| 1 |
// Start of copy.cl |
| 2 |
|
| 3 |
#define GEN_COPY_KERNEL(NAME, ELEM_TYPE) \ |
| 4 |
FUTHARK_KERNEL void lmad_copy_##NAME(SHARED_MEM_PARAM \ |
| 5 |
__global ELEM_TYPE *dst_mem, \ |
| 6 |
int64_t dst_offset, \ |
| 7 |
__global ELEM_TYPE *src_mem, \ |
| 8 |
int64_t src_offset, \ |
| 9 |
int64_t n, \ |
| 10 |
int r, \ |
| 11 |
int64_t shape0, int64_t dst_stride0, int64_t src_stride0, \ |
| 12 |
int64_t shape1, int64_t dst_stride1, int64_t src_stride1, \ |
| 13 |
int64_t shape2, int64_t dst_stride2, int64_t src_stride2, \ |
| 14 |
int64_t shape3, int64_t dst_stride3, int64_t src_stride3, \ |
| 15 |
int64_t shape4, int64_t dst_stride4, int64_t src_stride4, \ |
| 16 |
int64_t shape5, int64_t dst_stride5, int64_t src_stride5, \ |
| 17 |
int64_t shape6, int64_t dst_stride6, int64_t src_stride6, \ |
| 18 |
int64_t shape7, int64_t dst_stride7, int64_t src_stride7) { \ |
| 19 |
int64_t gtid = get_global_id(0); \ |
| 20 |
int64_t remainder = gtid; \ |
| 21 |
\ |
| 22 |
if (gtid >= n) { \ |
| 23 |
return; \ |
| 24 |
} \ |
| 25 |
\ |
| 26 |
if (r > 0) { \ |
| 27 |
int64_t i = remainder % shape0; \ |
| 28 |
dst_offset += i * dst_stride0; \ |
| 29 |
src_offset += i * src_stride0; \ |
| 30 |
remainder /= shape0; \ |
| 31 |
} \ |
| 32 |
if (r > 1) { \ |
| 33 |
int64_t i = remainder % shape1; \ |
| 34 |
dst_offset += i * dst_stride1; \ |
| 35 |
src_offset += i * src_stride1; \ |
| 36 |
remainder /= shape1; \ |
| 37 |
} \ |
| 38 |
if (r > 2) { \ |
| 39 |
int64_t i = remainder % shape2; \ |
| 40 |
dst_offset += i * dst_stride2; \ |
| 41 |
src_offset += i * src_stride2; \ |
| 42 |
remainder /= shape2; \ |
| 43 |
} \ |
| 44 |
if (r > 3) { \ |
| 45 |
int64_t i = remainder % shape3; \ |
| 46 |
dst_offset += i * dst_stride3; \ |
| 47 |
src_offset += i * src_stride3; \ |
| 48 |
remainder /= shape3; \ |
| 49 |
} \ |
| 50 |
if (r > 4) { \ |
| 51 |
int64_t i = remainder % shape4; \ |
| 52 |
dst_offset += i * dst_stride4; \ |
| 53 |
src_offset += i * src_stride4; \ |
| 54 |
remainder /= shape4; \ |
| 55 |
} \ |
| 56 |
if (r > 5) { \ |
| 57 |
int64_t i = remainder % shape5; \ |
| 58 |
dst_offset += i * dst_stride5; \ |
| 59 |
src_offset += i * src_stride5; \ |
| 60 |
remainder /= shape5; \ |
| 61 |
} \ |
| 62 |
if (r > 6) { \ |
| 63 |
int64_t i = remainder % shape6; \ |
| 64 |
dst_offset += i * dst_stride6; \ |
| 65 |
src_offset += i * src_stride6; \ |
| 66 |
remainder /= shape6; \ |
| 67 |
} \ |
| 68 |
if (r > 7) { \ |
| 69 |
int64_t i = remainder % shape7; \ |
| 70 |
dst_offset += i * dst_stride7; \ |
| 71 |
src_offset += i * src_stride7; \ |
| 72 |
remainder /= shape7; \ |
| 73 |
} \ |
| 74 |
\ |
| 75 |
dst_mem[dst_offset] = src_mem[src_offset]; \ |
| 76 |
} |
| 77 |
|
| 78 |
GEN_COPY_KERNEL(1b, uint8_t) |
| 79 |
GEN_COPY_KERNEL(2b, uint16_t) |
| 80 |
GEN_COPY_KERNEL(4b, uint32_t) |
| 81 |
GEN_COPY_KERNEL(8b, uint64_t) |
| 82 |
|
| 83 |
// End of copy.cl |