Compute Library
18.05
|
#include "helpers.h"
Go to the source code of this file.
Macros | |
#define | SWAP_ROW(u0, l0) |
#define | SWAP_4x4(u0, u1, u2, u3, l0, l1, l2, l3) |
#define | SWAP_8x8(u0, u1, u2, u3, u4, u5, u6, u7, l0, l1, l2, l3, l4, l5, l6, l7) |
#define | TRANSPOSE_4x4(u0, u1, u2, u3) |
#define | TRANSPOSE_8x8(u0, u1, u2, u3, u4, u5, u6, u7) |
#define | TRANSPOSE_16x16(u0, u1, u2, u3, u4, u5, u6, u7, u8, u9, u10, u11, u12, u13, u14, u15) |
Functions | |
__kernel void | transpose (__global uchar *src_ptr, uint src_stride_x, uint src_step_x, uint src_stride_y, uint src_step_y, uint src_offset_first_element_in_bytes, __global uchar *dst_ptr, uint dst_stride_x, uint dst_step_x, uint dst_stride_y, uint dst_step_y, uint dst_offset_first_element_in_bytes) |
This OpenCL kernel computes the matrix transposition of input matrix. More... | |
#define SWAP_4x4 | ( | u0, | |
u1, | |||
u2, | |||
u3, | |||
l0, | |||
l1, | |||
l2, | |||
l3 | |||
) |
Definition at line 33 of file transpose.cl.
#define SWAP_8x8 | ( | u0, | |
u1, | |||
u2, | |||
u3, | |||
u4, | |||
u5, | |||
u6, | |||
u7, | |||
l0, | |||
l1, | |||
l2, | |||
l3, | |||
l4, | |||
l5, | |||
l6, | |||
l7 | |||
) |
Definition at line 43 of file transpose.cl.
#define SWAP_ROW | ( | u0, | |
l0 | |||
) |
Definition at line 26 of file transpose.cl.
#define TRANSPOSE_16x16 | ( | u0, | |
u1, | |||
u2, | |||
u3, | |||
u4, | |||
u5, | |||
u6, | |||
u7, | |||
u8, | |||
u9, | |||
u10, | |||
u11, | |||
u12, | |||
u13, | |||
u14, | |||
u15 | |||
) |
Definition at line 89 of file transpose.cl.
#define TRANSPOSE_4x4 | ( | u0, | |
u1, | |||
u2, | |||
u3 | |||
) |
Definition at line 57 of file transpose.cl.
#define TRANSPOSE_8x8 | ( | u0, | |
u1, | |||
u2, | |||
u3, | |||
u4, | |||
u5, | |||
u6, | |||
u7 | |||
) |
Definition at line 80 of file transpose.cl.
__kernel void transpose | ( | __global uchar * | src_ptr, |
uint | src_stride_x, | ||
uint | src_step_x, | ||
uint | src_stride_y, | ||
uint | src_step_y, | ||
uint | src_offset_first_element_in_bytes, | ||
__global uchar * | dst_ptr, | ||
uint | dst_stride_x, | ||
uint | dst_step_x, | ||
uint | dst_stride_y, | ||
uint | dst_step_y, | ||
uint | dst_offset_first_element_in_bytes | ||
) |
This OpenCL kernel computes the matrix transposition of input matrix.
[in] | src_ptr | Pointer to the source matrix. Supported data types: U8/S8/U16/S16/F16/U32/S32/F32 |
[in] | src_stride_x | Stride of the source matrix in X dimension (in bytes) |
[in] | src_step_x | src_stride_x * number of elements along X processed per workitem(in bytes) |
[in] | src_stride_y | Stride of the source matrix in Y dimension (in bytes) |
[in] | src_step_y | src_stride_y * number of elements along Y processed per workitem(in bytes) |
[in] | src_offset_first_element_in_bytes | The offset of the first element in the source matrix |
[out] | dst_ptr | Pointer to the destination matrix Supported data type: same as src_ptr |
[in] | dst_stride_x | Stride of the destination matrix in X dimension (in bytes) |
[in] | dst_step_x | dst_gx_stride_x * number of elements along X processed per workitem(in bytes) |
[in] | dst_stride_y | Stride of the destination matrix in Y dimension (in bytes) |
[in] | dst_step_y | dst_gx_stride_y * number of elements along Y processed per workitem(in bytes) |
[in] | dst_offset_first_element_in_bytes | The offset of the first element in the destination matrix |
Definition at line 148 of file transpose.cl.
References CONVERT_TO_IMAGE_STRUCT, DATA_TYPE, offset(), arm_compute::test::validation::src, VEC_DATA_TYPE, VLOAD, and VSTORE.