diff options
| author | Sheri Zhang <sheri.zhang@arm.com> | 2021-06-29 17:34:06 +0100 |
|---|---|---|
| committer | Sheri Zhang <sheri.zhang@arm.com> | 2021-07-13 15:36:03 +0000 |
| commit | a387e271b1e02ffd5c2993702b9a21c1ed5c95fa (patch) | |
| tree | f53416756c70c85d962218168ad3cd3359d9f5c8 /src/core/CL/cl_kernels/elementwise_operation_quantized.cl | |
| parent | 6fc7d528382716de9e417c9dcf0fddf109446e9f (diff) | |
| download | ComputeLibrary-a387e271b1e02ffd5c2993702b9a21c1ed5c95fa.tar.gz | |
Add in-place calculation support for CL elementwise arithmetic kernels
- Add in-place calculation support in ClArithmeticKernel, ClSaturatedArithmeticKernel and ClMulKernel
- Add in-place test cases
Resolves: COMPMID-4431
Signed-off-by: Sheri Zhang <sheri.zhang@arm.com>
Change-Id: Id484bdb76b74478a33fedb471ae0c7f799c599f6
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5885
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/CL/cl_kernels/elementwise_operation_quantized.cl')
| -rw-r--r-- | src/core/CL/cl_kernels/elementwise_operation_quantized.cl | 21 |
1 files changed, 18 insertions, 3 deletions
diff --git a/src/core/CL/cl_kernels/elementwise_operation_quantized.cl b/src/core/CL/cl_kernels/elementwise_operation_quantized.cl index 0051babf03..a11be80875 100644 --- a/src/core/CL/cl_kernels/elementwise_operation_quantized.cl +++ b/src/core/CL/cl_kernels/elementwise_operation_quantized.cl @@ -86,8 +86,12 @@ */ __kernel void OP_FUN_NAME(OP)( TENSOR3D_DECLARATION(in1), - TENSOR3D_DECLARATION(in2), - TENSOR3D_DECLARATION(out)) + TENSOR3D_DECLARATION(in2) +#if !defined(IN_PLACE) + , + TENSOR3D_DECLARATION(out) +#endif // !defined(IN_PLACE) +) { #if VEC_SIZE_IN1 == 1 uint in1_x_offs = 0; @@ -99,12 +103,23 @@ __kernel void OP_FUN_NAME(OP)( #else // VEC_SIZE_IN2 == 1 uint in2_x_offs = max((int)(get_global_id(0) * VEC_SIZE_IN2 - (VEC_SIZE_IN2 - VEC_SIZE_LEFTOVER) % VEC_SIZE_IN2), 0); #endif // VEC_SIZE_IN2 == 1 +#if !defined(IN_PLACE) uint out_x_offs = max((int)(get_global_id(0) * VEC_SIZE_OUT - (VEC_SIZE_OUT - VEC_SIZE_LEFTOVER) % VEC_SIZE_OUT), 0); +#endif // !defined(IN_PLACE) // Get pixels pointer __global uchar *in1_addr = in1_ptr + in1_offset_first_element_in_bytes + in1_x_offs * sizeof(DATA_TYPE) + get_global_id(1) * in1_step_y + get_global_id(2) * in1_step_z; __global uchar *in2_addr = in2_ptr + in2_offset_first_element_in_bytes + in2_x_offs * sizeof(DATA_TYPE) + get_global_id(1) * in2_step_y + get_global_id(2) * in2_step_z; - __global uchar *out_addr = out_ptr + out_offset_first_element_in_bytes + out_x_offs * sizeof(DATA_TYPE) + get_global_id(1) * out_step_y + get_global_id(2) * out_step_z; + __global uchar * +#if !defined(IN_PLACE) + out_addr = out_ptr + out_offset_first_element_in_bytes + out_x_offs * sizeof(DATA_TYPE) + get_global_id(1) * out_step_y + get_global_id(2) * out_step_z; +#else // !defined(IN_PLACE) +#if defined(SRC1_IN_PLACE) + out_addr = in1_addr; +#else //defined(SRC1_IN_PLACE) + out_addr = in2_addr; +#endif //defined(SRC1_IN_PLACE) +#endif // !defined(IN_PLACE) VEC_INT in_a = CONVERT((VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE_OUT))(VLOAD(VEC_SIZE_IN1)(0, (__global DATA_TYPE *)in1_addr)), VEC_INT); VEC_INT in_b = CONVERT((VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE_OUT))(VLOAD(VEC_SIZE_IN2)(0, (__global DATA_TYPE *)in2_addr)), VEC_INT); |