|
| 1 | +#--- source.hlsl |
| 2 | +RWStructuredBuffer<uint> _participant_bit : register(u0); |
| 3 | +RWStructuredBuffer<uint> _wave_op_index : register(u1); |
| 4 | + |
| 5 | +[numthreads(4, 1, 1)] |
| 6 | +void main(uint3 tid : SV_DispatchThreadID) { |
| 7 | + uint result = 0; |
| 8 | + switch ((WaveGetLaneIndex() % 4)) { |
| 9 | + case 0: { |
| 10 | + if ((WaveGetLaneIndex() < 8)) { |
| 11 | + result = (result + WaveActiveSum(1)); |
| 12 | + uint temp = 0; |
| 13 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 14 | + _participant_bit[temp] = (9 << 6); |
| 15 | + uint4 ballot = WaveActiveBallot(1); |
| 16 | + _participant_bit[(temp + 1)] = ballot.x; |
| 17 | + _participant_bit[(temp + 2)] = ballot.y; |
| 18 | + } |
| 19 | + } |
| 20 | + case 1: { |
| 21 | + if (((WaveGetLaneIndex() % 2) == 0)) { |
| 22 | + result = (result + WaveActiveSum(2)); |
| 23 | + uint temp = 0; |
| 24 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 25 | + _participant_bit[temp] = (18 << 6); |
| 26 | + uint4 ballot = WaveActiveBallot(1); |
| 27 | + _participant_bit[(temp + 1)] = ballot.x; |
| 28 | + _participant_bit[(temp + 2)] = ballot.y; |
| 29 | + } |
| 30 | + break; |
| 31 | + } |
| 32 | + case 2: { |
| 33 | + for (uint i0 = 0; (i0 < 3); i0 = (i0 + 1)) { |
| 34 | + uint counter1 = 0; |
| 35 | + while ((counter1 < 3)) { |
| 36 | + counter1 = (counter1 + 1); |
| 37 | + switch ((WaveGetLaneIndex() % 3)) { |
| 38 | + case 0: { |
| 39 | + if ((WaveGetLaneIndex() < 8)) { |
| 40 | + result = (result + WaveActiveSum(1)); |
| 41 | + uint temp = 0; |
| 42 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 43 | + _participant_bit[temp] = (((43 << 6) | (i0 << 4)) | (counter1 << 2)); |
| 44 | + uint4 ballot = WaveActiveBallot(1); |
| 45 | + _participant_bit[(temp + 1)] = ballot.x; |
| 46 | + _participant_bit[(temp + 2)] = ballot.y; |
| 47 | + } |
| 48 | + break; |
| 49 | + } |
| 50 | + case 1: { |
| 51 | + if (((WaveGetLaneIndex() % 2) == 0)) { |
| 52 | + result = (result + WaveActiveSum(2)); |
| 53 | + uint temp = 0; |
| 54 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 55 | + _participant_bit[temp] = (((52 << 6) | (i0 << 4)) | (counter1 << 2)); |
| 56 | + uint4 ballot = WaveActiveBallot(1); |
| 57 | + _participant_bit[(temp + 1)] = ballot.x; |
| 58 | + _participant_bit[(temp + 2)] = ballot.y; |
| 59 | + } |
| 60 | + break; |
| 61 | + } |
| 62 | + case 2: { |
| 63 | + if (true) { |
| 64 | + result = (result + WaveActiveSum(3)); |
| 65 | + uint temp = 0; |
| 66 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 67 | + _participant_bit[temp] = (((57 << 6) | (i0 << 4)) | (counter1 << 2)); |
| 68 | + uint4 ballot = WaveActiveBallot(1); |
| 69 | + _participant_bit[(temp + 1)] = ballot.x; |
| 70 | + _participant_bit[(temp + 2)] = ballot.y; |
| 71 | + } |
| 72 | + break; |
| 73 | + } |
| 74 | + } |
| 75 | + if ((((WaveGetLaneIndex() == 0) || (WaveGetLaneIndex() == 2)) || (WaveGetLaneIndex() == 3))) { |
| 76 | + result = (result + WaveActiveMax((WaveGetLaneIndex() + 4))); |
| 77 | + uint temp = 0; |
| 78 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 79 | + _participant_bit[temp] = (((74 << 6) | (i0 << 4)) | (counter1 << 2)); |
| 80 | + uint4 ballot = WaveActiveBallot(1); |
| 81 | + _participant_bit[(temp + 1)] = ballot.x; |
| 82 | + _participant_bit[(temp + 2)] = ballot.y; |
| 83 | + } |
| 84 | + } |
| 85 | + if ((((WaveGetLaneIndex() == 0) || (WaveGetLaneIndex() == 2)) || (WaveGetLaneIndex() == 1))) { |
| 86 | + result = (result + WaveActiveMin(result)); |
| 87 | + uint temp = 0; |
| 88 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 89 | + _participant_bit[temp] = ((89 << 6) | (i0 << 4)); |
| 90 | + uint4 ballot = WaveActiveBallot(1); |
| 91 | + _participant_bit[(temp + 1)] = ballot.x; |
| 92 | + _participant_bit[(temp + 2)] = ballot.y; |
| 93 | + } |
| 94 | + if ((i0 == 2)) { |
| 95 | + break; |
| 96 | + } |
| 97 | + } |
| 98 | + } |
| 99 | + case 3: { |
| 100 | + uint counter2 = 0; |
| 101 | + while ((counter2 < 2)) { |
| 102 | + counter2 = (counter2 + 1); |
| 103 | + for (uint i3 = 0; (i3 < 3); i3 = (i3 + 1)) { |
| 104 | + switch ((WaveGetLaneIndex() % 4)) { |
| 105 | + case 0: { |
| 106 | + if ((WaveGetLaneIndex() < 8)) { |
| 107 | + result = (result + WaveActiveSum(1)); |
| 108 | + uint temp = 0; |
| 109 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 110 | + _participant_bit[temp] = (((117 << 6) | (counter2 << 4)) | (i3 << 2)); |
| 111 | + uint4 ballot = WaveActiveBallot(1); |
| 112 | + _participant_bit[(temp + 1)] = ballot.x; |
| 113 | + _participant_bit[(temp + 2)] = ballot.y; |
| 114 | + } |
| 115 | + break; |
| 116 | + } |
| 117 | + case 1: { |
| 118 | + if (((WaveGetLaneIndex() % 2) == 0)) { |
| 119 | + result = (result + WaveActiveSum(2)); |
| 120 | + uint temp = 0; |
| 121 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 122 | + _participant_bit[temp] = (((126 << 6) | (counter2 << 4)) | (i3 << 2)); |
| 123 | + uint4 ballot = WaveActiveBallot(1); |
| 124 | + _participant_bit[(temp + 1)] = ballot.x; |
| 125 | + _participant_bit[(temp + 2)] = ballot.y; |
| 126 | + } |
| 127 | + break; |
| 128 | + } |
| 129 | + case 2: { |
| 130 | + if (true) { |
| 131 | + result = (result + WaveActiveSum(3)); |
| 132 | + uint temp = 0; |
| 133 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 134 | + _participant_bit[temp] = (((131 << 6) | (counter2 << 4)) | (i3 << 2)); |
| 135 | + uint4 ballot = WaveActiveBallot(1); |
| 136 | + _participant_bit[(temp + 1)] = ballot.x; |
| 137 | + _participant_bit[(temp + 2)] = ballot.y; |
| 138 | + } |
| 139 | + break; |
| 140 | + } |
| 141 | + case 3: { |
| 142 | + if ((WaveGetLaneIndex() < 20)) { |
| 143 | + result = (result + WaveActiveSum(4)); |
| 144 | + uint temp = 0; |
| 145 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 146 | + _participant_bit[temp] = (((138 << 6) | (counter2 << 4)) | (i3 << 2)); |
| 147 | + uint4 ballot = WaveActiveBallot(1); |
| 148 | + _participant_bit[(temp + 1)] = ballot.x; |
| 149 | + _participant_bit[(temp + 2)] = ballot.y; |
| 150 | + } |
| 151 | + break; |
| 152 | + } |
| 153 | + default: { |
| 154 | + result = (result + WaveActiveSum(99)); |
| 155 | + uint temp = 0; |
| 156 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 157 | + _participant_bit[temp] = (((142 << 6) | (counter2 << 4)) | (i3 << 2)); |
| 158 | + uint4 ballot = WaveActiveBallot(1); |
| 159 | + _participant_bit[(temp + 1)] = ballot.x; |
| 160 | + _participant_bit[(temp + 2)] = ballot.y; |
| 161 | + break; |
| 162 | + } |
| 163 | + } |
| 164 | + if ((WaveGetLaneIndex() < 2)) { |
| 165 | + result = (result + WaveActiveMax(WaveGetLaneIndex())); |
| 166 | + uint temp = 0; |
| 167 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 168 | + _participant_bit[temp] = (((149 << 6) | (counter2 << 4)) | (i3 << 2)); |
| 169 | + uint4 ballot = WaveActiveBallot(1); |
| 170 | + _participant_bit[(temp + 1)] = ballot.x; |
| 171 | + _participant_bit[(temp + 2)] = ballot.y; |
| 172 | + } |
| 173 | + if ((i3 == 1)) { |
| 174 | + continue; |
| 175 | + } |
| 176 | + if ((i3 == 2)) { |
| 177 | + break; |
| 178 | + } |
| 179 | + } |
| 180 | + if ((WaveGetLaneIndex() < 2)) { |
| 181 | + result = (result + WaveActiveMin(result)); |
| 182 | + uint temp = 0; |
| 183 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 184 | + _participant_bit[temp] = ((162 << 6) | (counter2 << 4)); |
| 185 | + uint4 ballot = WaveActiveBallot(1); |
| 186 | + _participant_bit[(temp + 1)] = ballot.x; |
| 187 | + _participant_bit[(temp + 2)] = ballot.y; |
| 188 | + } |
| 189 | + if ((counter2 == 1)) { |
| 190 | + break; |
| 191 | + } |
| 192 | + } |
| 193 | + break; |
| 194 | + } |
| 195 | + } |
| 196 | + if (((WaveGetLaneIndex() == 0) || (WaveGetLaneIndex() == 3))) { |
| 197 | + if (((WaveGetLaneIndex() == 1) || (WaveGetLaneIndex() == 2))) { |
| 198 | + result = (result + WaveActiveMax(result)); |
| 199 | + uint temp = 0; |
| 200 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 201 | + _participant_bit[temp] = (183 << 6); |
| 202 | + uint4 ballot = WaveActiveBallot(1); |
| 203 | + _participant_bit[(temp + 1)] = ballot.x; |
| 204 | + _participant_bit[(temp + 2)] = ballot.y; |
| 205 | + } |
| 206 | + if (((WaveGetLaneIndex() == 1) || (WaveGetLaneIndex() == 3))) { |
| 207 | + result = (result + WaveActiveMax((WaveGetLaneIndex() + 4))); |
| 208 | + uint temp = 0; |
| 209 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 210 | + _participant_bit[temp] = (196 << 6); |
| 211 | + uint4 ballot = WaveActiveBallot(1); |
| 212 | + _participant_bit[(temp + 1)] = ballot.x; |
| 213 | + _participant_bit[(temp + 2)] = ballot.y; |
| 214 | + } |
| 215 | + } else { |
| 216 | + if ((((WaveGetLaneIndex() == 1) || (WaveGetLaneIndex() == 2)) || (WaveGetLaneIndex() == 1))) { |
| 217 | + result = (result + WaveActiveMin(result)); |
| 218 | + uint temp = 0; |
| 219 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 220 | + _participant_bit[temp] = (211 << 6); |
| 221 | + uint4 ballot = WaveActiveBallot(1); |
| 222 | + _participant_bit[(temp + 1)] = ballot.x; |
| 223 | + _participant_bit[(temp + 2)] = ballot.y; |
| 224 | + } |
| 225 | + for (uint i4 = 0; (i4 < 3); i4 = (i4 + 1)) { |
| 226 | + if (((WaveGetLaneIndex() < 1) || (WaveGetLaneIndex() >= 2))) { |
| 227 | + if (((WaveGetLaneIndex() < 1) || (WaveGetLaneIndex() >= 3))) { |
| 228 | + result = (result + WaveActiveMax(result)); |
| 229 | + uint temp = 0; |
| 230 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 231 | + _participant_bit[temp] = ((237 << 6) | (i4 << 4)); |
| 232 | + uint4 ballot = WaveActiveBallot(1); |
| 233 | + _participant_bit[(temp + 1)] = ballot.x; |
| 234 | + _participant_bit[(temp + 2)] = ballot.y; |
| 235 | + } |
| 236 | + if (((WaveGetLaneIndex() < 1) || (WaveGetLaneIndex() >= 2))) { |
| 237 | + result = (result + WaveActiveMax(result)); |
| 238 | + uint temp = 0; |
| 239 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 240 | + _participant_bit[temp] = ((248 << 6) | (i4 << 4)); |
| 241 | + uint4 ballot = WaveActiveBallot(1); |
| 242 | + _participant_bit[(temp + 1)] = ballot.x; |
| 243 | + _participant_bit[(temp + 2)] = ballot.y; |
| 244 | + } |
| 245 | + } |
| 246 | + if ((WaveGetLaneIndex() == 1)) { |
| 247 | + result = (result + WaveActiveMax(5)); |
| 248 | + uint temp = 0; |
| 249 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 250 | + _participant_bit[temp] = ((255 << 6) | (i4 << 4)); |
| 251 | + uint4 ballot = WaveActiveBallot(1); |
| 252 | + _participant_bit[(temp + 1)] = ballot.x; |
| 253 | + _participant_bit[(temp + 2)] = ballot.y; |
| 254 | + } |
| 255 | + if ((i4 == 1)) { |
| 256 | + continue; |
| 257 | + } |
| 258 | + } |
| 259 | + if ((((WaveGetLaneIndex() == 1) || (WaveGetLaneIndex() == 3)) || (WaveGetLaneIndex() == 3))) { |
| 260 | + result = (result + WaveActiveMin((WaveGetLaneIndex() + 4))); |
| 261 | + uint temp = 0; |
| 262 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 263 | + _participant_bit[temp] = (275 << 6); |
| 264 | + uint4 ballot = WaveActiveBallot(1); |
| 265 | + _participant_bit[(temp + 1)] = ballot.x; |
| 266 | + _participant_bit[(temp + 2)] = ballot.y; |
| 267 | + } |
| 268 | + } |
| 269 | + switch ((WaveGetLaneIndex() % 2)) { |
| 270 | + case 0: { |
| 271 | + if ((WaveGetLaneIndex() < 8)) { |
| 272 | + result = (result + WaveActiveSum(1)); |
| 273 | + uint temp = 0; |
| 274 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 275 | + _participant_bit[temp] = (285 << 6); |
| 276 | + uint4 ballot = WaveActiveBallot(1); |
| 277 | + _participant_bit[(temp + 1)] = ballot.x; |
| 278 | + _participant_bit[(temp + 2)] = ballot.y; |
| 279 | + } |
| 280 | + break; |
| 281 | + } |
| 282 | + case 1: { |
| 283 | + if (((WaveGetLaneIndex() % 2) == 0)) { |
| 284 | + result = (result + WaveActiveSum(2)); |
| 285 | + uint temp = 0; |
| 286 | + InterlockedAdd(_wave_op_index[0], 3, temp); |
| 287 | + _participant_bit[temp] = (294 << 6); |
| 288 | + uint4 ballot = WaveActiveBallot(1); |
| 289 | + _participant_bit[(temp + 1)] = ballot.x; |
| 290 | + _participant_bit[(temp + 2)] = ballot.y; |
| 291 | + } |
| 292 | + break; |
| 293 | + } |
| 294 | + } |
| 295 | +} |
| 296 | + |
| 297 | +#--- pipeline.yaml |
| 298 | +--- |
| 299 | +Shaders: |
| 300 | + - Stage: Compute |
| 301 | + Entry: main |
| 302 | + DispatchSize: [1, 1, 1] # Single dispatch for 4 threads |
| 303 | +Buffers: |
| 304 | + - Name: _participant_bit |
| 305 | + Format: UInt32 |
| 306 | + Stride: 4 |
| 307 | + Fill: 0 |
| 308 | + Size: 123 |
| 309 | + - Name: expected_bit_patterns |
| 310 | + Format: UInt32 |
| 311 | + Stride: 4 |
| 312 | + Data: [576, 1, 0, 1152, 1, 0, 3652, 4, 0, 3656, 4, 0, 3660, 4, 0, 3668, 4, 0, 3672, 4, 0, 3676, 4, 0, 3684, 4, 0, 3688, 4, 0, 3692, 4, 0, 4740, 4, 0, 4744, 4, 0, 4748, 4, 0, 4756, 4, 0, 4760, 4, 0, 4764, 4, 0, 4772, 4, 0, 4776, 4, 0, 4780, 4, 0, 5696, 4, 0, 5712, 4, 0, 5728, 4, 0, 8400, 4, 0, 8404, 4, 0, 8408, 4, 0, 8848, 8, 0, 8852, 8, 0, 8856, 8, 0, 12544, 8, 0, 13504, 6, 0, 13504, 6, 0, 15872, 4, 0, 15888, 4, 0, 15904, 4, 0, 16320, 2, 0, 16336, 2, 0, 16352, 2, 0, 17600, 2, 0, 18240, 5, 0, 18240, 5, 0] |
| 313 | + - Name: _wave_op_index |
| 314 | + Format: UInt32 |
| 315 | + Stride: 4 |
| 316 | + Data: [0] |
| 317 | +Results: |
| 318 | + - Result: BitTrackingValidation |
| 319 | + Rule: BufferParticipantPattern |
| 320 | + GroupSize: 3 |
| 321 | + Actual: _participant_bit |
| 322 | + Expected: expected_bit_patterns |
| 323 | +DescriptorSets: |
| 324 | + - Resources: |
| 325 | + - Name: _participant_bit |
| 326 | + Kind: RWStructuredBuffer |
| 327 | + DirectXBinding: |
| 328 | + Register: 0 |
| 329 | + Space: 0 |
| 330 | + VulkanBinding: |
| 331 | + Binding: 0 |
| 332 | + - Name: _wave_op_index |
| 333 | + Kind: RWStructuredBuffer |
| 334 | + DirectXBinding: |
| 335 | + Register: 1 |
| 336 | + Space: 0 |
| 337 | + VulkanBinding: |
| 338 | + Binding: 1 |
| 339 | +... |
| 340 | +#--- end |
| 341 | + |
| 342 | +# RUN: split-file %s %t |
| 343 | +# RUN: %dxc_target -T cs_6_0 -Fo %t.o %t/source.hlsl |
| 344 | +# RUN: %offloader %t/pipeline.yaml %t.o |
0 commit comments