | |
parent | 13b623e575ed2f1096c70560a2db4a9e03cf22f9 (diff) | |
download | ComputeLibrary-f67903b8ab8205b47f0ee2c27aeca8bed405c58e.tar.gz |
Add Dynamic Fusion Tests with BugFixes
- Allow fusing arbitrary number of existing elementwise operators
- Fix issues with 3D and 4D tensors in Elementwise Addition and Floor components
- Collapse the 3D/4D window in the same way as that used by Conv2d,
i.e. collapse dim 1 and dim 2 together
- Fix Floor component issues when used after other components
- Add Dynamic Fusion Tests (Floor + Div, Conv2d + Add + Div)
- Add Addition ElementWise Broadcasting Test
Resolves: [COMPMID-5356]
Change-Id: I58b93a90175bb0440d43531d18cac94b5f5c2689
Signed-off-by: Mohammed Suhail Munshi <MohammedSuhail.Munshi@arm.com>
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/c/VisualCompute/ComputeLibrary/+/433956
Tested-by: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Pablo Tello <pablo.tello@arm.com>
Comments-Addressed: bsgcomp <bsgcomp@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7957
Reviewed-by: SiCong Li <sicong.li@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/experimental/dynamic_fusion/ClKernelBuildingImpl/components/ClStoreKernelComponents.cpp')
-rw-r--r-- | src/core/experimental/dynamic_fusion/ClKernelBuildingImpl/components/ClStoreKernelComponents.cpp | 4 |
1 files changed, 4 insertions, 0 deletions
diff --git a/src/core/experimental/dynamic_fusion/ClKernelBuildingImpl/components/ClStoreKernelComponents.cpp b/src/core/experimental/dynamic_fusion/ClKernelBuildingImpl/components/ClStoreKernelComponents.cpp index 4ac27e007f..7c805d5368 100644 --- a/src/core/experimental/dynamic_fusion/ClKernelBuildingImpl/components/ClStoreKernelComponents.cpp +++ b/src/core/experimental/dynamic_fusion/ClKernelBuildingImpl/components/ClStoreKernelComponents.cpp @@ -108,6 +108,10 @@ std::string ClStoreIndirectWidthSelectKernelComponent::get_component_code() cons return R"_( //------------------ START KERNEL {{meta_kernel_id}} STORE --------------------- { + // This also follows NHWC layout + // cout maps to global_id(0) maps to Channel + // mout maps to global_id(1) maps to Height and Weight (Collapsed Window) + // bout maps to global_id(3) maps to N / Batch #define _IDST_WIDTH {{dst}}_w #define _IDST_HEIGHT {{dst}}_h TILE(uint, M0, 1, dst_indirect_y); |