| 1 |
#pragma clang diagnostic ignored "-Wmissing-prototypes" |
| 2 |
|
| 3 |
#include <metal_stdlib> |
| 4 |
#include <simd/simd.h> |
| 5 |
|
| 6 |
using namespace metal; |
| 7 |
|
| 8 |
struct main0_out |
| 9 |
{ |
| 10 |
float FragColor [[color(0)]]; |
| 11 |
}; |
| 12 |
|
| 13 |
template<typename T> |
| 14 |
inline T spvSubgroupBroadcast(T value, ushort lane) |
| 15 |
{ |
| 16 |
return simd_broadcast(value, lane); |
| 17 |
} |
| 18 |
|
| 19 |
template<> |
| 20 |
inline bool spvSubgroupBroadcast(bool value, ushort lane) |
| 21 |
{ |
| 22 |
return !!simd_broadcast((ushort)value, lane); |
| 23 |
} |
| 24 |
|
| 25 |
template<uint N> |
| 26 |
inline vec<bool, N> spvSubgroupBroadcast(vec<bool, N> value, ushort lane) |
| 27 |
{ |
| 28 |
return (vec<bool, N>)simd_broadcast((vec<ushort, N>)value, lane); |
| 29 |
} |
| 30 |
|
| 31 |
template<typename T> |
| 32 |
inline T spvSubgroupBroadcastFirst(T value) |
| 33 |
{ |
| 34 |
return simd_broadcast_first(value); |
| 35 |
} |
| 36 |
|
| 37 |
template<> |
| 38 |
inline bool spvSubgroupBroadcastFirst(bool value) |
| 39 |
{ |
| 40 |
return !!simd_broadcast_first((ushort)value); |
| 41 |
} |
| 42 |
|
| 43 |
template<uint N> |
| 44 |
inline vec<bool, N> spvSubgroupBroadcastFirst(vec<bool, N> value) |
| 45 |
{ |
| 46 |
return (vec<bool, N>)simd_broadcast_first((vec<ushort, N>)value); |
| 47 |
} |
| 48 |
|
| 49 |
inline uint4 spvSubgroupBallot(bool value) |
| 50 |
{ |
| 51 |
simd_vote vote = simd_ballot(value); |
| 52 |
// simd_ballot() returns a 64-bit integer-like object, but |
| 53 |
// SPIR-V callers expect a uint4. We must convert. |
| 54 |
// FIXME: This won't include higher bits if Apple ever supports |
| 55 |
// 128 lanes in an SIMD-group. |
| 56 |
return uint4((uint)((simd_vote::vote_t)vote & 0xFFFFFFFF), (uint)(((simd_vote::vote_t)vote >> 32) & 0xFFFFFFFF), 0, 0); |
| 57 |
} |
| 58 |
|
| 59 |
inline bool spvSubgroupBallotBitExtract(uint4 ballot, uint bit) |
| 60 |
{ |
| 61 |
return !!extract_bits(ballot[bit / 32], bit % 32, 1); |
| 62 |
} |
| 63 |
|
| 64 |
inline uint spvSubgroupBallotFindLSB(uint4 ballot, uint gl_SubgroupSize) |
| 65 |
{ |
| 66 |
uint4 mask = uint4(extract_bits(0xFFFFFFFF, 0, min(gl_SubgroupSize, 32u)), extract_bits(0xFFFFFFFF, 0, (uint)max((int)gl_SubgroupSize - 32, 0)), uint2(0)); |
| 67 |
ballot &= mask; |
| 68 |
return select(ctz(ballot.x), select(32 + ctz(ballot.y), select(64 + ctz(ballot.z), select(96 + ctz(ballot.w), uint(-1), ballot.w == 0), ballot.z == 0), ballot.y == 0), ballot.x == 0); |
| 69 |
} |
| 70 |
|
| 71 |
inline uint spvSubgroupBallotFindMSB(uint4 ballot, uint gl_SubgroupSize) |
| 72 |
{ |
| 73 |
uint4 mask = uint4(extract_bits(0xFFFFFFFF, 0, min(gl_SubgroupSize, 32u)), extract_bits(0xFFFFFFFF, 0, (uint)max((int)gl_SubgroupSize - 32, 0)), uint2(0)); |
| 74 |
ballot &= mask; |
| 75 |
return select(128 - (clz(ballot.w) + 1), select(96 - (clz(ballot.z) + 1), select(64 - (clz(ballot.y) + 1), select(32 - (clz(ballot.x) + 1), uint(-1), ballot.x == 0), ballot.y == 0), ballot.z == 0), ballot.w == 0); |
| 76 |
} |
| 77 |
|
| 78 |
inline uint spvPopCount4(uint4 ballot) |
| 79 |
{ |
| 80 |
return popcount(ballot.x) + popcount(ballot.y) + popcount(ballot.z) + popcount(ballot.w); |
| 81 |
} |
| 82 |
|
| 83 |
inline uint spvSubgroupBallotBitCount(uint4 ballot, uint gl_SubgroupSize) |
| 84 |
{ |
| 85 |
uint4 mask = uint4(extract_bits(0xFFFFFFFF, 0, min(gl_SubgroupSize, 32u)), extract_bits(0xFFFFFFFF, 0, (uint)max((int)gl_SubgroupSize - 32, 0)), uint2(0)); |
| 86 |
return spvPopCount4(ballot & mask); |
| 87 |
} |
| 88 |
|
| 89 |
inline uint spvSubgroupBallotInclusiveBitCount(uint4 ballot, uint gl_SubgroupInvocationID) |
| 90 |
{ |
| 91 |
uint4 mask = uint4(extract_bits(0xFFFFFFFF, 0, min(gl_SubgroupInvocationID + 1, 32u)), extract_bits(0xFFFFFFFF, 0, (uint)max((int)gl_SubgroupInvocationID + 1 - 32, 0)), uint2(0)); |
| 92 |
return spvPopCount4(ballot & mask); |
| 93 |
} |
| 94 |
|
| 95 |
inline uint spvSubgroupBallotExclusiveBitCount(uint4 ballot, uint gl_SubgroupInvocationID) |
| 96 |
{ |
| 97 |
uint4 mask = uint4(extract_bits(0xFFFFFFFF, 0, min(gl_SubgroupInvocationID, 32u)), extract_bits(0xFFFFFFFF, 0, (uint)max((int)gl_SubgroupInvocationID - 32, 0)), uint2(0)); |
| 98 |
return spvPopCount4(ballot & mask); |
| 99 |
} |
| 100 |
|
| 101 |
template<typename T> |
| 102 |
inline bool spvSubgroupAllEqual(T value) |
| 103 |
{ |
| 104 |
return simd_all(all(value == simd_broadcast_first(value))); |
| 105 |
} |
| 106 |
|
| 107 |
template<> |
| 108 |
inline bool spvSubgroupAllEqual(bool value) |
| 109 |
{ |
| 110 |
return simd_all(value) || !simd_any(value); |
| 111 |
} |
| 112 |
|
| 113 |
template<uint N> |
| 114 |
inline bool spvSubgroupAllEqual(vec<bool, N> value) |
| 115 |
{ |
| 116 |
return simd_all(all(value == (vec<bool, N>)simd_broadcast_first((vec<ushort, N>)value))); |
| 117 |
} |
| 118 |
|
| 119 |
template<typename T> |
| 120 |
inline T spvSubgroupShuffle(T value, ushort lane) |
| 121 |
{ |
| 122 |
return simd_shuffle(value, lane); |
| 123 |
} |
| 124 |
|
| 125 |
template<> |
| 126 |
inline bool spvSubgroupShuffle(bool value, ushort lane) |
| 127 |
{ |
| 128 |
return !!simd_shuffle((ushort)value, lane); |
| 129 |
} |
| 130 |
|
| 131 |
template<uint N> |
| 132 |
inline vec<bool, N> spvSubgroupShuffle(vec<bool, N> value, ushort lane) |
| 133 |
{ |
| 134 |
return (vec<bool, N>)simd_shuffle((vec<ushort, N>)value, lane); |
| 135 |
} |
| 136 |
|
| 137 |
template<typename T> |
| 138 |
inline T spvSubgroupShuffleXor(T value, ushort mask) |
| 139 |
{ |
| 140 |
return simd_shuffle_xor(value, mask); |
| 141 |
} |
| 142 |
|
| 143 |
template<> |
| 144 |
inline bool spvSubgroupShuffleXor(bool value, ushort mask) |
| 145 |
{ |
| 146 |
return !!simd_shuffle_xor((ushort)value, mask); |
| 147 |
} |
| 148 |
|
| 149 |
template<uint N> |
| 150 |
inline vec<bool, N> spvSubgroupShuffleXor(vec<bool, N> value, ushort mask) |
| 151 |
{ |
| 152 |
return (vec<bool, N>)simd_shuffle_xor((vec<ushort, N>)value, mask); |
| 153 |
} |
| 154 |
|
| 155 |
template<typename T> |
| 156 |
inline T spvSubgroupShuffleUp(T value, ushort delta) |
| 157 |
{ |
| 158 |
return simd_shuffle_up(value, delta); |
| 159 |
} |
| 160 |
|
| 161 |
template<> |
| 162 |
inline bool spvSubgroupShuffleUp(bool value, ushort delta) |
| 163 |
{ |
| 164 |
return !!simd_shuffle_up((ushort)value, delta); |
| 165 |
} |
| 166 |
|
| 167 |
template<uint N> |
| 168 |
inline vec<bool, N> spvSubgroupShuffleUp(vec<bool, N> value, ushort delta) |
| 169 |
{ |
| 170 |
return (vec<bool, N>)simd_shuffle_up((vec<ushort, N>)value, delta); |
| 171 |
} |
| 172 |
|
| 173 |
template<typename T> |
| 174 |
inline T spvSubgroupShuffleDown(T value, ushort delta) |
| 175 |
{ |
| 176 |
return simd_shuffle_down(value, delta); |
| 177 |
} |
| 178 |
|
| 179 |
template<> |
| 180 |
inline bool spvSubgroupShuffleDown(bool value, ushort delta) |
| 181 |
{ |
| 182 |
return !!simd_shuffle_down((ushort)value, delta); |
| 183 |
} |
| 184 |
|
| 185 |
template<uint N> |
| 186 |
inline vec<bool, N> spvSubgroupShuffleDown(vec<bool, N> value, ushort delta) |
| 187 |
{ |
| 188 |
return (vec<bool, N>)simd_shuffle_down((vec<ushort, N>)value, delta); |
| 189 |
} |
| 190 |
|
| 191 |
template<typename T> |
| 192 |
inline T spvQuadBroadcast(T value, uint lane) |
| 193 |
{ |
| 194 |
return quad_broadcast(value, lane); |
| 195 |
} |
| 196 |
|
| 197 |
template<> |
| 198 |
inline bool spvQuadBroadcast(bool value, uint lane) |
| 199 |
{ |
| 200 |
return !!quad_broadcast((ushort)value, lane); |
| 201 |
} |
| 202 |
|
| 203 |
template<uint N> |
| 204 |
inline vec<bool, N> spvQuadBroadcast(vec<bool, N> value, uint lane) |
| 205 |
{ |
| 206 |
return (vec<bool, N>)quad_broadcast((vec<ushort, N>)value, lane); |
| 207 |
} |
| 208 |
|
| 209 |
template<typename T> |
| 210 |
inline T spvQuadSwap(T value, uint dir) |
| 211 |
{ |
| 212 |
return quad_shuffle_xor(value, dir + 1); |
| 213 |
} |
| 214 |
|
| 215 |
template<> |
| 216 |
inline bool spvQuadSwap(bool value, uint dir) |
| 217 |
{ |
| 218 |
return !!quad_shuffle_xor((ushort)value, dir + 1); |
| 219 |
} |
| 220 |
|
| 221 |
template<uint N> |
| 222 |
inline vec<bool, N> spvQuadSwap(vec<bool, N> value, uint dir) |
| 223 |
{ |
| 224 |
return (vec<bool, N>)quad_shuffle_xor((vec<ushort, N>)value, dir + 1); |
| 225 |
} |
| 226 |
|
| 227 |
fragment main0_out main0() |
| 228 |
{ |
| 229 |
main0_out out = {}; |
| 230 |
uint gl_SubgroupSize = simd_sum(1); |
| 231 |
uint gl_SubgroupInvocationID = simd_prefix_exclusive_sum(1); |
| 232 |
uint4 gl_SubgroupEqMask = gl_SubgroupInvocationID >= 32 ? uint4(0, (1 << (gl_SubgroupInvocationID - 32)), uint2(0)) : uint4(1 << gl_SubgroupInvocationID, uint3(0)); |
| 233 |
uint4 gl_SubgroupGeMask = uint4(insert_bits(0u, 0xFFFFFFFF, min(gl_SubgroupInvocationID, 32u), (uint)max(min((int)gl_SubgroupSize, 32) - (int)gl_SubgroupInvocationID, 0)), insert_bits(0u, 0xFFFFFFFF, (uint)max((int)gl_SubgroupInvocationID - 32, 0), (uint)max((int)gl_SubgroupSize - (int)max(gl_SubgroupInvocationID, 32u), 0)), uint2(0)); |
| 234 |
uint4 gl_SubgroupGtMask = uint4(insert_bits(0u, 0xFFFFFFFF, min(gl_SubgroupInvocationID + 1, 32u), (uint)max(min((int)gl_SubgroupSize, 32) - (int)gl_SubgroupInvocationID - 1, 0)), insert_bits(0u, 0xFFFFFFFF, (uint)max((int)gl_SubgroupInvocationID + 1 - 32, 0), (uint)max((int)gl_SubgroupSize - (int)max(gl_SubgroupInvocationID + 1, 32u), 0)), uint2(0)); |
| 235 |
uint4 gl_SubgroupLeMask = uint4(extract_bits(0xFFFFFFFF, 0, min(gl_SubgroupInvocationID + 1, 32u)), extract_bits(0xFFFFFFFF, 0, (uint)max((int)gl_SubgroupInvocationID + 1 - 32, 0)), uint2(0)); |
| 236 |
uint4 gl_SubgroupLtMask = uint4(extract_bits(0xFFFFFFFF, 0, min(gl_SubgroupInvocationID, 32u)), extract_bits(0xFFFFFFFF, 0, (uint)max((int)gl_SubgroupInvocationID - 32, 0)), uint2(0)); |
| 237 |
out.FragColor = float(gl_SubgroupSize); |
| 238 |
out.FragColor = float(gl_SubgroupInvocationID); |
| 239 |
bool elected = simd_is_first(); |
| 240 |
out.FragColor = float4(gl_SubgroupEqMask).x; |
| 241 |
out.FragColor = float4(gl_SubgroupGeMask).x; |
| 242 |
out.FragColor = float4(gl_SubgroupGtMask).x; |
| 243 |
out.FragColor = float4(gl_SubgroupLeMask).x; |
| 244 |
out.FragColor = float4(gl_SubgroupLtMask).x; |
| 245 |
float4 broadcasted = spvSubgroupBroadcast(float4(10.0), 8u); |
| 246 |
bool2 broadcasted_bool = spvSubgroupBroadcast(bool2(true), 8u); |
| 247 |
float3 first = spvSubgroupBroadcastFirst(float3(20.0)); |
| 248 |
bool4 first_bool = spvSubgroupBroadcastFirst(bool4(false)); |
| 249 |
uint4 ballot_value = spvSubgroupBallot(true); |
| 250 |
bool inverse_ballot_value = spvSubgroupBallotBitExtract(ballot_value, gl_SubgroupInvocationID); |
| 251 |
bool bit_extracted = spvSubgroupBallotBitExtract(uint4(10u), 8u); |
| 252 |
uint bit_count = spvSubgroupBallotBitCount(ballot_value, gl_SubgroupSize); |
| 253 |
uint inclusive_bit_count = spvSubgroupBallotInclusiveBitCount(ballot_value, gl_SubgroupInvocationID); |
| 254 |
uint exclusive_bit_count = spvSubgroupBallotExclusiveBitCount(ballot_value, gl_SubgroupInvocationID); |
| 255 |
uint lsb = spvSubgroupBallotFindLSB(ballot_value, gl_SubgroupSize); |
| 256 |
uint msb = spvSubgroupBallotFindMSB(ballot_value, gl_SubgroupSize); |
| 257 |
uint shuffled = spvSubgroupShuffle(10u, 8u); |
| 258 |
bool shuffled_bool = spvSubgroupShuffle(true, 9u); |
| 259 |
uint shuffled_xor = spvSubgroupShuffleXor(30u, 8u); |
| 260 |
bool shuffled_xor_bool = spvSubgroupShuffleXor(false, 9u); |
| 261 |
uint shuffled_up = spvSubgroupShuffleUp(20u, 4u); |
| 262 |
bool shuffled_up_bool = spvSubgroupShuffleUp(true, 4u); |
| 263 |
uint shuffled_down = spvSubgroupShuffleDown(20u, 4u); |
| 264 |
bool shuffled_down_bool = spvSubgroupShuffleDown(false, 4u); |
| 265 |
bool has_all = simd_all(true); |
| 266 |
bool has_any = simd_any(true); |
| 267 |
bool has_equal = spvSubgroupAllEqual(0); |
| 268 |
has_equal = spvSubgroupAllEqual(true); |
| 269 |
has_equal = spvSubgroupAllEqual(float3(0.0, 1.0, 2.0)); |
| 270 |
has_equal = spvSubgroupAllEqual(bool4(true, true, false, true)); |
| 271 |
float4 added = simd_sum(float4(20.0)); |
| 272 |
int4 iadded = simd_sum(int4(20)); |
| 273 |
float4 multiplied = simd_product(float4(20.0)); |
| 274 |
int4 imultiplied = simd_product(int4(20)); |
| 275 |
float4 lo = simd_min(float4(20.0)); |
| 276 |
float4 hi = simd_max(float4(20.0)); |
| 277 |
int4 slo = simd_min(int4(20)); |
| 278 |
int4 shi = simd_max(int4(20)); |
| 279 |
uint4 ulo = simd_min(uint4(20u)); |
| 280 |
uint4 uhi = simd_max(uint4(20u)); |
| 281 |
uint4 anded = simd_and(ballot_value); |
| 282 |
uint4 ored = simd_or(ballot_value); |
| 283 |
uint4 xored = simd_xor(ballot_value); |
| 284 |
added = simd_prefix_inclusive_sum(added); |
| 285 |
iadded = simd_prefix_inclusive_sum(iadded); |
| 286 |
multiplied = simd_prefix_inclusive_product(multiplied); |
| 287 |
imultiplied = simd_prefix_inclusive_product(imultiplied); |
| 288 |
added = simd_prefix_exclusive_sum(multiplied); |
| 289 |
multiplied = simd_prefix_exclusive_product(multiplied); |
| 290 |
iadded = simd_prefix_exclusive_sum(imultiplied); |
| 291 |
imultiplied = simd_prefix_exclusive_product(imultiplied); |
| 292 |
added = quad_sum(added); |
| 293 |
multiplied = quad_product(multiplied); |
| 294 |
iadded = quad_sum(iadded); |
| 295 |
imultiplied = quad_product(imultiplied); |
| 296 |
lo = quad_min(lo); |
| 297 |
hi = quad_max(hi); |
| 298 |
ulo = quad_min(ulo); |
| 299 |
uhi = quad_max(uhi); |
| 300 |
slo = quad_min(slo); |
| 301 |
shi = quad_max(shi); |
| 302 |
anded = quad_and(anded); |
| 303 |
ored = quad_or(ored); |
| 304 |
xored = quad_xor(xored); |
| 305 |
float4 swap_horiz = spvQuadSwap(float4(20.0), 0u); |
| 306 |
bool4 swap_horiz_bool = spvQuadSwap(bool4(true), 0u); |
| 307 |
float4 swap_vertical = spvQuadSwap(float4(20.0), 1u); |
| 308 |
bool4 swap_vertical_bool = spvQuadSwap(bool4(true), 1u); |
| 309 |
float4 swap_diagonal = spvQuadSwap(float4(20.0), 2u); |
| 310 |
bool4 swap_diagonal_bool = spvQuadSwap(bool4(true), 2u); |
| 311 |
float4 quad_broadcast0 = spvQuadBroadcast(float4(20.0), 3u); |
| 312 |
bool4 quad_broadcast_bool = spvQuadBroadcast(bool4(true), 3u); |
| 313 |
return out; |
| 314 |
} |
| 315 |
|