vktSpvAsmInstructionTests.cpp revision 9a7e44c293be9b55281335c1a6532af9088c0d6d
1/*------------------------------------------------------------------------- 2 * Vulkan Conformance Tests 3 * ------------------------ 4 * 5 * Copyright (c) 2015 Google Inc. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a 8 * copy of this software and/or associated documentation files (the 9 * "Materials"), to deal in the Materials without restriction, including 10 * without limitation the rights to use, copy, modify, merge, publish, 11 * distribute, sublicense, and/or sell copies of the Materials, and to 12 * permit persons to whom the Materials are furnished to do so, subject to 13 * the following conditions: 14 * 15 * The above copyright notice(s) and this permission notice shall be 16 * included in all copies or substantial portions of the Materials. 17 * 18 * The Materials are Confidential Information as defined by the 19 * Khronos Membership Agreement until designated non-confidential by 20 * Khronos, at which point this condition clause shall be removed. 21 * 22 * THE MATERIALS ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 23 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 24 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 25 * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY 26 * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 27 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 28 * MATERIALS OR THE USE OR OTHER DEALINGS IN THE MATERIALS. 29 * 30 *//*! 31 * \file 32 * \brief SPIR-V Assembly Tests for Instructions (special opcode/operand) 33 *//*--------------------------------------------------------------------*/ 34 35#include "vktSpvAsmInstructionTests.hpp" 36 37#include "tcuCommandLine.hpp" 38#include "tcuFormatUtil.hpp" 39#include "tcuRGBA.hpp" 40#include "tcuStringTemplate.hpp" 41#include "tcuTestLog.hpp" 42#include "tcuVectorUtil.hpp" 43 44#include "vkDefs.hpp" 45#include "vkDeviceUtil.hpp" 46#include "vkMemUtil.hpp" 47#include "vkPlatform.hpp" 48#include "vkPrograms.hpp" 49#include "vkQueryUtil.hpp" 50#include "vkRef.hpp" 51#include "vkRefUtil.hpp" 52#include "vkStrUtil.hpp" 53#include "vkTypeUtil.hpp" 54 55#include "deRandom.hpp" 56#include "deStringUtil.hpp" 57#include "deUniquePtr.hpp" 58#include "tcuStringTemplate.hpp" 59 60#include "vktSpvAsmComputeShaderCase.hpp" 61#include "vktSpvAsmComputeShaderTestUtil.hpp" 62#include "vktTestCaseUtil.hpp" 63 64#include <map> 65#include <string> 66#include <sstream> 67 68namespace vkt 69{ 70namespace SpirVAssembly 71{ 72 73namespace 74{ 75 76using namespace vk; 77using std::map; 78using std::string; 79using std::vector; 80using tcu::IVec3; 81using tcu::IVec4; 82using tcu::RGBA; 83using tcu::TestLog; 84using tcu::TestStatus; 85using tcu::Vec4; 86using de::UniquePtr; 87using tcu::StringTemplate; 88 89typedef Unique<VkShaderModule> ModuleHandleUp; 90typedef de::SharedPtr<ModuleHandleUp> ModuleHandleSp; 91typedef Unique<VkShader> VkShaderUp; 92typedef de::SharedPtr<VkShaderUp> VkShaderSp; 93 94template<typename T> T randomScalar (de::Random& rnd, T minValue, T maxValue); 95template<> inline float randomScalar (de::Random& rnd, float minValue, float maxValue) { return rnd.getFloat(minValue, maxValue); } 96template<> inline deInt32 randomScalar (de::Random& rnd, deInt32 minValue, deInt32 maxValue) { return rnd.getInt(minValue, maxValue); } 97template<> inline deUint32 randomScalar (de::Random& rnd, deUint32 minValue, deUint32 maxValue) { return minValue + rnd.getUint32() % (maxValue - minValue + 1); } 98 99template<typename T> 100static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0) 101{ 102 T* const typedPtr = (T*)dst; 103 for (int ndx = 0; ndx < numValues; ndx++) 104 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue); 105} 106 107// \todo [2015-11-19 antiagainst] New rules for Vulkan pipeline interface requires that 108// all BuiltIn variables have to all be members in a block (struct with Block decoration). 109 110// Assembly code used for testing OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code: 111// 112// #version 430 113// 114// layout(std140, set = 0, binding = 0) readonly buffer Input { 115// float elements[]; 116// } input_data; 117// layout(std140, set = 0, binding = 1) writeonly buffer Output { 118// float elements[]; 119// } output_data; 120// 121// layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in; 122// 123// void main() { 124// uint x = gl_GlobalInvocationID.x; 125// output_data.elements[x] = -input_data.elements[x]; 126// } 127 128static const char* const s_ShaderPreamble = 129 "OpCapability Shader\n" 130 "OpMemoryModel Logical GLSL450\n" 131 "OpEntryPoint GLCompute %main \"main\" %id\n" 132 "OpExecutionMode %main LocalSize 1 1 1\n"; 133 134static const char* const s_CommonTypes = 135 "%bool = OpTypeBool\n" 136 "%void = OpTypeVoid\n" 137 "%voidf = OpTypeFunction %void\n" 138 "%u32 = OpTypeInt 32 0\n" 139 "%i32 = OpTypeInt 32 1\n" 140 "%f32 = OpTypeFloat 32\n" 141 "%uvec3 = OpTypeVector %u32 3\n" 142 "%uvec3ptr = OpTypePointer Input %uvec3\n" 143 "%f32ptr = OpTypePointer Uniform %f32\n" 144 "%f32arr = OpTypeRuntimeArray %f32\n"; 145 146// Declares two uniform variables (indata, outdata) of type "struct { float[] }". Depends on type "f32arr" (for "float[]"). 147static const char* const s_InputOutputBuffer = 148 "%inbuf = OpTypeStruct %f32arr\n" 149 "%inbufptr = OpTypePointer Uniform %inbuf\n" 150 "%indata = OpVariable %inbufptr Uniform\n" 151 "%outbuf = OpTypeStruct %f32arr\n" 152 "%outbufptr = OpTypePointer Uniform %outbuf\n" 153 "%outdata = OpVariable %outbufptr Uniform\n"; 154 155// Declares buffer type and layout for uniform variables indata and outdata. Both of them are SSBO bounded to descriptor set 0. 156// indata is at binding point 0, while outdata is at 1. 157static const char* const s_InputOutputBufferTraits = 158 "OpDecorate %inbuf BufferBlock\n" 159 "OpDecorate %indata DescriptorSet 0\n" 160 "OpDecorate %indata Binding 0\n" 161 "OpDecorate %outbuf BufferBlock\n" 162 "OpDecorate %outdata DescriptorSet 0\n" 163 "OpDecorate %outdata Binding 1\n" 164 "OpDecorate %f32arr ArrayStride 4\n" 165 "OpMemberDecorate %inbuf 0 Offset 0\n" 166 "OpMemberDecorate %outbuf 0 Offset 0\n"; 167 168tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx) 169{ 170 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction")); 171 ComputeShaderSpec spec; 172 de::Random rnd (deStringHash(group->getName())); 173 const int numElements = 100; 174 vector<float> positiveFloats (numElements, 0); 175 vector<float> negativeFloats (numElements, 0); 176 177 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements); 178 179 for (size_t ndx = 0; ndx < numElements; ++ndx) 180 negativeFloats[ndx] = -positiveFloats[ndx]; 181 182 spec.assembly = 183 "OpNop\n" // As the first instruction 184 185 + string(s_ShaderPreamble) + 186 187 "OpNop\n" // After OpEntryPoint but before any type definitions 188 189 "OpSource GLSL 430\n" 190 "OpName %main \"main\"\n" 191 "OpName %id \"gl_GlobalInvocationID\"\n" 192 193 "OpDecorate %id BuiltIn GlobalInvocationId\n" 194 195 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + 196 197 "OpNop\n" // In the middle of type definitions 198 199 + string(s_InputOutputBuffer) + 200 201 "%id = OpVariable %uvec3ptr Input\n" 202 "%zero = OpConstant %i32 0\n" 203 204 "%main = OpFunction %void None %voidf\n" 205 "%label = OpLabel\n" 206 "%idval = OpLoad %uvec3 %id\n" 207 "%x = OpCompositeExtract %u32 %idval 0\n" 208 209 " OpNop\n" // Inside a function body 210 211 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 212 "%inval = OpLoad %f32 %inloc\n" 213 "%neg = OpFNegate %f32 %inval\n" 214 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 215 " OpStore %outloc %neg\n" 216 " OpReturn\n" 217 " OpFunctionEnd\n"; 218 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats))); 219 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats))); 220 spec.numWorkGroups = IVec3(numElements, 1, 1); 221 222 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec)); 223 224 return group.release(); 225} 226 227tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx) 228{ 229 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction")); 230 ComputeShaderSpec spec; 231 de::Random rnd (deStringHash(group->getName())); 232 const int numElements = 100; 233 vector<float> positiveFloats (numElements, 0); 234 vector<float> negativeFloats (numElements, 0); 235 236 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements); 237 238 for (size_t ndx = 0; ndx < numElements; ++ndx) 239 negativeFloats[ndx] = -positiveFloats[ndx]; 240 241 spec.assembly = 242 string(s_ShaderPreamble) + 243 244 "%fname1 = OpString \"negateInputs.comp\"\n" 245 "%fname2 = OpString \"negateInputs\"\n" 246 247 "OpSource GLSL 430\n" 248 "OpName %main \"main\"\n" 249 "OpName %id \"gl_GlobalInvocationID\"\n" 250 251 "OpDecorate %id BuiltIn GlobalInvocationId\n" 252 253 + string(s_InputOutputBufferTraits) + 254 255 "OpLine %fname1 0 0\n" // At the earliest possible position 256 257 + string(s_CommonTypes) + string(s_InputOutputBuffer) + 258 259 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence 260 "OpLine %fname2 1 0\n" // Different filenames 261 "OpLine %fname1 1000 100000\n" 262 263 "%id = OpVariable %uvec3ptr Input\n" 264 "%zero = OpConstant %i32 0\n" 265 266 "OpLine %fname1 1 1\n" // Before a function 267 268 "%main = OpFunction %void None %voidf\n" 269 "%label = OpLabel\n" 270 271 "OpLine %fname1 1 1\n" // In a function 272 273 "%idval = OpLoad %uvec3 %id\n" 274 "%x = OpCompositeExtract %u32 %idval 0\n" 275 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 276 "%inval = OpLoad %f32 %inloc\n" 277 "%neg = OpFNegate %f32 %inval\n" 278 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 279 " OpStore %outloc %neg\n" 280 " OpReturn\n" 281 " OpFunctionEnd\n"; 282 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats))); 283 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats))); 284 spec.numWorkGroups = IVec3(numElements, 1, 1); 285 286 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec)); 287 288 return group.release(); 289} 290 291tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx) 292{ 293 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction")); 294 ComputeShaderSpec spec; 295 de::Random rnd (deStringHash(group->getName())); 296 const int numElements = 100; 297 vector<float> positiveFloats (numElements, 0); 298 vector<float> negativeFloats (numElements, 0); 299 300 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements); 301 302 for (size_t ndx = 0; ndx < numElements; ++ndx) 303 negativeFloats[ndx] = -positiveFloats[ndx]; 304 305 spec.assembly = 306 string(s_ShaderPreamble) + 307 308 "%fname = OpString \"negateInputs.comp\"\n" 309 310 "OpSource GLSL 430\n" 311 "OpName %main \"main\"\n" 312 "OpName %id \"gl_GlobalInvocationID\"\n" 313 314 "OpDecorate %id BuiltIn GlobalInvocationId\n" 315 316 + string(s_InputOutputBufferTraits) + 317 318 "OpNoLine\n" // At the earliest possible position, without preceding OpLine 319 320 + string(s_CommonTypes) + string(s_InputOutputBuffer) + 321 322 "OpLine %fname 0 1\n" 323 "OpNoLine\n" // Immediately following a preceding OpLine 324 325 "OpLine %fname 1000 1\n" 326 327 "%id = OpVariable %uvec3ptr Input\n" 328 "%zero = OpConstant %i32 0\n" 329 330 "OpNoLine\n" // Contents after the previous OpLine 331 332 "%main = OpFunction %void None %voidf\n" 333 "%label = OpLabel\n" 334 "%idval = OpLoad %uvec3 %id\n" 335 "%x = OpCompositeExtract %u32 %idval 0\n" 336 337 "OpNoLine\n" // Multiple OpNoLine 338 "OpNoLine\n" 339 "OpNoLine\n" 340 341 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 342 "%inval = OpLoad %f32 %inloc\n" 343 "%neg = OpFNegate %f32 %inval\n" 344 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 345 " OpStore %outloc %neg\n" 346 " OpReturn\n" 347 " OpFunctionEnd\n"; 348 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats))); 349 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats))); 350 spec.numWorkGroups = IVec3(numElements, 1, 1); 351 352 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec)); 353 354 return group.release(); 355} 356 357// Assembly code used for testing OpUnreachable is based on GLSL source code: 358// 359// #version 430 360// 361// layout(std140, set = 0, binding = 0) readonly buffer Input { 362// float elements[]; 363// } input_data; 364// layout(std140, set = 0, binding = 1) writeonly buffer Output { 365// float elements[]; 366// } output_data; 367// 368// void not_called_func() { 369// // place OpUnreachable here 370// } 371// 372// uint modulo4(uint val) { 373// switch (val % uint(4)) { 374// case 0: return 3; 375// case 1: return 2; 376// case 2: return 1; 377// case 3: return 0; 378// default: return 100; // place OpUnreachable here 379// } 380// } 381// 382// uint const5() { 383// return 5; 384// // place OpUnreachable here 385// } 386// 387// void main() { 388// uint x = gl_GlobalInvocationID.x; 389// if (const5() > modulo4(1000)) { 390// output_data.elements[x] = -input_data.elements[x]; 391// } else { 392// // place OpUnreachable here 393// output_data.elements[x] = input_data.elements[x]; 394// } 395// } 396 397tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx) 398{ 399 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction")); 400 ComputeShaderSpec spec; 401 de::Random rnd (deStringHash(group->getName())); 402 const int numElements = 100; 403 vector<float> positiveFloats (numElements, 0); 404 vector<float> negativeFloats (numElements, 0); 405 406 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements); 407 408 for (size_t ndx = 0; ndx < numElements; ++ndx) 409 negativeFloats[ndx] = -positiveFloats[ndx]; 410 411 spec.assembly = 412 string(s_ShaderPreamble) + 413 414 "OpSource GLSL 430\n" 415 "OpName %func_main \"main\"\n" 416 "OpName %func_not_called_func \"not_called_func(\"\n" 417 "OpName %func_modulo4 \"modulo4(u1;\"\n" 418 "OpName %func_const5 \"const5(\"\n" 419 "OpName %id \"gl_GlobalInvocationID\"\n" 420 421 "OpDecorate %id BuiltIn GlobalInvocationId\n" 422 423 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + 424 425 "%u32ptr = OpTypePointer Function %u32\n" 426 "%uintfuint = OpTypeFunction %u32 %u32ptr\n" 427 "%unitf = OpTypeFunction %u32\n" 428 429 "%id = OpVariable %uvec3ptr Input\n" 430 "%zero = OpConstant %u32 0\n" 431 "%one = OpConstant %u32 1\n" 432 "%two = OpConstant %u32 2\n" 433 "%three = OpConstant %u32 3\n" 434 "%four = OpConstant %u32 4\n" 435 "%five = OpConstant %u32 5\n" 436 "%hundred = OpConstant %u32 100\n" 437 "%thousand = OpConstant %u32 1000\n" 438 439 + string(s_InputOutputBuffer) + 440 441 // Main() 442 "%func_main = OpFunction %void None %voidf\n" 443 "%main_entry = OpLabel\n" 444 "%idval = OpLoad %uvec3 %id\n" 445 "%x = OpCompositeExtract %u32 %idval 0\n" 446 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 447 "%inval = OpLoad %f32 %inloc\n" 448 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 449 "%ret_const5 = OpFunctionCall %u32 %func_const5\n" 450 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %thousand\n" 451 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n" 452 " OpSelectionMerge %if_end None\n" 453 " OpBranchConditional %cmp_gt %if_true %if_false\n" 454 "%if_true = OpLabel\n" 455 "%negate = OpFNegate %f32 %inval\n" 456 " OpStore %outloc %negate\n" 457 " OpBranch %if_end\n" 458 "%if_false = OpLabel\n" 459 " OpUnreachable\n" // Unreachable else branch for if statement 460 "%if_end = OpLabel\n" 461 " OpReturn\n" 462 " OpFunctionEnd\n" 463 464 // not_called_function() 465 "%func_not_called_func = OpFunction %void None %voidf\n" 466 "%not_called_func_entry = OpLabel\n" 467 " OpUnreachable\n" // Unreachable entry block in not called static function 468 " OpFunctionEnd\n" 469 470 // modulo4() 471 "%func_modulo4 = OpFunction %u32 None %uintfuint\n" 472 "%valptr = OpFunctionParameter %u32ptr\n" 473 "%modulo4_entry = OpLabel\n" 474 "%val = OpLoad %u32 %valptr\n" 475 "%modulo = OpUMod %u32 %val %four\n" 476 " OpSelectionMerge %switch_merge None\n" 477 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n" 478 "%case0 = OpLabel\n" 479 " OpReturnValue %three\n" 480 "%case1 = OpLabel\n" 481 " OpReturnValue %two\n" 482 "%case2 = OpLabel\n" 483 " OpReturnValue %one\n" 484 "%case3 = OpLabel\n" 485 " OpReturnValue %zero\n" 486 "%default = OpLabel\n" 487 " OpUnreachable\n" // Unreachable default case for switch statement 488 "%switch_merge = OpLabel\n" 489 " OpUnreachable\n" // Unreachable merge block for switch statement 490 " OpFunctionEnd\n" 491 492 // const5() 493 "%func_const5 = OpFunction %u32 None %unitf\n" 494 "%const5_entry = OpLabel\n" 495 " OpReturnValue %five\n" 496 "%unreachable = OpLabel\n" 497 " OpUnreachable\n" // Unreachable block in function 498 " OpFunctionEnd\n"; 499 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats))); 500 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats))); 501 spec.numWorkGroups = IVec3(numElements, 1, 1); 502 503 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec)); 504 505 return group.release(); 506} 507 508// Assembly code used for testing decoration group is based on GLSL source code: 509// 510// #version 430 511// 512// layout(std140, set = 0, binding = 0) readonly buffer Input0 { 513// float elements[]; 514// } input_data0; 515// layout(std140, set = 0, binding = 1) readonly buffer Input1 { 516// float elements[]; 517// } input_data1; 518// layout(std140, set = 0, binding = 2) readonly buffer Input2 { 519// float elements[]; 520// } input_data2; 521// layout(std140, set = 0, binding = 3) readonly buffer Input3 { 522// float elements[]; 523// } input_data3; 524// layout(std140, set = 0, binding = 4) readonly buffer Input4 { 525// float elements[]; 526// } input_data4; 527// layout(std140, set = 0, binding = 5) writeonly buffer Output { 528// float elements[]; 529// } output_data; 530// 531// void main() { 532// uint x = gl_GlobalInvocationID.x; 533// output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x]; 534// } 535tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx) 536{ 537 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction")); 538 ComputeShaderSpec spec; 539 de::Random rnd (deStringHash(group->getName())); 540 const int numElements = 100; 541 vector<float> inputFloats0 (numElements, 0); 542 vector<float> inputFloats1 (numElements, 0); 543 vector<float> inputFloats2 (numElements, 0); 544 vector<float> inputFloats3 (numElements, 0); 545 vector<float> inputFloats4 (numElements, 0); 546 vector<float> outputFloats (numElements, 0); 547 548 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements); 549 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements); 550 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements); 551 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements); 552 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements); 553 554 for (size_t ndx = 0; ndx < numElements; ++ndx) 555 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx]; 556 557 spec.assembly = 558 string(s_ShaderPreamble) + 559 560 "OpSource GLSL 430\n" 561 "OpName %main \"main\"\n" 562 "OpName %id \"gl_GlobalInvocationID\"\n" 563 564 // Not using group decoration on variable. 565 "OpDecorate %id BuiltIn GlobalInvocationId\n" 566 // Not using group decoration on type. 567 "OpDecorate %f32arr ArrayStride 4\n" 568 569 "OpDecorate %groups BufferBlock\n" 570 "OpDecorate %groupm Offset 0\n" 571 "%groups = OpDecorationGroup\n" 572 "%groupm = OpDecorationGroup\n" 573 574 // Group decoration on multiple structs. 575 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n" 576 // Group decoration on multiple struct members. 577 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n" 578 579 "OpDecorate %group1 DescriptorSet 0\n" 580 "OpDecorate %group3 DescriptorSet 0\n" 581 "OpDecorate %group3 NonWritable\n" 582 "OpDecorate %group3 Restrict\n" 583 "%group0 = OpDecorationGroup\n" 584 "%group1 = OpDecorationGroup\n" 585 "%group3 = OpDecorationGroup\n" 586 587 // Applying the same decoration group multiple times. 588 "OpGroupDecorate %group1 %outdata\n" 589 "OpGroupDecorate %group1 %outdata\n" 590 "OpGroupDecorate %group1 %outdata\n" 591 "OpDecorate %outdata DescriptorSet 0\n" 592 "OpDecorate %outdata Binding 5\n" 593 // Applying decoration group containing nothing. 594 "OpGroupDecorate %group0 %indata0\n" 595 "OpDecorate %indata0 DescriptorSet 0\n" 596 "OpDecorate %indata0 Binding 0\n" 597 // Applying decoration group containing one decoration. 598 "OpGroupDecorate %group1 %indata1\n" 599 "OpDecorate %indata1 Binding 1\n" 600 // Applying decoration group containing multiple decorations. 601 "OpGroupDecorate %group3 %indata2 %indata3\n" 602 "OpDecorate %indata2 Binding 2\n" 603 "OpDecorate %indata3 Binding 3\n" 604 // Applying multiple decoration groups (with overlapping). 605 "OpGroupDecorate %group0 %indata4\n" 606 "OpGroupDecorate %group1 %indata4\n" 607 "OpGroupDecorate %group3 %indata4\n" 608 "OpDecorate %indata4 Binding 4\n" 609 610 + string(s_CommonTypes) + 611 612 "%id = OpVariable %uvec3ptr Input\n" 613 "%zero = OpConstant %i32 0\n" 614 615 "%outbuf = OpTypeStruct %f32arr\n" 616 "%outbufptr = OpTypePointer Uniform %outbuf\n" 617 "%outdata = OpVariable %outbufptr Uniform\n" 618 "%inbuf0 = OpTypeStruct %f32arr\n" 619 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n" 620 "%indata0 = OpVariable %inbuf0ptr Uniform\n" 621 "%inbuf1 = OpTypeStruct %f32arr\n" 622 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n" 623 "%indata1 = OpVariable %inbuf1ptr Uniform\n" 624 "%inbuf2 = OpTypeStruct %f32arr\n" 625 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n" 626 "%indata2 = OpVariable %inbuf2ptr Uniform\n" 627 "%inbuf3 = OpTypeStruct %f32arr\n" 628 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n" 629 "%indata3 = OpVariable %inbuf3ptr Uniform\n" 630 "%inbuf4 = OpTypeStruct %f32arr\n" 631 "%inbufptr = OpTypePointer Uniform %inbuf4\n" 632 "%indata4 = OpVariable %inbufptr Uniform\n" 633 634 "%main = OpFunction %void None %voidf\n" 635 "%label = OpLabel\n" 636 "%idval = OpLoad %uvec3 %id\n" 637 "%x = OpCompositeExtract %u32 %idval 0\n" 638 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n" 639 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n" 640 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n" 641 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n" 642 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n" 643 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 644 "%inval0 = OpLoad %f32 %inloc0\n" 645 "%inval1 = OpLoad %f32 %inloc1\n" 646 "%inval2 = OpLoad %f32 %inloc2\n" 647 "%inval3 = OpLoad %f32 %inloc3\n" 648 "%inval4 = OpLoad %f32 %inloc4\n" 649 "%add0 = OpFAdd %f32 %inval0 %inval1\n" 650 "%add1 = OpFAdd %f32 %add0 %inval2\n" 651 "%add2 = OpFAdd %f32 %add1 %inval3\n" 652 "%add = OpFAdd %f32 %add2 %inval4\n" 653 " OpStore %outloc %add\n" 654 " OpReturn\n" 655 " OpFunctionEnd\n"; 656 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0))); 657 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1))); 658 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2))); 659 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3))); 660 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4))); 661 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats))); 662 spec.numWorkGroups = IVec3(numElements, 1, 1); 663 664 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec)); 665 666 return group.release(); 667} 668 669tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx) 670{ 671 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction")); 672 ComputeShaderSpec spec; 673 de::Random rnd (deStringHash(group->getName())); 674 const int numElements = 100; 675 vector<float> inputFloats (numElements, 0); 676 vector<float> outputFloats (numElements, 0); 677 678 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements); 679 680 for (size_t ndx = 0; ndx < numElements; ++ndx) 681 { 682 switch (ndx % 3) 683 { 684 case 0: outputFloats[ndx] = inputFloats[ndx] + 5.5f; break; 685 case 1: outputFloats[ndx] = inputFloats[ndx] + 20.5f; break; 686 case 2: outputFloats[ndx] = inputFloats[ndx] + 1.75f; break; 687 default: break; 688 } 689 } 690 691 spec.assembly = 692 string(s_ShaderPreamble) + 693 694 "OpSource GLSL 430\n" 695 "OpName %main \"main\"\n" 696 "OpName %id \"gl_GlobalInvocationID\"\n" 697 698 "OpDecorate %id BuiltIn GlobalInvocationId\n" 699 700 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) + 701 702 "%id = OpVariable %uvec3ptr Input\n" 703 "%zero = OpConstant %i32 0\n" 704 "%three = OpConstant %u32 3\n" 705 "%constf5p5 = OpConstant %f32 5.5\n" 706 "%constf20p5 = OpConstant %f32 20.5\n" 707 "%constf1p75 = OpConstant %f32 1.75\n" 708 "%constf8p5 = OpConstant %f32 8.5\n" 709 "%constf6p5 = OpConstant %f32 6.5\n" 710 711 "%main = OpFunction %void None %voidf\n" 712 "%entry = OpLabel\n" 713 "%idval = OpLoad %uvec3 %id\n" 714 "%x = OpCompositeExtract %u32 %idval 0\n" 715 "%selector = OpUMod %u32 %x %three\n" 716 " OpSelectionMerge %default None\n" 717 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n" 718 719 // Case 1 before OpPhi. 720 "%case1 = OpLabel\n" 721 " OpBranch %phi\n" 722 723 "%default = OpLabel\n" 724 " OpUnreachable\n" 725 726 "%phi = OpLabel\n" 727 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0" // not in the order of blocks 728 " %constf8p5 %phi %constf6p5 %default\n" // from the same block & from an unreachable block 729 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 730 "%inval = OpLoad %f32 %inloc\n" 731 "%add = OpFAdd %f32 %inval %operand\n" 732 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 733 " OpStore %outloc %add\n" 734 " OpReturn\n" 735 736 // Case 0 after OpPhi. 737 "%case0 = OpLabel\n" 738 " OpBranch %phi\n" 739 740 741 // Case 2 after OpPhi. 742 "%case2 = OpLabel\n" 743 " OpBranch %phi\n" 744 745 " OpFunctionEnd\n"; 746 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats))); 747 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats))); 748 spec.numWorkGroups = IVec3(numElements, 1, 1); 749 750 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpPhi corner cases", spec)); 751 752 return group.release(); 753} 754 755// Assembly code used for testing block order is based on GLSL source code: 756// 757// #version 430 758// 759// layout(std140, set = 0, binding = 0) readonly buffer Input { 760// float elements[]; 761// } input_data; 762// layout(std140, set = 0, binding = 1) writeonly buffer Output { 763// float elements[]; 764// } output_data; 765// 766// void main() { 767// uint x = gl_GlobalInvocationID.x; 768// output_data.elements[x] = input_data.elements[x]; 769// if (x > uint(50)) { 770// switch (x % uint(3)) { 771// case 0: output_data.elements[x] += 1.5f; break; 772// case 1: output_data.elements[x] += 42.f; break; 773// case 2: output_data.elements[x] -= 27.f; break; 774// default: break; 775// } 776// } else { 777// output_data.elements[x] = -input_data.elements[x]; 778// } 779// } 780tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx) 781{ 782 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders")); 783 ComputeShaderSpec spec; 784 de::Random rnd (deStringHash(group->getName())); 785 const int numElements = 100; 786 vector<float> inputFloats (numElements, 0); 787 vector<float> outputFloats (numElements, 0); 788 789 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements); 790 791 for (size_t ndx = 0; ndx <= 50; ++ndx) 792 outputFloats[ndx] = -inputFloats[ndx]; 793 794 for (size_t ndx = 51; ndx < numElements; ++ndx) 795 { 796 switch (ndx % 3) 797 { 798 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break; 799 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break; 800 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break; 801 default: break; 802 } 803 } 804 805 spec.assembly = 806 string(s_ShaderPreamble) + 807 808 "OpSource GLSL 430\n" 809 "OpName %main \"main\"\n" 810 "OpName %id \"gl_GlobalInvocationID\"\n" 811 812 "OpDecorate %id BuiltIn GlobalInvocationId\n" 813 814 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + 815 816 "%u32ptr = OpTypePointer Function %u32\n" 817 "%u32ptr_input = OpTypePointer Input %u32\n" 818 819 + string(s_InputOutputBuffer) + 820 821 "%id = OpVariable %uvec3ptr Input\n" 822 "%zero = OpConstant %i32 0\n" 823 "%const3 = OpConstant %u32 3\n" 824 "%const50 = OpConstant %u32 50\n" 825 "%constf1p5 = OpConstant %f32 1.5\n" 826 "%constf27 = OpConstant %f32 27.0\n" 827 "%constf42 = OpConstant %f32 42.0\n" 828 829 "%main = OpFunction %void None %voidf\n" 830 831 // entry block. 832 "%entry = OpLabel\n" 833 834 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x. 835 "%xvar = OpVariable %u32ptr Function\n" 836 "%xptr = OpAccessChain %u32ptr_input %id %zero\n" 837 "%x = OpLoad %u32 %xptr\n" 838 " OpStore %xvar %x\n" 839 840 "%cmp = OpUGreaterThan %bool %x %const50\n" 841 " OpSelectionMerge %if_merge None\n" 842 " OpBranchConditional %cmp %if_true %if_false\n" 843 844 // Merge block for switch-statement: placed at the beginning. 845 "%switch_merge = OpLabel\n" 846 " OpBranch %if_merge\n" 847 848 // Case 1 for switch-statement. 849 "%case1 = OpLabel\n" 850 "%x_1 = OpLoad %u32 %xvar\n" 851 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n" 852 "%inval_1 = OpLoad %f32 %inloc_1\n" 853 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n" 854 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n" 855 " OpStore %outloc_1 %addf42\n" 856 " OpBranch %switch_merge\n" 857 858 // False branch for if-statement: placed in the middle of switch cases and before true branch. 859 "%if_false = OpLabel\n" 860 "%x_f = OpLoad %u32 %xvar\n" 861 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n" 862 "%inval_f = OpLoad %f32 %inloc_f\n" 863 "%negate = OpFNegate %f32 %inval_f\n" 864 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n" 865 " OpStore %outloc_f %negate\n" 866 " OpBranch %if_merge\n" 867 868 // Merge block for if-statement: placed in the middle of true and false branch. 869 "%if_merge = OpLabel\n" 870 " OpReturn\n" 871 872 // True branch for if-statement: placed in the middle of swtich cases and after the false branch. 873 "%if_true = OpLabel\n" 874 "%xval_t = OpLoad %u32 %xvar\n" 875 "%mod = OpUMod %u32 %xval_t %const3\n" 876 " OpSelectionMerge %switch_merge None\n" 877 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n" 878 879 // Case 2 for switch-statement. 880 "%case2 = OpLabel\n" 881 "%x_2 = OpLoad %u32 %xvar\n" 882 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n" 883 "%inval_2 = OpLoad %f32 %inloc_2\n" 884 "%subf27 = OpFSub %f32 %inval_2 %constf27\n" 885 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n" 886 " OpStore %outloc_2 %subf27\n" 887 " OpBranch %switch_merge\n" 888 889 // Default case for switch-statement: placed in the middle of normal cases. 890 "%default = OpLabel\n" 891 " OpBranch %switch_merge\n" 892 893 // Case 0 for switch-statement: out of order. 894 "%case0 = OpLabel\n" 895 "%x_0 = OpLoad %u32 %xvar\n" 896 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n" 897 "%inval_0 = OpLoad %f32 %inloc_0\n" 898 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n" 899 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n" 900 " OpStore %outloc_0 %addf1p5\n" 901 " OpBranch %switch_merge\n" 902 903 " OpFunctionEnd\n"; 904 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats))); 905 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats))); 906 spec.numWorkGroups = IVec3(numElements, 1, 1); 907 908 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec)); 909 910 return group.release(); 911} 912 913struct CaseParameter 914{ 915 const char* name; 916 string param; 917 918 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {} 919}; 920 921tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx) 922{ 923 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction")); 924 vector<CaseParameter> cases; 925 de::Random rnd (deStringHash(group->getName())); 926 const int numElements = 100; 927 vector<float> positiveFloats (numElements, 0); 928 vector<float> negativeFloats (numElements, 0); 929 const StringTemplate shaderTemplate ( 930 "OpCapability Shader\n" 931 "OpMemoryModel Logical GLSL450\n" 932 933 "OpEntryPoint GLCompute %main \"main\" %id\n" 934 "OpExecutionMode %main LocalSize 1 1 1\n" 935 936 "${SOURCE}\n" 937 938 "OpName %main \"main\"\n" 939 "OpName %id \"gl_GlobalInvocationID\"\n" 940 941 "OpDecorate %id BuiltIn GlobalInvocationId\n" 942 943 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) + 944 945 "%id = OpVariable %uvec3ptr Input\n" 946 "%zero = OpConstant %i32 0\n" 947 948 "%main = OpFunction %void None %voidf\n" 949 "%label = OpLabel\n" 950 "%idval = OpLoad %uvec3 %id\n" 951 "%x = OpCompositeExtract %u32 %idval 0\n" 952 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 953 "%inval = OpLoad %f32 %inloc\n" 954 "%neg = OpFNegate %f32 %inval\n" 955 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 956 " OpStore %outloc %neg\n" 957 " OpReturn\n" 958 " OpFunctionEnd\n"); 959 960 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0")); 961 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL 210")); 962 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n" 963 "OpSource GLSL 430 %fname")); 964 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n" 965 "OpSource GLSL 430 %fname")); 966 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n" 967 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"")); 968 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n" 969 "OpSource GLSL 430 %fname \"\"")); 970 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n" 971 "OpSource GLSL 430 %fname \"" + string(65530, 'x') + "\"")); // word count: 65535 972 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n" 973 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol 974 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n" 975 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n" 976 "OpSourceContinued \"id main() {}\"")); 977 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n" 978 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n" 979 "OpSourceContinued \"\"")); 980 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n" 981 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n" 982 "OpSourceContinued \"" + string(65533, 'x') + "\"")); // word count: 65535 983 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n" 984 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n" 985 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol 986 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n" 987 "OpSource GLSL 430 %fname \"#version 430\n\"\n" 988 "OpSourceContinued \"void\"\n" 989 "OpSourceContinued \"main()\"\n" 990 "OpSourceContinued \"{}\"")); 991 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n" 992 "OpSource GLSL 430 %fname \"\"\n" 993 "OpSourceContinued \"#version 430\nvoid main() {}\"")); 994 995 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements); 996 997 for (size_t ndx = 0; ndx < numElements; ++ndx) 998 negativeFloats[ndx] = -positiveFloats[ndx]; 999 1000 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx) 1001 { 1002 map<string, string> specializations; 1003 ComputeShaderSpec spec; 1004 1005 specializations["SOURCE"] = cases[caseNdx].param; 1006 spec.assembly = shaderTemplate.specialize(specializations); 1007 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats))); 1008 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats))); 1009 spec.numWorkGroups = IVec3(numElements, 1, 1); 1010 1011 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec)); 1012 } 1013 1014 return group.release(); 1015} 1016 1017tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx) 1018{ 1019 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction")); 1020 vector<CaseParameter> cases; 1021 de::Random rnd (deStringHash(group->getName())); 1022 const int numElements = 100; 1023 vector<float> inputFloats (numElements, 0); 1024 vector<float> outputFloats (numElements, 0); 1025 const StringTemplate shaderTemplate ( 1026 string(s_ShaderPreamble) + 1027 1028 "OpSourceExtension \"${EXTENSION}\"\n" 1029 1030 "OpName %main \"main\"\n" 1031 "OpName %id \"gl_GlobalInvocationID\"\n" 1032 1033 "OpDecorate %id BuiltIn GlobalInvocationId\n" 1034 1035 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) + 1036 1037 "%id = OpVariable %uvec3ptr Input\n" 1038 "%zero = OpConstant %i32 0\n" 1039 1040 "%main = OpFunction %void None %voidf\n" 1041 "%label = OpLabel\n" 1042 "%idval = OpLoad %uvec3 %id\n" 1043 "%x = OpCompositeExtract %u32 %idval 0\n" 1044 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 1045 "%inval = OpLoad %f32 %inloc\n" 1046 "%neg = OpFNegate %f32 %inval\n" 1047 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 1048 " OpStore %outloc %neg\n" 1049 " OpReturn\n" 1050 " OpFunctionEnd\n"); 1051 1052 cases.push_back(CaseParameter("empty_extension", "")); 1053 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle")); 1054 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension")); 1055 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85")); 1056 cases.push_back(CaseParameter("long_extension", string(65533, 'e'))); // word count: 65535 1057 1058 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements); 1059 1060 for (size_t ndx = 0; ndx < numElements; ++ndx) 1061 outputFloats[ndx] = -inputFloats[ndx]; 1062 1063 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx) 1064 { 1065 map<string, string> specializations; 1066 ComputeShaderSpec spec; 1067 1068 specializations["EXTENSION"] = cases[caseNdx].param; 1069 spec.assembly = shaderTemplate.specialize(specializations); 1070 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats))); 1071 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats))); 1072 spec.numWorkGroups = IVec3(numElements, 1, 1); 1073 1074 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec)); 1075 } 1076 1077 return group.release(); 1078} 1079 1080// Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it. 1081tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx) 1082{ 1083 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction")); 1084 vector<CaseParameter> cases; 1085 de::Random rnd (deStringHash(group->getName())); 1086 const int numElements = 100; 1087 vector<float> positiveFloats (numElements, 0); 1088 vector<float> negativeFloats (numElements, 0); 1089 const StringTemplate shaderTemplate ( 1090 string(s_ShaderPreamble) + 1091 1092 "OpSource GLSL 430\n" 1093 "OpName %main \"main\"\n" 1094 "OpName %id \"gl_GlobalInvocationID\"\n" 1095 1096 "OpDecorate %id BuiltIn GlobalInvocationId\n" 1097 1098 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) + 1099 1100 "${TYPE}\n" 1101 "%null = OpConstantNull %type\n" 1102 1103 "%id = OpVariable %uvec3ptr Input\n" 1104 "%zero = OpConstant %i32 0\n" 1105 1106 "%main = OpFunction %void None %voidf\n" 1107 "%label = OpLabel\n" 1108 "%idval = OpLoad %uvec3 %id\n" 1109 "%x = OpCompositeExtract %u32 %idval 0\n" 1110 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 1111 "%inval = OpLoad %f32 %inloc\n" 1112 "%neg = OpFNegate %f32 %inval\n" 1113 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 1114 " OpStore %outloc %neg\n" 1115 " OpReturn\n" 1116 " OpFunctionEnd\n"); 1117 1118 cases.push_back(CaseParameter("bool", "%type = OpTypeBool")); 1119 cases.push_back(CaseParameter("sint32", "%type = OpTypeInt 32 1")); 1120 cases.push_back(CaseParameter("uint32", "%type = OpTypeInt 32 0")); 1121 cases.push_back(CaseParameter("float32", "%type = OpTypeFloat 32")); 1122 cases.push_back(CaseParameter("vec4float32", "%type = OpTypeVector %f32 4")); 1123 cases.push_back(CaseParameter("vec3bool", "%type = OpTypeVector %bool 3")); 1124 cases.push_back(CaseParameter("vec2uint32", "%type = OpTypeVector %u32 2")); 1125 cases.push_back(CaseParameter("matrix", "%type = OpTypeMatrix %uvec3 3")); 1126 cases.push_back(CaseParameter("array", "%100 = OpConstant %u32 100\n" 1127 "%type = OpTypeArray %i32 %100")); 1128 cases.push_back(CaseParameter("runtimearray", "%type = OpTypeRuntimeArray %f32")); 1129 cases.push_back(CaseParameter("struct", "%type = OpTypeStruct %f32 %i32 %u32")); 1130 cases.push_back(CaseParameter("pointer", "%type = OpTypePointer Function %i32")); 1131 1132 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements); 1133 1134 for (size_t ndx = 0; ndx < numElements; ++ndx) 1135 negativeFloats[ndx] = -positiveFloats[ndx]; 1136 1137 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx) 1138 { 1139 map<string, string> specializations; 1140 ComputeShaderSpec spec; 1141 1142 specializations["TYPE"] = cases[caseNdx].param; 1143 spec.assembly = shaderTemplate.specialize(specializations); 1144 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats))); 1145 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats))); 1146 spec.numWorkGroups = IVec3(numElements, 1, 1); 1147 1148 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec)); 1149 } 1150 1151 return group.release(); 1152} 1153 1154// Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it. 1155tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx) 1156{ 1157 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction")); 1158 vector<CaseParameter> cases; 1159 de::Random rnd (deStringHash(group->getName())); 1160 const int numElements = 100; 1161 vector<float> positiveFloats (numElements, 0); 1162 vector<float> negativeFloats (numElements, 0); 1163 const StringTemplate shaderTemplate ( 1164 string(s_ShaderPreamble) + 1165 1166 "OpSource GLSL 430\n" 1167 "OpName %main \"main\"\n" 1168 "OpName %id \"gl_GlobalInvocationID\"\n" 1169 1170 "OpDecorate %id BuiltIn GlobalInvocationId\n" 1171 1172 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) + 1173 1174 "%id = OpVariable %uvec3ptr Input\n" 1175 "%zero = OpConstant %i32 0\n" 1176 1177 "${CONSTANT}\n" 1178 1179 "%main = OpFunction %void None %voidf\n" 1180 "%label = OpLabel\n" 1181 "%idval = OpLoad %uvec3 %id\n" 1182 "%x = OpCompositeExtract %u32 %idval 0\n" 1183 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 1184 "%inval = OpLoad %f32 %inloc\n" 1185 "%neg = OpFNegate %f32 %inval\n" 1186 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 1187 " OpStore %outloc %neg\n" 1188 " OpReturn\n" 1189 " OpFunctionEnd\n"); 1190 1191 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n" 1192 "%const = OpConstantComposite %uvec3 %five %zero %five")); 1193 cases.push_back(CaseParameter("matrix", "%m3uvec3 = OpTypeMatrix %uvec3 3\n" 1194 "%ten = OpConstant %u32 10\n" 1195 "%vec = OpConstantComposite %uvec3 %ten %zero %ten\n" 1196 "%mat = OpConstantComposite %m3uvec3 %vec %vec %vec")); 1197 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %uvec3 2\n" 1198 "%struct = OpTypeStruct %u32 %f32 %uvec3 %m2vec3\n" 1199 "%one = OpConstant %u32 1\n" 1200 "%point5 = OpConstant %f32 0.5\n" 1201 "%vec = OpConstantComposite %uvec3 %one %one %zero\n" 1202 "%mat = OpConstantComposite %m2vec3 %vec %vec\n" 1203 "%const = OpConstantComposite %one %point5 %vec %mat")); 1204 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n" 1205 "%st2 = OpTypeStruct %i32 %i32\n" 1206 "%struct = OpTypeStruct %st1 %st2\n" 1207 "%point5 = OpConstant %f32 0.5\n" 1208 "%one = OpConstant %u32 1\n" 1209 "%ten = OpConstant %i32 10\n" 1210 "%st1val = OpConstantComposite %one %point5\n" 1211 "%st2val = OpConstantComposite %ten %ten\n" 1212 "%const = OpConstantComposite %st1val %st2val")); 1213 1214 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements); 1215 1216 for (size_t ndx = 0; ndx < numElements; ++ndx) 1217 negativeFloats[ndx] = -positiveFloats[ndx]; 1218 1219 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx) 1220 { 1221 map<string, string> specializations; 1222 ComputeShaderSpec spec; 1223 1224 specializations["CONSTANT"] = cases[caseNdx].param; 1225 spec.assembly = shaderTemplate.specialize(specializations); 1226 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats))); 1227 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats))); 1228 spec.numWorkGroups = IVec3(numElements, 1, 1); 1229 1230 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec)); 1231 } 1232 1233 return group.release(); 1234} 1235 1236// Checks that constant null/composite values can be used in computation. 1237tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx) 1238{ 1239 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction")); 1240 ComputeShaderSpec spec; 1241 de::Random rnd (deStringHash(group->getName())); 1242 const int numElements = 100; 1243 vector<float> positiveFloats (numElements, 0); 1244 vector<float> negativeFloats (numElements, 0); 1245 1246 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements); 1247 1248 for (size_t ndx = 0; ndx < numElements; ++ndx) 1249 negativeFloats[ndx] = -positiveFloats[ndx]; 1250 1251 spec.assembly = 1252 "OpCapability Shader\n" 1253 "%std450 = OpExtInstImport \"GLSL.std.450\"\n" 1254 "OpMemoryModel Logical GLSL450\n" 1255 "OpEntryPoint GLCompute %main \"main\" %id\n" 1256 "OpExecutionMode %main LocalSize 1 1 1\n" 1257 1258 "OpSource GLSL 430\n" 1259 "OpName %main \"main\"\n" 1260 "OpName %id \"gl_GlobalInvocationID\"\n" 1261 1262 "OpDecorate %id BuiltIn GlobalInvocationId\n" 1263 1264 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + 1265 1266 "%fvec3 = OpTypeVector %f32 3\n" 1267 "%fmat = OpTypeMatrix %fvec3 3\n" 1268 "%ten = OpConstant %u32 10\n" 1269 "%f32arr10 = OpTypeArray %f32 %ten\n" 1270 "%fst = OpTypeStruct %f32 %f32\n" 1271 1272 + string(s_InputOutputBuffer) + 1273 1274 "%id = OpVariable %uvec3ptr Input\n" 1275 "%zero = OpConstant %i32 0\n" 1276 1277 // Create a bunch of null values 1278 "%unull = OpConstantNull %u32\n" 1279 "%fnull = OpConstantNull %f32\n" 1280 "%vnull = OpConstantNull %fvec3\n" 1281 "%mnull = OpConstantNull %fmat\n" 1282 "%anull = OpConstantNull %f32arr10\n" 1283 "%snull = OpConstantComposite %fst %fnull %fnull\n" 1284 1285 "%main = OpFunction %void None %voidf\n" 1286 "%label = OpLabel\n" 1287 "%idval = OpLoad %uvec3 %id\n" 1288 "%x = OpCompositeExtract %u32 %idval 0\n" 1289 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 1290 "%inval = OpLoad %f32 %inloc\n" 1291 "%neg = OpFNegate %f32 %inval\n" 1292 1293 // Get the abs() of (a certain element of) those null values 1294 "%unull_cov = OpConvertUToF %f32 %unull\n" 1295 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n" 1296 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n" 1297 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n" 1298 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n" 1299 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n" 1300 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n" 1301 "%anull_3 = OpCompositeExtract %f32 %anull 3\n" 1302 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n" 1303 "%snull_1 = OpCompositeExtract %f32 %snull 1\n" 1304 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n" 1305 1306 // Add them all 1307 "%add1 = OpFAdd %f32 %neg %unull_abs\n" 1308 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n" 1309 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n" 1310 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n" 1311 "%add5 = OpFAdd %f32 %add4 %anull_abs\n" 1312 "%final = OpFAdd %f32 %add5 %snull_abs\n" 1313 1314 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 1315 " OpStore %outloc %final\n" // write to output 1316 " OpReturn\n" 1317 " OpFunctionEnd\n"; 1318 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats))); 1319 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats))); 1320 spec.numWorkGroups = IVec3(numElements, 1, 1); 1321 1322 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec)); 1323 1324 return group.release(); 1325} 1326 1327// Assembly code used for testing loop control is based on GLSL source code: 1328// #version 430 1329// 1330// layout(std140, set = 0, binding = 0) readonly buffer Input { 1331// float elements[]; 1332// } input_data; 1333// layout(std140, set = 0, binding = 1) writeonly buffer Output { 1334// float elements[]; 1335// } output_data; 1336// 1337// void main() { 1338// uint x = gl_GlobalInvocationID.x; 1339// output_data.elements[x] = input_data.elements[x]; 1340// for (uint i = 0; i < 4; ++i) 1341// output_data.elements[x] += 1.f; 1342// } 1343tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx) 1344{ 1345 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases")); 1346 vector<CaseParameter> cases; 1347 de::Random rnd (deStringHash(group->getName())); 1348 const int numElements = 100; 1349 vector<float> inputFloats (numElements, 0); 1350 vector<float> outputFloats (numElements, 0); 1351 const StringTemplate shaderTemplate ( 1352 string(s_ShaderPreamble) + 1353 1354 "OpSource GLSL 430\n" 1355 "OpName %main \"main\"\n" 1356 "OpName %id \"gl_GlobalInvocationID\"\n" 1357 1358 "OpDecorate %id BuiltIn GlobalInvocationId\n" 1359 1360 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) + 1361 1362 "%u32ptr = OpTypePointer Function %u32\n" 1363 1364 "%id = OpVariable %uvec3ptr Input\n" 1365 "%zero = OpConstant %i32 0\n" 1366 "%one = OpConstant %i32 1\n" 1367 "%constf1 = OpConstant %f32 1.0\n" 1368 "%four = OpConstant %u32 4\n" 1369 1370 "%main = OpFunction %void None %voidf\n" 1371 "%entry = OpLabel\n" 1372 "%i = OpVariable %u32ptr Function\n" 1373 " OpStore %i %zero\n" 1374 1375 "%idval = OpLoad %uvec3 %id\n" 1376 "%x = OpCompositeExtract %u32 %idval 0\n" 1377 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 1378 "%inval = OpLoad %f32 %inloc\n" 1379 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 1380 " OpStore %outloc %inval\n" 1381 " OpBranch %loop_entry\n" 1382 1383 "%loop_entry = OpLabel\n" 1384 "%i_val = OpLoad %u32 %i\n" 1385 "%cmp_lt = OpULessThan %bool %i_val %four\n" 1386 " OpLoopMerge %loop_merge %loop_entry ${CONTROL}\n" 1387 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n" 1388 "%loop_body = OpLabel\n" 1389 "%outval = OpLoad %f32 %outloc\n" 1390 "%addf1 = OpFAdd %f32 %outval %constf1\n" 1391 " OpStore %outloc %addf1\n" 1392 "%new_i = OpIAdd %u32 %i_val %one\n" 1393 " OpStore %i %new_i\n" 1394 " OpBranch %loop_entry\n" 1395 "%loop_merge = OpLabel\n" 1396 " OpReturn\n" 1397 " OpFunctionEnd\n"); 1398 1399 cases.push_back(CaseParameter("none", "None")); 1400 cases.push_back(CaseParameter("unroll", "Unroll")); 1401 cases.push_back(CaseParameter("dont_unroll", "DontUnroll")); 1402 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll")); 1403 1404 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements); 1405 1406 for (size_t ndx = 0; ndx < numElements; ++ndx) 1407 outputFloats[ndx] = inputFloats[ndx] + 4.f; 1408 1409 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx) 1410 { 1411 map<string, string> specializations; 1412 ComputeShaderSpec spec; 1413 1414 specializations["CONTROL"] = cases[caseNdx].param; 1415 spec.assembly = shaderTemplate.specialize(specializations); 1416 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats))); 1417 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats))); 1418 spec.numWorkGroups = IVec3(numElements, 1, 1); 1419 1420 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec)); 1421 } 1422 1423 return group.release(); 1424} 1425 1426// Assembly code used for testing selection control is based on GLSL source code: 1427// #version 430 1428// 1429// layout(std140, set = 0, binding = 0) readonly buffer Input { 1430// float elements[]; 1431// } input_data; 1432// layout(std140, set = 0, binding = 1) writeonly buffer Output { 1433// float elements[]; 1434// } output_data; 1435// 1436// void main() { 1437// uint x = gl_GlobalInvocationID.x; 1438// float val = input_data.elements[x]; 1439// if (val > 10.f) 1440// output_data.elements[x] = val + 1.f; 1441// else 1442// output_data.elements[x] = val - 1.f; 1443// } 1444tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx) 1445{ 1446 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases")); 1447 vector<CaseParameter> cases; 1448 de::Random rnd (deStringHash(group->getName())); 1449 const int numElements = 100; 1450 vector<float> inputFloats (numElements, 0); 1451 vector<float> outputFloats (numElements, 0); 1452 const StringTemplate shaderTemplate ( 1453 string(s_ShaderPreamble) + 1454 1455 "OpSource GLSL 430\n" 1456 "OpName %main \"main\"\n" 1457 "OpName %id \"gl_GlobalInvocationID\"\n" 1458 1459 "OpDecorate %id BuiltIn GlobalInvocationId\n" 1460 1461 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) + 1462 1463 "%id = OpVariable %uvec3ptr Input\n" 1464 "%zero = OpConstant %i32 0\n" 1465 "%constf1 = OpConstant %f32 1.0\n" 1466 "%constf10 = OpConstant %f32 10.0\n" 1467 1468 "%main = OpFunction %void None %voidf\n" 1469 "%entry = OpLabel\n" 1470 "%idval = OpLoad %uvec3 %id\n" 1471 "%x = OpCompositeExtract %u32 %idval 0\n" 1472 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 1473 "%inval = OpLoad %f32 %inloc\n" 1474 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 1475 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n" 1476 1477 " OpSelectionMerge %if_end ${CONTROL}\n" 1478 " OpBranchConditional %cmp_gt %if_true %if_false\n" 1479 "%if_true = OpLabel\n" 1480 "%addf1 = OpFAdd %f32 %inval %constf1\n" 1481 " OpStore %outloc %addf1\n" 1482 " OpBranch %if_end\n" 1483 "%if_false = OpLabel\n" 1484 "%subf1 = OpFSub %f32 %inval %constf1\n" 1485 " OpStore %outloc %subf1\n" 1486 " OpBranch %if_end\n" 1487 "%if_end = OpLabel\n" 1488 " OpReturn\n" 1489 " OpFunctionEnd\n"); 1490 1491 cases.push_back(CaseParameter("none", "None")); 1492 cases.push_back(CaseParameter("flatten", "Flatten")); 1493 cases.push_back(CaseParameter("dont_flatten", "DontFlatten")); 1494 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten")); 1495 1496 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements); 1497 1498 for (size_t ndx = 0; ndx < numElements; ++ndx) 1499 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f); 1500 1501 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx) 1502 { 1503 map<string, string> specializations; 1504 ComputeShaderSpec spec; 1505 1506 specializations["CONTROL"] = cases[caseNdx].param; 1507 spec.assembly = shaderTemplate.specialize(specializations); 1508 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats))); 1509 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats))); 1510 spec.numWorkGroups = IVec3(numElements, 1, 1); 1511 1512 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec)); 1513 } 1514 1515 return group.release(); 1516} 1517 1518// Assembly code used for testing function control is based on GLSL source code: 1519// 1520// #version 430 1521// 1522// layout(std140, set = 0, binding = 0) readonly buffer Input { 1523// float elements[]; 1524// } input_data; 1525// layout(std140, set = 0, binding = 1) writeonly buffer Output { 1526// float elements[]; 1527// } output_data; 1528// 1529// float const10() { return 10.f; } 1530// 1531// void main() { 1532// uint x = gl_GlobalInvocationID.x; 1533// output_data.elements[x] = input_data.elements[x] + const10(); 1534// } 1535tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx) 1536{ 1537 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases")); 1538 vector<CaseParameter> cases; 1539 de::Random rnd (deStringHash(group->getName())); 1540 const int numElements = 100; 1541 vector<float> inputFloats (numElements, 0); 1542 vector<float> outputFloats (numElements, 0); 1543 const StringTemplate shaderTemplate ( 1544 string(s_ShaderPreamble) + 1545 1546 "OpSource GLSL 430\n" 1547 "OpName %main \"main\"\n" 1548 "OpName %func_const10 \"const10(\"\n" 1549 "OpName %id \"gl_GlobalInvocationID\"\n" 1550 1551 "OpDecorate %id BuiltIn GlobalInvocationId\n" 1552 1553 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) + 1554 1555 "%f32f = OpTypeFunction %f32\n" 1556 "%id = OpVariable %uvec3ptr Input\n" 1557 "%zero = OpConstant %i32 0\n" 1558 "%constf10 = OpConstant %f32 10.0\n" 1559 1560 "%main = OpFunction %void None %voidf\n" 1561 "%entry = OpLabel\n" 1562 "%idval = OpLoad %uvec3 %id\n" 1563 "%x = OpCompositeExtract %u32 %idval 0\n" 1564 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 1565 "%inval = OpLoad %f32 %inloc\n" 1566 "%ret_10 = OpFunctionCall %f32 %func_const10\n" 1567 "%fadd = OpFAdd %f32 %inval %ret_10\n" 1568 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 1569 " OpStore %outloc %fadd\n" 1570 " OpReturn\n" 1571 " OpFunctionEnd\n" 1572 1573 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n" 1574 "%label = OpLabel\n" 1575 " OpReturnValue %constf10\n" 1576 " OpFunctionEnd\n"); 1577 1578 cases.push_back(CaseParameter("none", "None")); 1579 cases.push_back(CaseParameter("inline", "Inline")); 1580 cases.push_back(CaseParameter("dont_inline", "DontInline")); 1581 cases.push_back(CaseParameter("pure", "Pure")); 1582 cases.push_back(CaseParameter("const", "Const")); 1583 cases.push_back(CaseParameter("inline_pure", "Inline|Pure")); 1584 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline")); 1585 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline")); 1586 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline")); 1587 1588 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements); 1589 1590 for (size_t ndx = 0; ndx < numElements; ++ndx) 1591 outputFloats[ndx] = inputFloats[ndx] + 10.f; 1592 1593 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx) 1594 { 1595 map<string, string> specializations; 1596 ComputeShaderSpec spec; 1597 1598 specializations["CONTROL"] = cases[caseNdx].param; 1599 spec.assembly = shaderTemplate.specialize(specializations); 1600 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats))); 1601 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats))); 1602 spec.numWorkGroups = IVec3(numElements, 1, 1); 1603 1604 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec)); 1605 } 1606 1607 return group.release(); 1608} 1609 1610// Checks that we can get undefined values for various types, without exercising a computation with it. 1611tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx) 1612{ 1613 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction")); 1614 vector<CaseParameter> cases; 1615 de::Random rnd (deStringHash(group->getName())); 1616 const int numElements = 100; 1617 vector<float> positiveFloats (numElements, 0); 1618 vector<float> negativeFloats (numElements, 0); 1619 const StringTemplate shaderTemplate ( 1620 string(s_ShaderPreamble) + 1621 1622 "OpSource GLSL 430\n" 1623 "OpName %main \"main\"\n" 1624 "OpName %id \"gl_GlobalInvocationID\"\n" 1625 1626 "OpDecorate %id BuiltIn GlobalInvocationId\n" 1627 1628 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) + 1629 1630 "${TYPE}\n" 1631 1632 "%id = OpVariable %uvec3ptr Input\n" 1633 "%zero = OpConstant %i32 0\n" 1634 1635 "%main = OpFunction %void None %voidf\n" 1636 "%label = OpLabel\n" 1637 1638 "%undef = OpUndef %type\n" 1639 1640 "%idval = OpLoad %uvec3 %id\n" 1641 "%x = OpCompositeExtract %u32 %idval 0\n" 1642 1643 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n" 1644 "%inval = OpLoad %f32 %inloc\n" 1645 "%neg = OpFNegate %f32 %inval\n" 1646 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n" 1647 " OpStore %outloc %neg\n" 1648 " OpReturn\n" 1649 " OpFunctionEnd\n"); 1650 1651 cases.push_back(CaseParameter("bool", "%type = OpTypeBool")); 1652 cases.push_back(CaseParameter("sint32", "%type = OpTypeInt 32 1")); 1653 cases.push_back(CaseParameter("uint32", "%type = OpTypeInt 32 0")); 1654 cases.push_back(CaseParameter("float32", "%type = OpTypeFloat 32")); 1655 cases.push_back(CaseParameter("vec4float32", "%type = OpTypeVector %f32 4")); 1656 cases.push_back(CaseParameter("vec2uint32", "%type = OpTypeVector %u32 2")); 1657 cases.push_back(CaseParameter("matrix", "%type = OpTypeMatrix %uvec3 3")); 1658 cases.push_back(CaseParameter("image", "%type = OpTypeImage %f32 2D 0 0 0 0 Unknown")); 1659 cases.push_back(CaseParameter("sampler", "%type = OpTypeSampler")); 1660 cases.push_back(CaseParameter("sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 0 Unknown\n" 1661 "%type = OpTypeSampledImage %img")); 1662 cases.push_back(CaseParameter("array", "%100 = OpConstant %u32 100\n" 1663 "%type = OpTypeArray %i32 %100")); 1664 cases.push_back(CaseParameter("runtimearray", "%type = OpTypeRuntimeArray %f32")); 1665 cases.push_back(CaseParameter("struct", "%type = OpTypeStruct %f32 %i32 %u32")); 1666 cases.push_back(CaseParameter("pointer", "%type = OpTypePointer Function %i32")); 1667 cases.push_back(CaseParameter("function", "%type = OpTypeFunction %void %i32 %f32")); 1668 1669 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements); 1670 1671 for (size_t ndx = 0; ndx < numElements; ++ndx) 1672 negativeFloats[ndx] = -positiveFloats[ndx]; 1673 1674 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx) 1675 { 1676 map<string, string> specializations; 1677 ComputeShaderSpec spec; 1678 1679 specializations["TYPE"] = cases[caseNdx].param; 1680 spec.assembly = shaderTemplate.specialize(specializations); 1681 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats))); 1682 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats))); 1683 spec.numWorkGroups = IVec3(numElements, 1, 1); 1684 1685 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec)); 1686 } 1687 1688 return group.release(); 1689} 1690 1691 1692typedef std::pair<std::string, VkShaderStage> EntryToStage; 1693typedef map<string, vector<EntryToStage> > ModuleMap; 1694 1695// Context for a specific test instantiation. For example, an instantiation 1696// may test colors yellow/magenta/cyan/mauve in a tesselation shader 1697// with an entry point named 'main_to_the_main' 1698struct InstanceContext 1699{ 1700 // Map of modules to what entry_points we care to use from those modules. 1701 ModuleMap moduleMap; 1702 RGBA inputColors[4]; 1703 RGBA outputColors[4]; 1704 // Concrete SPIR-V code to test via boilerplate specialization. 1705 map<string, string> testCodeFragments; 1706 1707 InstanceContext (const RGBA (&inputs)[4], const RGBA (&outputs)[4], const map<string, string>& testCodeFragments_) 1708 : testCodeFragments (testCodeFragments_) 1709 { 1710 inputColors[0] = inputs[0]; 1711 inputColors[1] = inputs[1]; 1712 inputColors[2] = inputs[2]; 1713 inputColors[3] = inputs[3]; 1714 1715 outputColors[0] = outputs[0]; 1716 outputColors[1] = outputs[1]; 1717 outputColors[2] = outputs[2]; 1718 outputColors[3] = outputs[3]; 1719 } 1720 1721 InstanceContext (const InstanceContext& other) 1722 : moduleMap (other.moduleMap) 1723 , testCodeFragments (other.testCodeFragments) 1724 { 1725 inputColors[0] = other.inputColors[0]; 1726 inputColors[1] = other.inputColors[1]; 1727 inputColors[2] = other.inputColors[2]; 1728 inputColors[3] = other.inputColors[3]; 1729 1730 outputColors[0] = other.outputColors[0]; 1731 outputColors[1] = other.outputColors[1]; 1732 outputColors[2] = other.outputColors[2]; 1733 outputColors[3] = other.outputColors[3]; 1734 } 1735}; 1736 1737// A description of a shader to be used for a single stage of the graphics pipeline. 1738struct ShaderElement 1739{ 1740 // The module that contains this shader entrypoint. 1741 const char* moduleName; 1742 1743 // The name of the entrypoint. 1744 const char* entryName; 1745 1746 // Which shader stage this entry point represents. 1747 VkShaderStage stage; 1748 1749 ShaderElement (const char* moduleName_, const char* entryPoint_, VkShaderStage shaderStage_) 1750 : moduleName(moduleName_) 1751 , entryName(entryPoint_) 1752 , stage(shaderStage_) 1753 { 1754 } 1755}; 1756 1757void getDefaultColors(RGBA (&colors)[4]) { 1758 colors[0] = RGBA::white(); 1759 colors[1] = RGBA::red(); 1760 colors[2] = RGBA::blue(); 1761 colors[3] = RGBA::green(); 1762} 1763 1764// Turns a statically sized array of ShaderElements into an instance-context 1765// by setting up the mapping of modules to their contained shaders and stages. 1766// The inputs and expected outputs are given by inputColors and outputColors 1767template<size_t N> 1768InstanceContext createInstanceContext (const ShaderElement (&elements)[N], const RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments) 1769{ 1770 InstanceContext ctx (inputColors, outputColors, testCodeFragments); 1771 for (size_t i = 0; i < N; ++i) 1772 { 1773 ctx.moduleMap[elements[i].moduleName].push_back(std::make_pair(elements[i].entryName, elements[i].stage)); 1774 } 1775 return ctx; 1776} 1777 1778// The same as createInstanceContext above, but with default colors. 1779template<size_t N> 1780InstanceContext createInstanceContext (const ShaderElement (&elements)[N], const map<string, string>& testCodeFragments) 1781{ 1782 RGBA defaultColors[4]; 1783 getDefaultColors(defaultColors); 1784 return createInstanceContext(elements, defaultColors, defaultColors, testCodeFragments); 1785} 1786 1787// For the current InstanceContext, constructs the required modules and shaders. 1788// Fills in the modules vector with all of the used modules, and stage_shaders with 1789// all stages and shaders that are to be used. 1790void createShaders (const DeviceInterface& vk, const VkDevice vkDevice, InstanceContext& instance, Context& context, vector<ModuleHandleSp>& modules, map<VkShaderStage, VkShaderSp>& stage_shaders) 1791{ 1792 for (ModuleMap::const_iterator moduleNdx = instance.moduleMap.begin(); moduleNdx != instance.moduleMap.end(); ++moduleNdx) 1793 { 1794 const ModuleHandleSp mod(new Unique<VkShaderModule>(createShaderModule(vk, vkDevice, context.getBinaryCollection().get(moduleNdx->first), 0))); 1795 modules.push_back(ModuleHandleSp(mod)); 1796 for (vector<EntryToStage>::const_iterator shaderNdx = moduleNdx->second.begin(); shaderNdx != moduleNdx->second.end(); ++shaderNdx) 1797 { 1798 const EntryToStage& stage = *shaderNdx; 1799 const VkShaderCreateInfo shaderParam = 1800 { 1801 VK_STRUCTURE_TYPE_SHADER_CREATE_INFO, // VkStructureType sType; 1802 DE_NULL, // const void* pNext; 1803 **modules.back(), // VkShaderModule module; 1804 stage.first.c_str(), // const char* pName; 1805 0u, // VkShaderCreateFlags flags; 1806 stage.second, // VkShaderStage stage; 1807 }; 1808 stage_shaders[stage.second] = VkShaderSp(new Unique<VkShader>(createShader(vk, vkDevice, &shaderParam))); 1809 } 1810 } 1811} 1812 1813#define SPIRV_ASSEMBLY_TYPES \ 1814 "%void = OpTypeVoid\n" \ 1815 "%bool = OpTypeBool\n" \ 1816 \ 1817 "%i32 = OpTypeInt 32 1\n" \ 1818 "%u32 = OpTypeInt 32 0\n" \ 1819 \ 1820 "%f32 = OpTypeFloat 32\n" \ 1821 "%v3f32 = OpTypeVector %f32 3\n" \ 1822 "%v4f32 = OpTypeVector %f32 4\n" \ 1823 \ 1824 "%v4f32_function = OpTypeFunction %v4f32 %v4f32\n" \ 1825 "%fun = OpTypeFunction %void\n" \ 1826 \ 1827 "%ip_f32 = OpTypePointer Input %f32\n" \ 1828 "%ip_i32 = OpTypePointer Input %i32\n" \ 1829 "%ip_v3f32 = OpTypePointer Input %v3f32\n" \ 1830 "%ip_v4f32 = OpTypePointer Input %v4f32\n" \ 1831 \ 1832 "%op_f32 = OpTypePointer Output %f32\n" \ 1833 "%op_v4f32 = OpTypePointer Output %v4f32\n" 1834 1835#define SPIRV_ASSEMBLY_CONSTANTS \ 1836 "%c_f32_1 = OpConstant %f32 1\n" \ 1837 "%c_i32_0 = OpConstant %i32 0\n" \ 1838 "%c_i32_1 = OpConstant %i32 1\n" \ 1839 "%c_i32_2 = OpConstant %i32 2\n" \ 1840 "%c_u32_0 = OpConstant %u32 0\n" \ 1841 "%c_u32_1 = OpConstant %u32 1\n" \ 1842 "%c_u32_2 = OpConstant %u32 2\n" \ 1843 "%c_u32_3 = OpConstant %u32 3\n" \ 1844 "%c_u32_32 = OpConstant %u32 32\n" \ 1845 "%c_u32_4 = OpConstant %u32 4\n" 1846 1847#define SPIRV_ASSEMBLY_ARRAYS \ 1848 "%a1f32 = OpTypeArray %f32 %c_u32_1\n" \ 1849 "%a2f32 = OpTypeArray %f32 %c_u32_2\n" \ 1850 "%a3v4f32 = OpTypeArray %v4f32 %c_u32_3\n" \ 1851 "%a4f32 = OpTypeArray %f32 %c_u32_4\n" \ 1852 "%a32v4f32 = OpTypeArray %v4f32 %c_u32_32\n" \ 1853 "%ip_a3v4f32 = OpTypePointer Input %a3v4f32\n" \ 1854 "%ip_a32v4f32 = OpTypePointer Input %a32v4f32\n" \ 1855 "%op_a2f32 = OpTypePointer Output %a2f32\n" \ 1856 "%op_a3v4f32 = OpTypePointer Output %a3v4f32\n" \ 1857 "%op_a4f32 = OpTypePointer Output %a4f32\n" 1858 1859// Creates vertex-shader assembly by specializing a boilerplate StringTemplate 1860// on fragments, which must (at least) map "testfun" to an OpFunction definition 1861// for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed 1862// with "BP_" to avoid collisions with fragments. 1863// 1864// It corresponds roughly to this GLSL: 1865//; 1866// layout(location = 0) in vec4 position; 1867// layout(location = 1) in vec4 color; 1868// layout(location = 1) out highp vec4 vtxColor; 1869// void main (void) { gl_Position = position; vtxColor = test_func(color); } 1870string makeVertexShaderAssembly(const map<string, string>& fragments) 1871{ 1872// \todo [2015-11-23 awoloszyn] Remove OpName once these have stabalized 1873// \todo [2015-11-23 awoloszyn] Remove Smooth decoration when we move to SPIR-V 1.0 1874 static const char vertexShaderBoilerplate[] = 1875 "OpCapability Shader\n" 1876 "OpMemoryModel Logical GLSL450\n" 1877 "OpEntryPoint Vertex %4 \"main\" %BP_Position %BP_vtxColor %BP_color " 1878 "%BP_vtxPosition %BP_vertex_id %BP_instance_id\n" 1879 "${debug:opt}\n" 1880 "OpName %main \"main\"\n" 1881 "OpName %BP_vtxPosition \"vtxPosition\"\n" 1882 "OpName %BP_Position \"position\"\n" 1883 "OpName %BP_vtxColor \"vtxColor\"\n" 1884 "OpName %BP_color \"color\"\n" 1885 "OpName %vertex_id \"gl_VertexID\"\n" 1886 "OpName %instance_id \"gl_InstanceID\"\n" 1887 "OpName %test_code \"testfun(vf4;\"\n" 1888 "OpDecorate %BP_vtxPosition Smooth\n" 1889 "OpDecorate %BP_vtxPosition Location 2\n" 1890 "OpDecorate %BP_Position Location 0\n" 1891 "OpDecorate %BP_vtxColor Smooth\n" 1892 "OpDecorate %BP_vtxColor Location 1\n" 1893 "OpDecorate %BP_color Location 1\n" 1894 "OpDecorate %BP_vertex_id BuiltIn VertexId\n" 1895 "OpDecorate %BP_instance_id BuiltIn InstanceId\n" 1896 SPIRV_ASSEMBLY_TYPES 1897 SPIRV_ASSEMBLY_CONSTANTS 1898 SPIRV_ASSEMBLY_ARRAYS 1899 "%BP_vtxPosition = OpVariable %op_v4f32 Output\n" 1900 "%BP_Position = OpVariable %ip_v4f32 Input\n" 1901 "%BP_vtxColor = OpVariable %op_v4f32 Output\n" 1902 "%BP_color = OpVariable %ip_v4f32 Input\n" 1903 "%BP_vertex_id = OpVariable %ip_i32 Input\n" 1904 "%BP_instance_id = OpVariable %ip_i32 Input\n" 1905 "%main = OpFunction %void None %fun\n" 1906 "%BP_label = OpLabel\n" 1907 "%BP_tmp_position = OpLoad %v4f32 %BP_Position\n" 1908 "OpStore %BP_vtxPosition %BP_tmp_position\n" 1909 "%BP_tmp_color = OpLoad %v4f32 %BP_color\n" 1910 "%BP_clr_transformed = OpFunctionCall %v4f32 %test_code %BP_tmp_color\n" 1911 "OpStore %BP_vtxColor %BP_clr_transformed\n" 1912 "OpReturn\n" 1913 "OpFunctionEnd\n" 1914 "${testfun}\n"; 1915 return tcu::StringTemplate(vertexShaderBoilerplate).specialize(fragments); 1916} 1917 1918// Creates tess-control-shader assembly by specializing a boilerplate 1919// StringTemplate on fragments, which must (at least) map "testfun" to an 1920// OpFunction definition for %test_code that takes and returns a %v4f32. 1921// Boilerplate IDs are prefixed with "BP_" to avoid collisions with fragments. 1922// 1923// It roughly corresponds to the following GLSL. 1924// 1925// #version 450 1926// layout(vertices = 3) out; 1927// layout(location = 1) in vec4 in_color[]; 1928// layout(location = 2) in vec4 in_position[]; 1929// layout(location = 1) out vec4 out_color[]; 1930// layout(location = 2) out vec4 out_position[]; 1931// 1932// void main() { 1933// out_color[gl_InvocationID] = testfun(in_color[gl_InvocationID]); 1934// out_position[gl_InvocationID] = in_position[gl_InvocationID]; 1935// if (gl_InvocationID == 0) { 1936// gl_TessLevelOuter[0] = 1.0; 1937// gl_TessLevelOuter[1] = 1.0; 1938// gl_TessLevelOuter[2] = 1.0; 1939// gl_TessLevelInner[0] = 1.0; 1940// } 1941// } 1942string makeTessControlShaderAssembly(const map<string, string>& fragments) 1943{ 1944 static const char tessControlShaderBoilerplate[] = 1945 "OpCapability Tessellation\n" 1946 "OpMemoryModel Logical GLSL450\n" 1947 "OpEntryPoint TessellationControl %BP_main \"main\" %BP_out_color %BP_gl_InvocationID %BP_in_color %BP_out_position %BP_in_position %BP_gl_TessLevelOuter %BP_gl_TessLevelInner\n" 1948 "OpExecutionMode %BP_main OutputVertices 3\n" 1949 "${debug:opt}\n" 1950 "OpName %BP_main \"main\"\n" 1951 "OpName %BP_out_color \"out_color\"\n" 1952 "OpName %BP_gl_InvocationID \"gl_InvocationID\"\n" 1953 "OpName %BP_in_color \"in_color\"\n" 1954 "OpName %BP_out_position \"out_position\"\n" 1955 "OpName %BP_in_position \"in_position\"\n" 1956 "OpName %BP_gl_TessLevelOuter \"gl_TessLevelOuter\"\n" 1957 "OpName %BP_gl_TessLevelInner \"gl_TessLevelInner\"\n" 1958 "OpName %test_code \"testfun(vf4;\"\n" 1959 "OpDecorate %BP_out_color Location 1\n" 1960 "OpDecorate %BP_gl_InvocationID BuiltIn InvocationId\n" 1961 "OpDecorate %BP_in_color Location 1\n" 1962 "OpDecorate %BP_out_position Location 2\n" 1963 "OpDecorate %BP_in_position Location 2\n" 1964 "OpDecorate %BP_gl_TessLevelOuter Patch\n" 1965 "OpDecorate %BP_gl_TessLevelOuter BuiltIn TessLevelOuter\n" 1966 "OpDecorate %BP_gl_TessLevelInner Patch\n" 1967 "OpDecorate %BP_gl_TessLevelInner BuiltIn TessLevelInner\n" 1968 SPIRV_ASSEMBLY_TYPES 1969 SPIRV_ASSEMBLY_CONSTANTS 1970 SPIRV_ASSEMBLY_ARRAYS 1971 "%BP_out_color = OpVariable %op_a3v4f32 Output\n" 1972 "%BP_gl_InvocationID = OpVariable %ip_i32 Input\n" 1973 "%BP_in_color = OpVariable %ip_a32v4f32 Input\n" 1974 "%BP_out_position = OpVariable %op_a3v4f32 Output\n" 1975 "%BP_in_position = OpVariable %ip_a32v4f32 Input\n" 1976 "%BP_gl_TessLevelOuter = OpVariable %op_a4f32 Output\n" 1977 "%BP_gl_TessLevelInner = OpVariable %op_a2f32 Output\n" 1978 1979 "%BP_main = OpFunction %void None %fun\n" 1980 "%BP_label = OpLabel\n" 1981 1982 "%BP_invocation_id = OpLoad %i32 %BP_gl_InvocationID\n" 1983 1984 "%BP_in_color_ptr = OpAccessChain %ip_v4f32 %BP_in_color %BP_invocation_id\n" 1985 "%BP_in_position_ptr = OpAccessChain %ip_v4f32 %BP_in_position %BP_invocation_id\n" 1986 1987 "%BP_in_color_val = OpLoad %v4f32 %BP_in_color_ptr\n" 1988 "%BP_in_position_val = OpLoad %v4f32 %BP_in_position_ptr\n" 1989 1990 "%BP_clr_transformed = OpFunctionCall %v4f32 %test_code %BP_in_color_val\n" 1991 1992 "%BP_out_color_ptr = OpAccessChain %op_v4f32 %BP_out_color %BP_invocation_id\n" 1993 "%BP_out_position_ptr = OpAccessChain %op_v4f32 %BP_out_position %BP_invocation_id\n" 1994 1995 "OpStore %BP_out_color_ptr %BP_clr_transformed\n" 1996 "OpStore %BP_out_position_ptr %BP_in_position_val\n" 1997 1998 "%BP_is_first_invocation = OpIEqual %bool %BP_invocation_id %c_i32_0\n" 1999 "OpSelectionMerge %BP_merge_label None\n" 2000 "OpBranchConditional %BP_is_first_invocation %BP_first_invocation %BP_merge_label\n" 2001 2002 "%BP_first_invocation = OpLabel\n" 2003 "%BP_tess_outer_0 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_0\n" 2004 "%BP_tess_outer_1 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_1\n" 2005 "%BP_tess_outer_2 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_2\n" 2006 "%BP_tess_inner = OpAccessChain %op_f32 %BP_gl_TessLevelInner %c_i32_0\n" 2007 2008 "OpStore %BP_tess_outer_0 %c_f32_1\n" 2009 "OpStore %BP_tess_outer_1 %c_f32_1\n" 2010 "OpStore %BP_tess_outer_2 %c_f32_1\n" 2011 "OpStore %BP_tess_inner %c_f32_1\n" 2012 2013 "OpBranch %BP_merge_label\n" 2014 "%BP_merge_label = OpLabel\n" 2015 "OpReturn\n" 2016 "OpFunctionEnd\n" 2017 "${testfun}\n"; 2018 return tcu::StringTemplate(tessControlShaderBoilerplate).specialize(fragments); 2019} 2020 2021// Creates tess-evaluation-shader assembly by specializing a boilerplate 2022// StringTemplate on fragments, which must (at least) map "testfun" to an 2023// OpFunction definition for %test_code that takes and returns a %v4f32. 2024// Boilerplate IDs are prefixed with "BP_" to avoid collisions with fragments. 2025// 2026// It roughly corresponds to the following glsl. 2027// 2028// #version 450 2029// 2030// layout(triangles, equal_spacing, ccw) in; 2031// layout(location = 1) in vec4 in_color[]; 2032// layout(location = 2) in vec4 in_position[]; 2033// layout(location = 1) out vec4 out_color; 2034// 2035// #define interpolate(val) 2036// vec4(gl_TessCoord.x) * val[0] + vec4(gl_TessCoord.y) * val[1] + 2037// vec4(gl_TessCoord.z) * val[2] 2038// 2039// void main() { 2040// gl_Position = vec4(gl_TessCoord.x) * in_position[0] + 2041// vec4(gl_TessCoord.y) * in_position[1] + 2042// vec4(gl_TessCoord.z) * in_position[2]; 2043// out_color = testfun(interpolate(in_color)); 2044// } 2045string makeTessEvalShaderAssembly(const map<string, string>& fragments) 2046{ 2047 static const char tessEvalBoilerplate[] = 2048 "OpCapability Tessellation\n" 2049 "OpMemoryModel Logical GLSL450\n" 2050 "OpEntryPoint TessellationEvaluation %BP_main \"main\" %BP_stream %BP_gl_tessCoord %BP_in_position %BP_out_color %BP_in_color \n" 2051 "OpExecutionMode %BP_main InputTriangles\n" 2052 "${debug:opt}\n" 2053 "OpName %BP_main \"main\"\n" 2054 "OpName %BP_per_vertex_out \"gl_PerVertex\"\n" 2055 "OpMemberName %BP_per_vertex_out 0 \"gl_Position\"\n" 2056 "OpMemberName %BP_per_vertex_out 1 \"gl_PointSize\"\n" 2057 "OpMemberName %BP_per_vertex_out 2 \"gl_ClipDistance\"\n" 2058 "OpMemberName %BP_per_vertex_out 3 \"gl_CullDistance\"\n" 2059 "OpName %BP_stream \"\"\n" 2060 "OpName %BP_gl_tessCoord \"gl_TessCoord\"\n" 2061 "OpName %BP_in_position \"in_position\"\n" 2062 "OpName %BP_out_color \"out_color\"\n" 2063 "OpName %BP_in_color \"in_color\"\n" 2064 "OpName %test_code \"testfun(vf4;\"\n" 2065 "OpMemberDecorate %BP_per_vertex_out 0 BuiltIn Position\n" 2066 "OpMemberDecorate %BP_per_vertex_out 1 BuiltIn PointSize\n" 2067 "OpMemberDecorate %BP_per_vertex_out 2 BuiltIn ClipDistance\n" 2068 "OpMemberDecorate %BP_per_vertex_out 3 BuiltIn CullDistance\n" 2069 "OpDecorate %BP_per_vertex_out Block\n" 2070 "OpDecorate %BP_gl_tessCoord BuiltIn TessCoord\n" 2071 "OpDecorate %BP_in_position Location 2\n" 2072 "OpDecorate %BP_out_color Location 1\n" 2073 "OpDecorate %BP_in_color Location 1\n" 2074 SPIRV_ASSEMBLY_TYPES 2075 SPIRV_ASSEMBLY_CONSTANTS 2076 SPIRV_ASSEMBLY_ARRAYS 2077 "%BP_per_vertex_out = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n" 2078 "%BP_op_per_vertex_out = OpTypePointer Output %BP_per_vertex_out\n" 2079 "%BP_stream = OpVariable %BP_op_per_vertex_out Output\n" 2080 "%BP_gl_tessCoord = OpVariable %ip_v3f32 Input\n" 2081 "%BP_in_position = OpVariable %ip_a32v4f32 Input\n" 2082 "%BP_out_color = OpVariable %op_v4f32 Output\n" 2083 "%BP_in_color = OpVariable %ip_a32v4f32 Input\n" 2084 "%BP_main = OpFunction %void None %fun\n" 2085 "%BP_label = OpLabel\n" 2086 "%BP_tc_0_ptr = OpAccessChain %ip_f32 %BP_gl_tessCoord %c_u32_0\n" 2087 "%BP_tc_1_ptr = OpAccessChain %ip_f32 %BP_gl_tessCoord %c_u32_1\n" 2088 "%BP_tc_2_ptr = OpAccessChain %ip_f32 %BP_gl_tessCoord %c_u32_2\n" 2089 2090 "%BP_tc_0 = OpLoad %f32 %BP_tc_0_ptr\n" 2091 "%BP_tc_1 = OpLoad %f32 %BP_tc_1_ptr\n" 2092 "%BP_tc_2 = OpLoad %f32 %BP_tc_2_ptr\n" 2093 2094 "%BP_in_pos_0_ptr = OpAccessChain %ip_v4f32 %BP_in_position %c_i32_0\n" 2095 "%BP_in_pos_1_ptr = OpAccessChain %ip_v4f32 %BP_in_position %c_i32_1\n" 2096 "%BP_in_pos_2_ptr = OpAccessChain %ip_v4f32 %BP_in_position %c_i32_2\n" 2097 2098 "%BP_in_pos_0 = OpLoad %v4f32 %BP_in_pos_0_ptr\n" 2099 "%BP_in_pos_1 = OpLoad %v4f32 %BP_in_pos_1_ptr\n" 2100 "%BP_in_pos_2 = OpLoad %v4f32 %BP_in_pos_2_ptr\n" 2101 2102 "%BP_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %BP_tc_0 %BP_in_pos_0\n" 2103 "%BP_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %BP_tc_1 %BP_in_pos_1\n" 2104 "%BP_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %BP_tc_2 %BP_in_pos_2\n" 2105 2106 "%BP_out_pos_ptr = OpAccessChain %op_v4f32 %BP_stream %c_i32_0\n" 2107 2108 "%BP_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %BP_in_pos_0_weighted %BP_in_pos_1_weighted\n" 2109 "%BP_computed_out = OpFAdd %v4f32 %BP_in_pos_0_plus_pos_1 %BP_in_pos_2_weighted\n" 2110 "OpStore %BP_out_pos_ptr %BP_computed_out\n" 2111 2112 "%BP_in_clr_0_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_0\n" 2113 "%BP_in_clr_1_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_1\n" 2114 "%BP_in_clr_2_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_2\n" 2115 2116 "%BP_in_clr_0 = OpLoad %v4f32 %BP_in_clr_0_ptr\n" 2117 "%BP_in_clr_1 = OpLoad %v4f32 %BP_in_clr_1_ptr\n" 2118 "%BP_in_clr_2 = OpLoad %v4f32 %BP_in_clr_2_ptr\n" 2119 2120 "%BP_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %BP_tc_0 %BP_in_clr_0\n" 2121 "%BP_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %BP_tc_1 %BP_in_clr_1\n" 2122 "%BP_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %BP_tc_2 %BP_in_clr_2\n" 2123 2124 "%BP_in_clr_0_plus_col_1 = OpFAdd %v4f32 %BP_in_clr_0_weighted %BP_in_clr_1_weighted\n" 2125 "%BP_computed_clr = OpFAdd %v4f32 %BP_in_clr_0_plus_col_1 %BP_in_clr_2_weighted\n" 2126 "%BP_clr_transformed = OpFunctionCall %v4f32 %test_code %BP_computed_clr\n" 2127 2128 "OpStore %BP_out_color %BP_clr_transformed\n" 2129 "OpReturn\n" 2130 "OpFunctionEnd\n" 2131 "${testfun}\n"; 2132 return tcu::StringTemplate(tessEvalBoilerplate).specialize(fragments); 2133} 2134 2135// Creates geometry-shader assembly by specializing a boilerplate StringTemplate 2136// on fragments, which must (at least) map "testfun" to an OpFunction definition 2137// for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed 2138// with "BP_" to avoid collisions with fragments. 2139// 2140// Derived from this GLSL: 2141// 2142// #version 450 2143// layout(triangles) in; 2144// layout(triangle_strip, max_vertices = 3) out; 2145// 2146// layout(location = 1) in vec4 in_color[]; 2147// layout(location = 1) out vec4 out_color; 2148// 2149// void main() { 2150// gl_Position = gl_in[0].gl_Position; 2151// out_color = test_fun(in_color[0]); 2152// EmitVertex(); 2153// gl_Position = gl_in[1].gl_Position; 2154// out_color = test_fun(in_color[1]); 2155// EmitVertex(); 2156// gl_Position = gl_in[2].gl_Position; 2157// out_color = test_fun(in_color[2]); 2158// EmitVertex(); 2159// EndPrimitive(); 2160// } 2161string makeGeometryShaderAssembly(const map<string, string>& fragments) 2162{ 2163 static const char geometryShaderBoilerplate[] = 2164 "OpCapability Geometry\n" 2165 "OpMemoryModel Logical GLSL450\n" 2166 "OpEntryPoint Geometry %BP_main \"main\" %BP_out_gl_position %BP_gl_in %BP_out_color %BP_in_color\n" 2167 "OpExecutionMode %BP_main InputTriangles\n" 2168 "OpExecutionMode %BP_main Invocations 0\n" 2169 "OpExecutionMode %BP_main OutputTriangleStrip\n" 2170 "OpExecutionMode %BP_main OutputVertices 3\n" 2171 "${debug:opt}\n" 2172 "OpName %BP_main \"main\"\n" 2173 "OpName %BP_per_vertex_in \"gl_PerVertex\"\n" 2174 "OpMemberName %BP_per_vertex_in 0 \"gl_Position\"\n" 2175 "OpMemberName %BP_per_vertex_in 1 \"gl_PointSize\"\n" 2176 "OpMemberName %BP_per_vertex_in 2 \"gl_ClipDistance\"\n" 2177 "OpMemberName %BP_per_vertex_in 3 \"gl_CullDistance\"\n" 2178 "OpName %BP_gl_in \"gl_in\"\n" 2179 "OpName %BP_out_color \"out_color\"\n" 2180 "OpName %BP_in_color \"in_color\"\n" 2181 "OpName %test_code \"testfun(vf4;\"\n" 2182 "OpDecorate %BP_out_gl_position BuiltIn Position\n" 2183 "OpMemberDecorate %BP_per_vertex_in 0 BuiltIn Position\n" 2184 "OpMemberDecorate %BP_per_vertex_in 1 BuiltIn PointSize\n" 2185 "OpMemberDecorate %BP_per_vertex_in 2 BuiltIn ClipDistance\n" 2186 "OpMemberDecorate %BP_per_vertex_in 3 BuiltIn CullDistance\n" 2187 "OpDecorate %BP_per_vertex_in Block\n" 2188 "OpDecorate %BP_out_color Location 1\n" 2189 "OpDecorate %BP_out_color Stream 0\n" 2190 "OpDecorate %BP_in_color Location 1\n" 2191 SPIRV_ASSEMBLY_TYPES 2192 SPIRV_ASSEMBLY_CONSTANTS 2193 SPIRV_ASSEMBLY_ARRAYS 2194 "%BP_per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n" 2195 "%BP_a3_per_vertex_in = OpTypeArray %BP_per_vertex_in %c_u32_3\n" 2196 "%BP_ip_a3_per_vertex_in = OpTypePointer Input %BP_a3_per_vertex_in\n" 2197 2198 "%BP_gl_in = OpVariable %BP_ip_a3_per_vertex_in Input\n" 2199 "%BP_out_color = OpVariable %op_v4f32 Output\n" 2200 "%BP_in_color = OpVariable %ip_a3v4f32 Input\n" 2201 "%BP_out_gl_position = OpVariable %op_v4f32 Output\n" 2202 2203 "%BP_main = OpFunction %void None %fun\n" 2204 "%BP_label = OpLabel\n" 2205 "%BP_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_0 %c_i32_0\n" 2206 "%BP_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_1 %c_i32_0\n" 2207 "%BP_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_2 %c_i32_0\n" 2208 2209 "%BP_in_position_0 = OpLoad %v4f32 %BP_gl_in_0_gl_position\n" 2210 "%BP_in_position_1 = OpLoad %v4f32 %BP_gl_in_1_gl_position\n" 2211 "%BP_in_position_2 = OpLoad %v4f32 %BP_gl_in_2_gl_position \n" 2212 2213 "%BP_in_color_0_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_0\n" 2214 "%BP_in_color_1_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_1\n" 2215 "%BP_in_color_2_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_2\n" 2216 2217 "%BP_in_color_0 = OpLoad %v4f32 %BP_in_color_0_ptr\n" 2218 "%BP_in_color_1 = OpLoad %v4f32 %BP_in_color_1_ptr\n" 2219 "%BP_in_color_2 = OpLoad %v4f32 %BP_in_color_2_ptr\n" 2220 2221 "%BP_transformed_in_color_0 = OpFunctionCall %v4f32 %test_code %BP_in_color_0\n" 2222 "%BP_transformed_in_color_1 = OpFunctionCall %v4f32 %test_code %BP_in_color_1\n" 2223 "%BP_transformed_in_color_2 = OpFunctionCall %v4f32 %test_code %BP_in_color_2\n" 2224 "OpStore %BP_out_gl_position %BP_in_position_0\n" 2225 "OpStore %BP_out_color %BP_transformed_in_color_0\n" 2226 "OpEmitVertex\n" 2227 2228 "OpStore %BP_out_gl_position %BP_in_position_1\n" 2229 "OpStore %BP_out_color %BP_transformed_in_color_1\n" 2230 "OpEmitVertex\n" 2231 2232 "OpStore %BP_out_gl_position %BP_in_position_2\n" 2233 "OpStore %BP_out_color %BP_transformed_in_color_2\n" 2234 "OpEmitVertex\n" 2235 2236 "OpEndPrimitive\n" 2237 "OpReturn\n" 2238 "OpFunctionEnd\n" 2239 "${testfun}\n"; 2240 return tcu::StringTemplate(geometryShaderBoilerplate).specialize(fragments); 2241} 2242 2243// Creates fragment-shader assembly by specializing a boilerplate StringTemplate 2244// on fragments, which must (at least) map "testfun" to an OpFunction definition 2245// for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed 2246// with "BP_" to avoid collisions with fragments. 2247// 2248// Derived from this GLSL: 2249// 2250// layout(location = 0) in highp vec4 vtxColor; 2251// layout(location = 1) out highp vec4 fragColor; 2252// highp vec4 testfun(highp vec4 x) { return x; } 2253// void main(void) { fragColor = testfun(vtxColor); } 2254// 2255// with modifications including passing vtxColor by value and ripping out 2256// testfun() definition. 2257string makeFragmentShaderAssembly(const map<string, string>& fragments) 2258{ 2259 static const char fragmentShaderBoilerplate[] = 2260 "OpCapability Shader\n" 2261 "OpMemoryModel Logical GLSL450\n" 2262 "OpEntryPoint Fragment %BP_main \"main\" %BP_vtxColor %BP_fragColor\n" 2263 "OpExecutionMode %BP_main OriginUpperLeft\n" 2264 "${debug:opt}\n" 2265 "OpName %BP_main \"main\"\n" 2266 "OpName %BP_fragColor \"fragColor\"\n" 2267 "OpName %BP_vtxColor \"vtxColor\"\n" 2268 "OpName %test_code \"testfun(vf4;\"\n" 2269 "OpDecorate %BP_fragColor Location 0\n" 2270 "OpDecorate %BP_vtxColor Smooth\n" 2271 "OpDecorate %BP_vtxColor Location 1\n" 2272 SPIRV_ASSEMBLY_TYPES 2273 SPIRV_ASSEMBLY_CONSTANTS 2274 SPIRV_ASSEMBLY_ARRAYS 2275 "%BP_fragColor = OpVariable %op_v4f32 Output\n" 2276 "%BP_vtxColor = OpVariable %ip_v4f32 Input\n" 2277 "%BP_main = OpFunction %void None %fun\n" 2278 "%BP_label_main = OpLabel\n" 2279 "%BP_tmp1 = OpLoad %v4f32 %BP_vtxColor\n" 2280 "%BP_tmp2 = OpFunctionCall %v4f32 %test_code %BP_tmp1\n" 2281 "OpStore %BP_fragColor %BP_tmp2\n" 2282 "OpReturn\n" 2283 "OpFunctionEnd\n" 2284 "${testfun}\n"; 2285 return tcu::StringTemplate(fragmentShaderBoilerplate).specialize(fragments); 2286} 2287 2288// Creates fragments that specialize into a simple pass-through shader (of any kind). 2289map<string, string> passthruFragments(void) 2290{ 2291 map<string, string> fragments; 2292 fragments["testfun"] = 2293 // A %test_code function that returns its argument unchanged. 2294 "%test_code = OpFunction %v4f32 None %v4f32_function\n" 2295 "%param1 = OpFunctionParameter %v4f32\n" 2296 "%label_testfun = OpLabel\n" 2297 "OpReturnValue %param1\n" 2298 "OpFunctionEnd\n"; 2299 return fragments; 2300} 2301 2302// Adds shader assembly text to dst.spirvAsmSources for all shader kinds. 2303// Vertex shader gets custom code from context, the rest are pass-through. 2304void addShaderCodeCustomVertex(vk::SourceCollections& dst, InstanceContext context) { 2305 map<string, string> passthru = passthruFragments(); 2306 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(context.testCodeFragments); 2307 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(passthru); 2308 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(passthru); 2309 dst.spirvAsmSources.add("geom") << makeGeometryShaderAssembly(passthru); 2310 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru); 2311} 2312 2313// Adds shader assembly text to dst.spirvAsmSources for all shader kinds. 2314// Tessellation control shader gets custom code from context, the rest are 2315// pass-through. 2316void addShaderCodeCustomTessControl(vk::SourceCollections& dst, InstanceContext context) { 2317 map<string, string> passthru = passthruFragments(); 2318 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru); 2319 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(context.testCodeFragments); 2320 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(passthru); 2321 dst.spirvAsmSources.add("geom") << makeGeometryShaderAssembly(passthru); 2322 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru); 2323} 2324 2325// Adds shader assembly text to dst.spirvAsmSources for all shader kinds. 2326// Tessellation evaluation shader gets custom code from context, the rest are 2327// pass-through. 2328void addShaderCodeCustomTessEval(vk::SourceCollections& dst, InstanceContext context) { 2329 map<string, string> passthru = passthruFragments(); 2330 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru); 2331 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(passthru); 2332 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(context.testCodeFragments); 2333 dst.spirvAsmSources.add("geom") << makeGeometryShaderAssembly(passthru); 2334 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru); 2335} 2336 2337// Adds shader assembly text to dst.spirvAsmSources for all shader kinds. 2338// Geometry shader gets custom code from context, the rest are pass-through. 2339void addShaderCodeCustomGeometry(vk::SourceCollections& dst, InstanceContext context) { 2340 map<string, string> passthru = passthruFragments(); 2341 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru); 2342 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(passthru); 2343 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(passthru); 2344 dst.spirvAsmSources.add("geom") << makeGeometryShaderAssembly(context.testCodeFragments); 2345 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru); 2346} 2347 2348// Adds shader assembly text to dst.spirvAsmSources for all shader kinds. 2349// Fragment shader gets custom code from context, the rest are pass-through. 2350void addShaderCodeCustomFragment(vk::SourceCollections& dst, InstanceContext context) { 2351 map<string, string> passthru = passthruFragments(); 2352 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru); 2353 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(passthru); 2354 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(passthru); 2355 dst.spirvAsmSources.add("geom") << makeGeometryShaderAssembly(passthru); 2356 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(context.testCodeFragments); 2357} 2358 2359// Sets up and runs a Vulkan pipeline, then spot-checks the resulting image. 2360// Feeds the pipeline a set of colored triangles, which then must occur in the 2361// rendered image. The surface is cleared before executing the pipeline, so 2362// whatever the shaders draw can be directly spot-checked. 2363TestStatus runAndVerifyDefaultPipeline (Context& context, InstanceContext instance) 2364{ 2365 const VkDevice vkDevice = context.getDevice(); 2366 const DeviceInterface& vk = context.getDeviceInterface(); 2367 const VkQueue queue = context.getUniversalQueue(); 2368 const deUint32 queueFamilyIndex = context.getUniversalQueueFamilyIndex(); 2369 const tcu::IVec2 renderSize (256, 256); 2370 vector<ModuleHandleSp> modules; 2371 map<VkShaderStage, VkShaderSp> shaders; 2372 const int testSpecificSeed = 31354125; 2373 const int seed = context.getTestContext().getCommandLine().getBaseSeed() ^ testSpecificSeed; 2374 de::Random(seed).shuffle(instance.inputColors, instance.inputColors+4); 2375 de::Random(seed).shuffle(instance.outputColors, instance.outputColors+4); 2376 const Vec4 vertexData[] = 2377 { 2378 // Upper left corner: 2379 Vec4(-1.0f, -1.0f, 0.0f, 1.0f), instance.inputColors[0].toVec(), 2380 Vec4(-0.5f, -1.0f, 0.0f, 1.0f), instance.inputColors[0].toVec(), 2381 Vec4(-1.0f, -0.5f, 0.0f, 1.0f), instance.inputColors[0].toVec(), 2382 2383 // Upper right corner: 2384 Vec4(+0.5f, -1.0f, 0.0f, 1.0f), instance.inputColors[1].toVec(), 2385 Vec4(+1.0f, -1.0f, 0.0f, 1.0f), instance.inputColors[1].toVec(), 2386 Vec4(+1.0f, -0.5f, 0.0f, 1.0f), instance.inputColors[1].toVec(), 2387 2388 // Lower left corner: 2389 Vec4(-1.0f, +0.5f, 0.0f, 1.0f), instance.inputColors[2].toVec(), 2390 Vec4(-0.5f, +1.0f, 0.0f, 1.0f), instance.inputColors[2].toVec(), 2391 Vec4(-1.0f, +1.0f, 0.0f, 1.0f), instance.inputColors[2].toVec(), 2392 2393 // Lower right corner: 2394 Vec4(+1.0f, +0.5f, 0.0f, 1.0f), instance.inputColors[3].toVec(), 2395 Vec4(+1.0f, +1.0f, 0.0f, 1.0f), instance.inputColors[3].toVec(), 2396 Vec4(+0.5f, +1.0f, 0.0f, 1.0f), instance.inputColors[3].toVec() 2397 }; 2398 const size_t singleVertexDataSize = 2 * sizeof(Vec4); 2399 const size_t vertexCount = sizeof(vertexData) / singleVertexDataSize; 2400 2401 const VkBufferCreateInfo vertexBufferParams = 2402 { 2403 VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType; 2404 DE_NULL, // const void* pNext; 2405 (VkDeviceSize)sizeof(vertexData), // VkDeviceSize size; 2406 VK_BUFFER_USAGE_VERTEX_BUFFER_BIT, // VkBufferUsageFlags usage; 2407 0u, // VkBufferCreateFlags flags; 2408 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode; 2409 1u, // deUint32 queueFamilyCount; 2410 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices; 2411 }; 2412 const Unique<VkBuffer> vertexBuffer (createBuffer(vk, vkDevice, &vertexBufferParams)); 2413 const UniquePtr<Allocation> vertexBufferMemory (context.getDefaultAllocator().allocate(getBufferMemoryRequirements(vk, vkDevice, *vertexBuffer), MemoryRequirement::HostVisible)); 2414 2415 VK_CHECK(vk.bindBufferMemory(vkDevice, *vertexBuffer, vertexBufferMemory->getMemory(), vertexBufferMemory->getOffset())); 2416 2417 const VkDeviceSize imageSizeBytes = (VkDeviceSize)(sizeof(deUint32)*renderSize.x()*renderSize.y()); 2418 const VkBufferCreateInfo readImageBufferParams = 2419 { 2420 VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType; 2421 DE_NULL, // const void* pNext; 2422 imageSizeBytes, // VkDeviceSize size; 2423 VK_BUFFER_USAGE_TRANSFER_DESTINATION_BIT, // VkBufferUsageFlags usage; 2424 0u, // VkBufferCreateFlags flags; 2425 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode; 2426 1u, // deUint32 queueFamilyCount; 2427 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices; 2428 }; 2429 const Unique<VkBuffer> readImageBuffer (createBuffer(vk, vkDevice, &readImageBufferParams)); 2430 const UniquePtr<Allocation> readImageBufferMemory (context.getDefaultAllocator().allocate(getBufferMemoryRequirements(vk, vkDevice, *readImageBuffer), MemoryRequirement::HostVisible)); 2431 2432 VK_CHECK(vk.bindBufferMemory(vkDevice, *readImageBuffer, readImageBufferMemory->getMemory(), readImageBufferMemory->getOffset())); 2433 2434 const VkImageCreateInfo imageParams = 2435 { 2436 VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, // VkStructureType sType; 2437 DE_NULL, // const void* pNext; 2438 VK_IMAGE_TYPE_2D, // VkImageType imageType; 2439 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format; 2440 { renderSize.x(), renderSize.y(), 1 }, // VkExtent3D extent; 2441 1u, // deUint32 mipLevels; 2442 1u, // deUint32 arraySize; 2443 1u, // deUint32 samples; 2444 VK_IMAGE_TILING_OPTIMAL, // VkImageTiling tiling; 2445 VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT|VK_IMAGE_USAGE_TRANSFER_SOURCE_BIT, // VkImageUsageFlags usage; 2446 0u, // VkImageCreateFlags flags; 2447 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode; 2448 1u, // deUint32 queueFamilyCount; 2449 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices; 2450 VK_IMAGE_LAYOUT_UNDEFINED, // VkImageLayout initialLayout; 2451 }; 2452 2453 const Unique<VkImage> image (createImage(vk, vkDevice, &imageParams)); 2454 const UniquePtr<Allocation> imageMemory (context.getDefaultAllocator().allocate(getImageMemoryRequirements(vk, vkDevice, *image), MemoryRequirement::Any)); 2455 2456 VK_CHECK(vk.bindImageMemory(vkDevice, *image, imageMemory->getMemory(), imageMemory->getOffset())); 2457 2458 const VkAttachmentDescription colorAttDesc = 2459 { 2460 VK_STRUCTURE_TYPE_ATTACHMENT_DESCRIPTION, // VkStructureType sType; 2461 DE_NULL, // const void* pNext; 2462 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format; 2463 1u, // deUint32 samples; 2464 VK_ATTACHMENT_LOAD_OP_CLEAR, // VkAttachmentLoadOp loadOp; 2465 VK_ATTACHMENT_STORE_OP_STORE, // VkAttachmentStoreOp storeOp; 2466 VK_ATTACHMENT_LOAD_OP_DONT_CARE, // VkAttachmentLoadOp stencilLoadOp; 2467 VK_ATTACHMENT_STORE_OP_DONT_CARE, // VkAttachmentStoreOp stencilStoreOp; 2468 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout initialLayout; 2469 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout finalLayout; 2470 0u, // VkAttachmentDescriptionFlags flags; 2471 }; 2472 const VkAttachmentReference colorAttRef = 2473 { 2474 0u, // deUint32 attachment; 2475 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout layout; 2476 }; 2477 const VkSubpassDescription subpassDesc = 2478 { 2479 VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION, // VkStructureType sType; 2480 DE_NULL, // const void* pNext; 2481 VK_PIPELINE_BIND_POINT_GRAPHICS, // VkPipelineBindPoint pipelineBindPoint; 2482 0u, // VkSubpassDescriptionFlags flags; 2483 0u, // deUint32 inputCount; 2484 DE_NULL, // const VkAttachmentReference* pInputAttachments; 2485 1u, // deUint32 colorCount; 2486 &colorAttRef, // const VkAttachmentReference* pColorAttachments; 2487 DE_NULL, // const VkAttachmentReference* pResolveAttachments; 2488 { VK_NO_ATTACHMENT, VK_IMAGE_LAYOUT_GENERAL }, // VkAttachmentReference depthStencilAttachment; 2489 0u, // deUint32 preserveCount; 2490 DE_NULL, // const VkAttachmentReference* pPreserveAttachments; 2491 2492 }; 2493 const VkRenderPassCreateInfo renderPassParams = 2494 { 2495 VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO, // VkStructureType sType; 2496 DE_NULL, // const void* pNext; 2497 1u, // deUint32 attachmentCount; 2498 &colorAttDesc, // const VkAttachmentDescription* pAttachments; 2499 1u, // deUint32 subpassCount; 2500 &subpassDesc, // const VkSubpassDescription* pSubpasses; 2501 0u, // deUint32 dependencyCount; 2502 DE_NULL, // const VkSubpassDependency* pDependencies; 2503 }; 2504 const Unique<VkRenderPass> renderPass (createRenderPass(vk, vkDevice, &renderPassParams)); 2505 2506 const VkImageViewCreateInfo colorAttViewParams = 2507 { 2508 VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, // VkStructureType sType; 2509 DE_NULL, // const void* pNext; 2510 *image, // VkImage image; 2511 VK_IMAGE_VIEW_TYPE_2D, // VkImageViewType viewType; 2512 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format; 2513 { 2514 VK_CHANNEL_SWIZZLE_R, 2515 VK_CHANNEL_SWIZZLE_G, 2516 VK_CHANNEL_SWIZZLE_B, 2517 VK_CHANNEL_SWIZZLE_A 2518 }, // VkChannelMapping channels; 2519 { 2520 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspectFlags aspectMask; 2521 0u, // deUint32 baseMipLevel; 2522 1u, // deUint32 mipLevels; 2523 0u, // deUint32 baseArrayLayer; 2524 1u, // deUint32 arraySize; 2525 }, // VkImageSubresourceRange subresourceRange; 2526 0u, // VkImageViewCreateFlags flags; 2527 }; 2528 const Unique<VkImageView> colorAttView (createImageView(vk, vkDevice, &colorAttViewParams)); 2529 2530 createShaders(vk, vkDevice, instance, context, modules, shaders); 2531 2532 // Pipeline layout 2533 const VkPipelineLayoutCreateInfo pipelineLayoutParams = 2534 { 2535 VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, // VkStructureType sType; 2536 DE_NULL, // const void* pNext; 2537 0u, // deUint32 descriptorSetCount; 2538 DE_NULL, // const VkDescriptorSetLayout* pSetLayouts; 2539 0u, // deUint32 pushConstantRangeCount; 2540 DE_NULL, // const VkPushConstantRange* pPushConstantRanges; 2541 }; 2542 const Unique<VkPipelineLayout> pipelineLayout (createPipelineLayout(vk, vkDevice, &pipelineLayoutParams)); 2543 2544 // Pipeline 2545 const VkSpecializationInfo emptyShaderSpecParams = 2546 { 2547 0u, // deUint32 mapEntryCount; 2548 DE_NULL, // const VkSpecializationMapEntry* pMap; 2549 0, // const deUintptr dataSize; 2550 DE_NULL, // const void* pData; 2551 }; 2552 vector<VkPipelineShaderStageCreateInfo> shaderStageParams; 2553 for(map<VkShaderStage, VkShaderSp>::const_iterator stage = shaders.begin(); stage != shaders.end(); ++stage) { 2554 VkPipelineShaderStageCreateInfo info = { 2555 VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, // VkStructureType sType; 2556 DE_NULL, // const void* pNext; 2557 stage->first, // VkShaderStage stage; 2558 **stage->second, // VkShader shader; 2559 &emptyShaderSpecParams, // const VkSpecializationInfo* pSpecializationInfo; 2560 }; 2561 shaderStageParams.push_back(info); 2562 } 2563 const VkPipelineDepthStencilStateCreateInfo depthStencilParams = 2564 { 2565 VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO, // VkStructureType sType; 2566 DE_NULL, // const void* pNext; 2567 DE_FALSE, // deUint32 depthTestEnable; 2568 DE_FALSE, // deUint32 depthWriteEnable; 2569 VK_COMPARE_OP_ALWAYS, // VkCompareOp depthCompareOp; 2570 DE_FALSE, // deUint32 depthBoundsTestEnable; 2571 DE_FALSE, // deUint32 stencilTestEnable; 2572 { 2573 VK_STENCIL_OP_KEEP, // VkStencilOp stencilFailOp; 2574 VK_STENCIL_OP_KEEP, // VkStencilOp stencilPassOp; 2575 VK_STENCIL_OP_KEEP, // VkStencilOp stencilDepthFailOp; 2576 VK_COMPARE_OP_ALWAYS, // VkCompareOp stencilCompareOp; 2577 0u, // deUint32 stencilCompareMask; 2578 0u, // deUint32 stencilWriteMask; 2579 0u, // deUint32 stencilReference; 2580 }, // VkStencilOpState front; 2581 { 2582 VK_STENCIL_OP_KEEP, // VkStencilOp stencilFailOp; 2583 VK_STENCIL_OP_KEEP, // VkStencilOp stencilPassOp; 2584 VK_STENCIL_OP_KEEP, // VkStencilOp stencilDepthFailOp; 2585 VK_COMPARE_OP_ALWAYS, // VkCompareOp stencilCompareOp; 2586 0u, // deUint32 stencilCompareMask; 2587 0u, // deUint32 stencilWriteMask; 2588 0u, // deUint32 stencilReference; 2589 }, // VkStencilOpState back; 2590 -1.0f, // float minDepthBounds; 2591 +1.0f, // float maxDepthBounds; 2592 }; 2593 const VkViewport viewport0 = 2594 { 2595 0.0f, // float originX; 2596 0.0f, // float originY; 2597 (float)renderSize.x(), // float width; 2598 (float)renderSize.y(), // float height; 2599 0.0f, // float minDepth; 2600 1.0f, // float maxDepth; 2601 }; 2602 const VkRect2D scissor0 = 2603 { 2604 { 2605 0u, // deInt32 x; 2606 0u, // deInt32 y; 2607 }, // VkOffset2D offset; 2608 { 2609 renderSize.x(), // deInt32 width; 2610 renderSize.y(), // deInt32 height; 2611 }, // VkExtent2D extent; 2612 }; 2613 const VkPipelineViewportStateCreateInfo viewportParams = 2614 { 2615 VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO, // VkStructureType sType; 2616 DE_NULL, // const void* pNext; 2617 1u, // deUint32 viewportCount; 2618 &viewport0, 2619 1u, 2620 &scissor0 2621 }; 2622 const VkSampleMask sampleMask = ~0u; 2623 const VkPipelineMultisampleStateCreateInfo multisampleParams = 2624 { 2625 VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO, // VkStructureType sType; 2626 DE_NULL, // const void* pNext; 2627 1u, // deUint32 rasterSamples; 2628 DE_FALSE, // deUint32 sampleShadingEnable; 2629 0.0f, // float minSampleShading; 2630 &sampleMask, // VkSampleMask sampleMask; 2631 }; 2632 const VkPipelineRasterStateCreateInfo rasterParams = 2633 { 2634 VK_STRUCTURE_TYPE_PIPELINE_RASTER_STATE_CREATE_INFO, // VkStructureType sType; 2635 DE_NULL, // const void* pNext; 2636 DE_TRUE, // deUint32 depthClipEnable; 2637 DE_FALSE, // deUint32 rasterizerDiscardEnable; 2638 VK_FILL_MODE_SOLID, // VkFillMode fillMode; 2639 VK_CULL_MODE_NONE, // VkCullMode cullMode; 2640 VK_FRONT_FACE_CCW, // VkFrontFace frontFace; 2641 VK_FALSE, // VkBool32 depthBiasEnable; 2642 0.0f, // float depthBias; 2643 0.0f, // float depthBiasClamp; 2644 0.0f, // float slopeScaledDepthBias; 2645 1.0f, // float lineWidth; 2646 }; 2647 const VkPipelineInputAssemblyStateCreateInfo inputAssemblyParams = 2648 { 2649 VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO, // VkStructureType sType; 2650 DE_NULL, // const void* pNext; 2651 VK_PRIMITIVE_TOPOLOGY_PATCH, // VkPrimitiveTopology topology; 2652 DE_FALSE, // deUint32 primitiveRestartEnable; 2653 }; 2654 const VkVertexInputBindingDescription vertexBinding0 = 2655 { 2656 0u, // deUint32 binding; 2657 deUint32(singleVertexDataSize), // deUint32 strideInBytes; 2658 VK_VERTEX_INPUT_STEP_RATE_VERTEX // VkVertexInputStepRate stepRate; 2659 }; 2660 const VkVertexInputAttributeDescription vertexAttrib0[2] = 2661 { 2662 { 2663 0u, // deUint32 location; 2664 0u, // deUint32 binding; 2665 VK_FORMAT_R32G32B32A32_SFLOAT, // VkFormat format; 2666 0u // deUint32 offsetInBytes; 2667 }, 2668 { 2669 1u, // deUint32 location; 2670 0u, // deUint32 binding; 2671 VK_FORMAT_R32G32B32A32_SFLOAT, // VkFormat format; 2672 sizeof(Vec4), // deUint32 offsetInBytes; 2673 } 2674 }; 2675 2676 const VkPipelineVertexInputStateCreateInfo vertexInputStateParams = 2677 { 2678 VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO, // VkStructureType sType; 2679 DE_NULL, // const void* pNext; 2680 1u, // deUint32 bindingCount; 2681 &vertexBinding0, // const VkVertexInputBindingDescription* pVertexBindingDescriptions; 2682 2u, // deUint32 attributeCount; 2683 vertexAttrib0, // const VkVertexInputAttributeDescription* pVertexAttributeDescriptions; 2684 }; 2685 const VkPipelineColorBlendAttachmentState attBlendParams = 2686 { 2687 DE_FALSE, // deUint32 blendEnable; 2688 VK_BLEND_ONE, // VkBlend srcBlendColor; 2689 VK_BLEND_ZERO, // VkBlend destBlendColor; 2690 VK_BLEND_OP_ADD, // VkBlendOp blendOpColor; 2691 VK_BLEND_ONE, // VkBlend srcBlendAlpha; 2692 VK_BLEND_ZERO, // VkBlend destBlendAlpha; 2693 VK_BLEND_OP_ADD, // VkBlendOp blendOpAlpha; 2694 VK_CHANNEL_R_BIT|VK_CHANNEL_G_BIT|VK_CHANNEL_B_BIT|VK_CHANNEL_A_BIT, // VkChannelFlags channelWriteMask; 2695 }; 2696 const VkPipelineColorBlendStateCreateInfo blendParams = 2697 { 2698 VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO, // VkStructureType sType; 2699 DE_NULL, // const void* pNext; 2700 DE_FALSE, // VkBool32 alphaToCoverageEnable; 2701 DE_FALSE, // VkBool32 alphaToOneEnable; 2702 DE_FALSE, // VkBool32 logicOpEnable; 2703 VK_LOGIC_OP_COPY, // VkLogicOp logicOp; 2704 1u, // deUint32 attachmentCount; 2705 &attBlendParams, // const VkPipelineColorBlendAttachmentState* pAttachments; 2706 { 0.0f, 0.0f, 0.0f, 0.0f }, // float blendConst[4]; 2707 }; 2708 const VkPipelineDynamicStateCreateInfo dynamicStateInfo = 2709 { 2710 VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO, // VkStructureType sType; 2711 DE_NULL, // const void* pNext; 2712 0u, // deUint32 dynamicStateCount; 2713 DE_NULL // const VkDynamicState* pDynamicStates; 2714 }; 2715 2716 const VkPipelineTessellationStateCreateInfo tessellationState = 2717 { 2718 VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO, 2719 DE_NULL, 2720 3u 2721 }; 2722 2723 const VkGraphicsPipelineCreateInfo pipelineParams = 2724 { 2725 VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, // VkStructureType sType; 2726 DE_NULL, // const void* pNext; 2727 (deUint32)shaderStageParams.size(), // deUint32 stageCount; 2728 &shaderStageParams[0], // const VkPipelineShaderStageCreateInfo* pStages; 2729 &vertexInputStateParams, // const VkPipelineVertexInputStateCreateInfo* pVertexInputState; 2730 &inputAssemblyParams, // const VkPipelineInputAssemblyStateCreateInfo* pInputAssemblyState; 2731 &tessellationState, // const VkPipelineTessellationStateCreateInfo* pTessellationState; 2732 &viewportParams, // const VkPipelineViewportStateCreateInfo* pViewportState; 2733 &rasterParams, // const VkPipelineRasterStateCreateInfo* pRasterState; 2734 &multisampleParams, // const VkPipelineMultisampleStateCreateInfo* pMultisampleState; 2735 &depthStencilParams, // const VkPipelineDepthStencilStateCreateInfo* pDepthStencilState; 2736 &blendParams, // const VkPipelineColorBlendStateCreateInfo* pColorBlendState; 2737 &dynamicStateInfo, // const VkPipelineDynamicStateCreateInfo* pDynamicState; 2738 0u, // VkPipelineCreateFlags flags; 2739 *pipelineLayout, // VkPipelineLayout layout; 2740 *renderPass, // VkRenderPass renderPass; 2741 0u, // deUint32 subpass; 2742 DE_NULL, // VkPipeline basePipelineHandle; 2743 0u, // deInt32 basePipelineIndex; 2744 }; 2745 2746 const Unique<VkPipeline> pipeline (createGraphicsPipeline(vk, vkDevice, DE_NULL, &pipelineParams)); 2747 2748 // Framebuffer 2749 const VkFramebufferCreateInfo framebufferParams = 2750 { 2751 VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO, // VkStructureType sType; 2752 DE_NULL, // const void* pNext; 2753 *renderPass, // VkRenderPass renderPass; 2754 1u, // deUint32 attachmentCount; 2755 &*colorAttView, // const VkImageView* pAttachments; 2756 (deUint32)renderSize.x(), // deUint32 width; 2757 (deUint32)renderSize.y(), // deUint32 height; 2758 1u, // deUint32 layers; 2759 }; 2760 const Unique<VkFramebuffer> framebuffer (createFramebuffer(vk, vkDevice, &framebufferParams)); 2761 2762 const VkCmdPoolCreateInfo cmdPoolParams = 2763 { 2764 VK_STRUCTURE_TYPE_CMD_POOL_CREATE_INFO, // VkStructureType sType; 2765 DE_NULL, // const void* pNext; 2766 queueFamilyIndex, // deUint32 queueFamilyIndex; 2767 VK_CMD_POOL_CREATE_RESET_COMMAND_BUFFER_BIT // VkCmdPoolCreateFlags flags; 2768 }; 2769 const Unique<VkCmdPool> cmdPool (createCommandPool(vk, vkDevice, &cmdPoolParams)); 2770 2771 // Command buffer 2772 const VkCmdBufferCreateInfo cmdBufParams = 2773 { 2774 VK_STRUCTURE_TYPE_CMD_BUFFER_CREATE_INFO, // VkStructureType sType; 2775 DE_NULL, // const void* pNext; 2776 *cmdPool, // VkCmdPool pool; 2777 VK_CMD_BUFFER_LEVEL_PRIMARY, // VkCmdBufferLevel level; 2778 0u, // VkCmdBufferCreateFlags flags; 2779 }; 2780 const Unique<VkCmdBuffer> cmdBuf (createCommandBuffer(vk, vkDevice, &cmdBufParams)); 2781 2782 const VkCmdBufferBeginInfo cmdBufBeginParams = 2783 { 2784 VK_STRUCTURE_TYPE_CMD_BUFFER_BEGIN_INFO, // VkStructureType sType; 2785 DE_NULL, // const void* pNext; 2786 0u, // VkCmdBufferOptimizeFlags flags; 2787 DE_NULL, // VkRenderPass renderPass; 2788 0u, // deUint32 subpass; 2789 DE_NULL, // VkFramebuffer framebuffer; 2790 }; 2791 2792 // Record commands 2793 VK_CHECK(vk.beginCommandBuffer(*cmdBuf, &cmdBufBeginParams)); 2794 2795 { 2796 const VkMemoryBarrier vertFlushBarrier = 2797 { 2798 VK_STRUCTURE_TYPE_MEMORY_BARRIER, // VkStructureType sType; 2799 DE_NULL, // const void* pNext; 2800 VK_MEMORY_OUTPUT_HOST_WRITE_BIT, // VkMemoryOutputFlags outputMask; 2801 VK_MEMORY_INPUT_VERTEX_ATTRIBUTE_FETCH_BIT, // VkMemoryInputFlags inputMask; 2802 }; 2803 const VkImageMemoryBarrier colorAttBarrier = 2804 { 2805 VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType; 2806 DE_NULL, // const void* pNext; 2807 0u, // VkMemoryOutputFlags outputMask; 2808 VK_MEMORY_INPUT_COLOR_ATTACHMENT_BIT, // VkMemoryInputFlags inputMask; 2809 VK_IMAGE_LAYOUT_UNDEFINED, // VkImageLayout oldLayout; 2810 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout newLayout; 2811 queueFamilyIndex, // deUint32 srcQueueFamilyIndex; 2812 queueFamilyIndex, // deUint32 destQueueFamilyIndex; 2813 *image, // VkImage image; 2814 { 2815 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspect aspect; 2816 0u, // deUint32 baseMipLevel; 2817 1u, // deUint32 mipLevels; 2818 0u, // deUint32 baseArraySlice; 2819 1u, // deUint32 arraySize; 2820 } // VkImageSubresourceRange subresourceRange; 2821 }; 2822 const void* barriers[] = { &vertFlushBarrier, &colorAttBarrier }; 2823 vk.cmdPipelineBarrier(*cmdBuf, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_ALL_GPU_COMMANDS, DE_FALSE, (deUint32)DE_LENGTH_OF_ARRAY(barriers), barriers); 2824 } 2825 2826 { 2827 const VkClearValue clearValue = makeClearValueColorF32(0.125f, 0.25f, 0.75f, 1.0f); 2828 const VkRenderPassBeginInfo passBeginParams = 2829 { 2830 VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO, // VkStructureType sType; 2831 DE_NULL, // const void* pNext; 2832 *renderPass, // VkRenderPass renderPass; 2833 *framebuffer, // VkFramebuffer framebuffer; 2834 { { 0, 0 }, { renderSize.x(), renderSize.y() } }, // VkRect2D renderArea; 2835 1u, // deUint32 clearValueCount; 2836 &clearValue, // const VkClearValue* pClearValues; 2837 }; 2838 vk.cmdBeginRenderPass(*cmdBuf, &passBeginParams, VK_RENDER_PASS_CONTENTS_INLINE); 2839 } 2840 2841 vk.cmdBindPipeline(*cmdBuf, VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline); 2842 { 2843 const VkDeviceSize bindingOffset = 0; 2844 vk.cmdBindVertexBuffers(*cmdBuf, 0u, 1u, &vertexBuffer.get(), &bindingOffset); 2845 } 2846 vk.cmdDraw(*cmdBuf, deUint32(vertexCount), 1u /*run pipeline once*/, 0u /*first vertex*/, 0u /*first instanceIndex*/); 2847 vk.cmdEndRenderPass(*cmdBuf); 2848 2849 { 2850 const VkImageMemoryBarrier renderFinishBarrier = 2851 { 2852 VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType; 2853 DE_NULL, // const void* pNext; 2854 VK_MEMORY_OUTPUT_COLOR_ATTACHMENT_BIT, // VkMemoryOutputFlags outputMask; 2855 VK_MEMORY_INPUT_TRANSFER_BIT, // VkMemoryInputFlags inputMask; 2856 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout oldLayout; 2857 VK_IMAGE_LAYOUT_TRANSFER_SOURCE_OPTIMAL, // VkImageLayout newLayout; 2858 queueFamilyIndex, // deUint32 srcQueueFamilyIndex; 2859 queueFamilyIndex, // deUint32 destQueueFamilyIndex; 2860 *image, // VkImage image; 2861 { 2862 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspectFlags aspectMask; 2863 0u, // deUint32 baseMipLevel; 2864 1u, // deUint32 mipLevels; 2865 0u, // deUint32 baseArraySlice; 2866 1u, // deUint32 arraySize; 2867 } // VkImageSubresourceRange subresourceRange; 2868 }; 2869 const void* barriers[] = { &renderFinishBarrier }; 2870 vk.cmdPipelineBarrier(*cmdBuf, VK_PIPELINE_STAGE_ALL_GRAPHICS, VK_PIPELINE_STAGE_TRANSFER_BIT, DE_FALSE, (deUint32)DE_LENGTH_OF_ARRAY(barriers), barriers); 2871 } 2872 2873 { 2874 const VkBufferImageCopy copyParams = 2875 { 2876 (VkDeviceSize)0u, // VkDeviceSize bufferOffset; 2877 (deUint32)renderSize.x(), // deUint32 bufferRowLength; 2878 (deUint32)renderSize.y(), // deUint32 bufferImageHeight; 2879 { 2880 VK_IMAGE_ASPECT_COLOR, // VkImageAspect aspect; 2881 0u, // deUint32 mipLevel; 2882 0u, // deUint32 arrayLayer; 2883 1u, // deUint32 arraySize; 2884 }, // VkImageSubresourceCopy imageSubresource; 2885 { 0u, 0u, 0u }, // VkOffset3D imageOffset; 2886 { renderSize.x(), renderSize.y(), 1u } // VkExtent3D imageExtent; 2887 }; 2888 vk.cmdCopyImageToBuffer(*cmdBuf, *image, VK_IMAGE_LAYOUT_TRANSFER_SOURCE_OPTIMAL, *readImageBuffer, 1u, ©Params); 2889 } 2890 2891 { 2892 const VkBufferMemoryBarrier copyFinishBarrier = 2893 { 2894 VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, // VkStructureType sType; 2895 DE_NULL, // const void* pNext; 2896 VK_MEMORY_OUTPUT_TRANSFER_BIT, // VkMemoryOutputFlags outputMask; 2897 VK_MEMORY_INPUT_HOST_READ_BIT, // VkMemoryInputFlags inputMask; 2898 queueFamilyIndex, // deUint32 srcQueueFamilyIndex; 2899 queueFamilyIndex, // deUint32 destQueueFamilyIndex; 2900 *readImageBuffer, // VkBuffer buffer; 2901 0u, // VkDeviceSize offset; 2902 imageSizeBytes // VkDeviceSize size; 2903 }; 2904 const void* barriers[] = { ©FinishBarrier }; 2905 vk.cmdPipelineBarrier(*cmdBuf, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, DE_FALSE, (deUint32)DE_LENGTH_OF_ARRAY(barriers), barriers); 2906 } 2907 2908 VK_CHECK(vk.endCommandBuffer(*cmdBuf)); 2909 2910 // Upload vertex data 2911 { 2912 const VkMappedMemoryRange range = 2913 { 2914 VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, // VkStructureType sType; 2915 DE_NULL, // const void* pNext; 2916 vertexBufferMemory->getMemory(), // VkDeviceMemory mem; 2917 0, // VkDeviceSize offset; 2918 (VkDeviceSize)sizeof(vertexData), // VkDeviceSize size; 2919 }; 2920 void* vertexBufPtr = vertexBufferMemory->getHostPtr(); 2921 2922 deMemcpy(vertexBufPtr, &vertexData[0], sizeof(vertexData)); 2923 VK_CHECK(vk.flushMappedMemoryRanges(vkDevice, 1u, &range)); 2924 } 2925 2926 // Submit & wait for completion 2927 { 2928 const VkFenceCreateInfo fenceParams = 2929 { 2930 VK_STRUCTURE_TYPE_FENCE_CREATE_INFO, // VkStructureType sType; 2931 DE_NULL, // const void* pNext; 2932 0u, // VkFenceCreateFlags flags; 2933 }; 2934 const Unique<VkFence> fence (createFence(vk, vkDevice, &fenceParams)); 2935 2936 VK_CHECK(vk.queueSubmit(queue, 1u, &cmdBuf.get(), *fence)); 2937 VK_CHECK(vk.waitForFences(vkDevice, 1u, &fence.get(), DE_TRUE, ~0ull)); 2938 } 2939 2940 const void* imagePtr = readImageBufferMemory->getHostPtr(); 2941 const tcu::ConstPixelBufferAccess pixelBuffer(tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNORM_INT8), 2942 renderSize.x(), renderSize.y(), 1, imagePtr); 2943 // Log image 2944 { 2945 const VkMappedMemoryRange range = 2946 { 2947 VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, // VkStructureType sType; 2948 DE_NULL, // const void* pNext; 2949 readImageBufferMemory->getMemory(), // VkDeviceMemory mem; 2950 0, // VkDeviceSize offset; 2951 imageSizeBytes, // VkDeviceSize size; 2952 }; 2953 2954 VK_CHECK(vk.invalidateMappedMemoryRanges(vkDevice, 1u, &range)); 2955 context.getTestContext().getLog() << TestLog::Image("Result", "Result", pixelBuffer); 2956 } 2957 2958 const RGBA threshold(1, 1, 1, 1); 2959 const RGBA upperLeft(pixelBuffer.getPixel(1, 1)); 2960 if (!tcu::compareThreshold(upperLeft, instance.outputColors[0], threshold)) 2961 return TestStatus::fail("Upper left corner mismatch"); 2962 2963 const RGBA upperRight(pixelBuffer.getPixel(pixelBuffer.getWidth() - 1, 1)); 2964 if (!tcu::compareThreshold(upperRight, instance.outputColors[1], threshold)) 2965 return TestStatus::fail("Upper right corner mismatch"); 2966 2967 const RGBA lowerLeft(pixelBuffer.getPixel(1, pixelBuffer.getHeight() - 1)); 2968 if (!tcu::compareThreshold(lowerLeft, instance.outputColors[2], threshold)) 2969 return TestStatus::fail("Lower left corner mismatch"); 2970 2971 const RGBA lowerRight(pixelBuffer.getPixel(pixelBuffer.getWidth() - 1, pixelBuffer.getHeight() - 1)); 2972 if (!tcu::compareThreshold(lowerRight, instance.outputColors[3], threshold)) 2973 return TestStatus::fail("Lower right corner mismatch"); 2974 2975 return TestStatus::pass("Rendered output matches input"); 2976} 2977 2978void createTestsForAllStages(const std::string& name, 2979 const RGBA (&inputColors)[4], 2980 const RGBA (&outputColors)[4], 2981 const map<string, string>& testCodeFragments, 2982 tcu::TestCaseGroup* tests) 2983{ 2984 const ShaderElement pipelineStages[] = 2985 { 2986 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX), 2987 ShaderElement("tessc", "main", VK_SHADER_STAGE_TESS_CONTROL), 2988 ShaderElement("tesse", "main", VK_SHADER_STAGE_TESS_EVALUATION), 2989 ShaderElement("geom", "main", VK_SHADER_STAGE_GEOMETRY), 2990 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT), 2991 }; 2992 2993 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "-vert", "", addShaderCodeCustomVertex, runAndVerifyDefaultPipeline, 2994 createInstanceContext(pipelineStages, inputColors, outputColors, testCodeFragments)); 2995 2996 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "-tessc", "", addShaderCodeCustomTessControl, runAndVerifyDefaultPipeline, 2997 createInstanceContext(pipelineStages, inputColors, outputColors, testCodeFragments)); 2998 2999 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "-tesse", "", addShaderCodeCustomTessEval, runAndVerifyDefaultPipeline, 3000 createInstanceContext(pipelineStages, inputColors, outputColors, testCodeFragments)); 3001 3002 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "-geom", "", addShaderCodeCustomGeometry, runAndVerifyDefaultPipeline, 3003 createInstanceContext(pipelineStages, inputColors, outputColors, testCodeFragments)); 3004 3005 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "-frag", "", addShaderCodeCustomFragment, runAndVerifyDefaultPipeline, 3006 createInstanceContext(pipelineStages, inputColors, outputColors, testCodeFragments)); 3007} 3008} // anonymous 3009 3010tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx) 3011{ 3012 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands")); 3013 3014 instructionTests->addChild(createOpNopGroup(testCtx)); 3015 instructionTests->addChild(createOpLineGroup(testCtx)); 3016 instructionTests->addChild(createOpNoLineGroup(testCtx)); 3017 instructionTests->addChild(createOpConstantNullGroup(testCtx)); 3018 instructionTests->addChild(createOpConstantCompositeGroup(testCtx)); 3019 instructionTests->addChild(createOpConstantUsageGroup(testCtx)); 3020 instructionTests->addChild(createOpSourceGroup(testCtx)); 3021 instructionTests->addChild(createOpSourceExtensionGroup(testCtx)); 3022 instructionTests->addChild(createDecorationGroupGroup(testCtx)); 3023 instructionTests->addChild(createOpPhiGroup(testCtx)); 3024 instructionTests->addChild(createLoopControlGroup(testCtx)); 3025 instructionTests->addChild(createFunctionControlGroup(testCtx)); 3026 instructionTests->addChild(createSelectionControlGroup(testCtx)); 3027 instructionTests->addChild(createBlockOrderGroup(testCtx)); 3028 instructionTests->addChild(createOpUndefGroup(testCtx)); 3029 instructionTests->addChild(createOpUnreachableGroup(testCtx)); 3030 3031 RGBA defaultColors[4]; 3032 getDefaultColors(defaultColors); 3033 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "graphics-assembly", "Test the graphics pipeline")); 3034 createTestsForAllStages("passthru", defaultColors, defaultColors, passthruFragments(), group.get()); 3035 instructionTests->addChild(group.release()); 3036 return instructionTests.release(); 3037} 3038 3039} // SpirVAssembly 3040} // vkt 3041