diff options
author | Viet-Hoa Do <viet-hoa.do@arm.com> | 2023-09-19 16:41:34 +0100 |
---|---|---|
committer | Viet-Hoa Do <viet-hoa.do@arm.com> | 2023-09-22 12:07:09 +0000 |
commit | cd1f03e765ad0f3ca3b68b1a7c1d0a1539cab439 (patch) | |
tree | 9cb78579e01e14501c316f5297c804ba13c8ad37 /compute_kernel_writer/include/ckw/KernelWriter.h | |
parent | 1f841a52f9a7f52948d676bc3807461bbed6f70a (diff) | |
download | ComputeLibrary-cd1f03e765ad0f3ca3b68b1a7c1d0a1539cab439.tar.gz |
Add row vector and scalar access support to tile operand
* Add the concept of tile view which refers to a specific rectangular
area of the tile object.
- The active area is added to TileOperand so that the user can access
part of the tile.
- Currently only row vector and scalar access are exposed to the user.
- All writing operations except load/store op support sub-tile.
* Add tests for sub-tile access.
Resolves: COMPMID-6557
Signed-off-by: Mohammed Suhail Munshi <MohammedSuhail.Munshi@arm.com>
Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Change-Id: Ica3f9eaf17f06e080c495d36c572f623b62c2910
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10354
Reviewed-by: Gunes Bayir <gunes.bayir@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 'compute_kernel_writer/include/ckw/KernelWriter.h')
-rw-r--r-- | compute_kernel_writer/include/ckw/KernelWriter.h | 8 |
1 files changed, 5 insertions, 3 deletions
diff --git a/compute_kernel_writer/include/ckw/KernelWriter.h b/compute_kernel_writer/include/ckw/KernelWriter.h index 93ae8aecd6..15c99fe652 100644 --- a/compute_kernel_writer/include/ckw/KernelWriter.h +++ b/compute_kernel_writer/include/ckw/KernelWriter.h @@ -34,6 +34,7 @@ #include <functional> #include <memory> #include <string> +#include <tuple> namespace ckw { @@ -42,6 +43,7 @@ namespace ckw class Kernel; class TensorInfo; class TensorSampler; +class TileArea; class TileInfo; enum class DataType; @@ -313,7 +315,7 @@ public: * @param[in] batch batch */ virtual void op_load_indirect(const TileOperand &tile_op, const TensorOperand &tensor_op, TensorSampler &sampler, - const TileOperand &x, const TileOperand &y, const TileOperand &z, const TileOperand &batch_op) = 0; + const TileOperand &x, const TileOperand &y, const TileOperand &z, const TileOperand &batch_op) = 0; protected: // ============================================================================================= @@ -355,8 +357,8 @@ protected: /** Create a new tile operand referring to the specified tile object. */ static TileOperand create_tile_operand(ITile &tile); - /** Get the reference to tile object from the tile operand. */ - static ITile &get_tile(const TileOperand &operand); + /** Get the reference to the tile object and the active area from the tile operand. */ + static std::tuple<ITile &, TileArea> get_tile(const TileOperand &operand); /** Create a new tensor operand from a tensor object. */ static TensorOperand create_tensor_operand(ITensor &tensor); |