|
| 1 | +//===----------------- generic/lib/misc/shuffle.cl ------------------------===// |
| 2 | +// |
| 3 | +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | +// See https://llvm.org/LICENSE.txt for license information. |
| 5 | +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | +// |
| 7 | +//===----------------------------------------------------------------------===// |
| 8 | + |
| 9 | +#include <spirv/spirv.h> |
| 10 | + |
| 11 | +#define _CLC_ELEMENT_CASES2(VAR) \ |
| 12 | + case 0: \ |
| 13 | + return VAR.s0; \ |
| 14 | + case 1: \ |
| 15 | + return VAR.s1; |
| 16 | + |
| 17 | +#define _CLC_ELEMENT_CASES4(VAR) \ |
| 18 | + _CLC_ELEMENT_CASES2(VAR) \ |
| 19 | + case 2: \ |
| 20 | + return VAR.s2; \ |
| 21 | + case 3: \ |
| 22 | + return VAR.s3; |
| 23 | + |
| 24 | +#define _CLC_ELEMENT_CASES8(VAR) \ |
| 25 | + _CLC_ELEMENT_CASES4(VAR) \ |
| 26 | + case 4: \ |
| 27 | + return VAR.s4; \ |
| 28 | + case 5: \ |
| 29 | + return VAR.s5; \ |
| 30 | + case 6: \ |
| 31 | + return VAR.s6; \ |
| 32 | + case 7: \ |
| 33 | + return VAR.s7; |
| 34 | + |
| 35 | +#define _CLC_ELEMENT_CASES16(VAR) \ |
| 36 | + _CLC_ELEMENT_CASES8(VAR) \ |
| 37 | + case 8: \ |
| 38 | + return VAR.s8; \ |
| 39 | + case 9: \ |
| 40 | + return VAR.s9; \ |
| 41 | + case 10: \ |
| 42 | + return VAR.sA; \ |
| 43 | + case 11: \ |
| 44 | + return VAR.sB; \ |
| 45 | + case 12: \ |
| 46 | + return VAR.sC; \ |
| 47 | + case 13: \ |
| 48 | + return VAR.sD; \ |
| 49 | + case 14: \ |
| 50 | + return VAR.sE; \ |
| 51 | + case 15: \ |
| 52 | + return VAR.sF; |
| 53 | + |
| 54 | +#define _CLC_GET_ELEMENT_DEFINE(ARGTYPE, ARGSIZE, IDXTYPE) \ |
| 55 | + inline ARGTYPE __clc_get_el_##ARGTYPE##ARGSIZE##_##IDXTYPE( \ |
| 56 | + ARGTYPE##ARGSIZE x, IDXTYPE idx) { \ |
| 57 | + switch (idx) { _CLC_ELEMENT_CASES##ARGSIZE(x) default : return 0; } \ |
| 58 | + } |
| 59 | + |
| 60 | +#define _CLC_SHUFFLE_SET_ONE_ELEMENT(ARGTYPE, ARGSIZE, INDEX, MASKTYPE) \ |
| 61 | + ret_val.s##INDEX = \ |
| 62 | + __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.s##INDEX); |
| 63 | + |
| 64 | +#define _CLC_SHUFFLE_SET_2_ELEMENTS(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 65 | + ret_val.s0 = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.s0); \ |
| 66 | + ret_val.s1 = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.s1); |
| 67 | + |
| 68 | +#define _CLC_SHUFFLE_SET_4_ELEMENTS(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 69 | + _CLC_SHUFFLE_SET_2_ELEMENTS(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 70 | + ret_val.s2 = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.s2); \ |
| 71 | + ret_val.s3 = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.s3); |
| 72 | + |
| 73 | +#define _CLC_SHUFFLE_SET_8_ELEMENTS(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 74 | + _CLC_SHUFFLE_SET_4_ELEMENTS(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 75 | + ret_val.s4 = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.s4); \ |
| 76 | + ret_val.s5 = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.s5); \ |
| 77 | + ret_val.s6 = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.s6); \ |
| 78 | + ret_val.s7 = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.s7); |
| 79 | + |
| 80 | +#define _CLC_SHUFFLE_SET_16_ELEMENTS(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 81 | + _CLC_SHUFFLE_SET_8_ELEMENTS(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 82 | + ret_val.s8 = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.s8); \ |
| 83 | + ret_val.s9 = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.s9); \ |
| 84 | + ret_val.sA = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.sA); \ |
| 85 | + ret_val.sB = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.sB); \ |
| 86 | + ret_val.sC = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.sC); \ |
| 87 | + ret_val.sD = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.sD); \ |
| 88 | + ret_val.sE = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.sE); \ |
| 89 | + ret_val.sF = __clc_get_el_##ARGTYPE##ARGSIZE##_##MASKTYPE(x, mask.sF); |
| 90 | + |
| 91 | +#define _CLC_SHUFFLE_DEFINE2(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 92 | + _CLC_DEF _CLC_OVERLOAD ARGTYPE##2 __spirv_ocl_shuffle(ARGTYPE##ARGSIZE x, \ |
| 93 | + MASKTYPE##2 mask) { \ |
| 94 | + ARGTYPE##2 ret_val; \ |
| 95 | + mask &= (MASKTYPE##2)(ARGSIZE - 1); \ |
| 96 | + _CLC_SHUFFLE_SET_2_ELEMENTS(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 97 | + return ret_val; \ |
| 98 | + } |
| 99 | + |
| 100 | +#define _CLC_SHUFFLE_DEFINE4(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 101 | + _CLC_DEF _CLC_OVERLOAD ARGTYPE##4 __spirv_ocl_shuffle(ARGTYPE##ARGSIZE x, \ |
| 102 | + MASKTYPE##4 mask) { \ |
| 103 | + ARGTYPE##4 ret_val; \ |
| 104 | + mask &= (MASKTYPE##4)(ARGSIZE - 1); \ |
| 105 | + _CLC_SHUFFLE_SET_4_ELEMENTS(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 106 | + return ret_val; \ |
| 107 | + } |
| 108 | + |
| 109 | +#define _CLC_SHUFFLE_DEFINE8(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 110 | + _CLC_DEF _CLC_OVERLOAD ARGTYPE##8 __spirv_ocl_shuffle(ARGTYPE##ARGSIZE x, \ |
| 111 | + MASKTYPE##8 mask) { \ |
| 112 | + ARGTYPE##8 ret_val; \ |
| 113 | + mask &= (MASKTYPE##8)(ARGSIZE - 1); \ |
| 114 | + _CLC_SHUFFLE_SET_8_ELEMENTS(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 115 | + return ret_val; \ |
| 116 | + } |
| 117 | + |
| 118 | +#define _CLC_SHUFFLE_DEFINE16(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 119 | + _CLC_DEF _CLC_OVERLOAD ARGTYPE##16 __spirv_ocl_shuffle(ARGTYPE##ARGSIZE x, \ |
| 120 | + MASKTYPE##16 mask) { \ |
| 121 | + ARGTYPE##16 ret_val; \ |
| 122 | + mask &= (MASKTYPE##16)(ARGSIZE - 1); \ |
| 123 | + _CLC_SHUFFLE_SET_16_ELEMENTS(ARGTYPE, ARGSIZE, MASKTYPE) \ |
| 124 | + return ret_val; \ |
| 125 | + } |
| 126 | + |
| 127 | +#define _CLC_VECTOR_SHUFFLE_MASKSIZE(INTYPE, ARGSIZE, MASKTYPE) \ |
| 128 | + _CLC_GET_ELEMENT_DEFINE(INTYPE, ARGSIZE, MASKTYPE) \ |
| 129 | + _CLC_SHUFFLE_DEFINE2(INTYPE, ARGSIZE, MASKTYPE) \ |
| 130 | + _CLC_SHUFFLE_DEFINE4(INTYPE, ARGSIZE, MASKTYPE) \ |
| 131 | + _CLC_SHUFFLE_DEFINE8(INTYPE, ARGSIZE, MASKTYPE) \ |
| 132 | + _CLC_SHUFFLE_DEFINE16(INTYPE, ARGSIZE, MASKTYPE) |
| 133 | + |
| 134 | +#define _CLC_VECTOR_SHUFFLE_INSIZE(TYPE, MASKTYPE) \ |
| 135 | + _CLC_VECTOR_SHUFFLE_MASKSIZE(TYPE, 2, MASKTYPE) \ |
| 136 | + _CLC_VECTOR_SHUFFLE_MASKSIZE(TYPE, 4, MASKTYPE) \ |
| 137 | + _CLC_VECTOR_SHUFFLE_MASKSIZE(TYPE, 8, MASKTYPE) \ |
| 138 | + _CLC_VECTOR_SHUFFLE_MASKSIZE(TYPE, 16, MASKTYPE) |
| 139 | + |
| 140 | +_CLC_VECTOR_SHUFFLE_INSIZE(char, uchar) |
| 141 | +_CLC_VECTOR_SHUFFLE_INSIZE(schar, uchar) |
| 142 | +_CLC_VECTOR_SHUFFLE_INSIZE(short, ushort) |
| 143 | +_CLC_VECTOR_SHUFFLE_INSIZE(int, uint) |
| 144 | +_CLC_VECTOR_SHUFFLE_INSIZE(long, ulong) |
| 145 | +_CLC_VECTOR_SHUFFLE_INSIZE(uchar, uchar) |
| 146 | +_CLC_VECTOR_SHUFFLE_INSIZE(ushort, ushort) |
| 147 | +_CLC_VECTOR_SHUFFLE_INSIZE(uint, uint) |
| 148 | +_CLC_VECTOR_SHUFFLE_INSIZE(ulong, ulong) |
| 149 | +_CLC_VECTOR_SHUFFLE_INSIZE(float, uint) |
| 150 | +#ifdef cl_khr_fp64 |
| 151 | +#pragma OPENCL EXTENSION cl_khr_fp64 : enable |
| 152 | +_CLC_VECTOR_SHUFFLE_INSIZE(double, ulong) |
| 153 | +#endif |
| 154 | +#ifdef cl_khr_fp16 |
| 155 | +#pragma OPENCL EXTENSION cl_khr_fp16 : enable |
| 156 | +_CLC_VECTOR_SHUFFLE_INSIZE(half, ushort) |
| 157 | +#endif |
| 158 | + |
| 159 | +#undef _CLC_ELEMENT_CASES2 |
| 160 | +#undef _CLC_ELEMENT_CASES4 |
| 161 | +#undef _CLC_ELEMENT_CASES8 |
| 162 | +#undef _CLC_ELEMENT_CASES16 |
| 163 | +#undef _CLC_GET_ELEMENT_DEFINE |
| 164 | +#undef _CLC_SHUFFLE_SET_ONE_ELEMENT |
| 165 | +#undef _CLC_SHUFFLE_SET_2_ELEMENTS |
| 166 | +#undef _CLC_SHUFFLE_SET_4_ELEMENTS |
| 167 | +#undef _CLC_SHUFFLE_SET_8_ELEMENTS |
| 168 | +#undef _CLC_SHUFFLE_SET_16_ELEMENTS |
| 169 | +#undef _CLC_SHUFFLE_DEFINE2 |
| 170 | +#undef _CLC_SHUFFLE_DEFINE4 |
| 171 | +#undef _CLC_SHUFFLE_DEFINE8 |
| 172 | +#undef _CLC_SHUFFLE_DEFINE16 |
| 173 | +#undef _CLC_VECTOR_SHUFFLE_MASKSIZE |
| 174 | +#undef _CLC_VECTOR_SHUFFLE_INSIZE |
0 commit comments