diff options
| author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2022-11-17 11:03:39 +0000 |
|---|---|---|
| committer | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2022-12-09 11:06:23 +0000 |
| commit | 76335eb8d8733b0bbc0110546797211540870c50 (patch) | |
| tree | 812fc44de593c9e1e45ac8b534094511b06163bf /src/core/CL/cl_kernels/tile_helpers.h | |
| parent | f16973b8b4605f12608bffa9f0ca6ed590202d41 (diff) | |
| download | ComputeLibrary-76335eb8d8733b0bbc0110546797211540870c50.tar.gz | |
Implement the OpenCL kernel to compute the indirect convolution
- Implement indirect convolution kernel
- Add operator support
- Add test
Resolves COMPMID-5709
Change-Id: I9272304163471a5a40da7fdec204599f3c1d8e32
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8701
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/CL/cl_kernels/tile_helpers.h')
| -rw-r--r-- | src/core/CL/cl_kernels/tile_helpers.h | 20 |
1 files changed, 17 insertions, 3 deletions
diff --git a/src/core/CL/cl_kernels/tile_helpers.h b/src/core/CL/cl_kernels/tile_helpers.h index 998bc9efb2..861ea63eca 100644 --- a/src/core/CL/cl_kernels/tile_helpers.h +++ b/src/core/CL/cl_kernels/tile_helpers.h @@ -653,13 +653,27 @@ }) \ }) -#define T_LOAD2D_INDIRECT(DATA_TYPE, TILE_AREA, TILE_CHANNELS, TENSOR_TYPE, TENSOR, B, Y, X, C, TENSOR_WIDTH, TENSOR_HEIGHT, STRIDE_Y, yi, dst) \ +/** Load a tile from global memory (tensor) using an indirect buffer for the Y coordinates + * + * @param[in] DATA_TYPE Data type + * @param[in] TILE_AREA Number of elements to load from Y (height) dimension * Number of elements to load from X (width) dimension + * @param[in] TILE_CHANNELS Number of elements to load from C (channel) dimension + * @param[in] TENSOR_TYPE Type of cl_type used to store the tensor in global memory (BUFFER=cl_buffer, IMAGE=cl_image). Currently BUFFER only is supported + * In case of cl_image, only TILE_CHANNELS multiples of 4 are supported (4, 8, 16) + * @param[in] TENSOR Tensor basename + * @param[in] C Starting C index + * @param[in] STRIDE_Y Stride Y (in bytes) + * @param[out] yi A tile with (TILE_WIDTH x TILE_HEIGHT) values with the indirect Y coordinate + * 16 is the maximum indirect buffer size. + * @param[out] dst Output tile + */ +#define T_LOAD2D_INDIRECT(DATA_TYPE, TILE_AREA, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, STRIDE_Y, yi, dst) \ ({ \ LOOP_UNROLLING(int, _i, 0, 1, TILE_AREA, \ { \ - if(yi[_i].v >= 0) \ + if(yi[0].s[_i] >= 0) \ { \ - dst[_i].v = V_LOAD(DATA_TYPE, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, yi[_i].v, STRIDE_Y); \ + dst[_i].v = V_LOAD(DATA_TYPE, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, yi[0].s[_i], STRIDE_Y); \ } \ }) \ }) |