1 /*-------------------------------------------------------------------------
2 * Vulkan Conformance Tests
3 * ------------------------
5 * Copyright (c) 2015 Google Inc.
7 * Licensed under the Apache License, Version 2.0 (the "License");
8 * you may not use this file except in compliance with the License.
9 * You may obtain a copy of the License at
11 * http://www.apache.org/licenses/LICENSE-2.0
13 * Unless required by applicable law or agreed to in writing, software
14 * distributed under the License is distributed on an "AS IS" BASIS,
15 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16 * See the License for the specific language governing permissions and
17 * limitations under the License.
21 * \brief SPIR-V Assembly Tests for Instructions (special opcode/operand)
22 *//*--------------------------------------------------------------------*/
24 #include "vktSpvAsmInstructionTests.hpp"
26 #include "tcuCommandLine.hpp"
27 #include "tcuFormatUtil.hpp"
28 #include "tcuFloat.hpp"
29 #include "tcuRGBA.hpp"
30 #include "tcuStringTemplate.hpp"
31 #include "tcuTestLog.hpp"
32 #include "tcuVectorUtil.hpp"
35 #include "vkDeviceUtil.hpp"
36 #include "vkMemUtil.hpp"
37 #include "vkPlatform.hpp"
38 #include "vkPrograms.hpp"
39 #include "vkQueryUtil.hpp"
41 #include "vkRefUtil.hpp"
42 #include "vkStrUtil.hpp"
43 #include "vkTypeUtil.hpp"
45 #include "deRandom.hpp"
46 #include "deStringUtil.hpp"
47 #include "deUniquePtr.hpp"
48 #include "tcuStringTemplate.hpp"
51 #include "vktSpvAsmComputeShaderCase.hpp"
52 #include "vktSpvAsmComputeShaderTestUtil.hpp"
53 #include "vktTestCaseUtil.hpp"
63 namespace SpirVAssembly
77 using tcu::TestStatus;
80 using tcu::StringTemplate;
83 typedef Unique<VkShaderModule> ModuleHandleUp;
84 typedef de::SharedPtr<ModuleHandleUp> ModuleHandleSp;
86 template<typename T> T randomScalar (de::Random& rnd, T minValue, T maxValue);
87 template<> inline float randomScalar (de::Random& rnd, float minValue, float maxValue) { return rnd.getFloat(minValue, maxValue); }
88 template<> inline deInt32 randomScalar (de::Random& rnd, deInt32 minValue, deInt32 maxValue) { return rnd.getInt(minValue, maxValue); }
91 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0)
93 T* const typedPtr = (T*)dst;
94 for (int ndx = 0; ndx < numValues; ndx++)
95 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue);
98 static void floorAll (vector<float>& values)
100 for (size_t i = 0; i < values.size(); i++)
101 values[i] = deFloatFloor(values[i]);
104 static void floorAll (vector<Vec4>& values)
106 for (size_t i = 0; i < values.size(); i++)
107 values[i] = floor(values[i]);
115 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {}
118 // Assembly code used for testing OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code:
122 // layout(std140, set = 0, binding = 0) readonly buffer Input {
125 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
129 // layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
132 // uint x = gl_GlobalInvocationID.x;
133 // output_data.elements[x] = -input_data.elements[x];
136 static const char* const s_ShaderPreamble =
137 "OpCapability Shader\n"
138 "OpMemoryModel Logical GLSL450\n"
139 "OpEntryPoint GLCompute %main \"main\" %id\n"
140 "OpExecutionMode %main LocalSize 1 1 1\n";
142 static const char* const s_CommonTypes =
143 "%bool = OpTypeBool\n"
144 "%void = OpTypeVoid\n"
145 "%voidf = OpTypeFunction %void\n"
146 "%u32 = OpTypeInt 32 0\n"
147 "%i32 = OpTypeInt 32 1\n"
148 "%f32 = OpTypeFloat 32\n"
149 "%uvec3 = OpTypeVector %u32 3\n"
150 "%fvec3 = OpTypeVector %f32 3\n"
151 "%uvec3ptr = OpTypePointer Input %uvec3\n"
152 "%f32ptr = OpTypePointer Uniform %f32\n"
153 "%boolptr = OpTypePointer Uniform %bool\n"
154 "%f32arr = OpTypeRuntimeArray %f32\n"
155 "%boolarr = OpTypeRuntimeArray %bool\n";
157 // Declares two uniform variables (indata, outdata) of type "struct { float[] }". Depends on type "f32arr" (for "float[]").
158 static const char* const s_InputOutputBuffer =
159 "%buf = OpTypeStruct %f32arr\n"
160 "%bufptr = OpTypePointer Uniform %buf\n"
161 "%indata = OpVariable %bufptr Uniform\n"
162 "%outdata = OpVariable %bufptr Uniform\n";
164 // Declares buffer type and layout for uniform variables indata and outdata. Both of them are SSBO bounded to descriptor set 0.
165 // indata is at binding point 0, while outdata is at 1.
166 static const char* const s_InputOutputBufferTraits =
167 "OpDecorate %buf BufferBlock\n"
168 "OpDecorate %indata DescriptorSet 0\n"
169 "OpDecorate %indata Binding 0\n"
170 "OpDecorate %outdata DescriptorSet 0\n"
171 "OpDecorate %outdata Binding 1\n"
172 "OpDecorate %f32arr ArrayStride 4\n"
173 "OpMemberDecorate %buf 0 Offset 0\n";
175 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
177 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
178 ComputeShaderSpec spec;
179 de::Random rnd (deStringHash(group->getName()));
180 const int numElements = 100;
181 vector<float> positiveFloats (numElements, 0);
182 vector<float> negativeFloats (numElements, 0);
184 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
186 for (size_t ndx = 0; ndx < numElements; ++ndx)
187 negativeFloats[ndx] = -positiveFloats[ndx];
190 string(s_ShaderPreamble) +
192 "OpSource GLSL 430\n"
193 "OpName %main \"main\"\n"
194 "OpName %id \"gl_GlobalInvocationID\"\n"
196 "OpDecorate %id BuiltIn GlobalInvocationId\n"
198 + string(s_InputOutputBufferTraits) + string(s_CommonTypes)
200 + string(s_InputOutputBuffer) +
202 "%id = OpVariable %uvec3ptr Input\n"
203 "%zero = OpConstant %i32 0\n"
205 "%main = OpFunction %void None %voidf\n"
207 "%idval = OpLoad %uvec3 %id\n"
208 "%x = OpCompositeExtract %u32 %idval 0\n"
210 " OpNop\n" // Inside a function body
212 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
213 "%inval = OpLoad %f32 %inloc\n"
214 "%neg = OpFNegate %f32 %inval\n"
215 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
216 " OpStore %outloc %neg\n"
219 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
220 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
221 spec.numWorkGroups = IVec3(numElements, 1, 1);
223 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
225 return group.release();
228 bool compareFUnord (const std::vector<BufferSp>& inputs, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
230 if (outputAllocs.size() != 1)
233 const BufferSp& expectedOutput = expectedOutputs[0];
234 const VkBool32* expectedOutputAsBool = static_cast<const VkBool32*>(expectedOutputs[0]->data());
235 const VkBool32* outputAsBool = static_cast<const VkBool32*>(outputAllocs[0]->getHostPtr());
236 const float* input1AsFloat = static_cast<const float*>(inputs[0]->data());
237 const float* input2AsFloat = static_cast<const float*>(inputs[1]->data());
238 bool returnValue = true;
240 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(VkBool32); ++idx)
242 if (outputAsBool[idx] != expectedOutputAsBool[idx])
244 log << TestLog::Message << "ERROR: Sub-case failed. inputs: " << input1AsFloat[idx] << "," << input2AsFloat[idx] << " output: " << outputAsBool[idx]<< " expected output: " << expectedOutputAsBool[idx] << TestLog::EndMessage;
251 typedef VkBool32 (*compareFuncType) (float, float);
257 compareFuncType compareFunc;
259 OpFUnordCase (const char* _name, const char* _opCode, compareFuncType _compareFunc)
262 , compareFunc (_compareFunc) {}
265 #define ADD_OPFUNORD_CASE(NAME, OPCODE, OPERATOR) \
267 struct compare_##NAME { static VkBool32 compare(float x, float y) { return (x OPERATOR y) ? VK_TRUE : VK_FALSE; } }; \
268 cases.push_back(OpFUnordCase(#NAME, OPCODE, compare_##NAME::compare)); \
269 } while (deGetFalse())
271 tcu::TestCaseGroup* createOpFUnordGroup (tcu::TestContext& testCtx)
273 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfunord", "Test the OpFUnord* opcodes"));
274 de::Random rnd (deStringHash(group->getName()));
275 const int numElements = 100;
276 vector<OpFUnordCase> cases;
278 const StringTemplate shaderTemplate (
280 string(s_ShaderPreamble) +
282 "OpSource GLSL 430\n"
283 "OpName %main \"main\"\n"
284 "OpName %id \"gl_GlobalInvocationID\"\n"
286 "OpDecorate %id BuiltIn GlobalInvocationId\n"
288 "OpDecorate %buf BufferBlock\n"
289 "OpDecorate %buf2 BufferBlock\n"
290 "OpDecorate %indata1 DescriptorSet 0\n"
291 "OpDecorate %indata1 Binding 0\n"
292 "OpDecorate %indata2 DescriptorSet 0\n"
293 "OpDecorate %indata2 Binding 1\n"
294 "OpDecorate %outdata DescriptorSet 0\n"
295 "OpDecorate %outdata Binding 2\n"
296 "OpDecorate %f32arr ArrayStride 4\n"
297 "OpDecorate %boolarr ArrayStride 4\n"
298 "OpMemberDecorate %buf 0 Offset 0\n"
299 "OpMemberDecorate %buf2 0 Offset 0\n"
301 + string(s_CommonTypes) +
303 "%buf = OpTypeStruct %f32arr\n"
304 "%bufptr = OpTypePointer Uniform %buf\n"
305 "%indata1 = OpVariable %bufptr Uniform\n"
306 "%indata2 = OpVariable %bufptr Uniform\n"
308 "%buf2 = OpTypeStruct %boolarr\n"
309 "%buf2ptr = OpTypePointer Uniform %buf2\n"
310 "%outdata = OpVariable %buf2ptr Uniform\n"
312 "%id = OpVariable %uvec3ptr Input\n"
313 "%zero = OpConstant %i32 0\n"
314 "%constf1 = OpConstant %f32 1.0\n"
316 "%main = OpFunction %void None %voidf\n"
318 "%idval = OpLoad %uvec3 %id\n"
319 "%x = OpCompositeExtract %u32 %idval 0\n"
321 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
322 "%inval1 = OpLoad %f32 %inloc1\n"
323 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
324 "%inval2 = OpLoad %f32 %inloc2\n"
325 "%outloc = OpAccessChain %boolptr %outdata %zero %x\n"
327 "%result = ${OPCODE} %bool %inval1 %inval2\n"
328 " OpStore %outloc %result\n"
333 ADD_OPFUNORD_CASE(equal, "OpFUnordEqual", ==);
334 ADD_OPFUNORD_CASE(less, "OpFUnordLessThan", <);
335 ADD_OPFUNORD_CASE(lessequal, "OpFUnordLessThanEqual", <=);
336 ADD_OPFUNORD_CASE(greater, "OpFUnordGreaterThan", >);
337 ADD_OPFUNORD_CASE(greaterequal, "OpFUnordGreaterThanEqual", >=);
338 ADD_OPFUNORD_CASE(notequal, "OpFUnordNotEqual", !=);
340 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
342 map<string, string> specializations;
343 ComputeShaderSpec spec;
344 const float NaN = std::numeric_limits<float>::quiet_NaN();
345 vector<float> inputFloats1 (numElements, 0);
346 vector<float> inputFloats2 (numElements, 0);
347 vector<VkBool32> expectedBools (numElements, VK_FALSE);
349 specializations["OPCODE"] = cases[caseNdx].opCode;
350 spec.assembly = shaderTemplate.specialize(specializations);
352 fillRandomScalars(rnd, 1.f, 100.f, &inputFloats1[0], numElements);
353 for (size_t ndx = 0; ndx < numElements; ++ndx)
357 case 0: inputFloats2[ndx] = inputFloats1[ndx] + 1.0f; break;
358 case 1: inputFloats2[ndx] = inputFloats1[ndx] - 1.0f; break;
359 case 2: inputFloats2[ndx] = inputFloats1[ndx]; break;
360 case 3: inputFloats2[ndx] = NaN; break;
361 case 4: inputFloats2[ndx] = inputFloats1[ndx]; inputFloats1[ndx] = NaN; break;
362 case 5: inputFloats2[ndx] = NaN; inputFloats1[ndx] = NaN; break;
364 expectedBools[ndx] = tcu::Float32(inputFloats1[ndx]).isNaN() || tcu::Float32(inputFloats2[ndx]).isNaN() || cases[caseNdx].compareFunc(inputFloats1[ndx], inputFloats2[ndx]);
367 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
368 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
369 spec.outputs.push_back(BufferSp(new BoolBuffer(expectedBools)));
370 spec.numWorkGroups = IVec3(numElements, 1, 1);
371 spec.verifyIO = &compareFUnord;
372 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
375 return group.release();
378 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
380 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
381 ComputeShaderSpec spec;
382 de::Random rnd (deStringHash(group->getName()));
383 const int numElements = 100;
384 vector<float> positiveFloats (numElements, 0);
385 vector<float> negativeFloats (numElements, 0);
387 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
389 for (size_t ndx = 0; ndx < numElements; ++ndx)
390 negativeFloats[ndx] = -positiveFloats[ndx];
393 string(s_ShaderPreamble) +
395 "%fname1 = OpString \"negateInputs.comp\"\n"
396 "%fname2 = OpString \"negateInputs\"\n"
398 "OpSource GLSL 430\n"
399 "OpName %main \"main\"\n"
400 "OpName %id \"gl_GlobalInvocationID\"\n"
402 "OpDecorate %id BuiltIn GlobalInvocationId\n"
404 + string(s_InputOutputBufferTraits) +
406 "OpLine %fname1 0 0\n" // At the earliest possible position
408 + string(s_CommonTypes) + string(s_InputOutputBuffer) +
410 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
411 "OpLine %fname2 1 0\n" // Different filenames
412 "OpLine %fname1 1000 100000\n"
414 "%id = OpVariable %uvec3ptr Input\n"
415 "%zero = OpConstant %i32 0\n"
417 "OpLine %fname1 1 1\n" // Before a function
419 "%main = OpFunction %void None %voidf\n"
422 "OpLine %fname1 1 1\n" // In a function
424 "%idval = OpLoad %uvec3 %id\n"
425 "%x = OpCompositeExtract %u32 %idval 0\n"
426 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
427 "%inval = OpLoad %f32 %inloc\n"
428 "%neg = OpFNegate %f32 %inval\n"
429 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
430 " OpStore %outloc %neg\n"
433 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
434 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
435 spec.numWorkGroups = IVec3(numElements, 1, 1);
437 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
439 return group.release();
442 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
444 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
445 ComputeShaderSpec spec;
446 de::Random rnd (deStringHash(group->getName()));
447 const int numElements = 100;
448 vector<float> positiveFloats (numElements, 0);
449 vector<float> negativeFloats (numElements, 0);
451 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
453 for (size_t ndx = 0; ndx < numElements; ++ndx)
454 negativeFloats[ndx] = -positiveFloats[ndx];
457 string(s_ShaderPreamble) +
459 "%fname = OpString \"negateInputs.comp\"\n"
461 "OpSource GLSL 430\n"
462 "OpName %main \"main\"\n"
463 "OpName %id \"gl_GlobalInvocationID\"\n"
465 "OpDecorate %id BuiltIn GlobalInvocationId\n"
467 + string(s_InputOutputBufferTraits) +
469 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
471 + string(s_CommonTypes) + string(s_InputOutputBuffer) +
473 "OpLine %fname 0 1\n"
474 "OpNoLine\n" // Immediately following a preceding OpLine
476 "OpLine %fname 1000 1\n"
478 "%id = OpVariable %uvec3ptr Input\n"
479 "%zero = OpConstant %i32 0\n"
481 "OpNoLine\n" // Contents after the previous OpLine
483 "%main = OpFunction %void None %voidf\n"
485 "%idval = OpLoad %uvec3 %id\n"
486 "%x = OpCompositeExtract %u32 %idval 0\n"
488 "OpNoLine\n" // Multiple OpNoLine
492 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
493 "%inval = OpLoad %f32 %inloc\n"
494 "%neg = OpFNegate %f32 %inval\n"
495 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
496 " OpStore %outloc %neg\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);
503 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
505 return group.release();
508 // Compare instruction for the contraction compute case.
509 // Returns true if the output is what is expected from the test case.
510 bool compareNoContractCase(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
512 if (outputAllocs.size() != 1)
515 // We really just need this for size because we are not comparing the exact values.
516 const BufferSp& expectedOutput = expectedOutputs[0];
517 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
519 for(size_t i = 0; i < expectedOutput->getNumBytes() / sizeof(float); ++i) {
520 if (outputAsFloat[i] != 0.f &&
521 outputAsFloat[i] != -ldexp(1, -24)) {
529 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
531 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
532 vector<CaseParameter> cases;
533 const int numElements = 100;
534 vector<float> inputFloats1 (numElements, 0);
535 vector<float> inputFloats2 (numElements, 0);
536 vector<float> outputFloats (numElements, 0);
537 const StringTemplate shaderTemplate (
538 string(s_ShaderPreamble) +
540 "OpName %main \"main\"\n"
541 "OpName %id \"gl_GlobalInvocationID\"\n"
543 "OpDecorate %id BuiltIn GlobalInvocationId\n"
547 "OpDecorate %buf BufferBlock\n"
548 "OpDecorate %indata1 DescriptorSet 0\n"
549 "OpDecorate %indata1 Binding 0\n"
550 "OpDecorate %indata2 DescriptorSet 0\n"
551 "OpDecorate %indata2 Binding 1\n"
552 "OpDecorate %outdata DescriptorSet 0\n"
553 "OpDecorate %outdata Binding 2\n"
554 "OpDecorate %f32arr ArrayStride 4\n"
555 "OpMemberDecorate %buf 0 Offset 0\n"
557 + string(s_CommonTypes) +
559 "%buf = OpTypeStruct %f32arr\n"
560 "%bufptr = OpTypePointer Uniform %buf\n"
561 "%indata1 = OpVariable %bufptr Uniform\n"
562 "%indata2 = OpVariable %bufptr Uniform\n"
563 "%outdata = OpVariable %bufptr Uniform\n"
565 "%id = OpVariable %uvec3ptr Input\n"
566 "%zero = OpConstant %i32 0\n"
567 "%c_f_m1 = OpConstant %f32 -1.\n"
569 "%main = OpFunction %void None %voidf\n"
571 "%idval = OpLoad %uvec3 %id\n"
572 "%x = OpCompositeExtract %u32 %idval 0\n"
573 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
574 "%inval1 = OpLoad %f32 %inloc1\n"
575 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
576 "%inval2 = OpLoad %f32 %inloc2\n"
577 "%mul = OpFMul %f32 %inval1 %inval2\n"
578 "%add = OpFAdd %f32 %mul %c_f_m1\n"
579 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
580 " OpStore %outloc %add\n"
584 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
585 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
586 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
588 for (size_t ndx = 0; ndx < numElements; ++ndx)
590 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
591 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
592 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
593 // conducted separately and the result is rounded to 1, or 0x1.fffffcp-1
594 // So the final result will be 0.f or 0x1p-24.
595 // If the operation is combined into a precise fused multiply-add, then the result would be
596 // 2^-46 (0xa8800000).
597 outputFloats[ndx] = 0.f;
600 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
602 map<string, string> specializations;
603 ComputeShaderSpec spec;
605 specializations["DECORATION"] = cases[caseNdx].param;
606 spec.assembly = shaderTemplate.specialize(specializations);
607 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
608 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
609 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
610 spec.numWorkGroups = IVec3(numElements, 1, 1);
611 // Check against the two possible answers based on rounding mode.
612 spec.verifyIO = &compareNoContractCase;
614 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
616 return group.release();
619 bool compareFRem(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
621 if (outputAllocs.size() != 1)
624 const BufferSp& expectedOutput = expectedOutputs[0];
625 const float *expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
626 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
628 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
630 const float f0 = expectedOutputAsFloat[idx];
631 const float f1 = outputAsFloat[idx];
632 // \todo relative error needs to be fairly high because FRem may be implemented as
633 // (roughly) frac(a/b)*b, so LSB errors can be magnified. But this should be fine for now.
634 if (deFloatAbs((f1 - f0) / f0) > 0.02)
641 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
643 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
644 ComputeShaderSpec spec;
645 de::Random rnd (deStringHash(group->getName()));
646 const int numElements = 200;
647 vector<float> inputFloats1 (numElements, 0);
648 vector<float> inputFloats2 (numElements, 0);
649 vector<float> outputFloats (numElements, 0);
651 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
652 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
654 for (size_t ndx = 0; ndx < numElements; ++ndx)
656 // Guard against divisors near zero.
657 if (std::fabs(inputFloats2[ndx]) < 1e-3)
658 inputFloats2[ndx] = 8.f;
660 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
661 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
665 string(s_ShaderPreamble) +
667 "OpName %main \"main\"\n"
668 "OpName %id \"gl_GlobalInvocationID\"\n"
670 "OpDecorate %id BuiltIn GlobalInvocationId\n"
672 "OpDecorate %buf BufferBlock\n"
673 "OpDecorate %indata1 DescriptorSet 0\n"
674 "OpDecorate %indata1 Binding 0\n"
675 "OpDecorate %indata2 DescriptorSet 0\n"
676 "OpDecorate %indata2 Binding 1\n"
677 "OpDecorate %outdata DescriptorSet 0\n"
678 "OpDecorate %outdata Binding 2\n"
679 "OpDecorate %f32arr ArrayStride 4\n"
680 "OpMemberDecorate %buf 0 Offset 0\n"
682 + string(s_CommonTypes) +
684 "%buf = OpTypeStruct %f32arr\n"
685 "%bufptr = OpTypePointer Uniform %buf\n"
686 "%indata1 = OpVariable %bufptr Uniform\n"
687 "%indata2 = OpVariable %bufptr Uniform\n"
688 "%outdata = OpVariable %bufptr Uniform\n"
690 "%id = OpVariable %uvec3ptr Input\n"
691 "%zero = OpConstant %i32 0\n"
693 "%main = OpFunction %void None %voidf\n"
695 "%idval = OpLoad %uvec3 %id\n"
696 "%x = OpCompositeExtract %u32 %idval 0\n"
697 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
698 "%inval1 = OpLoad %f32 %inloc1\n"
699 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
700 "%inval2 = OpLoad %f32 %inloc2\n"
701 "%rem = OpFRem %f32 %inval1 %inval2\n"
702 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
703 " OpStore %outloc %rem\n"
707 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
708 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
709 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
710 spec.numWorkGroups = IVec3(numElements, 1, 1);
711 spec.verifyIO = &compareFRem;
713 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
715 return group.release();
718 // Copy contents in the input buffer to the output buffer.
719 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
721 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
722 de::Random rnd (deStringHash(group->getName()));
723 const int numElements = 100;
725 // The following case adds vec4(0., 0.5, 1.5, 2.5) to each of the elements in the input buffer and writes output to the output buffer.
726 ComputeShaderSpec spec1;
727 vector<Vec4> inputFloats1 (numElements);
728 vector<Vec4> outputFloats1 (numElements);
730 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
732 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
733 floorAll(inputFloats1);
735 for (size_t ndx = 0; ndx < numElements; ++ndx)
736 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
739 string(s_ShaderPreamble) +
741 "OpName %main \"main\"\n"
742 "OpName %id \"gl_GlobalInvocationID\"\n"
744 "OpDecorate %id BuiltIn GlobalInvocationId\n"
745 "OpDecorate %vec4arr ArrayStride 16\n"
747 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
749 "%vec4 = OpTypeVector %f32 4\n"
750 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
751 "%vec4ptr_f = OpTypePointer Function %vec4\n"
752 "%vec4arr = OpTypeRuntimeArray %vec4\n"
753 "%buf = OpTypeStruct %vec4arr\n"
754 "%bufptr = OpTypePointer Uniform %buf\n"
755 "%indata = OpVariable %bufptr Uniform\n"
756 "%outdata = OpVariable %bufptr Uniform\n"
758 "%id = OpVariable %uvec3ptr Input\n"
759 "%zero = OpConstant %i32 0\n"
760 "%c_f_0 = OpConstant %f32 0.\n"
761 "%c_f_0_5 = OpConstant %f32 0.5\n"
762 "%c_f_1_5 = OpConstant %f32 1.5\n"
763 "%c_f_2_5 = OpConstant %f32 2.5\n"
764 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
766 "%main = OpFunction %void None %voidf\n"
768 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
769 "%idval = OpLoad %uvec3 %id\n"
770 "%x = OpCompositeExtract %u32 %idval 0\n"
771 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
772 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
773 " OpCopyMemory %v_vec4 %inloc\n"
774 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
775 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
776 " OpStore %outloc %add\n"
780 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
781 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
782 spec1.numWorkGroups = IVec3(numElements, 1, 1);
784 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
786 // The following case copies a float[100] variable from the input buffer to the output buffer.
787 ComputeShaderSpec spec2;
788 vector<float> inputFloats2 (numElements);
789 vector<float> outputFloats2 (numElements);
791 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
793 for (size_t ndx = 0; ndx < numElements; ++ndx)
794 outputFloats2[ndx] = inputFloats2[ndx];
797 string(s_ShaderPreamble) +
799 "OpName %main \"main\"\n"
800 "OpName %id \"gl_GlobalInvocationID\"\n"
802 "OpDecorate %id BuiltIn GlobalInvocationId\n"
803 "OpDecorate %f32arr100 ArrayStride 4\n"
805 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
807 "%hundred = OpConstant %u32 100\n"
808 "%f32arr100 = OpTypeArray %f32 %hundred\n"
809 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
810 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
811 "%buf = OpTypeStruct %f32arr100\n"
812 "%bufptr = OpTypePointer Uniform %buf\n"
813 "%indata = OpVariable %bufptr Uniform\n"
814 "%outdata = OpVariable %bufptr Uniform\n"
816 "%id = OpVariable %uvec3ptr Input\n"
817 "%zero = OpConstant %i32 0\n"
819 "%main = OpFunction %void None %voidf\n"
821 "%var = OpVariable %f32arr100ptr_f Function\n"
822 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
823 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
824 " OpCopyMemory %var %inarr\n"
825 " OpCopyMemory %outarr %var\n"
829 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
830 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
831 spec2.numWorkGroups = IVec3(1, 1, 1);
833 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
835 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
836 ComputeShaderSpec spec3;
837 vector<float> inputFloats3 (16);
838 vector<float> outputFloats3 (16);
840 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
842 for (size_t ndx = 0; ndx < 16; ++ndx)
843 outputFloats3[ndx] = inputFloats3[ndx];
846 string(s_ShaderPreamble) +
848 "OpName %main \"main\"\n"
849 "OpName %id \"gl_GlobalInvocationID\"\n"
851 "OpDecorate %id BuiltIn GlobalInvocationId\n"
852 "OpMemberDecorate %buf 0 Offset 0\n"
853 "OpMemberDecorate %buf 1 Offset 16\n"
854 "OpMemberDecorate %buf 2 Offset 32\n"
855 "OpMemberDecorate %buf 3 Offset 48\n"
857 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
859 "%vec4 = OpTypeVector %f32 4\n"
860 "%buf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
861 "%bufptr = OpTypePointer Uniform %buf\n"
862 "%indata = OpVariable %bufptr Uniform\n"
863 "%outdata = OpVariable %bufptr Uniform\n"
864 "%vec4stptr = OpTypePointer Function %buf\n"
866 "%id = OpVariable %uvec3ptr Input\n"
867 "%zero = OpConstant %i32 0\n"
869 "%main = OpFunction %void None %voidf\n"
871 "%var = OpVariable %vec4stptr Function\n"
872 " OpCopyMemory %var %indata\n"
873 " OpCopyMemory %outdata %var\n"
877 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
878 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
879 spec3.numWorkGroups = IVec3(1, 1, 1);
881 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
883 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
884 ComputeShaderSpec spec4;
885 vector<float> inputFloats4 (numElements);
886 vector<float> outputFloats4 (numElements);
888 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
890 for (size_t ndx = 0; ndx < numElements; ++ndx)
891 outputFloats4[ndx] = -inputFloats4[ndx];
894 string(s_ShaderPreamble) +
896 "OpName %main \"main\"\n"
897 "OpName %id \"gl_GlobalInvocationID\"\n"
899 "OpDecorate %id BuiltIn GlobalInvocationId\n"
901 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
903 "%f32ptr_f = OpTypePointer Function %f32\n"
904 "%id = OpVariable %uvec3ptr Input\n"
905 "%zero = OpConstant %i32 0\n"
907 "%main = OpFunction %void None %voidf\n"
909 "%var = OpVariable %f32ptr_f Function\n"
910 "%idval = OpLoad %uvec3 %id\n"
911 "%x = OpCompositeExtract %u32 %idval 0\n"
912 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
913 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
914 " OpCopyMemory %var %inloc\n"
915 "%val = OpLoad %f32 %var\n"
916 "%neg = OpFNegate %f32 %val\n"
917 " OpStore %outloc %neg\n"
921 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
922 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
923 spec4.numWorkGroups = IVec3(numElements, 1, 1);
925 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
927 return group.release();
930 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
932 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
933 ComputeShaderSpec spec;
934 de::Random rnd (deStringHash(group->getName()));
935 const int numElements = 100;
936 vector<float> inputFloats (numElements, 0);
937 vector<float> outputFloats (numElements, 0);
939 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
941 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
942 floorAll(inputFloats);
944 for (size_t ndx = 0; ndx < numElements; ++ndx)
945 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
948 string(s_ShaderPreamble) +
950 "OpName %main \"main\"\n"
951 "OpName %id \"gl_GlobalInvocationID\"\n"
953 "OpDecorate %id BuiltIn GlobalInvocationId\n"
955 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
957 "%fmat = OpTypeMatrix %fvec3 3\n"
958 "%three = OpConstant %u32 3\n"
959 "%farr = OpTypeArray %f32 %three\n"
960 "%fst = OpTypeStruct %f32 %f32\n"
962 + string(s_InputOutputBuffer) +
964 "%id = OpVariable %uvec3ptr Input\n"
965 "%zero = OpConstant %i32 0\n"
966 "%c_f = OpConstant %f32 1.5\n"
967 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
968 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
969 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
970 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
972 "%main = OpFunction %void None %voidf\n"
974 "%c_f_copy = OpCopyObject %f32 %c_f\n"
975 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
976 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
977 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
978 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
979 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
980 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
981 "%farr_elem = OpCompositeExtract %f32 %c_farr_copy 2\n"
982 "%fst_elem = OpCompositeExtract %f32 %c_fst_copy 1\n"
983 // Add up. 1.5 * 5 = 7.5.
984 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
985 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
986 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
987 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
989 "%idval = OpLoad %uvec3 %id\n"
990 "%x = OpCompositeExtract %u32 %idval 0\n"
991 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
992 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
993 "%inval = OpLoad %f32 %inloc\n"
994 "%add = OpFAdd %f32 %add4 %inval\n"
995 " OpStore %outloc %add\n"
998 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
999 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1000 spec.numWorkGroups = IVec3(numElements, 1, 1);
1002 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
1004 return group.release();
1006 // Assembly code used for testing OpUnreachable is based on GLSL source code:
1010 // layout(std140, set = 0, binding = 0) readonly buffer Input {
1011 // float elements[];
1013 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
1014 // float elements[];
1017 // void not_called_func() {
1018 // // place OpUnreachable here
1021 // uint modulo4(uint val) {
1022 // switch (val % uint(4)) {
1023 // case 0: return 3;
1024 // case 1: return 2;
1025 // case 2: return 1;
1026 // case 3: return 0;
1027 // default: return 100; // place OpUnreachable here
1033 // // place OpUnreachable here
1037 // uint x = gl_GlobalInvocationID.x;
1038 // if (const5() > modulo4(1000)) {
1039 // output_data.elements[x] = -input_data.elements[x];
1041 // // place OpUnreachable here
1042 // output_data.elements[x] = input_data.elements[x];
1046 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
1048 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
1049 ComputeShaderSpec spec;
1050 de::Random rnd (deStringHash(group->getName()));
1051 const int numElements = 100;
1052 vector<float> positiveFloats (numElements, 0);
1053 vector<float> negativeFloats (numElements, 0);
1055 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
1057 for (size_t ndx = 0; ndx < numElements; ++ndx)
1058 negativeFloats[ndx] = -positiveFloats[ndx];
1061 string(s_ShaderPreamble) +
1063 "OpSource GLSL 430\n"
1064 "OpName %main \"main\"\n"
1065 "OpName %func_not_called_func \"not_called_func(\"\n"
1066 "OpName %func_modulo4 \"modulo4(u1;\"\n"
1067 "OpName %func_const5 \"const5(\"\n"
1068 "OpName %id \"gl_GlobalInvocationID\"\n"
1070 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1072 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1074 "%u32ptr = OpTypePointer Function %u32\n"
1075 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
1076 "%unitf = OpTypeFunction %u32\n"
1078 "%id = OpVariable %uvec3ptr Input\n"
1079 "%zero = OpConstant %u32 0\n"
1080 "%one = OpConstant %u32 1\n"
1081 "%two = OpConstant %u32 2\n"
1082 "%three = OpConstant %u32 3\n"
1083 "%four = OpConstant %u32 4\n"
1084 "%five = OpConstant %u32 5\n"
1085 "%hundred = OpConstant %u32 100\n"
1086 "%thousand = OpConstant %u32 1000\n"
1088 + string(s_InputOutputBuffer) +
1091 "%main = OpFunction %void None %voidf\n"
1092 "%main_entry = OpLabel\n"
1093 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
1094 "%idval = OpLoad %uvec3 %id\n"
1095 "%x = OpCompositeExtract %u32 %idval 0\n"
1096 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1097 "%inval = OpLoad %f32 %inloc\n"
1098 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1099 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
1100 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
1101 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
1102 " OpSelectionMerge %if_end None\n"
1103 " OpBranchConditional %cmp_gt %if_true %if_false\n"
1104 "%if_true = OpLabel\n"
1105 "%negate = OpFNegate %f32 %inval\n"
1106 " OpStore %outloc %negate\n"
1107 " OpBranch %if_end\n"
1108 "%if_false = OpLabel\n"
1109 " OpUnreachable\n" // Unreachable else branch for if statement
1110 "%if_end = OpLabel\n"
1114 // not_called_function()
1115 "%func_not_called_func = OpFunction %void None %voidf\n"
1116 "%not_called_func_entry = OpLabel\n"
1117 " OpUnreachable\n" // Unreachable entry block in not called static function
1121 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
1122 "%valptr = OpFunctionParameter %u32ptr\n"
1123 "%modulo4_entry = OpLabel\n"
1124 "%val = OpLoad %u32 %valptr\n"
1125 "%modulo = OpUMod %u32 %val %four\n"
1126 " OpSelectionMerge %switch_merge None\n"
1127 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
1128 "%case0 = OpLabel\n"
1129 " OpReturnValue %three\n"
1130 "%case1 = OpLabel\n"
1131 " OpReturnValue %two\n"
1132 "%case2 = OpLabel\n"
1133 " OpReturnValue %one\n"
1134 "%case3 = OpLabel\n"
1135 " OpReturnValue %zero\n"
1136 "%default = OpLabel\n"
1137 " OpUnreachable\n" // Unreachable default case for switch statement
1138 "%switch_merge = OpLabel\n"
1139 " OpUnreachable\n" // Unreachable merge block for switch statement
1143 "%func_const5 = OpFunction %u32 None %unitf\n"
1144 "%const5_entry = OpLabel\n"
1145 " OpReturnValue %five\n"
1146 "%unreachable = OpLabel\n"
1147 " OpUnreachable\n" // Unreachable block in function
1149 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
1150 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
1151 spec.numWorkGroups = IVec3(numElements, 1, 1);
1153 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
1155 return group.release();
1158 // Assembly code used for testing decoration group is based on GLSL source code:
1162 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
1163 // float elements[];
1165 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
1166 // float elements[];
1168 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
1169 // float elements[];
1171 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
1172 // float elements[];
1174 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
1175 // float elements[];
1177 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
1178 // float elements[];
1182 // uint x = gl_GlobalInvocationID.x;
1183 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
1185 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
1187 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
1188 ComputeShaderSpec spec;
1189 de::Random rnd (deStringHash(group->getName()));
1190 const int numElements = 100;
1191 vector<float> inputFloats0 (numElements, 0);
1192 vector<float> inputFloats1 (numElements, 0);
1193 vector<float> inputFloats2 (numElements, 0);
1194 vector<float> inputFloats3 (numElements, 0);
1195 vector<float> inputFloats4 (numElements, 0);
1196 vector<float> outputFloats (numElements, 0);
1198 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
1199 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
1200 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
1201 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
1202 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
1204 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1205 floorAll(inputFloats0);
1206 floorAll(inputFloats1);
1207 floorAll(inputFloats2);
1208 floorAll(inputFloats3);
1209 floorAll(inputFloats4);
1211 for (size_t ndx = 0; ndx < numElements; ++ndx)
1212 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
1215 string(s_ShaderPreamble) +
1217 "OpSource GLSL 430\n"
1218 "OpName %main \"main\"\n"
1219 "OpName %id \"gl_GlobalInvocationID\"\n"
1221 // Not using group decoration on variable.
1222 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1223 // Not using group decoration on type.
1224 "OpDecorate %f32arr ArrayStride 4\n"
1226 "OpDecorate %groups BufferBlock\n"
1227 "OpDecorate %groupm Offset 0\n"
1228 "%groups = OpDecorationGroup\n"
1229 "%groupm = OpDecorationGroup\n"
1231 // Group decoration on multiple structs.
1232 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
1233 // Group decoration on multiple struct members.
1234 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
1236 "OpDecorate %group1 DescriptorSet 0\n"
1237 "OpDecorate %group3 DescriptorSet 0\n"
1238 "OpDecorate %group3 NonWritable\n"
1239 "OpDecorate %group3 Restrict\n"
1240 "%group0 = OpDecorationGroup\n"
1241 "%group1 = OpDecorationGroup\n"
1242 "%group3 = OpDecorationGroup\n"
1244 // Applying the same decoration group multiple times.
1245 "OpGroupDecorate %group1 %outdata\n"
1246 "OpGroupDecorate %group1 %outdata\n"
1247 "OpGroupDecorate %group1 %outdata\n"
1248 "OpDecorate %outdata DescriptorSet 0\n"
1249 "OpDecorate %outdata Binding 5\n"
1250 // Applying decoration group containing nothing.
1251 "OpGroupDecorate %group0 %indata0\n"
1252 "OpDecorate %indata0 DescriptorSet 0\n"
1253 "OpDecorate %indata0 Binding 0\n"
1254 // Applying decoration group containing one decoration.
1255 "OpGroupDecorate %group1 %indata1\n"
1256 "OpDecorate %indata1 Binding 1\n"
1257 // Applying decoration group containing multiple decorations.
1258 "OpGroupDecorate %group3 %indata2 %indata3\n"
1259 "OpDecorate %indata2 Binding 2\n"
1260 "OpDecorate %indata3 Binding 3\n"
1261 // Applying multiple decoration groups (with overlapping).
1262 "OpGroupDecorate %group0 %indata4\n"
1263 "OpGroupDecorate %group1 %indata4\n"
1264 "OpGroupDecorate %group3 %indata4\n"
1265 "OpDecorate %indata4 Binding 4\n"
1267 + string(s_CommonTypes) +
1269 "%id = OpVariable %uvec3ptr Input\n"
1270 "%zero = OpConstant %i32 0\n"
1272 "%outbuf = OpTypeStruct %f32arr\n"
1273 "%outbufptr = OpTypePointer Uniform %outbuf\n"
1274 "%outdata = OpVariable %outbufptr Uniform\n"
1275 "%inbuf0 = OpTypeStruct %f32arr\n"
1276 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
1277 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
1278 "%inbuf1 = OpTypeStruct %f32arr\n"
1279 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
1280 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
1281 "%inbuf2 = OpTypeStruct %f32arr\n"
1282 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
1283 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
1284 "%inbuf3 = OpTypeStruct %f32arr\n"
1285 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
1286 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
1287 "%inbuf4 = OpTypeStruct %f32arr\n"
1288 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
1289 "%indata4 = OpVariable %inbufptr Uniform\n"
1291 "%main = OpFunction %void None %voidf\n"
1292 "%label = OpLabel\n"
1293 "%idval = OpLoad %uvec3 %id\n"
1294 "%x = OpCompositeExtract %u32 %idval 0\n"
1295 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
1296 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1297 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1298 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1299 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
1300 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1301 "%inval0 = OpLoad %f32 %inloc0\n"
1302 "%inval1 = OpLoad %f32 %inloc1\n"
1303 "%inval2 = OpLoad %f32 %inloc2\n"
1304 "%inval3 = OpLoad %f32 %inloc3\n"
1305 "%inval4 = OpLoad %f32 %inloc4\n"
1306 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
1307 "%add1 = OpFAdd %f32 %add0 %inval2\n"
1308 "%add2 = OpFAdd %f32 %add1 %inval3\n"
1309 "%add = OpFAdd %f32 %add2 %inval4\n"
1310 " OpStore %outloc %add\n"
1313 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
1314 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1315 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1316 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1317 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
1318 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1319 spec.numWorkGroups = IVec3(numElements, 1, 1);
1321 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
1323 return group.release();
1326 struct SpecConstantTwoIntCase
1328 const char* caseName;
1329 const char* scDefinition0;
1330 const char* scDefinition1;
1331 const char* scResultType;
1332 const char* scOperation;
1333 deInt32 scActualValue0;
1334 deInt32 scActualValue1;
1335 const char* resultOperation;
1336 vector<deInt32> expectedOutput;
1338 SpecConstantTwoIntCase (const char* name,
1339 const char* definition0,
1340 const char* definition1,
1341 const char* resultType,
1342 const char* operation,
1345 const char* resultOp,
1346 const vector<deInt32>& output)
1348 , scDefinition0 (definition0)
1349 , scDefinition1 (definition1)
1350 , scResultType (resultType)
1351 , scOperation (operation)
1352 , scActualValue0 (value0)
1353 , scActualValue1 (value1)
1354 , resultOperation (resultOp)
1355 , expectedOutput (output) {}
1358 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
1360 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
1361 vector<SpecConstantTwoIntCase> cases;
1362 de::Random rnd (deStringHash(group->getName()));
1363 const int numElements = 100;
1364 vector<deInt32> inputInts (numElements, 0);
1365 vector<deInt32> outputInts1 (numElements, 0);
1366 vector<deInt32> outputInts2 (numElements, 0);
1367 vector<deInt32> outputInts3 (numElements, 0);
1368 vector<deInt32> outputInts4 (numElements, 0);
1369 const StringTemplate shaderTemplate (
1370 string(s_ShaderPreamble) +
1372 "OpName %main \"main\"\n"
1373 "OpName %id \"gl_GlobalInvocationID\"\n"
1375 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1376 "OpDecorate %sc_0 SpecId 0\n"
1377 "OpDecorate %sc_1 SpecId 1\n"
1378 "OpDecorate %i32arr ArrayStride 4\n"
1380 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1382 "%i32ptr = OpTypePointer Uniform %i32\n"
1383 "%i32arr = OpTypeRuntimeArray %i32\n"
1384 "%buf = OpTypeStruct %i32arr\n"
1385 "%bufptr = OpTypePointer Uniform %buf\n"
1386 "%indata = OpVariable %bufptr Uniform\n"
1387 "%outdata = OpVariable %bufptr Uniform\n"
1389 "%id = OpVariable %uvec3ptr Input\n"
1390 "%zero = OpConstant %i32 0\n"
1392 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
1393 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
1394 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
1396 "%main = OpFunction %void None %voidf\n"
1397 "%label = OpLabel\n"
1398 "%idval = OpLoad %uvec3 %id\n"
1399 "%x = OpCompositeExtract %u32 %idval 0\n"
1400 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
1401 "%inval = OpLoad %i32 %inloc\n"
1402 "%final = ${GEN_RESULT}\n"
1403 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1404 " OpStore %outloc %final\n"
1406 " OpFunctionEnd\n");
1408 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
1410 for (size_t ndx = 0; ndx < numElements; ++ndx)
1412 outputInts1[ndx] = inputInts[ndx] + 42;
1413 outputInts2[ndx] = inputInts[ndx];
1414 outputInts3[ndx] = inputInts[ndx] - 11200;
1415 outputInts4[ndx] = inputInts[ndx] + 1;
1418 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
1419 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
1420 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
1422 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
1423 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
1424 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
1425 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
1426 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
1427 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
1428 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
1429 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
1430 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
1431 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
1432 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
1433 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
1434 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
1435 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
1436 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
1437 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
1438 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
1439 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
1440 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
1441 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
1442 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
1443 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
1444 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
1445 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
1446 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
1447 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
1448 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
1449 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
1450 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
1451 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
1452 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
1453 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
1455 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
1457 map<string, string> specializations;
1458 ComputeShaderSpec spec;
1460 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
1461 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
1462 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
1463 specializations["SC_OP"] = cases[caseNdx].scOperation;
1464 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
1466 spec.assembly = shaderTemplate.specialize(specializations);
1467 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
1468 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
1469 spec.numWorkGroups = IVec3(numElements, 1, 1);
1470 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
1471 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
1473 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec));
1476 ComputeShaderSpec spec;
1479 string(s_ShaderPreamble) +
1481 "OpName %main \"main\"\n"
1482 "OpName %id \"gl_GlobalInvocationID\"\n"
1484 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1485 "OpDecorate %sc_0 SpecId 0\n"
1486 "OpDecorate %sc_1 SpecId 1\n"
1487 "OpDecorate %sc_2 SpecId 2\n"
1488 "OpDecorate %i32arr ArrayStride 4\n"
1490 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1492 "%ivec3 = OpTypeVector %i32 3\n"
1493 "%i32ptr = OpTypePointer Uniform %i32\n"
1494 "%i32arr = OpTypeRuntimeArray %i32\n"
1495 "%buf = OpTypeStruct %i32arr\n"
1496 "%bufptr = OpTypePointer Uniform %buf\n"
1497 "%indata = OpVariable %bufptr Uniform\n"
1498 "%outdata = OpVariable %bufptr Uniform\n"
1500 "%id = OpVariable %uvec3ptr Input\n"
1501 "%zero = OpConstant %i32 0\n"
1502 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
1504 "%sc_0 = OpSpecConstant %i32 0\n"
1505 "%sc_1 = OpSpecConstant %i32 0\n"
1506 "%sc_2 = OpSpecConstant %i32 0\n"
1507 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
1508 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
1509 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
1510 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %sc_vec3_1 1 0 4\n" // (0, sc_0, sc_1)
1511 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2 5 1 2\n" // (sc_2, sc_0, sc_1)
1512 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
1513 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
1514 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
1515 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
1516 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
1518 "%main = OpFunction %void None %voidf\n"
1519 "%label = OpLabel\n"
1520 "%idval = OpLoad %uvec3 %id\n"
1521 "%x = OpCompositeExtract %u32 %idval 0\n"
1522 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
1523 "%inval = OpLoad %i32 %inloc\n"
1524 "%final = OpIAdd %i32 %inval %sc_final\n"
1525 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1526 " OpStore %outloc %final\n"
1529 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
1530 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
1531 spec.numWorkGroups = IVec3(numElements, 1, 1);
1532 spec.specConstants.push_back(123);
1533 spec.specConstants.push_back(56);
1534 spec.specConstants.push_back(-77);
1536 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
1538 return group.release();
1541 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
1543 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
1544 ComputeShaderSpec spec1;
1545 ComputeShaderSpec spec2;
1546 ComputeShaderSpec spec3;
1547 de::Random rnd (deStringHash(group->getName()));
1548 const int numElements = 100;
1549 vector<float> inputFloats (numElements, 0);
1550 vector<float> outputFloats1 (numElements, 0);
1551 vector<float> outputFloats2 (numElements, 0);
1552 vector<float> outputFloats3 (numElements, 0);
1554 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
1556 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1557 floorAll(inputFloats);
1559 for (size_t ndx = 0; ndx < numElements; ++ndx)
1563 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
1564 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
1565 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
1568 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
1569 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
1573 string(s_ShaderPreamble) +
1575 "OpSource GLSL 430\n"
1576 "OpName %main \"main\"\n"
1577 "OpName %id \"gl_GlobalInvocationID\"\n"
1579 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1581 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1583 "%id = OpVariable %uvec3ptr Input\n"
1584 "%zero = OpConstant %i32 0\n"
1585 "%three = OpConstant %u32 3\n"
1586 "%constf5p5 = OpConstant %f32 5.5\n"
1587 "%constf20p5 = OpConstant %f32 20.5\n"
1588 "%constf1p75 = OpConstant %f32 1.75\n"
1589 "%constf8p5 = OpConstant %f32 8.5\n"
1590 "%constf6p5 = OpConstant %f32 6.5\n"
1592 "%main = OpFunction %void None %voidf\n"
1593 "%entry = OpLabel\n"
1594 "%idval = OpLoad %uvec3 %id\n"
1595 "%x = OpCompositeExtract %u32 %idval 0\n"
1596 "%selector = OpUMod %u32 %x %three\n"
1597 " OpSelectionMerge %phi None\n"
1598 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
1600 // Case 1 before OpPhi.
1601 "%case1 = OpLabel\n"
1604 "%default = OpLabel\n"
1608 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
1609 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1610 "%inval = OpLoad %f32 %inloc\n"
1611 "%add = OpFAdd %f32 %inval %operand\n"
1612 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1613 " OpStore %outloc %add\n"
1616 // Case 0 after OpPhi.
1617 "%case0 = OpLabel\n"
1621 // Case 2 after OpPhi.
1622 "%case2 = OpLabel\n"
1626 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1627 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
1628 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1630 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
1633 string(s_ShaderPreamble) +
1635 "OpName %main \"main\"\n"
1636 "OpName %id \"gl_GlobalInvocationID\"\n"
1638 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1640 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1642 "%id = OpVariable %uvec3ptr Input\n"
1643 "%zero = OpConstant %i32 0\n"
1644 "%one = OpConstant %i32 1\n"
1645 "%three = OpConstant %i32 3\n"
1646 "%constf6p5 = OpConstant %f32 6.5\n"
1648 "%main = OpFunction %void None %voidf\n"
1649 "%entry = OpLabel\n"
1650 "%idval = OpLoad %uvec3 %id\n"
1651 "%x = OpCompositeExtract %u32 %idval 0\n"
1652 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1653 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1654 "%inval = OpLoad %f32 %inloc\n"
1658 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
1659 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
1660 "%step_next = OpIAdd %i32 %step %one\n"
1661 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
1662 "%still_loop = OpSLessThan %bool %step %three\n"
1663 " OpLoopMerge %exit %phi None\n"
1664 " OpBranchConditional %still_loop %phi %exit\n"
1667 " OpStore %outloc %accum\n"
1670 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1671 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1672 spec2.numWorkGroups = IVec3(numElements, 1, 1);
1674 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
1677 string(s_ShaderPreamble) +
1679 "OpName %main \"main\"\n"
1680 "OpName %id \"gl_GlobalInvocationID\"\n"
1682 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1684 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1686 "%f32ptr_f = OpTypePointer Function %f32\n"
1687 "%id = OpVariable %uvec3ptr Input\n"
1688 "%true = OpConstantTrue %bool\n"
1689 "%false = OpConstantFalse %bool\n"
1690 "%zero = OpConstant %i32 0\n"
1691 "%constf8p5 = OpConstant %f32 8.5\n"
1693 "%main = OpFunction %void None %voidf\n"
1694 "%entry = OpLabel\n"
1695 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
1696 "%idval = OpLoad %uvec3 %id\n"
1697 "%x = OpCompositeExtract %u32 %idval 0\n"
1698 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1699 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1700 "%a_init = OpLoad %f32 %inloc\n"
1701 "%b_init = OpLoad %f32 %b\n"
1705 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
1706 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
1707 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
1708 " OpLoopMerge %exit %phi None\n"
1709 " OpBranchConditional %still_loop %phi %exit\n"
1712 "%sub = OpFSub %f32 %a_next %b_next\n"
1713 " OpStore %outloc %sub\n"
1716 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1717 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
1718 spec3.numWorkGroups = IVec3(numElements, 1, 1);
1720 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
1722 return group.release();
1725 // Assembly code used for testing block order is based on GLSL source code:
1729 // layout(std140, set = 0, binding = 0) readonly buffer Input {
1730 // float elements[];
1732 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
1733 // float elements[];
1737 // uint x = gl_GlobalInvocationID.x;
1738 // output_data.elements[x] = input_data.elements[x];
1739 // if (x > uint(50)) {
1740 // switch (x % uint(3)) {
1741 // case 0: output_data.elements[x] += 1.5f; break;
1742 // case 1: output_data.elements[x] += 42.f; break;
1743 // case 2: output_data.elements[x] -= 27.f; break;
1747 // output_data.elements[x] = -input_data.elements[x];
1750 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
1752 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
1753 ComputeShaderSpec spec;
1754 de::Random rnd (deStringHash(group->getName()));
1755 const int numElements = 100;
1756 vector<float> inputFloats (numElements, 0);
1757 vector<float> outputFloats (numElements, 0);
1759 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
1761 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1762 floorAll(inputFloats);
1764 for (size_t ndx = 0; ndx <= 50; ++ndx)
1765 outputFloats[ndx] = -inputFloats[ndx];
1767 for (size_t ndx = 51; ndx < numElements; ++ndx)
1771 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
1772 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
1773 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
1779 string(s_ShaderPreamble) +
1781 "OpSource GLSL 430\n"
1782 "OpName %main \"main\"\n"
1783 "OpName %id \"gl_GlobalInvocationID\"\n"
1785 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1787 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1789 "%u32ptr = OpTypePointer Function %u32\n"
1790 "%u32ptr_input = OpTypePointer Input %u32\n"
1792 + string(s_InputOutputBuffer) +
1794 "%id = OpVariable %uvec3ptr Input\n"
1795 "%zero = OpConstant %i32 0\n"
1796 "%const3 = OpConstant %u32 3\n"
1797 "%const50 = OpConstant %u32 50\n"
1798 "%constf1p5 = OpConstant %f32 1.5\n"
1799 "%constf27 = OpConstant %f32 27.0\n"
1800 "%constf42 = OpConstant %f32 42.0\n"
1802 "%main = OpFunction %void None %voidf\n"
1805 "%entry = OpLabel\n"
1807 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
1808 "%xvar = OpVariable %u32ptr Function\n"
1809 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
1810 "%x = OpLoad %u32 %xptr\n"
1811 " OpStore %xvar %x\n"
1813 "%cmp = OpUGreaterThan %bool %x %const50\n"
1814 " OpSelectionMerge %if_merge None\n"
1815 " OpBranchConditional %cmp %if_true %if_false\n"
1817 // Merge block for switch-statement: placed at the beginning.
1818 "%switch_merge = OpLabel\n"
1819 " OpBranch %if_merge\n"
1821 // Case 1 for switch-statement.
1822 "%case1 = OpLabel\n"
1823 "%x_1 = OpLoad %u32 %xvar\n"
1824 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
1825 "%inval_1 = OpLoad %f32 %inloc_1\n"
1826 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
1827 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
1828 " OpStore %outloc_1 %addf42\n"
1829 " OpBranch %switch_merge\n"
1831 // False branch for if-statement: placed in the middle of switch cases and before true branch.
1832 "%if_false = OpLabel\n"
1833 "%x_f = OpLoad %u32 %xvar\n"
1834 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
1835 "%inval_f = OpLoad %f32 %inloc_f\n"
1836 "%negate = OpFNegate %f32 %inval_f\n"
1837 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
1838 " OpStore %outloc_f %negate\n"
1839 " OpBranch %if_merge\n"
1841 // Merge block for if-statement: placed in the middle of true and false branch.
1842 "%if_merge = OpLabel\n"
1845 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
1846 "%if_true = OpLabel\n"
1847 "%xval_t = OpLoad %u32 %xvar\n"
1848 "%mod = OpUMod %u32 %xval_t %const3\n"
1849 " OpSelectionMerge %switch_merge None\n"
1850 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
1852 // Case 2 for switch-statement.
1853 "%case2 = OpLabel\n"
1854 "%x_2 = OpLoad %u32 %xvar\n"
1855 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
1856 "%inval_2 = OpLoad %f32 %inloc_2\n"
1857 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
1858 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
1859 " OpStore %outloc_2 %subf27\n"
1860 " OpBranch %switch_merge\n"
1862 // Default case for switch-statement: placed in the middle of normal cases.
1863 "%default = OpLabel\n"
1864 " OpBranch %switch_merge\n"
1866 // Case 0 for switch-statement: out of order.
1867 "%case0 = OpLabel\n"
1868 "%x_0 = OpLoad %u32 %xvar\n"
1869 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
1870 "%inval_0 = OpLoad %f32 %inloc_0\n"
1871 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
1872 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
1873 " OpStore %outloc_0 %addf1p5\n"
1874 " OpBranch %switch_merge\n"
1877 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1878 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1879 spec.numWorkGroups = IVec3(numElements, 1, 1);
1881 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
1883 return group.release();
1886 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
1888 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
1889 ComputeShaderSpec spec1;
1890 ComputeShaderSpec spec2;
1891 de::Random rnd (deStringHash(group->getName()));
1892 const int numElements = 100;
1893 vector<float> inputFloats (numElements, 0);
1894 vector<float> outputFloats1 (numElements, 0);
1895 vector<float> outputFloats2 (numElements, 0);
1896 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
1898 for (size_t ndx = 0; ndx < numElements; ++ndx)
1900 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
1901 outputFloats2[ndx] = -inputFloats[ndx];
1904 const string assembly(
1905 "OpCapability Shader\n"
1906 "OpCapability ClipDistance\n"
1907 "OpMemoryModel Logical GLSL450\n"
1908 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
1909 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
1910 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
1911 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
1912 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
1913 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
1915 "OpName %comp_main1 \"entrypoint1\"\n"
1916 "OpName %comp_main2 \"entrypoint2\"\n"
1917 "OpName %vert_main \"entrypoint2\"\n"
1918 "OpName %id \"gl_GlobalInvocationID\"\n"
1919 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
1920 "OpName %vertexIndex \"gl_VertexIndex\"\n"
1921 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
1922 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
1923 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
1924 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
1926 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1927 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
1928 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
1929 "OpDecorate %vert_builtin_st Block\n"
1930 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
1931 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
1932 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
1934 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1936 "%zero = OpConstant %i32 0\n"
1937 "%one = OpConstant %u32 1\n"
1938 "%c_f32_1 = OpConstant %f32 1\n"
1940 "%i32ptr = OpTypePointer Input %i32\n"
1941 "%vec4 = OpTypeVector %f32 4\n"
1942 "%vec4ptr = OpTypePointer Output %vec4\n"
1943 "%f32arr1 = OpTypeArray %f32 %one\n"
1944 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
1945 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
1946 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
1948 "%id = OpVariable %uvec3ptr Input\n"
1949 "%vertexIndex = OpVariable %i32ptr Input\n"
1950 "%instanceIndex = OpVariable %i32ptr Input\n"
1951 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
1953 // gl_Position = vec4(1.);
1954 "%vert_main = OpFunction %void None %voidf\n"
1955 "%vert_entry = OpLabel\n"
1956 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
1957 " OpStore %position %c_vec4_1\n"
1962 "%comp_main1 = OpFunction %void None %voidf\n"
1963 "%comp1_entry = OpLabel\n"
1964 "%idval1 = OpLoad %uvec3 %id\n"
1965 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
1966 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
1967 "%inval1 = OpLoad %f32 %inloc1\n"
1968 "%add = OpFAdd %f32 %inval1 %inval1\n"
1969 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
1970 " OpStore %outloc1 %add\n"
1975 "%comp_main2 = OpFunction %void None %voidf\n"
1976 "%comp2_entry = OpLabel\n"
1977 "%idval2 = OpLoad %uvec3 %id\n"
1978 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
1979 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
1980 "%inval2 = OpLoad %f32 %inloc2\n"
1981 "%neg = OpFNegate %f32 %inval2\n"
1982 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
1983 " OpStore %outloc2 %neg\n"
1985 " OpFunctionEnd\n");
1987 spec1.assembly = assembly;
1988 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1989 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
1990 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1991 spec1.entryPoint = "entrypoint1";
1993 spec2.assembly = assembly;
1994 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1995 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1996 spec2.numWorkGroups = IVec3(numElements, 1, 1);
1997 spec2.entryPoint = "entrypoint2";
1999 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
2000 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
2002 return group.release();
2005 inline std::string makeLongUTF8String (size_t num4ByteChars)
2007 // An example of a longest valid UTF-8 character. Be explicit about the
2008 // character type because Microsoft compilers can otherwise interpret the
2009 // character string as being over wide (16-bit) characters. Ideally, we
2010 // would just use a C++11 UTF-8 string literal, but we want to support older
2011 // Microsoft compilers.
2012 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
2013 std::string longString;
2014 longString.reserve(num4ByteChars * 4);
2015 for (size_t count = 0; count < num4ByteChars; count++)
2017 longString += earthAfrica;
2022 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
2024 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
2025 vector<CaseParameter> cases;
2026 de::Random rnd (deStringHash(group->getName()));
2027 const int numElements = 100;
2028 vector<float> positiveFloats (numElements, 0);
2029 vector<float> negativeFloats (numElements, 0);
2030 const StringTemplate shaderTemplate (
2031 "OpCapability Shader\n"
2032 "OpMemoryModel Logical GLSL450\n"
2034 "OpEntryPoint GLCompute %main \"main\" %id\n"
2035 "OpExecutionMode %main LocalSize 1 1 1\n"
2039 "OpName %main \"main\"\n"
2040 "OpName %id \"gl_GlobalInvocationID\"\n"
2042 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2044 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2046 "%id = OpVariable %uvec3ptr Input\n"
2047 "%zero = OpConstant %i32 0\n"
2049 "%main = OpFunction %void None %voidf\n"
2050 "%label = OpLabel\n"
2051 "%idval = OpLoad %uvec3 %id\n"
2052 "%x = OpCompositeExtract %u32 %idval 0\n"
2053 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2054 "%inval = OpLoad %f32 %inloc\n"
2055 "%neg = OpFNegate %f32 %inval\n"
2056 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2057 " OpStore %outloc %neg\n"
2059 " OpFunctionEnd\n");
2061 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
2062 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
2063 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
2064 "OpSource GLSL 430 %fname"));
2065 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
2066 "OpSource GLSL 430 %fname"));
2067 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
2068 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
2069 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
2070 "OpSource GLSL 430 %fname \"\""));
2071 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
2072 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
2073 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
2074 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
2075 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
2076 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
2077 "OpSourceContinued \"id main() {}\""));
2078 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
2079 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
2080 "OpSourceContinued \"\""));
2081 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
2082 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
2083 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
2084 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
2085 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
2086 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
2087 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
2088 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
2089 "OpSourceContinued \"void\"\n"
2090 "OpSourceContinued \"main()\"\n"
2091 "OpSourceContinued \"{}\""));
2092 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
2093 "OpSource GLSL 430 %fname \"\"\n"
2094 "OpSourceContinued \"#version 430\nvoid main() {}\""));
2096 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2098 for (size_t ndx = 0; ndx < numElements; ++ndx)
2099 negativeFloats[ndx] = -positiveFloats[ndx];
2101 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2103 map<string, string> specializations;
2104 ComputeShaderSpec spec;
2106 specializations["SOURCE"] = cases[caseNdx].param;
2107 spec.assembly = shaderTemplate.specialize(specializations);
2108 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2109 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2110 spec.numWorkGroups = IVec3(numElements, 1, 1);
2112 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2115 return group.release();
2118 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
2120 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
2121 vector<CaseParameter> cases;
2122 de::Random rnd (deStringHash(group->getName()));
2123 const int numElements = 100;
2124 vector<float> inputFloats (numElements, 0);
2125 vector<float> outputFloats (numElements, 0);
2126 const StringTemplate shaderTemplate (
2127 string(s_ShaderPreamble) +
2129 "OpSourceExtension \"${EXTENSION}\"\n"
2131 "OpName %main \"main\"\n"
2132 "OpName %id \"gl_GlobalInvocationID\"\n"
2134 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2136 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2138 "%id = OpVariable %uvec3ptr Input\n"
2139 "%zero = OpConstant %i32 0\n"
2141 "%main = OpFunction %void None %voidf\n"
2142 "%label = OpLabel\n"
2143 "%idval = OpLoad %uvec3 %id\n"
2144 "%x = OpCompositeExtract %u32 %idval 0\n"
2145 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2146 "%inval = OpLoad %f32 %inloc\n"
2147 "%neg = OpFNegate %f32 %inval\n"
2148 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2149 " OpStore %outloc %neg\n"
2151 " OpFunctionEnd\n");
2153 cases.push_back(CaseParameter("empty_extension", ""));
2154 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
2155 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
2156 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
2157 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
2159 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
2161 for (size_t ndx = 0; ndx < numElements; ++ndx)
2162 outputFloats[ndx] = -inputFloats[ndx];
2164 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2166 map<string, string> specializations;
2167 ComputeShaderSpec spec;
2169 specializations["EXTENSION"] = cases[caseNdx].param;
2170 spec.assembly = shaderTemplate.specialize(specializations);
2171 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2172 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2173 spec.numWorkGroups = IVec3(numElements, 1, 1);
2175 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2178 return group.release();
2181 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
2182 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
2184 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
2185 vector<CaseParameter> cases;
2186 de::Random rnd (deStringHash(group->getName()));
2187 const int numElements = 100;
2188 vector<float> positiveFloats (numElements, 0);
2189 vector<float> negativeFloats (numElements, 0);
2190 const StringTemplate shaderTemplate (
2191 string(s_ShaderPreamble) +
2193 "OpSource GLSL 430\n"
2194 "OpName %main \"main\"\n"
2195 "OpName %id \"gl_GlobalInvocationID\"\n"
2197 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2199 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2202 "%null = OpConstantNull %type\n"
2204 "%id = OpVariable %uvec3ptr Input\n"
2205 "%zero = OpConstant %i32 0\n"
2207 "%main = OpFunction %void None %voidf\n"
2208 "%label = OpLabel\n"
2209 "%idval = OpLoad %uvec3 %id\n"
2210 "%x = OpCompositeExtract %u32 %idval 0\n"
2211 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2212 "%inval = OpLoad %f32 %inloc\n"
2213 "%neg = OpFNegate %f32 %inval\n"
2214 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2215 " OpStore %outloc %neg\n"
2217 " OpFunctionEnd\n");
2219 cases.push_back(CaseParameter("bool", "%type = OpTypeBool"));
2220 cases.push_back(CaseParameter("sint32", "%type = OpTypeInt 32 1"));
2221 cases.push_back(CaseParameter("uint32", "%type = OpTypeInt 32 0"));
2222 cases.push_back(CaseParameter("float32", "%type = OpTypeFloat 32"));
2223 cases.push_back(CaseParameter("vec4float32", "%type = OpTypeVector %f32 4"));
2224 cases.push_back(CaseParameter("vec3bool", "%type = OpTypeVector %bool 3"));
2225 cases.push_back(CaseParameter("vec2uint32", "%type = OpTypeVector %u32 2"));
2226 cases.push_back(CaseParameter("matrix", "%type = OpTypeMatrix %fvec3 3"));
2227 cases.push_back(CaseParameter("array", "%100 = OpConstant %u32 100\n"
2228 "%type = OpTypeArray %i32 %100"));
2229 cases.push_back(CaseParameter("struct", "%type = OpTypeStruct %f32 %i32 %u32"));
2230 cases.push_back(CaseParameter("pointer", "%type = OpTypePointer Function %i32"));
2232 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2234 for (size_t ndx = 0; ndx < numElements; ++ndx)
2235 negativeFloats[ndx] = -positiveFloats[ndx];
2237 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2239 map<string, string> specializations;
2240 ComputeShaderSpec spec;
2242 specializations["TYPE"] = cases[caseNdx].param;
2243 spec.assembly = shaderTemplate.specialize(specializations);
2244 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2245 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2246 spec.numWorkGroups = IVec3(numElements, 1, 1);
2248 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2251 return group.release();
2254 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
2255 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
2257 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
2258 vector<CaseParameter> cases;
2259 de::Random rnd (deStringHash(group->getName()));
2260 const int numElements = 100;
2261 vector<float> positiveFloats (numElements, 0);
2262 vector<float> negativeFloats (numElements, 0);
2263 const StringTemplate shaderTemplate (
2264 string(s_ShaderPreamble) +
2266 "OpSource GLSL 430\n"
2267 "OpName %main \"main\"\n"
2268 "OpName %id \"gl_GlobalInvocationID\"\n"
2270 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2272 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2274 "%id = OpVariable %uvec3ptr Input\n"
2275 "%zero = OpConstant %i32 0\n"
2279 "%main = OpFunction %void None %voidf\n"
2280 "%label = OpLabel\n"
2281 "%idval = OpLoad %uvec3 %id\n"
2282 "%x = OpCompositeExtract %u32 %idval 0\n"
2283 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2284 "%inval = OpLoad %f32 %inloc\n"
2285 "%neg = OpFNegate %f32 %inval\n"
2286 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2287 " OpStore %outloc %neg\n"
2289 " OpFunctionEnd\n");
2291 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
2292 "%const = OpConstantComposite %uvec3 %five %zero %five"));
2293 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
2294 "%ten = OpConstant %f32 10.\n"
2295 "%fzero = OpConstant %f32 0.\n"
2296 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
2297 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
2298 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
2299 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
2300 "%fzero = OpConstant %f32 0.\n"
2301 "%one = OpConstant %f32 1.\n"
2302 "%point5 = OpConstant %f32 0.5\n"
2303 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
2304 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
2305 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
2306 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
2307 "%st2 = OpTypeStruct %i32 %i32\n"
2308 "%struct = OpTypeStruct %st1 %st2\n"
2309 "%point5 = OpConstant %f32 0.5\n"
2310 "%one = OpConstant %u32 1\n"
2311 "%ten = OpConstant %i32 10\n"
2312 "%st1val = OpConstantComposite %st1 %one %point5\n"
2313 "%st2val = OpConstantComposite %st2 %ten %ten\n"
2314 "%const = OpConstantComposite %struct %st1val %st2val"));
2316 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2318 for (size_t ndx = 0; ndx < numElements; ++ndx)
2319 negativeFloats[ndx] = -positiveFloats[ndx];
2321 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2323 map<string, string> specializations;
2324 ComputeShaderSpec spec;
2326 specializations["CONSTANT"] = cases[caseNdx].param;
2327 spec.assembly = shaderTemplate.specialize(specializations);
2328 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2329 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2330 spec.numWorkGroups = IVec3(numElements, 1, 1);
2332 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2335 return group.release();
2338 // Creates a floating point number with the given exponent, and significand
2339 // bits set. It can only create normalized numbers. Only the least significant
2340 // 24 bits of the significand will be examined. The final bit of the
2341 // significand will also be ignored. This allows alignment to be written
2342 // similarly to C99 hex-floats.
2343 // For example if you wanted to write 0x1.7f34p-12 you would call
2344 // constructNormalizedFloat(-12, 0x7f3400)
2345 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
2349 for (deInt32 idx = 0; idx < 23; ++idx)
2351 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
2355 return std::ldexp(f, exponent);
2358 // Compare instruction for the OpQuantizeF16 compute exact case.
2359 // Returns true if the output is what is expected from the test case.
2360 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
2362 if (outputAllocs.size() != 1)
2365 // We really just need this for size because we cannot compare Nans.
2366 const BufferSp& expectedOutput = expectedOutputs[0];
2367 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
2369 if (expectedOutput->getNumBytes() != 4*sizeof(float)) {
2373 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
2374 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
2379 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
2380 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
2385 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
2386 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
2391 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
2392 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
2399 // Checks that every output from a test-case is a float NaN.
2400 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
2402 if (outputAllocs.size() != 1)
2405 // We really just need this for size because we cannot compare Nans.
2406 const BufferSp& expectedOutput = expectedOutputs[0];
2407 const float* output_as_float = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
2409 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
2411 if (!isnan(output_as_float[idx]))
2420 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
2421 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
2423 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
2425 const std::string shader (
2426 string(s_ShaderPreamble) +
2428 "OpSource GLSL 430\n"
2429 "OpName %main \"main\"\n"
2430 "OpName %id \"gl_GlobalInvocationID\"\n"
2432 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2434 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2436 "%id = OpVariable %uvec3ptr Input\n"
2437 "%zero = OpConstant %i32 0\n"
2439 "%main = OpFunction %void None %voidf\n"
2440 "%label = OpLabel\n"
2441 "%idval = OpLoad %uvec3 %id\n"
2442 "%x = OpCompositeExtract %u32 %idval 0\n"
2443 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2444 "%inval = OpLoad %f32 %inloc\n"
2445 "%quant = OpQuantizeToF16 %f32 %inval\n"
2446 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2447 " OpStore %outloc %quant\n"
2449 " OpFunctionEnd\n");
2452 ComputeShaderSpec spec;
2453 const deUint32 numElements = 100;
2454 vector<float> infinities;
2455 vector<float> results;
2457 infinities.reserve(numElements);
2458 results.reserve(numElements);
2460 for (size_t idx = 0; idx < numElements; ++idx)
2465 infinities.push_back(std::numeric_limits<float>::infinity());
2466 results.push_back(std::numeric_limits<float>::infinity());
2469 infinities.push_back(-std::numeric_limits<float>::infinity());
2470 results.push_back(-std::numeric_limits<float>::infinity());
2473 infinities.push_back(std::ldexp(1.0f, 16));
2474 results.push_back(std::numeric_limits<float>::infinity());
2477 infinities.push_back(std::ldexp(-1.0f, 32));
2478 results.push_back(-std::numeric_limits<float>::infinity());
2483 spec.assembly = shader;
2484 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
2485 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
2486 spec.numWorkGroups = IVec3(numElements, 1, 1);
2488 group->addChild(new SpvAsmComputeShaderCase(
2489 testCtx, "infinities", "Check that infinities propagated and created", spec));
2493 ComputeShaderSpec spec;
2495 const deUint32 numElements = 100;
2497 nans.reserve(numElements);
2499 for (size_t idx = 0; idx < numElements; ++idx)
2503 nans.push_back(std::numeric_limits<float>::quiet_NaN());
2507 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
2511 spec.assembly = shader;
2512 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
2513 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
2514 spec.numWorkGroups = IVec3(numElements, 1, 1);
2515 spec.verifyIO = &compareNan;
2517 group->addChild(new SpvAsmComputeShaderCase(
2518 testCtx, "propagated_nans", "Check that nans are propagated", spec));
2522 ComputeShaderSpec spec;
2523 vector<float> small;
2524 vector<float> zeros;
2525 const deUint32 numElements = 100;
2527 small.reserve(numElements);
2528 zeros.reserve(numElements);
2530 for (size_t idx = 0; idx < numElements; ++idx)
2535 small.push_back(0.f);
2536 zeros.push_back(0.f);
2539 small.push_back(-0.f);
2540 zeros.push_back(-0.f);
2543 small.push_back(std::ldexp(1.0f, -16));
2544 zeros.push_back(0.f);
2547 small.push_back(std::ldexp(-1.0f, -32));
2548 zeros.push_back(-0.f);
2551 small.push_back(std::ldexp(1.0f, -127));
2552 zeros.push_back(0.f);
2555 small.push_back(-std::ldexp(1.0f, -128));
2556 zeros.push_back(-0.f);
2561 spec.assembly = shader;
2562 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
2563 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
2564 spec.numWorkGroups = IVec3(numElements, 1, 1);
2566 group->addChild(new SpvAsmComputeShaderCase(
2567 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
2571 ComputeShaderSpec spec;
2572 vector<float> exact;
2573 const deUint32 numElements = 200;
2575 exact.reserve(numElements);
2577 for (size_t idx = 0; idx < numElements; ++idx)
2578 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
2580 spec.assembly = shader;
2581 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
2582 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
2583 spec.numWorkGroups = IVec3(numElements, 1, 1);
2585 group->addChild(new SpvAsmComputeShaderCase(
2586 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
2590 ComputeShaderSpec spec;
2591 vector<float> inputs;
2592 const deUint32 numElements = 4;
2594 inputs.push_back(constructNormalizedFloat(8, 0x300300));
2595 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
2596 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
2597 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
2599 spec.assembly = shader;
2600 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
2601 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2602 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
2603 spec.numWorkGroups = IVec3(numElements, 1, 1);
2605 group->addChild(new SpvAsmComputeShaderCase(
2606 testCtx, "rounded", "Check that are rounded when needed", spec));
2609 return group.release();
2612 // Performs a bitwise copy of source to the destination type Dest.
2613 template <typename Dest, typename Src>
2614 Dest bitwiseCast(Src source)
2617 DE_STATIC_ASSERT(sizeof(source) == sizeof(dest));
2618 deMemcpy(&dest, &source, sizeof(dest));
2622 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
2624 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
2626 const std::string shader (
2627 string(s_ShaderPreamble) +
2629 "OpName %main \"main\"\n"
2630 "OpName %id \"gl_GlobalInvocationID\"\n"
2632 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2634 "OpDecorate %sc_0 SpecId 0\n"
2635 "OpDecorate %sc_1 SpecId 1\n"
2636 "OpDecorate %sc_2 SpecId 2\n"
2637 "OpDecorate %sc_3 SpecId 3\n"
2638 "OpDecorate %sc_4 SpecId 4\n"
2639 "OpDecorate %sc_5 SpecId 5\n"
2641 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2643 "%id = OpVariable %uvec3ptr Input\n"
2644 "%zero = OpConstant %i32 0\n"
2645 "%c_u32_6 = OpConstant %u32 6\n"
2647 "%sc_0 = OpSpecConstant %f32 0.\n"
2648 "%sc_1 = OpSpecConstant %f32 0.\n"
2649 "%sc_2 = OpSpecConstant %f32 0.\n"
2650 "%sc_3 = OpSpecConstant %f32 0.\n"
2651 "%sc_4 = OpSpecConstant %f32 0.\n"
2652 "%sc_5 = OpSpecConstant %f32 0.\n"
2654 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
2655 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
2656 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
2657 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
2658 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
2659 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
2661 "%main = OpFunction %void None %voidf\n"
2662 "%label = OpLabel\n"
2663 "%idval = OpLoad %uvec3 %id\n"
2664 "%x = OpCompositeExtract %u32 %idval 0\n"
2665 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2666 "%selector = OpUMod %u32 %x %c_u32_6\n"
2667 " OpSelectionMerge %exit None\n"
2668 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
2670 "%case0 = OpLabel\n"
2671 " OpStore %outloc %sc_0_quant\n"
2674 "%case1 = OpLabel\n"
2675 " OpStore %outloc %sc_1_quant\n"
2678 "%case2 = OpLabel\n"
2679 " OpStore %outloc %sc_2_quant\n"
2682 "%case3 = OpLabel\n"
2683 " OpStore %outloc %sc_3_quant\n"
2686 "%case4 = OpLabel\n"
2687 " OpStore %outloc %sc_4_quant\n"
2690 "%case5 = OpLabel\n"
2691 " OpStore %outloc %sc_5_quant\n"
2697 " OpFunctionEnd\n");
2700 ComputeShaderSpec spec;
2701 const deUint8 numCases = 4;
2702 vector<float> inputs (numCases, 0.f);
2703 vector<float> outputs;
2705 spec.assembly = shader;
2706 spec.numWorkGroups = IVec3(numCases, 1, 1);
2708 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
2709 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
2710 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
2711 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
2713 outputs.push_back(std::numeric_limits<float>::infinity());
2714 outputs.push_back(-std::numeric_limits<float>::infinity());
2715 outputs.push_back(std::numeric_limits<float>::infinity());
2716 outputs.push_back(-std::numeric_limits<float>::infinity());
2718 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2719 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2721 group->addChild(new SpvAsmComputeShaderCase(
2722 testCtx, "infinities", "Check that infinities propagated and created", spec));
2726 ComputeShaderSpec spec;
2727 const deUint8 numCases = 2;
2728 vector<float> inputs (numCases, 0.f);
2729 vector<float> outputs;
2731 spec.assembly = shader;
2732 spec.numWorkGroups = IVec3(numCases, 1, 1);
2733 spec.verifyIO = &compareNan;
2735 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
2736 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
2738 for (deUint8 idx = 0; idx < numCases; ++idx)
2739 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
2741 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2742 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2744 group->addChild(new SpvAsmComputeShaderCase(
2745 testCtx, "propagated_nans", "Check that nans are propagated", spec));
2749 ComputeShaderSpec spec;
2750 const deUint8 numCases = 6;
2751 vector<float> inputs (numCases, 0.f);
2752 vector<float> outputs;
2754 spec.assembly = shader;
2755 spec.numWorkGroups = IVec3(numCases, 1, 1);
2757 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
2758 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
2759 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
2760 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
2761 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
2762 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
2764 outputs.push_back(0.f);
2765 outputs.push_back(-0.f);
2766 outputs.push_back(0.f);
2767 outputs.push_back(-0.f);
2768 outputs.push_back(0.f);
2769 outputs.push_back(-0.f);
2771 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2772 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2774 group->addChild(new SpvAsmComputeShaderCase(
2775 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
2779 ComputeShaderSpec spec;
2780 const deUint8 numCases = 6;
2781 vector<float> inputs (numCases, 0.f);
2782 vector<float> outputs;
2784 spec.assembly = shader;
2785 spec.numWorkGroups = IVec3(numCases, 1, 1);
2787 for (deUint8 idx = 0; idx < 6; ++idx)
2789 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
2790 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
2791 outputs.push_back(f);
2794 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2795 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2797 group->addChild(new SpvAsmComputeShaderCase(
2798 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
2802 ComputeShaderSpec spec;
2803 const deUint8 numCases = 4;
2804 vector<float> inputs (numCases, 0.f);
2805 vector<float> outputs;
2807 spec.assembly = shader;
2808 spec.numWorkGroups = IVec3(numCases, 1, 1);
2809 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
2811 outputs.push_back(constructNormalizedFloat(8, 0x300300));
2812 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
2813 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
2814 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
2816 for (deUint8 idx = 0; idx < numCases; ++idx)
2817 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
2819 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2820 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2822 group->addChild(new SpvAsmComputeShaderCase(
2823 testCtx, "rounded", "Check that are rounded when needed", spec));
2826 return group.release();
2829 // Checks that constant null/composite values can be used in computation.
2830 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
2832 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
2833 ComputeShaderSpec spec;
2834 de::Random rnd (deStringHash(group->getName()));
2835 const int numElements = 100;
2836 vector<float> positiveFloats (numElements, 0);
2837 vector<float> negativeFloats (numElements, 0);
2839 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2841 for (size_t ndx = 0; ndx < numElements; ++ndx)
2842 negativeFloats[ndx] = -positiveFloats[ndx];
2845 "OpCapability Shader\n"
2846 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
2847 "OpMemoryModel Logical GLSL450\n"
2848 "OpEntryPoint GLCompute %main \"main\" %id\n"
2849 "OpExecutionMode %main LocalSize 1 1 1\n"
2851 "OpSource GLSL 430\n"
2852 "OpName %main \"main\"\n"
2853 "OpName %id \"gl_GlobalInvocationID\"\n"
2855 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2857 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
2859 "%fmat = OpTypeMatrix %fvec3 3\n"
2860 "%ten = OpConstant %u32 10\n"
2861 "%f32arr10 = OpTypeArray %f32 %ten\n"
2862 "%fst = OpTypeStruct %f32 %f32\n"
2864 + string(s_InputOutputBuffer) +
2866 "%id = OpVariable %uvec3ptr Input\n"
2867 "%zero = OpConstant %i32 0\n"
2869 // Create a bunch of null values
2870 "%unull = OpConstantNull %u32\n"
2871 "%fnull = OpConstantNull %f32\n"
2872 "%vnull = OpConstantNull %fvec3\n"
2873 "%mnull = OpConstantNull %fmat\n"
2874 "%anull = OpConstantNull %f32arr10\n"
2875 "%snull = OpConstantComposite %fst %fnull %fnull\n"
2877 "%main = OpFunction %void None %voidf\n"
2878 "%label = OpLabel\n"
2879 "%idval = OpLoad %uvec3 %id\n"
2880 "%x = OpCompositeExtract %u32 %idval 0\n"
2881 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2882 "%inval = OpLoad %f32 %inloc\n"
2883 "%neg = OpFNegate %f32 %inval\n"
2885 // Get the abs() of (a certain element of) those null values
2886 "%unull_cov = OpConvertUToF %f32 %unull\n"
2887 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
2888 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
2889 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
2890 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
2891 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
2892 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
2893 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
2894 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
2895 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
2896 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
2899 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
2900 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
2901 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
2902 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
2903 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
2904 "%final = OpFAdd %f32 %add5 %snull_abs\n"
2906 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2907 " OpStore %outloc %final\n" // write to output
2910 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2911 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2912 spec.numWorkGroups = IVec3(numElements, 1, 1);
2914 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
2916 return group.release();
2919 // Assembly code used for testing loop control is based on GLSL source code:
2922 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2923 // float elements[];
2925 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2926 // float elements[];
2930 // uint x = gl_GlobalInvocationID.x;
2931 // output_data.elements[x] = input_data.elements[x];
2932 // for (uint i = 0; i < 4; ++i)
2933 // output_data.elements[x] += 1.f;
2935 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
2937 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
2938 vector<CaseParameter> cases;
2939 de::Random rnd (deStringHash(group->getName()));
2940 const int numElements = 100;
2941 vector<float> inputFloats (numElements, 0);
2942 vector<float> outputFloats (numElements, 0);
2943 const StringTemplate shaderTemplate (
2944 string(s_ShaderPreamble) +
2946 "OpSource GLSL 430\n"
2947 "OpName %main \"main\"\n"
2948 "OpName %id \"gl_GlobalInvocationID\"\n"
2950 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2952 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2954 "%u32ptr = OpTypePointer Function %u32\n"
2956 "%id = OpVariable %uvec3ptr Input\n"
2957 "%zero = OpConstant %i32 0\n"
2958 "%uzero = OpConstant %u32 0\n"
2959 "%one = OpConstant %i32 1\n"
2960 "%constf1 = OpConstant %f32 1.0\n"
2961 "%four = OpConstant %u32 4\n"
2963 "%main = OpFunction %void None %voidf\n"
2964 "%entry = OpLabel\n"
2965 "%i = OpVariable %u32ptr Function\n"
2966 " OpStore %i %uzero\n"
2968 "%idval = OpLoad %uvec3 %id\n"
2969 "%x = OpCompositeExtract %u32 %idval 0\n"
2970 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2971 "%inval = OpLoad %f32 %inloc\n"
2972 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2973 " OpStore %outloc %inval\n"
2974 " OpBranch %loop_entry\n"
2976 "%loop_entry = OpLabel\n"
2977 "%i_val = OpLoad %u32 %i\n"
2978 "%cmp_lt = OpULessThan %bool %i_val %four\n"
2979 " OpLoopMerge %loop_merge %loop_entry ${CONTROL}\n"
2980 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
2981 "%loop_body = OpLabel\n"
2982 "%outval = OpLoad %f32 %outloc\n"
2983 "%addf1 = OpFAdd %f32 %outval %constf1\n"
2984 " OpStore %outloc %addf1\n"
2985 "%new_i = OpIAdd %u32 %i_val %one\n"
2986 " OpStore %i %new_i\n"
2987 " OpBranch %loop_entry\n"
2988 "%loop_merge = OpLabel\n"
2990 " OpFunctionEnd\n");
2992 cases.push_back(CaseParameter("none", "None"));
2993 cases.push_back(CaseParameter("unroll", "Unroll"));
2994 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
2995 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
2997 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2999 for (size_t ndx = 0; ndx < numElements; ++ndx)
3000 outputFloats[ndx] = inputFloats[ndx] + 4.f;
3002 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3004 map<string, string> specializations;
3005 ComputeShaderSpec spec;
3007 specializations["CONTROL"] = cases[caseNdx].param;
3008 spec.assembly = shaderTemplate.specialize(specializations);
3009 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3010 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3011 spec.numWorkGroups = IVec3(numElements, 1, 1);
3013 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3016 return group.release();
3019 // Assembly code used for testing selection control is based on GLSL source code:
3022 // layout(std140, set = 0, binding = 0) readonly buffer Input {
3023 // float elements[];
3025 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
3026 // float elements[];
3030 // uint x = gl_GlobalInvocationID.x;
3031 // float val = input_data.elements[x];
3033 // output_data.elements[x] = val + 1.f;
3035 // output_data.elements[x] = val - 1.f;
3037 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
3039 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
3040 vector<CaseParameter> cases;
3041 de::Random rnd (deStringHash(group->getName()));
3042 const int numElements = 100;
3043 vector<float> inputFloats (numElements, 0);
3044 vector<float> outputFloats (numElements, 0);
3045 const StringTemplate shaderTemplate (
3046 string(s_ShaderPreamble) +
3048 "OpSource GLSL 430\n"
3049 "OpName %main \"main\"\n"
3050 "OpName %id \"gl_GlobalInvocationID\"\n"
3052 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3054 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
3056 "%id = OpVariable %uvec3ptr Input\n"
3057 "%zero = OpConstant %i32 0\n"
3058 "%constf1 = OpConstant %f32 1.0\n"
3059 "%constf10 = OpConstant %f32 10.0\n"
3061 "%main = OpFunction %void None %voidf\n"
3062 "%entry = OpLabel\n"
3063 "%idval = OpLoad %uvec3 %id\n"
3064 "%x = OpCompositeExtract %u32 %idval 0\n"
3065 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3066 "%inval = OpLoad %f32 %inloc\n"
3067 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3068 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
3070 " OpSelectionMerge %if_end ${CONTROL}\n"
3071 " OpBranchConditional %cmp_gt %if_true %if_false\n"
3072 "%if_true = OpLabel\n"
3073 "%addf1 = OpFAdd %f32 %inval %constf1\n"
3074 " OpStore %outloc %addf1\n"
3075 " OpBranch %if_end\n"
3076 "%if_false = OpLabel\n"
3077 "%subf1 = OpFSub %f32 %inval %constf1\n"
3078 " OpStore %outloc %subf1\n"
3079 " OpBranch %if_end\n"
3080 "%if_end = OpLabel\n"
3082 " OpFunctionEnd\n");
3084 cases.push_back(CaseParameter("none", "None"));
3085 cases.push_back(CaseParameter("flatten", "Flatten"));
3086 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
3087 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
3089 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3091 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3092 floorAll(inputFloats);
3094 for (size_t ndx = 0; ndx < numElements; ++ndx)
3095 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
3097 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3099 map<string, string> specializations;
3100 ComputeShaderSpec spec;
3102 specializations["CONTROL"] = cases[caseNdx].param;
3103 spec.assembly = shaderTemplate.specialize(specializations);
3104 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3105 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3106 spec.numWorkGroups = IVec3(numElements, 1, 1);
3108 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3111 return group.release();
3114 // Assembly code used for testing function control is based on GLSL source code:
3118 // layout(std140, set = 0, binding = 0) readonly buffer Input {
3119 // float elements[];
3121 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
3122 // float elements[];
3125 // float const10() { return 10.f; }
3128 // uint x = gl_GlobalInvocationID.x;
3129 // output_data.elements[x] = input_data.elements[x] + const10();
3131 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
3133 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
3134 vector<CaseParameter> cases;
3135 de::Random rnd (deStringHash(group->getName()));
3136 const int numElements = 100;
3137 vector<float> inputFloats (numElements, 0);
3138 vector<float> outputFloats (numElements, 0);
3139 const StringTemplate shaderTemplate (
3140 string(s_ShaderPreamble) +
3142 "OpSource GLSL 430\n"
3143 "OpName %main \"main\"\n"
3144 "OpName %func_const10 \"const10(\"\n"
3145 "OpName %id \"gl_GlobalInvocationID\"\n"
3147 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3149 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
3151 "%f32f = OpTypeFunction %f32\n"
3152 "%id = OpVariable %uvec3ptr Input\n"
3153 "%zero = OpConstant %i32 0\n"
3154 "%constf10 = OpConstant %f32 10.0\n"
3156 "%main = OpFunction %void None %voidf\n"
3157 "%entry = OpLabel\n"
3158 "%idval = OpLoad %uvec3 %id\n"
3159 "%x = OpCompositeExtract %u32 %idval 0\n"
3160 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3161 "%inval = OpLoad %f32 %inloc\n"
3162 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
3163 "%fadd = OpFAdd %f32 %inval %ret_10\n"
3164 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3165 " OpStore %outloc %fadd\n"
3169 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
3170 "%label = OpLabel\n"
3171 " OpReturnValue %constf10\n"
3172 " OpFunctionEnd\n");
3174 cases.push_back(CaseParameter("none", "None"));
3175 cases.push_back(CaseParameter("inline", "Inline"));
3176 cases.push_back(CaseParameter("dont_inline", "DontInline"));
3177 cases.push_back(CaseParameter("pure", "Pure"));
3178 cases.push_back(CaseParameter("const", "Const"));
3179 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
3180 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
3181 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
3182 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
3184 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3186 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3187 floorAll(inputFloats);
3189 for (size_t ndx = 0; ndx < numElements; ++ndx)
3190 outputFloats[ndx] = inputFloats[ndx] + 10.f;
3192 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3194 map<string, string> specializations;
3195 ComputeShaderSpec spec;
3197 specializations["CONTROL"] = cases[caseNdx].param;
3198 spec.assembly = shaderTemplate.specialize(specializations);
3199 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3200 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3201 spec.numWorkGroups = IVec3(numElements, 1, 1);
3203 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3206 return group.release();
3209 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
3211 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
3212 vector<CaseParameter> cases;
3213 de::Random rnd (deStringHash(group->getName()));
3214 const int numElements = 100;
3215 vector<float> inputFloats (numElements, 0);
3216 vector<float> outputFloats (numElements, 0);
3217 const StringTemplate shaderTemplate (
3218 string(s_ShaderPreamble) +
3220 "OpSource GLSL 430\n"
3221 "OpName %main \"main\"\n"
3222 "OpName %id \"gl_GlobalInvocationID\"\n"
3224 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3226 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
3228 "%f32ptr_f = OpTypePointer Function %f32\n"
3230 "%id = OpVariable %uvec3ptr Input\n"
3231 "%zero = OpConstant %i32 0\n"
3232 "%four = OpConstant %i32 4\n"
3234 "%main = OpFunction %void None %voidf\n"
3235 "%label = OpLabel\n"
3236 "%copy = OpVariable %f32ptr_f Function\n"
3237 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
3238 "%x = OpCompositeExtract %u32 %idval 0\n"
3239 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3240 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3241 " OpCopyMemory %copy %inloc ${ACCESS}\n"
3242 "%val1 = OpLoad %f32 %copy\n"
3243 "%val2 = OpLoad %f32 %inloc\n"
3244 "%add = OpFAdd %f32 %val1 %val2\n"
3245 " OpStore %outloc %add ${ACCESS}\n"
3247 " OpFunctionEnd\n");
3249 cases.push_back(CaseParameter("null", ""));
3250 cases.push_back(CaseParameter("none", "None"));
3251 cases.push_back(CaseParameter("volatile", "Volatile"));
3252 cases.push_back(CaseParameter("aligned", "Aligned 4"));
3253 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
3254 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
3255 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
3257 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3259 for (size_t ndx = 0; ndx < numElements; ++ndx)
3260 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
3262 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3264 map<string, string> specializations;
3265 ComputeShaderSpec spec;
3267 specializations["ACCESS"] = cases[caseNdx].param;
3268 spec.assembly = shaderTemplate.specialize(specializations);
3269 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3270 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3271 spec.numWorkGroups = IVec3(numElements, 1, 1);
3273 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3276 return group.release();
3279 // Checks that we can get undefined values for various types, without exercising a computation with it.
3280 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
3282 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
3283 vector<CaseParameter> cases;
3284 de::Random rnd (deStringHash(group->getName()));
3285 const int numElements = 100;
3286 vector<float> positiveFloats (numElements, 0);
3287 vector<float> negativeFloats (numElements, 0);
3288 const StringTemplate shaderTemplate (
3289 string(s_ShaderPreamble) +
3291 "OpSource GLSL 430\n"
3292 "OpName %main \"main\"\n"
3293 "OpName %id \"gl_GlobalInvocationID\"\n"
3295 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3297 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
3301 "%id = OpVariable %uvec3ptr Input\n"
3302 "%zero = OpConstant %i32 0\n"
3304 "%main = OpFunction %void None %voidf\n"
3305 "%label = OpLabel\n"
3307 "%undef = OpUndef %type\n"
3309 "%idval = OpLoad %uvec3 %id\n"
3310 "%x = OpCompositeExtract %u32 %idval 0\n"
3312 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3313 "%inval = OpLoad %f32 %inloc\n"
3314 "%neg = OpFNegate %f32 %inval\n"
3315 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3316 " OpStore %outloc %neg\n"
3318 " OpFunctionEnd\n");
3320 cases.push_back(CaseParameter("bool", "%type = OpTypeBool"));
3321 cases.push_back(CaseParameter("sint32", "%type = OpTypeInt 32 1"));
3322 cases.push_back(CaseParameter("uint32", "%type = OpTypeInt 32 0"));
3323 cases.push_back(CaseParameter("float32", "%type = OpTypeFloat 32"));
3324 cases.push_back(CaseParameter("vec4float32", "%type = OpTypeVector %f32 4"));
3325 cases.push_back(CaseParameter("vec2uint32", "%type = OpTypeVector %u32 2"));
3326 cases.push_back(CaseParameter("matrix", "%type = OpTypeMatrix %fvec3 3"));
3327 cases.push_back(CaseParameter("image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown"));
3328 cases.push_back(CaseParameter("sampler", "%type = OpTypeSampler"));
3329 cases.push_back(CaseParameter("sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
3330 "%type = OpTypeSampledImage %img"));
3331 cases.push_back(CaseParameter("array", "%100 = OpConstant %u32 100\n"
3332 "%type = OpTypeArray %i32 %100"));
3333 cases.push_back(CaseParameter("runtimearray", "%type = OpTypeRuntimeArray %f32"));
3334 cases.push_back(CaseParameter("struct", "%type = OpTypeStruct %f32 %i32 %u32"));
3335 cases.push_back(CaseParameter("pointer", "%type = OpTypePointer Function %i32"));
3337 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3339 for (size_t ndx = 0; ndx < numElements; ++ndx)
3340 negativeFloats[ndx] = -positiveFloats[ndx];
3342 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3344 map<string, string> specializations;
3345 ComputeShaderSpec spec;
3347 specializations["TYPE"] = cases[caseNdx].param;
3348 spec.assembly = shaderTemplate.specialize(specializations);
3349 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3350 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3351 spec.numWorkGroups = IVec3(numElements, 1, 1);
3353 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3356 return group.release();
3358 typedef std::pair<std::string, VkShaderStageFlagBits> EntryToStage;
3359 typedef map<string, vector<EntryToStage> > ModuleMap;
3360 typedef map<VkShaderStageFlagBits, vector<deInt32> > StageToSpecConstantMap;
3362 // Context for a specific test instantiation. For example, an instantiation
3363 // may test colors yellow/magenta/cyan/mauve in a tesselation shader
3364 // with an entry point named 'main_to_the_main'
3365 struct InstanceContext
3367 // Map of modules to what entry_points we care to use from those modules.
3368 ModuleMap moduleMap;
3369 RGBA inputColors[4];
3370 RGBA outputColors[4];
3371 // Concrete SPIR-V code to test via boilerplate specialization.
3372 map<string, string> testCodeFragments;
3373 StageToSpecConstantMap specConstants;
3374 bool hasTessellation;
3375 VkShaderStageFlagBits requiredStages;
3377 InstanceContext (const RGBA (&inputs)[4], const RGBA (&outputs)[4], const map<string, string>& testCodeFragments_, const StageToSpecConstantMap& specConstants_)
3378 : testCodeFragments (testCodeFragments_)
3379 , specConstants (specConstants_)
3380 , hasTessellation (false)
3381 , requiredStages (static_cast<VkShaderStageFlagBits>(0))
3383 inputColors[0] = inputs[0];
3384 inputColors[1] = inputs[1];
3385 inputColors[2] = inputs[2];
3386 inputColors[3] = inputs[3];
3388 outputColors[0] = outputs[0];
3389 outputColors[1] = outputs[1];
3390 outputColors[2] = outputs[2];
3391 outputColors[3] = outputs[3];
3394 InstanceContext (const InstanceContext& other)
3395 : moduleMap (other.moduleMap)
3396 , testCodeFragments (other.testCodeFragments)
3397 , specConstants (other.specConstants)
3398 , hasTessellation (other.hasTessellation)
3399 , requiredStages (other.requiredStages)
3401 inputColors[0] = other.inputColors[0];
3402 inputColors[1] = other.inputColors[1];
3403 inputColors[2] = other.inputColors[2];
3404 inputColors[3] = other.inputColors[3];
3406 outputColors[0] = other.outputColors[0];
3407 outputColors[1] = other.outputColors[1];
3408 outputColors[2] = other.outputColors[2];
3409 outputColors[3] = other.outputColors[3];
3413 // A description of a shader to be used for a single stage of the graphics pipeline.
3414 struct ShaderElement
3416 // The module that contains this shader entrypoint.
3419 // The name of the entrypoint.
3422 // Which shader stage this entry point represents.
3423 VkShaderStageFlagBits stage;
3425 ShaderElement (const string& moduleName_, const string& entryPoint_, VkShaderStageFlagBits shaderStage_)
3426 : moduleName(moduleName_)
3427 , entryName(entryPoint_)
3428 , stage(shaderStage_)
3433 void getDefaultColors (RGBA (&colors)[4])
3435 colors[0] = RGBA::white();
3436 colors[1] = RGBA::red();
3437 colors[2] = RGBA::green();
3438 colors[3] = RGBA::blue();
3441 void getHalfColorsFullAlpha (RGBA (&colors)[4])
3443 colors[0] = RGBA(127, 127, 127, 255);
3444 colors[1] = RGBA(127, 0, 0, 255);
3445 colors[2] = RGBA(0, 127, 0, 255);
3446 colors[3] = RGBA(0, 0, 127, 255);
3449 void getInvertedDefaultColors (RGBA (&colors)[4])
3451 colors[0] = RGBA(0, 0, 0, 255);
3452 colors[1] = RGBA(0, 255, 255, 255);
3453 colors[2] = RGBA(255, 0, 255, 255);
3454 colors[3] = RGBA(255, 255, 0, 255);
3457 // Turns a statically sized array of ShaderElements into an instance-context
3458 // by setting up the mapping of modules to their contained shaders and stages.
3459 // The inputs and expected outputs are given by inputColors and outputColors
3461 InstanceContext createInstanceContext (const ShaderElement (&elements)[N], const RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments, const StageToSpecConstantMap& specConstants)
3463 InstanceContext ctx (inputColors, outputColors, testCodeFragments, specConstants);
3464 for (size_t i = 0; i < N; ++i)
3466 ctx.moduleMap[elements[i].moduleName].push_back(std::make_pair(elements[i].entryName, elements[i].stage));
3467 ctx.requiredStages = static_cast<VkShaderStageFlagBits>(ctx.requiredStages | elements[i].stage);
3473 inline InstanceContext createInstanceContext (const ShaderElement (&elements)[N], RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments)
3475 return createInstanceContext(elements, inputColors, outputColors, testCodeFragments, StageToSpecConstantMap());
3478 // The same as createInstanceContext above, but with default colors.
3480 InstanceContext createInstanceContext (const ShaderElement (&elements)[N], const map<string, string>& testCodeFragments)
3482 RGBA defaultColors[4];
3483 getDefaultColors(defaultColors);
3484 return createInstanceContext(elements, defaultColors, defaultColors, testCodeFragments);
3487 // For the current InstanceContext, constructs the required modules and shader stage create infos.
3488 void createPipelineShaderStages (const DeviceInterface& vk, const VkDevice vkDevice, InstanceContext& instance, Context& context, vector<ModuleHandleSp>& modules, vector<VkPipelineShaderStageCreateInfo>& createInfos)
3490 for (ModuleMap::const_iterator moduleNdx = instance.moduleMap.begin(); moduleNdx != instance.moduleMap.end(); ++moduleNdx)
3492 const ModuleHandleSp mod(new Unique<VkShaderModule>(createShaderModule(vk, vkDevice, context.getBinaryCollection().get(moduleNdx->first), 0)));
3493 modules.push_back(ModuleHandleSp(mod));
3494 for (vector<EntryToStage>::const_iterator shaderNdx = moduleNdx->second.begin(); shaderNdx != moduleNdx->second.end(); ++shaderNdx)
3496 const EntryToStage& stage = *shaderNdx;
3497 const VkPipelineShaderStageCreateInfo shaderParam =
3499 VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, // VkStructureType sType;
3500 DE_NULL, // const void* pNext;
3501 (VkPipelineShaderStageCreateFlags)0,
3502 stage.second, // VkShaderStageFlagBits stage;
3503 **modules.back(), // VkShaderModule module;
3504 stage.first.c_str(), // const char* pName;
3505 (const VkSpecializationInfo*)DE_NULL,
3507 createInfos.push_back(shaderParam);
3512 #define SPIRV_ASSEMBLY_TYPES \
3513 "%void = OpTypeVoid\n" \
3514 "%bool = OpTypeBool\n" \
3516 "%i32 = OpTypeInt 32 1\n" \
3517 "%u32 = OpTypeInt 32 0\n" \
3519 "%f32 = OpTypeFloat 32\n" \
3520 "%v3f32 = OpTypeVector %f32 3\n" \
3521 "%v4f32 = OpTypeVector %f32 4\n" \
3522 "%v4bool = OpTypeVector %bool 4\n" \
3524 "%v4f32_function = OpTypeFunction %v4f32 %v4f32\n" \
3525 "%fun = OpTypeFunction %void\n" \
3527 "%ip_f32 = OpTypePointer Input %f32\n" \
3528 "%ip_i32 = OpTypePointer Input %i32\n" \
3529 "%ip_v3f32 = OpTypePointer Input %v3f32\n" \
3530 "%ip_v4f32 = OpTypePointer Input %v4f32\n" \
3532 "%op_f32 = OpTypePointer Output %f32\n" \
3533 "%op_v4f32 = OpTypePointer Output %v4f32\n" \
3535 "%fp_f32 = OpTypePointer Function %f32\n" \
3536 "%fp_i32 = OpTypePointer Function %i32\n" \
3537 "%fp_v4f32 = OpTypePointer Function %v4f32\n"
3539 #define SPIRV_ASSEMBLY_CONSTANTS \
3540 "%c_f32_1 = OpConstant %f32 1.0\n" \
3541 "%c_f32_0 = OpConstant %f32 0.0\n" \
3542 "%c_f32_0_5 = OpConstant %f32 0.5\n" \
3543 "%c_f32_n1 = OpConstant %f32 -1.\n" \
3544 "%c_f32_7 = OpConstant %f32 7.0\n" \
3545 "%c_f32_8 = OpConstant %f32 8.0\n" \
3546 "%c_i32_0 = OpConstant %i32 0\n" \
3547 "%c_i32_1 = OpConstant %i32 1\n" \
3548 "%c_i32_2 = OpConstant %i32 2\n" \
3549 "%c_i32_3 = OpConstant %i32 3\n" \
3550 "%c_i32_4 = OpConstant %i32 4\n" \
3551 "%c_u32_0 = OpConstant %u32 0\n" \
3552 "%c_u32_1 = OpConstant %u32 1\n" \
3553 "%c_u32_2 = OpConstant %u32 2\n" \
3554 "%c_u32_3 = OpConstant %u32 3\n" \
3555 "%c_u32_32 = OpConstant %u32 32\n" \
3556 "%c_u32_4 = OpConstant %u32 4\n" \
3557 "%c_u32_31_bits = OpConstant %u32 0x7FFFFFFF\n" \
3558 "%c_v4f32_1_1_1_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n" \
3559 "%c_v4f32_1_0_0_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_1\n" \
3560 "%c_v4f32_0_5_0_5_0_5_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5\n"
3562 #define SPIRV_ASSEMBLY_ARRAYS \
3563 "%a1f32 = OpTypeArray %f32 %c_u32_1\n" \
3564 "%a2f32 = OpTypeArray %f32 %c_u32_2\n" \
3565 "%a3v4f32 = OpTypeArray %v4f32 %c_u32_3\n" \
3566 "%a4f32 = OpTypeArray %f32 %c_u32_4\n" \
3567 "%a32v4f32 = OpTypeArray %v4f32 %c_u32_32\n" \
3568 "%ip_a3v4f32 = OpTypePointer Input %a3v4f32\n" \
3569 "%ip_a32v4f32 = OpTypePointer Input %a32v4f32\n" \
3570 "%op_a2f32 = OpTypePointer Output %a2f32\n" \
3571 "%op_a3v4f32 = OpTypePointer Output %a3v4f32\n" \
3572 "%op_a4f32 = OpTypePointer Output %a4f32\n"
3574 // Creates vertex-shader assembly by specializing a boilerplate StringTemplate
3575 // on fragments, which must (at least) map "testfun" to an OpFunction definition
3576 // for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed
3577 // with "BP_" to avoid collisions with fragments.
3579 // It corresponds roughly to this GLSL:
3581 // layout(location = 0) in vec4 position;
3582 // layout(location = 1) in vec4 color;
3583 // layout(location = 1) out highp vec4 vtxColor;
3584 // void main (void) { gl_Position = position; vtxColor = test_func(color); }
3585 string makeVertexShaderAssembly(const map<string, string>& fragments)
3587 // \todo [2015-11-23 awoloszyn] Remove OpName once these have stabalized
3588 static const char vertexShaderBoilerplate[] =
3589 "OpCapability Shader\n"
3590 "OpCapability ClipDistance\n"
3591 "OpCapability CullDistance\n"
3592 "OpMemoryModel Logical GLSL450\n"
3593 "OpEntryPoint Vertex %main \"main\" %BP_stream %BP_position %BP_vtx_color %BP_color %BP_gl_VertexIndex %BP_gl_InstanceIndex\n"
3595 "OpName %main \"main\"\n"
3596 "OpName %BP_gl_PerVertex \"gl_PerVertex\"\n"
3597 "OpMemberName %BP_gl_PerVertex 0 \"gl_Position\"\n"
3598 "OpMemberName %BP_gl_PerVertex 1 \"gl_PointSize\"\n"
3599 "OpMemberName %BP_gl_PerVertex 2 \"gl_ClipDistance\"\n"
3600 "OpMemberName %BP_gl_PerVertex 3 \"gl_CullDistance\"\n"
3601 "OpName %test_code \"testfun(vf4;\"\n"
3602 "OpName %BP_stream \"\"\n"
3603 "OpName %BP_position \"position\"\n"
3604 "OpName %BP_vtx_color \"vtxColor\"\n"
3605 "OpName %BP_color \"color\"\n"
3606 "OpName %BP_gl_VertexIndex \"gl_VertexIndex\"\n"
3607 "OpName %BP_gl_InstanceIndex \"gl_InstanceIndex\"\n"
3608 "OpMemberDecorate %BP_gl_PerVertex 0 BuiltIn Position\n"
3609 "OpMemberDecorate %BP_gl_PerVertex 1 BuiltIn PointSize\n"
3610 "OpMemberDecorate %BP_gl_PerVertex 2 BuiltIn ClipDistance\n"
3611 "OpMemberDecorate %BP_gl_PerVertex 3 BuiltIn CullDistance\n"
3612 "OpDecorate %BP_gl_PerVertex Block\n"
3613 "OpDecorate %BP_position Location 0\n"
3614 "OpDecorate %BP_vtx_color Location 1\n"
3615 "OpDecorate %BP_color Location 1\n"
3616 "OpDecorate %BP_gl_VertexIndex BuiltIn VertexIndex\n"
3617 "OpDecorate %BP_gl_InstanceIndex BuiltIn InstanceIndex\n"
3618 "${decoration:opt}\n"
3619 SPIRV_ASSEMBLY_TYPES
3620 SPIRV_ASSEMBLY_CONSTANTS
3621 SPIRV_ASSEMBLY_ARRAYS
3622 "%BP_gl_PerVertex = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3623 "%BP_op_gl_PerVertex = OpTypePointer Output %BP_gl_PerVertex\n"
3624 "%BP_stream = OpVariable %BP_op_gl_PerVertex Output\n"
3625 "%BP_position = OpVariable %ip_v4f32 Input\n"
3626 "%BP_vtx_color = OpVariable %op_v4f32 Output\n"
3627 "%BP_color = OpVariable %ip_v4f32 Input\n"
3628 "%BP_gl_VertexIndex = OpVariable %ip_i32 Input\n"
3629 "%BP_gl_InstanceIndex = OpVariable %ip_i32 Input\n"
3631 "%main = OpFunction %void None %fun\n"
3632 "%BP_label = OpLabel\n"
3633 "%BP_pos = OpLoad %v4f32 %BP_position\n"
3634 "%BP_gl_pos = OpAccessChain %op_v4f32 %BP_stream %c_i32_0\n"
3635 "OpStore %BP_gl_pos %BP_pos\n"
3636 "%BP_col = OpLoad %v4f32 %BP_color\n"
3637 "%BP_col_transformed = OpFunctionCall %v4f32 %test_code %BP_col\n"
3638 "OpStore %BP_vtx_color %BP_col_transformed\n"
3642 return tcu::StringTemplate(vertexShaderBoilerplate).specialize(fragments);
3645 // Creates tess-control-shader assembly by specializing a boilerplate
3646 // StringTemplate on fragments, which must (at least) map "testfun" to an
3647 // OpFunction definition for %test_code that takes and returns a %v4f32.
3648 // Boilerplate IDs are prefixed with "BP_" to avoid collisions with fragments.
3650 // It roughly corresponds to the following GLSL.
3653 // layout(vertices = 3) out;
3654 // layout(location = 1) in vec4 in_color[];
3655 // layout(location = 1) out vec4 out_color[];
3658 // out_color[gl_InvocationID] = testfun(in_color[gl_InvocationID]);
3659 // gl_out[gl_InvocationID].gl_Position = gl_in[gl_InvocationID].gl_Position;
3660 // if (gl_InvocationID == 0) {
3661 // gl_TessLevelOuter[0] = 1.0;
3662 // gl_TessLevelOuter[1] = 1.0;
3663 // gl_TessLevelOuter[2] = 1.0;
3664 // gl_TessLevelInner[0] = 1.0;
3667 string makeTessControlShaderAssembly (const map<string, string>& fragments)
3669 static const char tessControlShaderBoilerplate[] =
3670 "OpCapability Tessellation\n"
3671 "OpCapability ClipDistance\n"
3672 "OpCapability CullDistance\n"
3673 "OpMemoryModel Logical GLSL450\n"
3674 "OpEntryPoint TessellationControl %BP_main \"main\" %BP_out_color %BP_gl_InvocationID %BP_in_color %BP_gl_out %BP_gl_in %BP_gl_TessLevelOuter %BP_gl_TessLevelInner\n"
3675 "OpExecutionMode %BP_main OutputVertices 3\n"
3677 "OpName %BP_main \"main\"\n"
3678 "OpName %test_code \"testfun(vf4;\"\n"
3679 "OpName %BP_out_color \"out_color\"\n"
3680 "OpName %BP_gl_InvocationID \"gl_InvocationID\"\n"
3681 "OpName %BP_in_color \"in_color\"\n"
3682 "OpName %BP_gl_PerVertex \"gl_PerVertex\"\n"
3683 "OpMemberName %BP_gl_PerVertex 0 \"gl_Position\"\n"
3684 "OpMemberName %BP_gl_PerVertex 1 \"gl_PointSize\"\n"
3685 "OpMemberName %BP_gl_PerVertex 2 \"gl_ClipDistance\"\n"
3686 "OpMemberName %BP_gl_PerVertex 3 \"gl_CullDistance\"\n"
3687 "OpName %BP_gl_out \"gl_out\"\n"
3688 "OpName %BP_gl_PVOut \"gl_PerVertex\"\n"
3689 "OpMemberName %BP_gl_PVOut 0 \"gl_Position\"\n"
3690 "OpMemberName %BP_gl_PVOut 1 \"gl_PointSize\"\n"
3691 "OpMemberName %BP_gl_PVOut 2 \"gl_ClipDistance\"\n"
3692 "OpMemberName %BP_gl_PVOut 3 \"gl_CullDistance\"\n"
3693 "OpName %BP_gl_in \"gl_in\"\n"
3694 "OpName %BP_gl_TessLevelOuter \"gl_TessLevelOuter\"\n"
3695 "OpName %BP_gl_TessLevelInner \"gl_TessLevelInner\"\n"
3696 "OpDecorate %BP_out_color Location 1\n"
3697 "OpDecorate %BP_gl_InvocationID BuiltIn InvocationId\n"
3698 "OpDecorate %BP_in_color Location 1\n"
3699 "OpMemberDecorate %BP_gl_PerVertex 0 BuiltIn Position\n"
3700 "OpMemberDecorate %BP_gl_PerVertex 1 BuiltIn PointSize\n"
3701 "OpMemberDecorate %BP_gl_PerVertex 2 BuiltIn ClipDistance\n"
3702 "OpMemberDecorate %BP_gl_PerVertex 3 BuiltIn CullDistance\n"
3703 "OpDecorate %BP_gl_PerVertex Block\n"
3704 "OpMemberDecorate %BP_gl_PVOut 0 BuiltIn Position\n"
3705 "OpMemberDecorate %BP_gl_PVOut 1 BuiltIn PointSize\n"
3706 "OpMemberDecorate %BP_gl_PVOut 2 BuiltIn ClipDistance\n"
3707 "OpMemberDecorate %BP_gl_PVOut 3 BuiltIn CullDistance\n"
3708 "OpDecorate %BP_gl_PVOut Block\n"
3709 "OpDecorate %BP_gl_TessLevelOuter Patch\n"
3710 "OpDecorate %BP_gl_TessLevelOuter BuiltIn TessLevelOuter\n"
3711 "OpDecorate %BP_gl_TessLevelInner Patch\n"
3712 "OpDecorate %BP_gl_TessLevelInner BuiltIn TessLevelInner\n"
3713 "${decoration:opt}\n"
3714 SPIRV_ASSEMBLY_TYPES
3715 SPIRV_ASSEMBLY_CONSTANTS
3716 SPIRV_ASSEMBLY_ARRAYS
3717 "%BP_out_color = OpVariable %op_a3v4f32 Output\n"
3718 "%BP_gl_InvocationID = OpVariable %ip_i32 Input\n"
3719 "%BP_in_color = OpVariable %ip_a32v4f32 Input\n"
3720 "%BP_gl_PerVertex = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3721 "%BP_a3_gl_PerVertex = OpTypeArray %BP_gl_PerVertex %c_u32_3\n"
3722 "%BP_op_a3_gl_PerVertex = OpTypePointer Output %BP_a3_gl_PerVertex\n"
3723 "%BP_gl_out = OpVariable %BP_op_a3_gl_PerVertex Output\n"
3724 "%BP_gl_PVOut = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3725 "%BP_a32_gl_PVOut = OpTypeArray %BP_gl_PVOut %c_u32_32\n"
3726 "%BP_ip_a32_gl_PVOut = OpTypePointer Input %BP_a32_gl_PVOut\n"
3727 "%BP_gl_in = OpVariable %BP_ip_a32_gl_PVOut Input\n"
3728 "%BP_gl_TessLevelOuter = OpVariable %op_a4f32 Output\n"
3729 "%BP_gl_TessLevelInner = OpVariable %op_a2f32 Output\n"
3732 "%BP_main = OpFunction %void None %fun\n"
3733 "%BP_label = OpLabel\n"
3735 "%BP_gl_Invoc = OpLoad %i32 %BP_gl_InvocationID\n"
3737 "%BP_in_col_loc = OpAccessChain %ip_v4f32 %BP_in_color %BP_gl_Invoc\n"
3738 "%BP_out_col_loc = OpAccessChain %op_v4f32 %BP_out_color %BP_gl_Invoc\n"
3739 "%BP_in_col_val = OpLoad %v4f32 %BP_in_col_loc\n"
3740 "%BP_clr_transformed = OpFunctionCall %v4f32 %test_code %BP_in_col_val\n"
3741 "OpStore %BP_out_col_loc %BP_clr_transformed\n"
3743 "%BP_in_pos_loc = OpAccessChain %ip_v4f32 %BP_gl_in %BP_gl_Invoc %c_i32_0\n"
3744 "%BP_out_pos_loc = OpAccessChain %op_v4f32 %BP_gl_out %BP_gl_Invoc %c_i32_0\n"
3745 "%BP_in_pos_val = OpLoad %v4f32 %BP_in_pos_loc\n"
3746 "OpStore %BP_out_pos_loc %BP_in_pos_val\n"
3748 "%BP_cmp = OpIEqual %bool %BP_gl_Invoc %c_i32_0\n"
3749 "OpSelectionMerge %BP_merge_label None\n"
3750 "OpBranchConditional %BP_cmp %BP_if_label %BP_merge_label\n"
3751 "%BP_if_label = OpLabel\n"
3752 "%BP_gl_TessLevelOuterPos_0 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_0\n"
3753 "%BP_gl_TessLevelOuterPos_1 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_1\n"
3754 "%BP_gl_TessLevelOuterPos_2 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_2\n"
3755 "%BP_gl_TessLevelInnerPos_0 = OpAccessChain %op_f32 %BP_gl_TessLevelInner %c_i32_0\n"
3756 "OpStore %BP_gl_TessLevelOuterPos_0 %c_f32_1\n"
3757 "OpStore %BP_gl_TessLevelOuterPos_1 %c_f32_1\n"
3758 "OpStore %BP_gl_TessLevelOuterPos_2 %c_f32_1\n"
3759 "OpStore %BP_gl_TessLevelInnerPos_0 %c_f32_1\n"
3760 "OpBranch %BP_merge_label\n"
3761 "%BP_merge_label = OpLabel\n"
3765 return tcu::StringTemplate(tessControlShaderBoilerplate).specialize(fragments);
3768 // Creates tess-evaluation-shader assembly by specializing a boilerplate
3769 // StringTemplate on fragments, which must (at least) map "testfun" to an
3770 // OpFunction definition for %test_code that takes and returns a %v4f32.
3771 // Boilerplate IDs are prefixed with "BP_" to avoid collisions with fragments.
3773 // It roughly corresponds to the following glsl.
3777 // layout(triangles, equal_spacing, ccw) in;
3778 // layout(location = 1) in vec4 in_color[];
3779 // layout(location = 1) out vec4 out_color;
3781 // #define interpolate(val)
3782 // vec4(gl_TessCoord.x) * val[0] + vec4(gl_TessCoord.y) * val[1] +
3783 // vec4(gl_TessCoord.z) * val[2]
3786 // gl_Position = vec4(gl_TessCoord.x) * gl_in[0].gl_Position +
3787 // vec4(gl_TessCoord.y) * gl_in[1].gl_Position +
3788 // vec4(gl_TessCoord.z) * gl_in[2].gl_Position;
3789 // out_color = testfun(interpolate(in_color));
3791 string makeTessEvalShaderAssembly(const map<string, string>& fragments)
3793 static const char tessEvalBoilerplate[] =
3794 "OpCapability Tessellation\n"
3795 "OpCapability ClipDistance\n"
3796 "OpCapability CullDistance\n"
3797 "OpMemoryModel Logical GLSL450\n"
3798 "OpEntryPoint TessellationEvaluation %BP_main \"main\" %BP_stream %BP_gl_TessCoord %BP_gl_in %BP_out_color %BP_in_color\n"
3799 "OpExecutionMode %BP_main Triangles\n"
3800 "OpExecutionMode %BP_main SpacingEqual\n"
3801 "OpExecutionMode %BP_main VertexOrderCcw\n"
3803 "OpName %BP_main \"main\"\n"
3804 "OpName %test_code \"testfun(vf4;\"\n"
3805 "OpName %BP_gl_PerVertexOut \"gl_PerVertex\"\n"
3806 "OpMemberName %BP_gl_PerVertexOut 0 \"gl_Position\"\n"
3807 "OpMemberName %BP_gl_PerVertexOut 1 \"gl_PointSize\"\n"
3808 "OpMemberName %BP_gl_PerVertexOut 2 \"gl_ClipDistance\"\n"
3809 "OpMemberName %BP_gl_PerVertexOut 3 \"gl_CullDistance\"\n"
3810 "OpName %BP_stream \"\"\n"
3811 "OpName %BP_gl_TessCoord \"gl_TessCoord\"\n"
3812 "OpName %BP_gl_PerVertexIn \"gl_PerVertex\"\n"
3813 "OpMemberName %BP_gl_PerVertexIn 0 \"gl_Position\"\n"
3814 "OpMemberName %BP_gl_PerVertexIn 1 \"gl_PointSize\"\n"
3815 "OpMemberName %BP_gl_PerVertexIn 2 \"gl_ClipDistance\"\n"
3816 "OpMemberName %BP_gl_PerVertexIn 3 \"gl_CullDistance\"\n"
3817 "OpName %BP_gl_in \"gl_in\"\n"
3818 "OpName %BP_out_color \"out_color\"\n"
3819 "OpName %BP_in_color \"in_color\"\n"
3820 "OpMemberDecorate %BP_gl_PerVertexOut 0 BuiltIn Position\n"
3821 "OpMemberDecorate %BP_gl_PerVertexOut 1 BuiltIn PointSize\n"
3822 "OpMemberDecorate %BP_gl_PerVertexOut 2 BuiltIn ClipDistance\n"
3823 "OpMemberDecorate %BP_gl_PerVertexOut 3 BuiltIn CullDistance\n"
3824 "OpDecorate %BP_gl_PerVertexOut Block\n"
3825 "OpDecorate %BP_gl_TessCoord BuiltIn TessCoord\n"
3826 "OpMemberDecorate %BP_gl_PerVertexIn 0 BuiltIn Position\n"
3827 "OpMemberDecorate %BP_gl_PerVertexIn 1 BuiltIn PointSize\n"
3828 "OpMemberDecorate %BP_gl_PerVertexIn 2 BuiltIn ClipDistance\n"
3829 "OpMemberDecorate %BP_gl_PerVertexIn 3 BuiltIn CullDistance\n"
3830 "OpDecorate %BP_gl_PerVertexIn Block\n"
3831 "OpDecorate %BP_out_color Location 1\n"
3832 "OpDecorate %BP_in_color Location 1\n"
3833 "${decoration:opt}\n"
3834 SPIRV_ASSEMBLY_TYPES
3835 SPIRV_ASSEMBLY_CONSTANTS
3836 SPIRV_ASSEMBLY_ARRAYS
3837 "%BP_gl_PerVertexOut = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3838 "%BP_op_gl_PerVertexOut = OpTypePointer Output %BP_gl_PerVertexOut\n"
3839 "%BP_stream = OpVariable %BP_op_gl_PerVertexOut Output\n"
3840 "%BP_gl_TessCoord = OpVariable %ip_v3f32 Input\n"
3841 "%BP_gl_PerVertexIn = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3842 "%BP_a32_gl_PerVertexIn = OpTypeArray %BP_gl_PerVertexIn %c_u32_32\n"
3843 "%BP_ip_a32_gl_PerVertexIn = OpTypePointer Input %BP_a32_gl_PerVertexIn\n"
3844 "%BP_gl_in = OpVariable %BP_ip_a32_gl_PerVertexIn Input\n"
3845 "%BP_out_color = OpVariable %op_v4f32 Output\n"
3846 "%BP_in_color = OpVariable %ip_a32v4f32 Input\n"
3848 "%BP_main = OpFunction %void None %fun\n"
3849 "%BP_label = OpLabel\n"
3850 "%BP_gl_TC_0 = OpAccessChain %ip_f32 %BP_gl_TessCoord %c_u32_0\n"
3851 "%BP_gl_TC_1 = OpAccessChain %ip_f32 %BP_gl_TessCoord %c_u32_1\n"
3852 "%BP_gl_TC_2 = OpAccessChain %ip_f32 %BP_gl_TessCoord %c_u32_2\n"
3853 "%BP_gl_in_gl_Pos_0 = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_0 %c_i32_0\n"
3854 "%BP_gl_in_gl_Pos_1 = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_1 %c_i32_0\n"
3855 "%BP_gl_in_gl_Pos_2 = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_2 %c_i32_0\n"
3857 "%BP_gl_OPos = OpAccessChain %op_v4f32 %BP_stream %c_i32_0\n"
3858 "%BP_in_color_0 = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_0\n"
3859 "%BP_in_color_1 = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_1\n"
3860 "%BP_in_color_2 = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_2\n"
3862 "%BP_TC_W_0 = OpLoad %f32 %BP_gl_TC_0\n"
3863 "%BP_TC_W_1 = OpLoad %f32 %BP_gl_TC_1\n"
3864 "%BP_TC_W_2 = OpLoad %f32 %BP_gl_TC_2\n"
3865 "%BP_v4f32_TC_0 = OpCompositeConstruct %v4f32 %BP_TC_W_0 %BP_TC_W_0 %BP_TC_W_0 %BP_TC_W_0\n"
3866 "%BP_v4f32_TC_1 = OpCompositeConstruct %v4f32 %BP_TC_W_1 %BP_TC_W_1 %BP_TC_W_1 %BP_TC_W_1\n"
3867 "%BP_v4f32_TC_2 = OpCompositeConstruct %v4f32 %BP_TC_W_2 %BP_TC_W_2 %BP_TC_W_2 %BP_TC_W_2\n"
3869 "%BP_gl_IP_0 = OpLoad %v4f32 %BP_gl_in_gl_Pos_0\n"
3870 "%BP_gl_IP_1 = OpLoad %v4f32 %BP_gl_in_gl_Pos_1\n"
3871 "%BP_gl_IP_2 = OpLoad %v4f32 %BP_gl_in_gl_Pos_2\n"
3873 "%BP_IP_W_0 = OpFMul %v4f32 %BP_v4f32_TC_0 %BP_gl_IP_0\n"
3874 "%BP_IP_W_1 = OpFMul %v4f32 %BP_v4f32_TC_1 %BP_gl_IP_1\n"
3875 "%BP_IP_W_2 = OpFMul %v4f32 %BP_v4f32_TC_2 %BP_gl_IP_2\n"
3877 "%BP_pos_sum_0 = OpFAdd %v4f32 %BP_IP_W_0 %BP_IP_W_1\n"
3878 "%BP_pos_sum_1 = OpFAdd %v4f32 %BP_pos_sum_0 %BP_IP_W_2\n"
3880 "OpStore %BP_gl_OPos %BP_pos_sum_1\n"
3882 "%BP_IC_0 = OpLoad %v4f32 %BP_in_color_0\n"
3883 "%BP_IC_1 = OpLoad %v4f32 %BP_in_color_1\n"
3884 "%BP_IC_2 = OpLoad %v4f32 %BP_in_color_2\n"
3886 "%BP_IC_W_0 = OpFMul %v4f32 %BP_v4f32_TC_0 %BP_IC_0\n"
3887 "%BP_IC_W_1 = OpFMul %v4f32 %BP_v4f32_TC_1 %BP_IC_1\n"
3888 "%BP_IC_W_2 = OpFMul %v4f32 %BP_v4f32_TC_2 %BP_IC_2\n"
3890 "%BP_col_sum_0 = OpFAdd %v4f32 %BP_IC_W_0 %BP_IC_W_1\n"
3891 "%BP_col_sum_1 = OpFAdd %v4f32 %BP_col_sum_0 %BP_IC_W_2\n"
3893 "%BP_clr_transformed = OpFunctionCall %v4f32 %test_code %BP_col_sum_1\n"
3895 "OpStore %BP_out_color %BP_clr_transformed\n"
3899 return tcu::StringTemplate(tessEvalBoilerplate).specialize(fragments);
3902 // Creates geometry-shader assembly by specializing a boilerplate StringTemplate
3903 // on fragments, which must (at least) map "testfun" to an OpFunction definition
3904 // for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed
3905 // with "BP_" to avoid collisions with fragments.
3907 // Derived from this GLSL:
3910 // layout(triangles) in;
3911 // layout(triangle_strip, max_vertices = 3) out;
3913 // layout(location = 1) in vec4 in_color[];
3914 // layout(location = 1) out vec4 out_color;
3917 // gl_Position = gl_in[0].gl_Position;
3918 // out_color = test_fun(in_color[0]);
3920 // gl_Position = gl_in[1].gl_Position;
3921 // out_color = test_fun(in_color[1]);
3923 // gl_Position = gl_in[2].gl_Position;
3924 // out_color = test_fun(in_color[2]);
3928 string makeGeometryShaderAssembly(const map<string, string>& fragments)
3930 static const char geometryShaderBoilerplate[] =
3931 "OpCapability Geometry\n"
3932 "OpCapability ClipDistance\n"
3933 "OpCapability CullDistance\n"
3934 "OpMemoryModel Logical GLSL450\n"
3935 "OpEntryPoint Geometry %BP_main \"main\" %BP_out_gl_position %BP_gl_in %BP_out_color %BP_in_color\n"
3936 "OpExecutionMode %BP_main Triangles\n"
3937 "OpExecutionMode %BP_main OutputTriangleStrip\n"
3938 "OpExecutionMode %BP_main OutputVertices 3\n"
3940 "OpName %BP_main \"main\"\n"
3941 "OpName %BP_per_vertex_in \"gl_PerVertex\"\n"
3942 "OpMemberName %BP_per_vertex_in 0 \"gl_Position\"\n"
3943 "OpMemberName %BP_per_vertex_in 1 \"gl_PointSize\"\n"
3944 "OpMemberName %BP_per_vertex_in 2 \"gl_ClipDistance\"\n"
3945 "OpMemberName %BP_per_vertex_in 3 \"gl_CullDistance\"\n"
3946 "OpName %BP_gl_in \"gl_in\"\n"
3947 "OpName %BP_out_color \"out_color\"\n"
3948 "OpName %BP_in_color \"in_color\"\n"
3949 "OpName %test_code \"testfun(vf4;\"\n"
3950 "OpDecorate %BP_out_gl_position BuiltIn Position\n"
3951 "OpMemberDecorate %BP_per_vertex_in 0 BuiltIn Position\n"
3952 "OpMemberDecorate %BP_per_vertex_in 1 BuiltIn PointSize\n"
3953 "OpMemberDecorate %BP_per_vertex_in 2 BuiltIn ClipDistance\n"
3954 "OpMemberDecorate %BP_per_vertex_in 3 BuiltIn CullDistance\n"
3955 "OpDecorate %BP_per_vertex_in Block\n"
3956 "OpDecorate %BP_out_color Location 1\n"
3957 "OpDecorate %BP_in_color Location 1\n"
3958 "${decoration:opt}\n"
3959 SPIRV_ASSEMBLY_TYPES
3960 SPIRV_ASSEMBLY_CONSTANTS
3961 SPIRV_ASSEMBLY_ARRAYS
3962 "%BP_per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3963 "%BP_a3_per_vertex_in = OpTypeArray %BP_per_vertex_in %c_u32_3\n"
3964 "%BP_ip_a3_per_vertex_in = OpTypePointer Input %BP_a3_per_vertex_in\n"
3966 "%BP_gl_in = OpVariable %BP_ip_a3_per_vertex_in Input\n"
3967 "%BP_out_color = OpVariable %op_v4f32 Output\n"
3968 "%BP_in_color = OpVariable %ip_a3v4f32 Input\n"
3969 "%BP_out_gl_position = OpVariable %op_v4f32 Output\n"
3972 "%BP_main = OpFunction %void None %fun\n"
3973 "%BP_label = OpLabel\n"
3974 "%BP_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_0 %c_i32_0\n"
3975 "%BP_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_1 %c_i32_0\n"
3976 "%BP_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_2 %c_i32_0\n"
3978 "%BP_in_position_0 = OpLoad %v4f32 %BP_gl_in_0_gl_position\n"
3979 "%BP_in_position_1 = OpLoad %v4f32 %BP_gl_in_1_gl_position\n"
3980 "%BP_in_position_2 = OpLoad %v4f32 %BP_gl_in_2_gl_position \n"
3982 "%BP_in_color_0_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_0\n"
3983 "%BP_in_color_1_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_1\n"
3984 "%BP_in_color_2_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_2\n"
3986 "%BP_in_color_0 = OpLoad %v4f32 %BP_in_color_0_ptr\n"
3987 "%BP_in_color_1 = OpLoad %v4f32 %BP_in_color_1_ptr\n"
3988 "%BP_in_color_2 = OpLoad %v4f32 %BP_in_color_2_ptr\n"
3990 "%BP_transformed_in_color_0 = OpFunctionCall %v4f32 %test_code %BP_in_color_0\n"
3991 "%BP_transformed_in_color_1 = OpFunctionCall %v4f32 %test_code %BP_in_color_1\n"
3992 "%BP_transformed_in_color_2 = OpFunctionCall %v4f32 %test_code %BP_in_color_2\n"
3995 "OpStore %BP_out_gl_position %BP_in_position_0\n"
3996 "OpStore %BP_out_color %BP_transformed_in_color_0\n"
3999 "OpStore %BP_out_gl_position %BP_in_position_1\n"
4000 "OpStore %BP_out_color %BP_transformed_in_color_1\n"
4003 "OpStore %BP_out_gl_position %BP_in_position_2\n"
4004 "OpStore %BP_out_color %BP_transformed_in_color_2\n"
4011 return tcu::StringTemplate(geometryShaderBoilerplate).specialize(fragments);
4014 // Creates fragment-shader assembly by specializing a boilerplate StringTemplate
4015 // on fragments, which must (at least) map "testfun" to an OpFunction definition
4016 // for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed
4017 // with "BP_" to avoid collisions with fragments.
4019 // Derived from this GLSL:
4021 // layout(location = 1) in highp vec4 vtxColor;
4022 // layout(location = 0) out highp vec4 fragColor;
4023 // highp vec4 testfun(highp vec4 x) { return x; }
4024 // void main(void) { fragColor = testfun(vtxColor); }
4026 // with modifications including passing vtxColor by value and ripping out
4027 // testfun() definition.
4028 string makeFragmentShaderAssembly(const map<string, string>& fragments)
4030 static const char fragmentShaderBoilerplate[] =
4031 "OpCapability Shader\n"
4032 "OpMemoryModel Logical GLSL450\n"
4033 "OpEntryPoint Fragment %BP_main \"main\" %BP_vtxColor %BP_fragColor\n"
4034 "OpExecutionMode %BP_main OriginUpperLeft\n"
4036 "OpName %BP_main \"main\"\n"
4037 "OpName %BP_fragColor \"fragColor\"\n"
4038 "OpName %BP_vtxColor \"vtxColor\"\n"
4039 "OpName %test_code \"testfun(vf4;\"\n"
4040 "OpDecorate %BP_fragColor Location 0\n"
4041 "OpDecorate %BP_vtxColor Location 1\n"
4042 "${decoration:opt}\n"
4043 SPIRV_ASSEMBLY_TYPES
4044 SPIRV_ASSEMBLY_CONSTANTS
4045 SPIRV_ASSEMBLY_ARRAYS
4046 "%BP_fragColor = OpVariable %op_v4f32 Output\n"
4047 "%BP_vtxColor = OpVariable %ip_v4f32 Input\n"
4049 "%BP_main = OpFunction %void None %fun\n"
4050 "%BP_label_main = OpLabel\n"
4051 "%BP_tmp1 = OpLoad %v4f32 %BP_vtxColor\n"
4052 "%BP_tmp2 = OpFunctionCall %v4f32 %test_code %BP_tmp1\n"
4053 "OpStore %BP_fragColor %BP_tmp2\n"
4057 return tcu::StringTemplate(fragmentShaderBoilerplate).specialize(fragments);
4060 // Creates fragments that specialize into a simple pass-through shader (of any kind).
4061 map<string, string> passthruFragments(void)
4063 map<string, string> fragments;
4064 fragments["testfun"] =
4065 // A %test_code function that returns its argument unchanged.
4066 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4067 "%param1 = OpFunctionParameter %v4f32\n"
4068 "%label_testfun = OpLabel\n"
4069 "OpReturnValue %param1\n"
4074 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
4075 // Vertex shader gets custom code from context, the rest are pass-through.
4076 void addShaderCodeCustomVertex(vk::SourceCollections& dst, InstanceContext context)
4078 map<string, string> passthru = passthruFragments();
4079 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(context.testCodeFragments);
4080 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
4083 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
4084 // Tessellation control shader gets custom code from context, the rest are
4086 void addShaderCodeCustomTessControl(vk::SourceCollections& dst, InstanceContext context)
4088 map<string, string> passthru = passthruFragments();
4089 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
4090 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(context.testCodeFragments);
4091 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(passthru);
4092 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
4095 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
4096 // Tessellation evaluation shader gets custom code from context, the rest are
4098 void addShaderCodeCustomTessEval(vk::SourceCollections& dst, InstanceContext context)
4100 map<string, string> passthru = passthruFragments();
4101 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
4102 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(passthru);
4103 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(context.testCodeFragments);
4104 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
4107 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
4108 // Geometry shader gets custom code from context, the rest are pass-through.
4109 void addShaderCodeCustomGeometry(vk::SourceCollections& dst, InstanceContext context)
4111 map<string, string> passthru = passthruFragments();
4112 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
4113 dst.spirvAsmSources.add("geom") << makeGeometryShaderAssembly(context.testCodeFragments);
4114 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
4117 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
4118 // Fragment shader gets custom code from context, the rest are pass-through.
4119 void addShaderCodeCustomFragment(vk::SourceCollections& dst, InstanceContext context)
4121 map<string, string> passthru = passthruFragments();
4122 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
4123 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(context.testCodeFragments);
4126 void createCombinedModule(vk::SourceCollections& dst, InstanceContext)
4128 // \todo [2015-12-07 awoloszyn] Make tessellation / geometry conditional
4129 // \todo [2015-12-07 awoloszyn] Remove OpName and OpMemberName at some point
4130 dst.spirvAsmSources.add("module") <<
4131 "OpCapability Shader\n"
4132 "OpCapability ClipDistance\n"
4133 "OpCapability CullDistance\n"
4134 "OpCapability Geometry\n"
4135 "OpCapability Tessellation\n"
4136 "OpMemoryModel Logical GLSL450\n"
4138 "OpEntryPoint Vertex %vert_main \"main\" %vert_Position %vert_vtxColor %vert_color %vert_vtxPosition %vert_vertex_id %vert_instance_id\n"
4139 "OpEntryPoint Geometry %geom_main \"main\" %geom_out_gl_position %geom_gl_in %geom_out_color %geom_in_color\n"
4140 "OpEntryPoint TessellationControl %tessc_main \"main\" %tessc_out_color %tessc_gl_InvocationID %tessc_in_color %tessc_out_position %tessc_in_position %tessc_gl_TessLevelOuter %tessc_gl_TessLevelInner\n"
4141 "OpEntryPoint TessellationEvaluation %tesse_main \"main\" %tesse_stream %tesse_gl_tessCoord %tesse_in_position %tesse_out_color %tesse_in_color \n"
4142 "OpEntryPoint Fragment %frag_main \"main\" %frag_vtxColor %frag_fragColor\n"
4144 "OpExecutionMode %geom_main Triangles\n"
4145 "OpExecutionMode %geom_main OutputTriangleStrip\n"
4146 "OpExecutionMode %geom_main OutputVertices 3\n"
4148 "OpExecutionMode %tessc_main OutputVertices 3\n"
4150 "OpExecutionMode %tesse_main Triangles\n"
4152 "OpExecutionMode %frag_main OriginUpperLeft\n"
4154 "OpName %vert_main \"main\"\n"
4155 "OpName %vert_vtxPosition \"vtxPosition\"\n"
4156 "OpName %vert_Position \"position\"\n"
4157 "OpName %vert_vtxColor \"vtxColor\"\n"
4158 "OpName %vert_color \"color\"\n"
4159 "OpName %vert_vertex_id \"gl_VertexIndex\"\n"
4160 "OpName %vert_instance_id \"gl_InstanceIndex\"\n"
4161 "OpName %geom_main \"main\"\n"
4162 "OpName %geom_per_vertex_in \"gl_PerVertex\"\n"
4163 "OpMemberName %geom_per_vertex_in 0 \"gl_Position\"\n"
4164 "OpMemberName %geom_per_vertex_in 1 \"gl_PointSize\"\n"
4165 "OpMemberName %geom_per_vertex_in 2 \"gl_ClipDistance\"\n"
4166 "OpMemberName %geom_per_vertex_in 3 \"gl_CullDistance\"\n"
4167 "OpName %geom_gl_in \"gl_in\"\n"
4168 "OpName %geom_out_color \"out_color\"\n"
4169 "OpName %geom_in_color \"in_color\"\n"
4170 "OpName %tessc_main \"main\"\n"
4171 "OpName %tessc_out_color \"out_color\"\n"
4172 "OpName %tessc_gl_InvocationID \"gl_InvocationID\"\n"
4173 "OpName %tessc_in_color \"in_color\"\n"
4174 "OpName %tessc_out_position \"out_position\"\n"
4175 "OpName %tessc_in_position \"in_position\"\n"
4176 "OpName %tessc_gl_TessLevelOuter \"gl_TessLevelOuter\"\n"
4177 "OpName %tessc_gl_TessLevelInner \"gl_TessLevelInner\"\n"
4178 "OpName %tesse_main \"main\"\n"
4179 "OpName %tesse_per_vertex_out \"gl_PerVertex\"\n"
4180 "OpMemberName %tesse_per_vertex_out 0 \"gl_Position\"\n"
4181 "OpMemberName %tesse_per_vertex_out 1 \"gl_PointSize\"\n"
4182 "OpMemberName %tesse_per_vertex_out 2 \"gl_ClipDistance\"\n"
4183 "OpMemberName %tesse_per_vertex_out 3 \"gl_CullDistance\"\n"
4184 "OpName %tesse_stream \"\"\n"
4185 "OpName %tesse_gl_tessCoord \"gl_TessCoord\"\n"
4186 "OpName %tesse_in_position \"in_position\"\n"
4187 "OpName %tesse_out_color \"out_color\"\n"
4188 "OpName %tesse_in_color \"in_color\"\n"
4189 "OpName %frag_main \"main\"\n"
4190 "OpName %frag_fragColor \"fragColor\"\n"
4191 "OpName %frag_vtxColor \"vtxColor\"\n"
4193 "; Vertex decorations\n"
4194 "OpDecorate %vert_vtxPosition Location 2\n"
4195 "OpDecorate %vert_Position Location 0\n"
4196 "OpDecorate %vert_vtxColor Location 1\n"
4197 "OpDecorate %vert_color Location 1\n"
4198 "OpDecorate %vert_vertex_id BuiltIn VertexIndex\n"
4199 "OpDecorate %vert_instance_id BuiltIn InstanceIndex\n"
4201 "; Geometry decorations\n"
4202 "OpDecorate %geom_out_gl_position BuiltIn Position\n"
4203 "OpMemberDecorate %geom_per_vertex_in 0 BuiltIn Position\n"
4204 "OpMemberDecorate %geom_per_vertex_in 1 BuiltIn PointSize\n"
4205 "OpMemberDecorate %geom_per_vertex_in 2 BuiltIn ClipDistance\n"
4206 "OpMemberDecorate %geom_per_vertex_in 3 BuiltIn CullDistance\n"
4207 "OpDecorate %geom_per_vertex_in Block\n"
4208 "OpDecorate %geom_out_color Location 1\n"
4209 "OpDecorate %geom_in_color Location 1\n"
4211 "; Tessellation Control decorations\n"
4212 "OpDecorate %tessc_out_color Location 1\n"
4213 "OpDecorate %tessc_gl_InvocationID BuiltIn InvocationId\n"
4214 "OpDecorate %tessc_in_color Location 1\n"
4215 "OpDecorate %tessc_out_position Location 2\n"
4216 "OpDecorate %tessc_in_position Location 2\n"
4217 "OpDecorate %tessc_gl_TessLevelOuter Patch\n"
4218 "OpDecorate %tessc_gl_TessLevelOuter BuiltIn TessLevelOuter\n"
4219 "OpDecorate %tessc_gl_TessLevelInner Patch\n"
4220 "OpDecorate %tessc_gl_TessLevelInner BuiltIn TessLevelInner\n"
4222 "; Tessellation Evaluation decorations\n"
4223 "OpMemberDecorate %tesse_per_vertex_out 0 BuiltIn Position\n"
4224 "OpMemberDecorate %tesse_per_vertex_out 1 BuiltIn PointSize\n"
4225 "OpMemberDecorate %tesse_per_vertex_out 2 BuiltIn ClipDistance\n"
4226 "OpMemberDecorate %tesse_per_vertex_out 3 BuiltIn CullDistance\n"
4227 "OpDecorate %tesse_per_vertex_out Block\n"
4228 "OpDecorate %tesse_gl_tessCoord BuiltIn TessCoord\n"
4229 "OpDecorate %tesse_in_position Location 2\n"
4230 "OpDecorate %tesse_out_color Location 1\n"
4231 "OpDecorate %tesse_in_color Location 1\n"
4233 "; Fragment decorations\n"
4234 "OpDecorate %frag_fragColor Location 0\n"
4235 "OpDecorate %frag_vtxColor Location 1\n"
4237 SPIRV_ASSEMBLY_TYPES
4238 SPIRV_ASSEMBLY_CONSTANTS
4239 SPIRV_ASSEMBLY_ARRAYS
4241 "; Vertex Variables\n"
4242 "%vert_vtxPosition = OpVariable %op_v4f32 Output\n"
4243 "%vert_Position = OpVariable %ip_v4f32 Input\n"
4244 "%vert_vtxColor = OpVariable %op_v4f32 Output\n"
4245 "%vert_color = OpVariable %ip_v4f32 Input\n"
4246 "%vert_vertex_id = OpVariable %ip_i32 Input\n"
4247 "%vert_instance_id = OpVariable %ip_i32 Input\n"
4249 "; Geometry Variables\n"
4250 "%geom_per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4251 "%geom_a3_per_vertex_in = OpTypeArray %geom_per_vertex_in %c_u32_3\n"
4252 "%geom_ip_a3_per_vertex_in = OpTypePointer Input %geom_a3_per_vertex_in\n"
4253 "%geom_gl_in = OpVariable %geom_ip_a3_per_vertex_in Input\n"
4254 "%geom_out_color = OpVariable %op_v4f32 Output\n"
4255 "%geom_in_color = OpVariable %ip_a3v4f32 Input\n"
4256 "%geom_out_gl_position = OpVariable %op_v4f32 Output\n"
4258 "; Tessellation Control Variables\n"
4259 "%tessc_out_color = OpVariable %op_a3v4f32 Output\n"
4260 "%tessc_gl_InvocationID = OpVariable %ip_i32 Input\n"
4261 "%tessc_in_color = OpVariable %ip_a32v4f32 Input\n"
4262 "%tessc_out_position = OpVariable %op_a3v4f32 Output\n"
4263 "%tessc_in_position = OpVariable %ip_a32v4f32 Input\n"
4264 "%tessc_gl_TessLevelOuter = OpVariable %op_a4f32 Output\n"
4265 "%tessc_gl_TessLevelInner = OpVariable %op_a2f32 Output\n"
4267 "; Tessellation Evaluation Decorations\n"
4268 "%tesse_per_vertex_out = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4269 "%tesse_op_per_vertex_out = OpTypePointer Output %tesse_per_vertex_out\n"
4270 "%tesse_stream = OpVariable %tesse_op_per_vertex_out Output\n"
4271 "%tesse_gl_tessCoord = OpVariable %ip_v3f32 Input\n"
4272 "%tesse_in_position = OpVariable %ip_a32v4f32 Input\n"
4273 "%tesse_out_color = OpVariable %op_v4f32 Output\n"
4274 "%tesse_in_color = OpVariable %ip_a32v4f32 Input\n"
4276 "; Fragment Variables\n"
4277 "%frag_fragColor = OpVariable %op_v4f32 Output\n"
4278 "%frag_vtxColor = OpVariable %ip_v4f32 Input\n"
4281 "%vert_main = OpFunction %void None %fun\n"
4282 "%vert_label = OpLabel\n"
4283 "%vert_tmp_position = OpLoad %v4f32 %vert_Position\n"
4284 "OpStore %vert_vtxPosition %vert_tmp_position\n"
4285 "%vert_tmp_color = OpLoad %v4f32 %vert_color\n"
4286 "OpStore %vert_vtxColor %vert_tmp_color\n"
4290 "; Geometry Entry\n"
4291 "%geom_main = OpFunction %void None %fun\n"
4292 "%geom_label = OpLabel\n"
4293 "%geom_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %geom_gl_in %c_i32_0 %c_i32_0\n"
4294 "%geom_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %geom_gl_in %c_i32_1 %c_i32_0\n"
4295 "%geom_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %geom_gl_in %c_i32_2 %c_i32_0\n"
4296 "%geom_in_position_0 = OpLoad %v4f32 %geom_gl_in_0_gl_position\n"
4297 "%geom_in_position_1 = OpLoad %v4f32 %geom_gl_in_1_gl_position\n"
4298 "%geom_in_position_2 = OpLoad %v4f32 %geom_gl_in_2_gl_position \n"
4299 "%geom_in_color_0_ptr = OpAccessChain %ip_v4f32 %geom_in_color %c_i32_0\n"
4300 "%geom_in_color_1_ptr = OpAccessChain %ip_v4f32 %geom_in_color %c_i32_1\n"
4301 "%geom_in_color_2_ptr = OpAccessChain %ip_v4f32 %geom_in_color %c_i32_2\n"
4302 "%geom_in_color_0 = OpLoad %v4f32 %geom_in_color_0_ptr\n"
4303 "%geom_in_color_1 = OpLoad %v4f32 %geom_in_color_1_ptr\n"
4304 "%geom_in_color_2 = OpLoad %v4f32 %geom_in_color_2_ptr\n"
4305 "OpStore %geom_out_gl_position %geom_in_position_0\n"
4306 "OpStore %geom_out_color %geom_in_color_0\n"
4308 "OpStore %geom_out_gl_position %geom_in_position_1\n"
4309 "OpStore %geom_out_color %geom_in_color_1\n"
4311 "OpStore %geom_out_gl_position %geom_in_position_2\n"
4312 "OpStore %geom_out_color %geom_in_color_2\n"
4318 "; Tessellation Control Entry\n"
4319 "%tessc_main = OpFunction %void None %fun\n"
4320 "%tessc_label = OpLabel\n"
4321 "%tessc_invocation_id = OpLoad %i32 %tessc_gl_InvocationID\n"
4322 "%tessc_in_color_ptr = OpAccessChain %ip_v4f32 %tessc_in_color %tessc_invocation_id\n"
4323 "%tessc_in_position_ptr = OpAccessChain %ip_v4f32 %tessc_in_position %tessc_invocation_id\n"
4324 "%tessc_in_color_val = OpLoad %v4f32 %tessc_in_color_ptr\n"
4325 "%tessc_in_position_val = OpLoad %v4f32 %tessc_in_position_ptr\n"
4326 "%tessc_out_color_ptr = OpAccessChain %op_v4f32 %tessc_out_color %tessc_invocation_id\n"
4327 "%tessc_out_position_ptr = OpAccessChain %op_v4f32 %tessc_out_position %tessc_invocation_id\n"
4328 "OpStore %tessc_out_color_ptr %tessc_in_color_val\n"
4329 "OpStore %tessc_out_position_ptr %tessc_in_position_val\n"
4330 "%tessc_is_first_invocation = OpIEqual %bool %tessc_invocation_id %c_i32_0\n"
4331 "OpSelectionMerge %tessc_merge_label None\n"
4332 "OpBranchConditional %tessc_is_first_invocation %tessc_first_invocation %tessc_merge_label\n"
4333 "%tessc_first_invocation = OpLabel\n"
4334 "%tessc_tess_outer_0 = OpAccessChain %op_f32 %tessc_gl_TessLevelOuter %c_i32_0\n"
4335 "%tessc_tess_outer_1 = OpAccessChain %op_f32 %tessc_gl_TessLevelOuter %c_i32_1\n"
4336 "%tessc_tess_outer_2 = OpAccessChain %op_f32 %tessc_gl_TessLevelOuter %c_i32_2\n"
4337 "%tessc_tess_inner = OpAccessChain %op_f32 %tessc_gl_TessLevelInner %c_i32_0\n"
4338 "OpStore %tessc_tess_outer_0 %c_f32_1\n"
4339 "OpStore %tessc_tess_outer_1 %c_f32_1\n"
4340 "OpStore %tessc_tess_outer_2 %c_f32_1\n"
4341 "OpStore %tessc_tess_inner %c_f32_1\n"
4342 "OpBranch %tessc_merge_label\n"
4343 "%tessc_merge_label = OpLabel\n"
4347 "; Tessellation Evaluation Entry\n"
4348 "%tesse_main = OpFunction %void None %fun\n"
4349 "%tesse_label = OpLabel\n"
4350 "%tesse_tc_0_ptr = OpAccessChain %ip_f32 %tesse_gl_tessCoord %c_u32_0\n"
4351 "%tesse_tc_1_ptr = OpAccessChain %ip_f32 %tesse_gl_tessCoord %c_u32_1\n"
4352 "%tesse_tc_2_ptr = OpAccessChain %ip_f32 %tesse_gl_tessCoord %c_u32_2\n"
4353 "%tesse_tc_0 = OpLoad %f32 %tesse_tc_0_ptr\n"
4354 "%tesse_tc_1 = OpLoad %f32 %tesse_tc_1_ptr\n"
4355 "%tesse_tc_2 = OpLoad %f32 %tesse_tc_2_ptr\n"
4356 "%tesse_in_pos_0_ptr = OpAccessChain %ip_v4f32 %tesse_in_position %c_i32_0\n"
4357 "%tesse_in_pos_1_ptr = OpAccessChain %ip_v4f32 %tesse_in_position %c_i32_1\n"
4358 "%tesse_in_pos_2_ptr = OpAccessChain %ip_v4f32 %tesse_in_position %c_i32_2\n"
4359 "%tesse_in_pos_0 = OpLoad %v4f32 %tesse_in_pos_0_ptr\n"
4360 "%tesse_in_pos_1 = OpLoad %v4f32 %tesse_in_pos_1_ptr\n"
4361 "%tesse_in_pos_2 = OpLoad %v4f32 %tesse_in_pos_2_ptr\n"
4362 "%tesse_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_0 %tesse_in_pos_0\n"
4363 "%tesse_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_1 %tesse_in_pos_1\n"
4364 "%tesse_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_2 %tesse_in_pos_2\n"
4365 "%tesse_out_pos_ptr = OpAccessChain %op_v4f32 %tesse_stream %c_i32_0\n"
4366 "%tesse_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %tesse_in_pos_0_weighted %tesse_in_pos_1_weighted\n"
4367 "%tesse_computed_out = OpFAdd %v4f32 %tesse_in_pos_0_plus_pos_1 %tesse_in_pos_2_weighted\n"
4368 "OpStore %tesse_out_pos_ptr %tesse_computed_out\n"
4369 "%tesse_in_clr_0_ptr = OpAccessChain %ip_v4f32 %tesse_in_color %c_i32_0\n"
4370 "%tesse_in_clr_1_ptr = OpAccessChain %ip_v4f32 %tesse_in_color %c_i32_1\n"
4371 "%tesse_in_clr_2_ptr = OpAccessChain %ip_v4f32 %tesse_in_color %c_i32_2\n"
4372 "%tesse_in_clr_0 = OpLoad %v4f32 %tesse_in_clr_0_ptr\n"
4373 "%tesse_in_clr_1 = OpLoad %v4f32 %tesse_in_clr_1_ptr\n"
4374 "%tesse_in_clr_2 = OpLoad %v4f32 %tesse_in_clr_2_ptr\n"
4375 "%tesse_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_0 %tesse_in_clr_0\n"
4376 "%tesse_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_1 %tesse_in_clr_1\n"
4377 "%tesse_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_2 %tesse_in_clr_2\n"
4378 "%tesse_in_clr_0_plus_col_1 = OpFAdd %v4f32 %tesse_in_clr_0_weighted %tesse_in_clr_1_weighted\n"
4379 "%tesse_computed_clr = OpFAdd %v4f32 %tesse_in_clr_0_plus_col_1 %tesse_in_clr_2_weighted\n"
4380 "OpStore %tesse_out_color %tesse_computed_clr\n"
4384 "; Fragment Entry\n"
4385 "%frag_main = OpFunction %void None %fun\n"
4386 "%frag_label_main = OpLabel\n"
4387 "%frag_tmp1 = OpLoad %v4f32 %frag_vtxColor\n"
4388 "OpStore %frag_fragColor %frag_tmp1\n"
4393 // This has two shaders of each stage. The first
4394 // is a passthrough, the second inverts the color.
4395 void createMultipleEntries(vk::SourceCollections& dst, InstanceContext)
4397 dst.spirvAsmSources.add("vert") <<
4398 // This module contains 2 vertex shaders. One that is a passthrough
4399 // and a second that inverts the color of the output (1.0 - color).
4400 "OpCapability Shader\n"
4401 "OpMemoryModel Logical GLSL450\n"
4402 "OpEntryPoint Vertex %main \"vert1\" %Position %vtxColor %color %vtxPosition %vertex_id %instance_id\n"
4403 "OpEntryPoint Vertex %main2 \"vert2\" %Position %vtxColor %color %vtxPosition %vertex_id %instance_id\n"
4405 "OpName %main \"vert1\"\n"
4406 "OpName %main2 \"vert2\"\n"
4407 "OpName %vtxPosition \"vtxPosition\"\n"
4408 "OpName %Position \"position\"\n"
4409 "OpName %vtxColor \"vtxColor\"\n"
4410 "OpName %color \"color\"\n"
4411 "OpName %vertex_id \"gl_VertexIndex\"\n"
4412 "OpName %instance_id \"gl_InstanceIndex\"\n"
4414 "OpDecorate %vtxPosition Location 2\n"
4415 "OpDecorate %Position Location 0\n"
4416 "OpDecorate %vtxColor Location 1\n"
4417 "OpDecorate %color Location 1\n"
4418 "OpDecorate %vertex_id BuiltIn VertexIndex\n"
4419 "OpDecorate %instance_id BuiltIn InstanceIndex\n"
4420 SPIRV_ASSEMBLY_TYPES
4421 SPIRV_ASSEMBLY_CONSTANTS
4422 SPIRV_ASSEMBLY_ARRAYS
4423 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4424 "%vtxPosition = OpVariable %op_v4f32 Output\n"
4425 "%Position = OpVariable %ip_v4f32 Input\n"
4426 "%vtxColor = OpVariable %op_v4f32 Output\n"
4427 "%color = OpVariable %ip_v4f32 Input\n"
4428 "%vertex_id = OpVariable %ip_i32 Input\n"
4429 "%instance_id = OpVariable %ip_i32 Input\n"
4431 "%main = OpFunction %void None %fun\n"
4432 "%label = OpLabel\n"
4433 "%tmp_position = OpLoad %v4f32 %Position\n"
4434 "OpStore %vtxPosition %tmp_position\n"
4435 "%tmp_color = OpLoad %v4f32 %color\n"
4436 "OpStore %vtxColor %tmp_color\n"
4440 "%main2 = OpFunction %void None %fun\n"
4441 "%label2 = OpLabel\n"
4442 "%tmp_position2 = OpLoad %v4f32 %Position\n"
4443 "OpStore %vtxPosition %tmp_position2\n"
4444 "%tmp_color2 = OpLoad %v4f32 %color\n"
4445 "%tmp_color3 = OpFSub %v4f32 %cval %tmp_color2\n"
4446 "%tmp_color4 = OpVectorInsertDynamic %v4f32 %tmp_color3 %c_f32_1 %c_i32_3\n"
4447 "OpStore %vtxColor %tmp_color4\n"
4451 dst.spirvAsmSources.add("frag") <<
4452 // This is a single module that contains 2 fragment shaders.
4453 // One that passes color through and the other that inverts the output
4454 // color (1.0 - color).
4455 "OpCapability Shader\n"
4456 "OpMemoryModel Logical GLSL450\n"
4457 "OpEntryPoint Fragment %main \"frag1\" %vtxColor %fragColor\n"
4458 "OpEntryPoint Fragment %main2 \"frag2\" %vtxColor %fragColor\n"
4459 "OpExecutionMode %main OriginUpperLeft\n"
4460 "OpExecutionMode %main2 OriginUpperLeft\n"
4462 "OpName %main \"frag1\"\n"
4463 "OpName %main2 \"frag2\"\n"
4464 "OpName %fragColor \"fragColor\"\n"
4465 "OpName %vtxColor \"vtxColor\"\n"
4466 "OpDecorate %fragColor Location 0\n"
4467 "OpDecorate %vtxColor Location 1\n"
4468 SPIRV_ASSEMBLY_TYPES
4469 SPIRV_ASSEMBLY_CONSTANTS
4470 SPIRV_ASSEMBLY_ARRAYS
4471 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4472 "%fragColor = OpVariable %op_v4f32 Output\n"
4473 "%vtxColor = OpVariable %ip_v4f32 Input\n"
4475 "%main = OpFunction %void None %fun\n"
4476 "%label_main = OpLabel\n"
4477 "%tmp1 = OpLoad %v4f32 %vtxColor\n"
4478 "OpStore %fragColor %tmp1\n"
4482 "%main2 = OpFunction %void None %fun\n"
4483 "%label_main2 = OpLabel\n"
4484 "%tmp2 = OpLoad %v4f32 %vtxColor\n"
4485 "%tmp3 = OpFSub %v4f32 %cval %tmp2\n"
4486 "%tmp4 = OpVectorInsertDynamic %v4f32 %tmp3 %c_f32_1 %c_i32_3\n"
4487 "OpStore %fragColor %tmp4\n"
4491 dst.spirvAsmSources.add("geom") <<
4492 "OpCapability Geometry\n"
4493 "OpCapability ClipDistance\n"
4494 "OpCapability CullDistance\n"
4495 "OpMemoryModel Logical GLSL450\n"
4496 "OpEntryPoint Geometry %geom1_main \"geom1\" %out_gl_position %gl_in %out_color %in_color\n"
4497 "OpEntryPoint Geometry %geom2_main \"geom2\" %out_gl_position %gl_in %out_color %in_color\n"
4498 "OpExecutionMode %geom1_main Triangles\n"
4499 "OpExecutionMode %geom2_main Triangles\n"
4500 "OpExecutionMode %geom1_main OutputTriangleStrip\n"
4501 "OpExecutionMode %geom2_main OutputTriangleStrip\n"
4502 "OpExecutionMode %geom1_main OutputVertices 3\n"
4503 "OpExecutionMode %geom2_main OutputVertices 3\n"
4504 "OpName %geom1_main \"geom1\"\n"
4505 "OpName %geom2_main \"geom2\"\n"
4506 "OpName %per_vertex_in \"gl_PerVertex\"\n"
4507 "OpMemberName %per_vertex_in 0 \"gl_Position\"\n"
4508 "OpMemberName %per_vertex_in 1 \"gl_PointSize\"\n"
4509 "OpMemberName %per_vertex_in 2 \"gl_ClipDistance\"\n"
4510 "OpMemberName %per_vertex_in 3 \"gl_CullDistance\"\n"
4511 "OpName %gl_in \"gl_in\"\n"
4512 "OpName %out_color \"out_color\"\n"
4513 "OpName %in_color \"in_color\"\n"
4514 "OpDecorate %out_gl_position BuiltIn Position\n"
4515 "OpMemberDecorate %per_vertex_in 0 BuiltIn Position\n"
4516 "OpMemberDecorate %per_vertex_in 1 BuiltIn PointSize\n"
4517 "OpMemberDecorate %per_vertex_in 2 BuiltIn ClipDistance\n"
4518 "OpMemberDecorate %per_vertex_in 3 BuiltIn CullDistance\n"
4519 "OpDecorate %per_vertex_in Block\n"
4520 "OpDecorate %out_color Location 1\n"
4521 "OpDecorate %in_color Location 1\n"
4522 SPIRV_ASSEMBLY_TYPES
4523 SPIRV_ASSEMBLY_CONSTANTS
4524 SPIRV_ASSEMBLY_ARRAYS
4525 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4526 "%per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4527 "%a3_per_vertex_in = OpTypeArray %per_vertex_in %c_u32_3\n"
4528 "%ip_a3_per_vertex_in = OpTypePointer Input %a3_per_vertex_in\n"
4529 "%gl_in = OpVariable %ip_a3_per_vertex_in Input\n"
4530 "%out_color = OpVariable %op_v4f32 Output\n"
4531 "%in_color = OpVariable %ip_a3v4f32 Input\n"
4532 "%out_gl_position = OpVariable %op_v4f32 Output\n"
4534 "%geom1_main = OpFunction %void None %fun\n"
4535 "%geom1_label = OpLabel\n"
4536 "%geom1_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_0 %c_i32_0\n"
4537 "%geom1_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_1 %c_i32_0\n"
4538 "%geom1_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_2 %c_i32_0\n"
4539 "%geom1_in_position_0 = OpLoad %v4f32 %geom1_gl_in_0_gl_position\n"
4540 "%geom1_in_position_1 = OpLoad %v4f32 %geom1_gl_in_1_gl_position\n"
4541 "%geom1_in_position_2 = OpLoad %v4f32 %geom1_gl_in_2_gl_position \n"
4542 "%geom1_in_color_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4543 "%geom1_in_color_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4544 "%geom1_in_color_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4545 "%geom1_in_color_0 = OpLoad %v4f32 %geom1_in_color_0_ptr\n"
4546 "%geom1_in_color_1 = OpLoad %v4f32 %geom1_in_color_1_ptr\n"
4547 "%geom1_in_color_2 = OpLoad %v4f32 %geom1_in_color_2_ptr\n"
4548 "OpStore %out_gl_position %geom1_in_position_0\n"
4549 "OpStore %out_color %geom1_in_color_0\n"
4551 "OpStore %out_gl_position %geom1_in_position_1\n"
4552 "OpStore %out_color %geom1_in_color_1\n"
4554 "OpStore %out_gl_position %geom1_in_position_2\n"
4555 "OpStore %out_color %geom1_in_color_2\n"
4561 "%geom2_main = OpFunction %void None %fun\n"
4562 "%geom2_label = OpLabel\n"
4563 "%geom2_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_0 %c_i32_0\n"
4564 "%geom2_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_1 %c_i32_0\n"
4565 "%geom2_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_2 %c_i32_0\n"
4566 "%geom2_in_position_0 = OpLoad %v4f32 %geom2_gl_in_0_gl_position\n"
4567 "%geom2_in_position_1 = OpLoad %v4f32 %geom2_gl_in_1_gl_position\n"
4568 "%geom2_in_position_2 = OpLoad %v4f32 %geom2_gl_in_2_gl_position \n"
4569 "%geom2_in_color_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4570 "%geom2_in_color_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4571 "%geom2_in_color_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4572 "%geom2_in_color_0 = OpLoad %v4f32 %geom2_in_color_0_ptr\n"
4573 "%geom2_in_color_1 = OpLoad %v4f32 %geom2_in_color_1_ptr\n"
4574 "%geom2_in_color_2 = OpLoad %v4f32 %geom2_in_color_2_ptr\n"
4575 "%geom2_transformed_in_color_0 = OpFSub %v4f32 %cval %geom2_in_color_0\n"
4576 "%geom2_transformed_in_color_1 = OpFSub %v4f32 %cval %geom2_in_color_1\n"
4577 "%geom2_transformed_in_color_2 = OpFSub %v4f32 %cval %geom2_in_color_2\n"
4578 "%geom2_transformed_in_color_0_a = OpVectorInsertDynamic %v4f32 %geom2_transformed_in_color_0 %c_f32_1 %c_i32_3\n"
4579 "%geom2_transformed_in_color_1_a = OpVectorInsertDynamic %v4f32 %geom2_transformed_in_color_1 %c_f32_1 %c_i32_3\n"
4580 "%geom2_transformed_in_color_2_a = OpVectorInsertDynamic %v4f32 %geom2_transformed_in_color_2 %c_f32_1 %c_i32_3\n"
4581 "OpStore %out_gl_position %geom2_in_position_0\n"
4582 "OpStore %out_color %geom2_transformed_in_color_0_a\n"
4584 "OpStore %out_gl_position %geom2_in_position_1\n"
4585 "OpStore %out_color %geom2_transformed_in_color_1_a\n"
4587 "OpStore %out_gl_position %geom2_in_position_2\n"
4588 "OpStore %out_color %geom2_transformed_in_color_2_a\n"
4594 dst.spirvAsmSources.add("tessc") <<
4595 "OpCapability Tessellation\n"
4596 "OpMemoryModel Logical GLSL450\n"
4597 "OpEntryPoint TessellationControl %tessc1_main \"tessc1\" %out_color %gl_InvocationID %in_color %out_position %in_position %gl_TessLevelOuter %gl_TessLevelInner\n"
4598 "OpEntryPoint TessellationControl %tessc2_main \"tessc2\" %out_color %gl_InvocationID %in_color %out_position %in_position %gl_TessLevelOuter %gl_TessLevelInner\n"
4599 "OpExecutionMode %tessc1_main OutputVertices 3\n"
4600 "OpExecutionMode %tessc2_main OutputVertices 3\n"
4601 "OpName %tessc1_main \"tessc1\"\n"
4602 "OpName %tessc2_main \"tessc2\"\n"
4603 "OpName %out_color \"out_color\"\n"
4604 "OpName %gl_InvocationID \"gl_InvocationID\"\n"
4605 "OpName %in_color \"in_color\"\n"
4606 "OpName %out_position \"out_position\"\n"
4607 "OpName %in_position \"in_position\"\n"
4608 "OpName %gl_TessLevelOuter \"gl_TessLevelOuter\"\n"
4609 "OpName %gl_TessLevelInner \"gl_TessLevelInner\"\n"
4610 "OpDecorate %out_color Location 1\n"
4611 "OpDecorate %gl_InvocationID BuiltIn InvocationId\n"
4612 "OpDecorate %in_color Location 1\n"
4613 "OpDecorate %out_position Location 2\n"
4614 "OpDecorate %in_position Location 2\n"
4615 "OpDecorate %gl_TessLevelOuter Patch\n"
4616 "OpDecorate %gl_TessLevelOuter BuiltIn TessLevelOuter\n"
4617 "OpDecorate %gl_TessLevelInner Patch\n"
4618 "OpDecorate %gl_TessLevelInner BuiltIn TessLevelInner\n"
4619 SPIRV_ASSEMBLY_TYPES
4620 SPIRV_ASSEMBLY_CONSTANTS
4621 SPIRV_ASSEMBLY_ARRAYS
4622 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4623 "%out_color = OpVariable %op_a3v4f32 Output\n"
4624 "%gl_InvocationID = OpVariable %ip_i32 Input\n"
4625 "%in_color = OpVariable %ip_a32v4f32 Input\n"
4626 "%out_position = OpVariable %op_a3v4f32 Output\n"
4627 "%in_position = OpVariable %ip_a32v4f32 Input\n"
4628 "%gl_TessLevelOuter = OpVariable %op_a4f32 Output\n"
4629 "%gl_TessLevelInner = OpVariable %op_a2f32 Output\n"
4631 "%tessc1_main = OpFunction %void None %fun\n"
4632 "%tessc1_label = OpLabel\n"
4633 "%tessc1_invocation_id = OpLoad %i32 %gl_InvocationID\n"
4634 "%tessc1_in_color_ptr = OpAccessChain %ip_v4f32 %in_color %tessc1_invocation_id\n"
4635 "%tessc1_in_position_ptr = OpAccessChain %ip_v4f32 %in_position %tessc1_invocation_id\n"
4636 "%tessc1_in_color_val = OpLoad %v4f32 %tessc1_in_color_ptr\n"
4637 "%tessc1_in_position_val = OpLoad %v4f32 %tessc1_in_position_ptr\n"
4638 "%tessc1_out_color_ptr = OpAccessChain %op_v4f32 %out_color %tessc1_invocation_id\n"
4639 "%tessc1_out_position_ptr = OpAccessChain %op_v4f32 %out_position %tessc1_invocation_id\n"
4640 "OpStore %tessc1_out_color_ptr %tessc1_in_color_val\n"
4641 "OpStore %tessc1_out_position_ptr %tessc1_in_position_val\n"
4642 "%tessc1_is_first_invocation = OpIEqual %bool %tessc1_invocation_id %c_i32_0\n"
4643 "OpSelectionMerge %tessc1_merge_label None\n"
4644 "OpBranchConditional %tessc1_is_first_invocation %tessc1_first_invocation %tessc1_merge_label\n"
4645 "%tessc1_first_invocation = OpLabel\n"
4646 "%tessc1_tess_outer_0 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_0\n"
4647 "%tessc1_tess_outer_1 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_1\n"
4648 "%tessc1_tess_outer_2 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_2\n"
4649 "%tessc1_tess_inner = OpAccessChain %op_f32 %gl_TessLevelInner %c_i32_0\n"
4650 "OpStore %tessc1_tess_outer_0 %c_f32_1\n"
4651 "OpStore %tessc1_tess_outer_1 %c_f32_1\n"
4652 "OpStore %tessc1_tess_outer_2 %c_f32_1\n"
4653 "OpStore %tessc1_tess_inner %c_f32_1\n"
4654 "OpBranch %tessc1_merge_label\n"
4655 "%tessc1_merge_label = OpLabel\n"
4659 "%tessc2_main = OpFunction %void None %fun\n"
4660 "%tessc2_label = OpLabel\n"
4661 "%tessc2_invocation_id = OpLoad %i32 %gl_InvocationID\n"
4662 "%tessc2_in_color_ptr = OpAccessChain %ip_v4f32 %in_color %tessc2_invocation_id\n"
4663 "%tessc2_in_position_ptr = OpAccessChain %ip_v4f32 %in_position %tessc2_invocation_id\n"
4664 "%tessc2_in_color_val = OpLoad %v4f32 %tessc2_in_color_ptr\n"
4665 "%tessc2_in_position_val = OpLoad %v4f32 %tessc2_in_position_ptr\n"
4666 "%tessc2_out_color_ptr = OpAccessChain %op_v4f32 %out_color %tessc2_invocation_id\n"
4667 "%tessc2_out_position_ptr = OpAccessChain %op_v4f32 %out_position %tessc2_invocation_id\n"
4668 "%tessc2_transformed_color = OpFSub %v4f32 %cval %tessc2_in_color_val\n"
4669 "%tessc2_transformed_color_a = OpVectorInsertDynamic %v4f32 %tessc2_transformed_color %c_f32_1 %c_i32_3\n"
4670 "OpStore %tessc2_out_color_ptr %tessc2_transformed_color_a\n"
4671 "OpStore %tessc2_out_position_ptr %tessc2_in_position_val\n"
4672 "%tessc2_is_first_invocation = OpIEqual %bool %tessc2_invocation_id %c_i32_0\n"
4673 "OpSelectionMerge %tessc2_merge_label None\n"
4674 "OpBranchConditional %tessc2_is_first_invocation %tessc2_first_invocation %tessc2_merge_label\n"
4675 "%tessc2_first_invocation = OpLabel\n"
4676 "%tessc2_tess_outer_0 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_0\n"
4677 "%tessc2_tess_outer_1 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_1\n"
4678 "%tessc2_tess_outer_2 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_2\n"
4679 "%tessc2_tess_inner = OpAccessChain %op_f32 %gl_TessLevelInner %c_i32_0\n"
4680 "OpStore %tessc2_tess_outer_0 %c_f32_1\n"
4681 "OpStore %tessc2_tess_outer_1 %c_f32_1\n"
4682 "OpStore %tessc2_tess_outer_2 %c_f32_1\n"
4683 "OpStore %tessc2_tess_inner %c_f32_1\n"
4684 "OpBranch %tessc2_merge_label\n"
4685 "%tessc2_merge_label = OpLabel\n"
4689 dst.spirvAsmSources.add("tesse") <<
4690 "OpCapability Tessellation\n"
4691 "OpCapability ClipDistance\n"
4692 "OpCapability CullDistance\n"
4693 "OpMemoryModel Logical GLSL450\n"
4694 "OpEntryPoint TessellationEvaluation %tesse1_main \"tesse1\" %stream %gl_tessCoord %in_position %out_color %in_color \n"
4695 "OpEntryPoint TessellationEvaluation %tesse2_main \"tesse2\" %stream %gl_tessCoord %in_position %out_color %in_color \n"
4696 "OpExecutionMode %tesse1_main Triangles\n"
4697 "OpExecutionMode %tesse2_main Triangles\n"
4698 "OpName %tesse1_main \"tesse1\"\n"
4699 "OpName %tesse2_main \"tesse2\"\n"
4700 "OpName %per_vertex_out \"gl_PerVertex\"\n"
4701 "OpMemberName %per_vertex_out 0 \"gl_Position\"\n"
4702 "OpMemberName %per_vertex_out 1 \"gl_PointSize\"\n"
4703 "OpMemberName %per_vertex_out 2 \"gl_ClipDistance\"\n"
4704 "OpMemberName %per_vertex_out 3 \"gl_CullDistance\"\n"
4705 "OpName %stream \"\"\n"
4706 "OpName %gl_tessCoord \"gl_TessCoord\"\n"
4707 "OpName %in_position \"in_position\"\n"
4708 "OpName %out_color \"out_color\"\n"
4709 "OpName %in_color \"in_color\"\n"
4710 "OpMemberDecorate %per_vertex_out 0 BuiltIn Position\n"
4711 "OpMemberDecorate %per_vertex_out 1 BuiltIn PointSize\n"
4712 "OpMemberDecorate %per_vertex_out 2 BuiltIn ClipDistance\n"
4713 "OpMemberDecorate %per_vertex_out 3 BuiltIn CullDistance\n"
4714 "OpDecorate %per_vertex_out Block\n"
4715 "OpDecorate %gl_tessCoord BuiltIn TessCoord\n"
4716 "OpDecorate %in_position Location 2\n"
4717 "OpDecorate %out_color Location 1\n"
4718 "OpDecorate %in_color Location 1\n"
4719 SPIRV_ASSEMBLY_TYPES
4720 SPIRV_ASSEMBLY_CONSTANTS
4721 SPIRV_ASSEMBLY_ARRAYS
4722 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4723 "%per_vertex_out = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4724 "%op_per_vertex_out = OpTypePointer Output %per_vertex_out\n"
4725 "%stream = OpVariable %op_per_vertex_out Output\n"
4726 "%gl_tessCoord = OpVariable %ip_v3f32 Input\n"
4727 "%in_position = OpVariable %ip_a32v4f32 Input\n"
4728 "%out_color = OpVariable %op_v4f32 Output\n"
4729 "%in_color = OpVariable %ip_a32v4f32 Input\n"
4731 "%tesse1_main = OpFunction %void None %fun\n"
4732 "%tesse1_label = OpLabel\n"
4733 "%tesse1_tc_0_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_0\n"
4734 "%tesse1_tc_1_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_1\n"
4735 "%tesse1_tc_2_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_2\n"
4736 "%tesse1_tc_0 = OpLoad %f32 %tesse1_tc_0_ptr\n"
4737 "%tesse1_tc_1 = OpLoad %f32 %tesse1_tc_1_ptr\n"
4738 "%tesse1_tc_2 = OpLoad %f32 %tesse1_tc_2_ptr\n"
4739 "%tesse1_in_pos_0_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_0\n"
4740 "%tesse1_in_pos_1_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_1\n"
4741 "%tesse1_in_pos_2_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_2\n"
4742 "%tesse1_in_pos_0 = OpLoad %v4f32 %tesse1_in_pos_0_ptr\n"
4743 "%tesse1_in_pos_1 = OpLoad %v4f32 %tesse1_in_pos_1_ptr\n"
4744 "%tesse1_in_pos_2 = OpLoad %v4f32 %tesse1_in_pos_2_ptr\n"
4745 "%tesse1_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_0 %tesse1_in_pos_0\n"
4746 "%tesse1_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_1 %tesse1_in_pos_1\n"
4747 "%tesse1_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_2 %tesse1_in_pos_2\n"
4748 "%tesse1_out_pos_ptr = OpAccessChain %op_v4f32 %stream %c_i32_0\n"
4749 "%tesse1_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %tesse1_in_pos_0_weighted %tesse1_in_pos_1_weighted\n"
4750 "%tesse1_computed_out = OpFAdd %v4f32 %tesse1_in_pos_0_plus_pos_1 %tesse1_in_pos_2_weighted\n"
4751 "OpStore %tesse1_out_pos_ptr %tesse1_computed_out\n"
4752 "%tesse1_in_clr_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4753 "%tesse1_in_clr_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4754 "%tesse1_in_clr_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4755 "%tesse1_in_clr_0 = OpLoad %v4f32 %tesse1_in_clr_0_ptr\n"
4756 "%tesse1_in_clr_1 = OpLoad %v4f32 %tesse1_in_clr_1_ptr\n"
4757 "%tesse1_in_clr_2 = OpLoad %v4f32 %tesse1_in_clr_2_ptr\n"
4758 "%tesse1_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_0 %tesse1_in_clr_0\n"
4759 "%tesse1_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_1 %tesse1_in_clr_1\n"
4760 "%tesse1_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_2 %tesse1_in_clr_2\n"
4761 "%tesse1_in_clr_0_plus_col_1 = OpFAdd %v4f32 %tesse1_in_clr_0_weighted %tesse1_in_clr_1_weighted\n"
4762 "%tesse1_computed_clr = OpFAdd %v4f32 %tesse1_in_clr_0_plus_col_1 %tesse1_in_clr_2_weighted\n"
4763 "OpStore %out_color %tesse1_computed_clr\n"
4767 "%tesse2_main = OpFunction %void None %fun\n"
4768 "%tesse2_label = OpLabel\n"
4769 "%tesse2_tc_0_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_0\n"
4770 "%tesse2_tc_1_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_1\n"
4771 "%tesse2_tc_2_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_2\n"
4772 "%tesse2_tc_0 = OpLoad %f32 %tesse2_tc_0_ptr\n"
4773 "%tesse2_tc_1 = OpLoad %f32 %tesse2_tc_1_ptr\n"
4774 "%tesse2_tc_2 = OpLoad %f32 %tesse2_tc_2_ptr\n"
4775 "%tesse2_in_pos_0_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_0\n"
4776 "%tesse2_in_pos_1_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_1\n"
4777 "%tesse2_in_pos_2_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_2\n"
4778 "%tesse2_in_pos_0 = OpLoad %v4f32 %tesse2_in_pos_0_ptr\n"
4779 "%tesse2_in_pos_1 = OpLoad %v4f32 %tesse2_in_pos_1_ptr\n"
4780 "%tesse2_in_pos_2 = OpLoad %v4f32 %tesse2_in_pos_2_ptr\n"
4781 "%tesse2_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_0 %tesse2_in_pos_0\n"
4782 "%tesse2_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_1 %tesse2_in_pos_1\n"
4783 "%tesse2_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_2 %tesse2_in_pos_2\n"
4784 "%tesse2_out_pos_ptr = OpAccessChain %op_v4f32 %stream %c_i32_0\n"
4785 "%tesse2_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %tesse2_in_pos_0_weighted %tesse2_in_pos_1_weighted\n"
4786 "%tesse2_computed_out = OpFAdd %v4f32 %tesse2_in_pos_0_plus_pos_1 %tesse2_in_pos_2_weighted\n"
4787 "OpStore %tesse2_out_pos_ptr %tesse2_computed_out\n"
4788 "%tesse2_in_clr_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4789 "%tesse2_in_clr_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4790 "%tesse2_in_clr_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4791 "%tesse2_in_clr_0 = OpLoad %v4f32 %tesse2_in_clr_0_ptr\n"
4792 "%tesse2_in_clr_1 = OpLoad %v4f32 %tesse2_in_clr_1_ptr\n"
4793 "%tesse2_in_clr_2 = OpLoad %v4f32 %tesse2_in_clr_2_ptr\n"
4794 "%tesse2_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_0 %tesse2_in_clr_0\n"
4795 "%tesse2_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_1 %tesse2_in_clr_1\n"
4796 "%tesse2_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_2 %tesse2_in_clr_2\n"
4797 "%tesse2_in_clr_0_plus_col_1 = OpFAdd %v4f32 %tesse2_in_clr_0_weighted %tesse2_in_clr_1_weighted\n"
4798 "%tesse2_computed_clr = OpFAdd %v4f32 %tesse2_in_clr_0_plus_col_1 %tesse2_in_clr_2_weighted\n"
4799 "%tesse2_clr_transformed = OpFSub %v4f32 %cval %tesse2_computed_clr\n"
4800 "%tesse2_clr_transformed_a = OpVectorInsertDynamic %v4f32 %tesse2_clr_transformed %c_f32_1 %c_i32_3\n"
4801 "OpStore %out_color %tesse2_clr_transformed_a\n"
4806 // Sets up and runs a Vulkan pipeline, then spot-checks the resulting image.
4807 // Feeds the pipeline a set of colored triangles, which then must occur in the
4808 // rendered image. The surface is cleared before executing the pipeline, so
4809 // whatever the shaders draw can be directly spot-checked.
4810 TestStatus runAndVerifyDefaultPipeline (Context& context, InstanceContext instance)
4812 const VkDevice vkDevice = context.getDevice();
4813 const DeviceInterface& vk = context.getDeviceInterface();
4814 const VkQueue queue = context.getUniversalQueue();
4815 const deUint32 queueFamilyIndex = context.getUniversalQueueFamilyIndex();
4816 const tcu::UVec2 renderSize (256, 256);
4817 vector<ModuleHandleSp> modules;
4818 map<VkShaderStageFlagBits, VkShaderModule> moduleByStage;
4819 const int testSpecificSeed = 31354125;
4820 const int seed = context.getTestContext().getCommandLine().getBaseSeed() ^ testSpecificSeed;
4821 bool supportsGeometry = false;
4822 bool supportsTessellation = false;
4823 bool hasTessellation = false;
4825 const VkPhysicalDeviceFeatures& features = context.getDeviceFeatures();
4826 supportsGeometry = features.geometryShader == VK_TRUE;
4827 supportsTessellation = features.tessellationShader == VK_TRUE;
4828 hasTessellation = (instance.requiredStages & VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT) ||
4829 (instance.requiredStages & VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT);
4831 if (hasTessellation && !supportsTessellation)
4833 throw tcu::NotSupportedError(std::string("Tessellation not supported"));
4836 if ((instance.requiredStages & VK_SHADER_STAGE_GEOMETRY_BIT) &&
4839 throw tcu::NotSupportedError(std::string("Geometry not supported"));
4842 de::Random(seed).shuffle(instance.inputColors, instance.inputColors+4);
4843 de::Random(seed).shuffle(instance.outputColors, instance.outputColors+4);
4844 const Vec4 vertexData[] =
4846 // Upper left corner:
4847 Vec4(-1.0f, -1.0f, 0.0f, 1.0f), instance.inputColors[0].toVec(),
4848 Vec4(-0.5f, -1.0f, 0.0f, 1.0f), instance.inputColors[0].toVec(),
4849 Vec4(-1.0f, -0.5f, 0.0f, 1.0f), instance.inputColors[0].toVec(),
4851 // Upper right corner:
4852 Vec4(+0.5f, -1.0f, 0.0f, 1.0f), instance.inputColors[1].toVec(),
4853 Vec4(+1.0f, -1.0f, 0.0f, 1.0f), instance.inputColors[1].toVec(),
4854 Vec4(+1.0f, -0.5f, 0.0f, 1.0f), instance.inputColors[1].toVec(),
4856 // Lower left corner:
4857 Vec4(-1.0f, +0.5f, 0.0f, 1.0f), instance.inputColors[2].toVec(),
4858 Vec4(-0.5f, +1.0f, 0.0f, 1.0f), instance.inputColors[2].toVec(),
4859 Vec4(-1.0f, +1.0f, 0.0f, 1.0f), instance.inputColors[2].toVec(),
4861 // Lower right corner:
4862 Vec4(+1.0f, +0.5f, 0.0f, 1.0f), instance.inputColors[3].toVec(),
4863 Vec4(+1.0f, +1.0f, 0.0f, 1.0f), instance.inputColors[3].toVec(),
4864 Vec4(+0.5f, +1.0f, 0.0f, 1.0f), instance.inputColors[3].toVec()
4866 const size_t singleVertexDataSize = 2 * sizeof(Vec4);
4867 const size_t vertexCount = sizeof(vertexData) / singleVertexDataSize;
4869 const VkBufferCreateInfo vertexBufferParams =
4871 VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType;
4872 DE_NULL, // const void* pNext;
4873 0u, // VkBufferCreateFlags flags;
4874 (VkDeviceSize)sizeof(vertexData), // VkDeviceSize size;
4875 VK_BUFFER_USAGE_VERTEX_BUFFER_BIT, // VkBufferUsageFlags usage;
4876 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
4877 1u, // deUint32 queueFamilyCount;
4878 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices;
4880 const Unique<VkBuffer> vertexBuffer (createBuffer(vk, vkDevice, &vertexBufferParams));
4881 const UniquePtr<Allocation> vertexBufferMemory (context.getDefaultAllocator().allocate(getBufferMemoryRequirements(vk, vkDevice, *vertexBuffer), MemoryRequirement::HostVisible));
4883 VK_CHECK(vk.bindBufferMemory(vkDevice, *vertexBuffer, vertexBufferMemory->getMemory(), vertexBufferMemory->getOffset()));
4885 const VkDeviceSize imageSizeBytes = (VkDeviceSize)(sizeof(deUint32)*renderSize.x()*renderSize.y());
4886 const VkBufferCreateInfo readImageBufferParams =
4888 VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType;
4889 DE_NULL, // const void* pNext;
4890 0u, // VkBufferCreateFlags flags;
4891 imageSizeBytes, // VkDeviceSize size;
4892 VK_BUFFER_USAGE_TRANSFER_DST_BIT, // VkBufferUsageFlags usage;
4893 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
4894 1u, // deUint32 queueFamilyCount;
4895 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices;
4897 const Unique<VkBuffer> readImageBuffer (createBuffer(vk, vkDevice, &readImageBufferParams));
4898 const UniquePtr<Allocation> readImageBufferMemory (context.getDefaultAllocator().allocate(getBufferMemoryRequirements(vk, vkDevice, *readImageBuffer), MemoryRequirement::HostVisible));
4900 VK_CHECK(vk.bindBufferMemory(vkDevice, *readImageBuffer, readImageBufferMemory->getMemory(), readImageBufferMemory->getOffset()));
4902 const VkImageCreateInfo imageParams =
4904 VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, // VkStructureType sType;
4905 DE_NULL, // const void* pNext;
4906 0u, // VkImageCreateFlags flags;
4907 VK_IMAGE_TYPE_2D, // VkImageType imageType;
4908 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format;
4909 { renderSize.x(), renderSize.y(), 1 }, // VkExtent3D extent;
4910 1u, // deUint32 mipLevels;
4911 1u, // deUint32 arraySize;
4912 VK_SAMPLE_COUNT_1_BIT, // deUint32 samples;
4913 VK_IMAGE_TILING_OPTIMAL, // VkImageTiling tiling;
4914 VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT|VK_IMAGE_USAGE_TRANSFER_SRC_BIT, // VkImageUsageFlags usage;
4915 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
4916 1u, // deUint32 queueFamilyCount;
4917 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices;
4918 VK_IMAGE_LAYOUT_UNDEFINED, // VkImageLayout initialLayout;
4921 const Unique<VkImage> image (createImage(vk, vkDevice, &imageParams));
4922 const UniquePtr<Allocation> imageMemory (context.getDefaultAllocator().allocate(getImageMemoryRequirements(vk, vkDevice, *image), MemoryRequirement::Any));
4924 VK_CHECK(vk.bindImageMemory(vkDevice, *image, imageMemory->getMemory(), imageMemory->getOffset()));
4926 const VkAttachmentDescription colorAttDesc =
4928 0u, // VkAttachmentDescriptionFlags flags;
4929 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format;
4930 VK_SAMPLE_COUNT_1_BIT, // deUint32 samples;
4931 VK_ATTACHMENT_LOAD_OP_CLEAR, // VkAttachmentLoadOp loadOp;
4932 VK_ATTACHMENT_STORE_OP_STORE, // VkAttachmentStoreOp storeOp;
4933 VK_ATTACHMENT_LOAD_OP_DONT_CARE, // VkAttachmentLoadOp stencilLoadOp;
4934 VK_ATTACHMENT_STORE_OP_DONT_CARE, // VkAttachmentStoreOp stencilStoreOp;
4935 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout initialLayout;
4936 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout finalLayout;
4938 const VkAttachmentReference colorAttRef =
4940 0u, // deUint32 attachment;
4941 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout layout;
4943 const VkSubpassDescription subpassDesc =
4945 0u, // VkSubpassDescriptionFlags flags;
4946 VK_PIPELINE_BIND_POINT_GRAPHICS, // VkPipelineBindPoint pipelineBindPoint;
4947 0u, // deUint32 inputCount;
4948 DE_NULL, // const VkAttachmentReference* pInputAttachments;
4949 1u, // deUint32 colorCount;
4950 &colorAttRef, // const VkAttachmentReference* pColorAttachments;
4951 DE_NULL, // const VkAttachmentReference* pResolveAttachments;
4952 DE_NULL, // const VkAttachmentReference* pDepthStencilAttachment;
4953 0u, // deUint32 preserveCount;
4954 DE_NULL, // const VkAttachmentReference* pPreserveAttachments;
4957 const VkRenderPassCreateInfo renderPassParams =
4959 VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO, // VkStructureType sType;
4960 DE_NULL, // const void* pNext;
4961 (VkRenderPassCreateFlags)0,
4962 1u, // deUint32 attachmentCount;
4963 &colorAttDesc, // const VkAttachmentDescription* pAttachments;
4964 1u, // deUint32 subpassCount;
4965 &subpassDesc, // const VkSubpassDescription* pSubpasses;
4966 0u, // deUint32 dependencyCount;
4967 DE_NULL, // const VkSubpassDependency* pDependencies;
4969 const Unique<VkRenderPass> renderPass (createRenderPass(vk, vkDevice, &renderPassParams));
4971 const VkImageViewCreateInfo colorAttViewParams =
4973 VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, // VkStructureType sType;
4974 DE_NULL, // const void* pNext;
4975 0u, // VkImageViewCreateFlags flags;
4976 *image, // VkImage image;
4977 VK_IMAGE_VIEW_TYPE_2D, // VkImageViewType viewType;
4978 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format;
4980 VK_COMPONENT_SWIZZLE_R,
4981 VK_COMPONENT_SWIZZLE_G,
4982 VK_COMPONENT_SWIZZLE_B,
4983 VK_COMPONENT_SWIZZLE_A
4984 }, // VkChannelMapping channels;
4986 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspectFlags aspectMask;
4987 0u, // deUint32 baseMipLevel;
4988 1u, // deUint32 mipLevels;
4989 0u, // deUint32 baseArrayLayer;
4990 1u, // deUint32 arraySize;
4991 }, // VkImageSubresourceRange subresourceRange;
4993 const Unique<VkImageView> colorAttView (createImageView(vk, vkDevice, &colorAttViewParams));
4997 const VkPipelineLayoutCreateInfo pipelineLayoutParams =
4999 VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, // VkStructureType sType;
5000 DE_NULL, // const void* pNext;
5001 (VkPipelineLayoutCreateFlags)0,
5002 0u, // deUint32 descriptorSetCount;
5003 DE_NULL, // const VkDescriptorSetLayout* pSetLayouts;
5004 0u, // deUint32 pushConstantRangeCount;
5005 DE_NULL, // const VkPushConstantRange* pPushConstantRanges;
5007 const Unique<VkPipelineLayout> pipelineLayout (createPipelineLayout(vk, vkDevice, &pipelineLayoutParams));
5010 vector<VkPipelineShaderStageCreateInfo> shaderStageParams;
5011 // We need these vectors to make sure that information about specialization constants for each stage can outlive createGraphicsPipeline().
5012 vector<vector<VkSpecializationMapEntry> > specConstantEntries;
5013 vector<VkSpecializationInfo> specializationInfos;
5014 createPipelineShaderStages(vk, vkDevice, instance, context, modules, shaderStageParams);
5016 // And we don't want the reallocation of these vectors to invalidate pointers pointing to their contents.
5017 specConstantEntries.reserve(shaderStageParams.size());
5018 specializationInfos.reserve(shaderStageParams.size());
5020 // Patch the specialization info field in PipelineShaderStageCreateInfos.
5021 for (vector<VkPipelineShaderStageCreateInfo>::iterator stageInfo = shaderStageParams.begin(); stageInfo != shaderStageParams.end(); ++stageInfo)
5023 const StageToSpecConstantMap::const_iterator stageIt = instance.specConstants.find(stageInfo->stage);
5025 if (stageIt != instance.specConstants.end())
5027 const size_t numSpecConstants = stageIt->second.size();
5028 vector<VkSpecializationMapEntry> entries;
5029 VkSpecializationInfo specInfo;
5031 entries.resize(numSpecConstants);
5033 // Only support 32-bit integers as spec constants now. And their constant IDs are numbered sequentially starting from 0.
5034 for (size_t ndx = 0; ndx < numSpecConstants; ++ndx)
5036 entries[ndx].constantID = (deUint32)ndx;
5037 entries[ndx].offset = deUint32(ndx * sizeof(deInt32));
5038 entries[ndx].size = sizeof(deInt32);
5041 specConstantEntries.push_back(entries);
5043 specInfo.mapEntryCount = (deUint32)numSpecConstants;
5044 specInfo.pMapEntries = specConstantEntries.back().data();
5045 specInfo.dataSize = numSpecConstants * sizeof(deInt32);
5046 specInfo.pData = stageIt->second.data();
5047 specializationInfos.push_back(specInfo);
5049 stageInfo->pSpecializationInfo = &specializationInfos.back();
5052 const VkPipelineDepthStencilStateCreateInfo depthStencilParams =
5054 VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO, // VkStructureType sType;
5055 DE_NULL, // const void* pNext;
5056 (VkPipelineDepthStencilStateCreateFlags)0,
5057 DE_FALSE, // deUint32 depthTestEnable;
5058 DE_FALSE, // deUint32 depthWriteEnable;
5059 VK_COMPARE_OP_ALWAYS, // VkCompareOp depthCompareOp;
5060 DE_FALSE, // deUint32 depthBoundsTestEnable;
5061 DE_FALSE, // deUint32 stencilTestEnable;
5063 VK_STENCIL_OP_KEEP, // VkStencilOp stencilFailOp;
5064 VK_STENCIL_OP_KEEP, // VkStencilOp stencilPassOp;
5065 VK_STENCIL_OP_KEEP, // VkStencilOp stencilDepthFailOp;
5066 VK_COMPARE_OP_ALWAYS, // VkCompareOp stencilCompareOp;
5067 0u, // deUint32 stencilCompareMask;
5068 0u, // deUint32 stencilWriteMask;
5069 0u, // deUint32 stencilReference;
5070 }, // VkStencilOpState front;
5072 VK_STENCIL_OP_KEEP, // VkStencilOp stencilFailOp;
5073 VK_STENCIL_OP_KEEP, // VkStencilOp stencilPassOp;
5074 VK_STENCIL_OP_KEEP, // VkStencilOp stencilDepthFailOp;
5075 VK_COMPARE_OP_ALWAYS, // VkCompareOp stencilCompareOp;
5076 0u, // deUint32 stencilCompareMask;
5077 0u, // deUint32 stencilWriteMask;
5078 0u, // deUint32 stencilReference;
5079 }, // VkStencilOpState back;
5080 -1.0f, // float minDepthBounds;
5081 +1.0f, // float maxDepthBounds;
5083 const VkViewport viewport0 =
5085 0.0f, // float originX;
5086 0.0f, // float originY;
5087 (float)renderSize.x(), // float width;
5088 (float)renderSize.y(), // float height;
5089 0.0f, // float minDepth;
5090 1.0f, // float maxDepth;
5092 const VkRect2D scissor0 =
5097 }, // VkOffset2D offset;
5099 renderSize.x(), // deInt32 width;
5100 renderSize.y(), // deInt32 height;
5101 }, // VkExtent2D extent;
5103 const VkPipelineViewportStateCreateInfo viewportParams =
5105 VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO, // VkStructureType sType;
5106 DE_NULL, // const void* pNext;
5107 (VkPipelineViewportStateCreateFlags)0,
5108 1u, // deUint32 viewportCount;
5113 const VkSampleMask sampleMask = ~0u;
5114 const VkPipelineMultisampleStateCreateInfo multisampleParams =
5116 VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO, // VkStructureType sType;
5117 DE_NULL, // const void* pNext;
5118 (VkPipelineMultisampleStateCreateFlags)0,
5119 VK_SAMPLE_COUNT_1_BIT, // VkSampleCountFlagBits rasterSamples;
5120 DE_FALSE, // deUint32 sampleShadingEnable;
5121 0.0f, // float minSampleShading;
5122 &sampleMask, // const VkSampleMask* pSampleMask;
5123 DE_FALSE, // VkBool32 alphaToCoverageEnable;
5124 DE_FALSE, // VkBool32 alphaToOneEnable;
5126 const VkPipelineRasterizationStateCreateInfo rasterParams =
5128 VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO, // VkStructureType sType;
5129 DE_NULL, // const void* pNext;
5130 (VkPipelineRasterizationStateCreateFlags)0,
5131 DE_TRUE, // deUint32 depthClipEnable;
5132 DE_FALSE, // deUint32 rasterizerDiscardEnable;
5133 VK_POLYGON_MODE_FILL, // VkFillMode fillMode;
5134 VK_CULL_MODE_NONE, // VkCullMode cullMode;
5135 VK_FRONT_FACE_COUNTER_CLOCKWISE, // VkFrontFace frontFace;
5136 VK_FALSE, // VkBool32 depthBiasEnable;
5137 0.0f, // float depthBias;
5138 0.0f, // float depthBiasClamp;
5139 0.0f, // float slopeScaledDepthBias;
5140 1.0f, // float lineWidth;
5142 const VkPrimitiveTopology topology = hasTessellation? VK_PRIMITIVE_TOPOLOGY_PATCH_LIST: VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
5143 const VkPipelineInputAssemblyStateCreateInfo inputAssemblyParams =
5145 VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO, // VkStructureType sType;
5146 DE_NULL, // const void* pNext;
5147 (VkPipelineInputAssemblyStateCreateFlags)0,
5148 topology, // VkPrimitiveTopology topology;
5149 DE_FALSE, // deUint32 primitiveRestartEnable;
5151 const VkVertexInputBindingDescription vertexBinding0 =
5153 0u, // deUint32 binding;
5154 deUint32(singleVertexDataSize), // deUint32 strideInBytes;
5155 VK_VERTEX_INPUT_RATE_VERTEX // VkVertexInputStepRate stepRate;
5157 const VkVertexInputAttributeDescription vertexAttrib0[2] =
5160 0u, // deUint32 location;
5161 0u, // deUint32 binding;
5162 VK_FORMAT_R32G32B32A32_SFLOAT, // VkFormat format;
5163 0u // deUint32 offsetInBytes;
5166 1u, // deUint32 location;
5167 0u, // deUint32 binding;
5168 VK_FORMAT_R32G32B32A32_SFLOAT, // VkFormat format;
5169 sizeof(Vec4), // deUint32 offsetInBytes;
5173 const VkPipelineVertexInputStateCreateInfo vertexInputStateParams =
5175 VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO, // VkStructureType sType;
5176 DE_NULL, // const void* pNext;
5177 (VkPipelineVertexInputStateCreateFlags)0,
5178 1u, // deUint32 bindingCount;
5179 &vertexBinding0, // const VkVertexInputBindingDescription* pVertexBindingDescriptions;
5180 2u, // deUint32 attributeCount;
5181 vertexAttrib0, // const VkVertexInputAttributeDescription* pVertexAttributeDescriptions;
5183 const VkPipelineColorBlendAttachmentState attBlendParams =
5185 DE_FALSE, // deUint32 blendEnable;
5186 VK_BLEND_FACTOR_ONE, // VkBlend srcBlendColor;
5187 VK_BLEND_FACTOR_ZERO, // VkBlend destBlendColor;
5188 VK_BLEND_OP_ADD, // VkBlendOp blendOpColor;
5189 VK_BLEND_FACTOR_ONE, // VkBlend srcBlendAlpha;
5190 VK_BLEND_FACTOR_ZERO, // VkBlend destBlendAlpha;
5191 VK_BLEND_OP_ADD, // VkBlendOp blendOpAlpha;
5192 (VK_COLOR_COMPONENT_R_BIT|
5193 VK_COLOR_COMPONENT_G_BIT|
5194 VK_COLOR_COMPONENT_B_BIT|
5195 VK_COLOR_COMPONENT_A_BIT), // VkChannelFlags channelWriteMask;
5197 const VkPipelineColorBlendStateCreateInfo blendParams =
5199 VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO, // VkStructureType sType;
5200 DE_NULL, // const void* pNext;
5201 (VkPipelineColorBlendStateCreateFlags)0,
5202 DE_FALSE, // VkBool32 logicOpEnable;
5203 VK_LOGIC_OP_COPY, // VkLogicOp logicOp;
5204 1u, // deUint32 attachmentCount;
5205 &attBlendParams, // const VkPipelineColorBlendAttachmentState* pAttachments;
5206 { 0.0f, 0.0f, 0.0f, 0.0f }, // float blendConst[4];
5208 const VkPipelineTessellationStateCreateInfo tessellationState =
5210 VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO,
5212 (VkPipelineTessellationStateCreateFlags)0,
5216 const VkPipelineTessellationStateCreateInfo* tessellationInfo = hasTessellation ? &tessellationState: DE_NULL;
5217 const VkGraphicsPipelineCreateInfo pipelineParams =
5219 VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, // VkStructureType sType;
5220 DE_NULL, // const void* pNext;
5221 0u, // VkPipelineCreateFlags flags;
5222 (deUint32)shaderStageParams.size(), // deUint32 stageCount;
5223 &shaderStageParams[0], // const VkPipelineShaderStageCreateInfo* pStages;
5224 &vertexInputStateParams, // const VkPipelineVertexInputStateCreateInfo* pVertexInputState;
5225 &inputAssemblyParams, // const VkPipelineInputAssemblyStateCreateInfo* pInputAssemblyState;
5226 tessellationInfo, // const VkPipelineTessellationStateCreateInfo* pTessellationState;
5227 &viewportParams, // const VkPipelineViewportStateCreateInfo* pViewportState;
5228 &rasterParams, // const VkPipelineRasterStateCreateInfo* pRasterState;
5229 &multisampleParams, // const VkPipelineMultisampleStateCreateInfo* pMultisampleState;
5230 &depthStencilParams, // const VkPipelineDepthStencilStateCreateInfo* pDepthStencilState;
5231 &blendParams, // const VkPipelineColorBlendStateCreateInfo* pColorBlendState;
5232 (const VkPipelineDynamicStateCreateInfo*)DE_NULL, // const VkPipelineDynamicStateCreateInfo* pDynamicState;
5233 *pipelineLayout, // VkPipelineLayout layout;
5234 *renderPass, // VkRenderPass renderPass;
5235 0u, // deUint32 subpass;
5236 DE_NULL, // VkPipeline basePipelineHandle;
5237 0u, // deInt32 basePipelineIndex;
5240 const Unique<VkPipeline> pipeline (createGraphicsPipeline(vk, vkDevice, DE_NULL, &pipelineParams));
5243 const VkFramebufferCreateInfo framebufferParams =
5245 VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO, // VkStructureType sType;
5246 DE_NULL, // const void* pNext;
5247 (VkFramebufferCreateFlags)0,
5248 *renderPass, // VkRenderPass renderPass;
5249 1u, // deUint32 attachmentCount;
5250 &*colorAttView, // const VkImageView* pAttachments;
5251 (deUint32)renderSize.x(), // deUint32 width;
5252 (deUint32)renderSize.y(), // deUint32 height;
5253 1u, // deUint32 layers;
5255 const Unique<VkFramebuffer> framebuffer (createFramebuffer(vk, vkDevice, &framebufferParams));
5257 const VkCommandPoolCreateInfo cmdPoolParams =
5259 VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO, // VkStructureType sType;
5260 DE_NULL, // const void* pNext;
5261 VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, // VkCmdPoolCreateFlags flags;
5262 queueFamilyIndex, // deUint32 queueFamilyIndex;
5264 const Unique<VkCommandPool> cmdPool (createCommandPool(vk, vkDevice, &cmdPoolParams));
5267 const VkCommandBufferAllocateInfo cmdBufParams =
5269 VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO, // VkStructureType sType;
5270 DE_NULL, // const void* pNext;
5271 *cmdPool, // VkCmdPool pool;
5272 VK_COMMAND_BUFFER_LEVEL_PRIMARY, // VkCmdBufferLevel level;
5273 1u, // deUint32 count;
5275 const Unique<VkCommandBuffer> cmdBuf (allocateCommandBuffer(vk, vkDevice, &cmdBufParams));
5277 const VkCommandBufferBeginInfo cmdBufBeginParams =
5279 VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, // VkStructureType sType;
5280 DE_NULL, // const void* pNext;
5281 (VkCommandBufferUsageFlags)0,
5282 (const VkCommandBufferInheritanceInfo*)DE_NULL,
5286 VK_CHECK(vk.beginCommandBuffer(*cmdBuf, &cmdBufBeginParams));
5289 const VkMemoryBarrier vertFlushBarrier =
5291 VK_STRUCTURE_TYPE_MEMORY_BARRIER, // VkStructureType sType;
5292 DE_NULL, // const void* pNext;
5293 VK_ACCESS_HOST_WRITE_BIT, // VkMemoryOutputFlags outputMask;
5294 VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT, // VkMemoryInputFlags inputMask;
5296 const VkImageMemoryBarrier colorAttBarrier =
5298 VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType;
5299 DE_NULL, // const void* pNext;
5300 0u, // VkMemoryOutputFlags outputMask;
5301 VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, // VkMemoryInputFlags inputMask;
5302 VK_IMAGE_LAYOUT_UNDEFINED, // VkImageLayout oldLayout;
5303 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout newLayout;
5304 queueFamilyIndex, // deUint32 srcQueueFamilyIndex;
5305 queueFamilyIndex, // deUint32 destQueueFamilyIndex;
5306 *image, // VkImage image;
5308 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspect aspect;
5309 0u, // deUint32 baseMipLevel;
5310 1u, // deUint32 mipLevels;
5311 0u, // deUint32 baseArraySlice;
5312 1u, // deUint32 arraySize;
5313 } // VkImageSubresourceRange subresourceRange;
5315 vk.cmdPipelineBarrier(*cmdBuf, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, (VkDependencyFlags)0, 1, &vertFlushBarrier, 0, (const VkBufferMemoryBarrier*)DE_NULL, 1, &colorAttBarrier);
5319 const VkClearValue clearValue = makeClearValueColorF32(0.125f, 0.25f, 0.75f, 1.0f);
5320 const VkRenderPassBeginInfo passBeginParams =
5322 VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO, // VkStructureType sType;
5323 DE_NULL, // const void* pNext;
5324 *renderPass, // VkRenderPass renderPass;
5325 *framebuffer, // VkFramebuffer framebuffer;
5326 { { 0, 0 }, { renderSize.x(), renderSize.y() } }, // VkRect2D renderArea;
5327 1u, // deUint32 clearValueCount;
5328 &clearValue, // const VkClearValue* pClearValues;
5330 vk.cmdBeginRenderPass(*cmdBuf, &passBeginParams, VK_SUBPASS_CONTENTS_INLINE);
5333 vk.cmdBindPipeline(*cmdBuf, VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline);
5335 const VkDeviceSize bindingOffset = 0;
5336 vk.cmdBindVertexBuffers(*cmdBuf, 0u, 1u, &vertexBuffer.get(), &bindingOffset);
5338 vk.cmdDraw(*cmdBuf, deUint32(vertexCount), 1u /*run pipeline once*/, 0u /*first vertex*/, 0u /*first instanceIndex*/);
5339 vk.cmdEndRenderPass(*cmdBuf);
5342 const VkImageMemoryBarrier renderFinishBarrier =
5344 VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType;
5345 DE_NULL, // const void* pNext;
5346 VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, // VkMemoryOutputFlags outputMask;
5347 VK_ACCESS_TRANSFER_READ_BIT, // VkMemoryInputFlags inputMask;
5348 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout oldLayout;
5349 VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, // VkImageLayout newLayout;
5350 queueFamilyIndex, // deUint32 srcQueueFamilyIndex;
5351 queueFamilyIndex, // deUint32 destQueueFamilyIndex;
5352 *image, // VkImage image;
5354 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspectFlags aspectMask;
5355 0u, // deUint32 baseMipLevel;
5356 1u, // deUint32 mipLevels;
5357 0u, // deUint32 baseArraySlice;
5358 1u, // deUint32 arraySize;
5359 } // VkImageSubresourceRange subresourceRange;
5361 vk.cmdPipelineBarrier(*cmdBuf, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 0, (const VkBufferMemoryBarrier*)DE_NULL, 1, &renderFinishBarrier);
5365 const VkBufferImageCopy copyParams =
5367 (VkDeviceSize)0u, // VkDeviceSize bufferOffset;
5368 (deUint32)renderSize.x(), // deUint32 bufferRowLength;
5369 (deUint32)renderSize.y(), // deUint32 bufferImageHeight;
5371 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspect aspect;
5372 0u, // deUint32 mipLevel;
5373 0u, // deUint32 arrayLayer;
5374 1u, // deUint32 arraySize;
5375 }, // VkImageSubresourceCopy imageSubresource;
5376 { 0u, 0u, 0u }, // VkOffset3D imageOffset;
5377 { renderSize.x(), renderSize.y(), 1u } // VkExtent3D imageExtent;
5379 vk.cmdCopyImageToBuffer(*cmdBuf, *image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, *readImageBuffer, 1u, ©Params);
5383 const VkBufferMemoryBarrier copyFinishBarrier =
5385 VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, // VkStructureType sType;
5386 DE_NULL, // const void* pNext;
5387 VK_ACCESS_TRANSFER_WRITE_BIT, // VkMemoryOutputFlags outputMask;
5388 VK_ACCESS_HOST_READ_BIT, // VkMemoryInputFlags inputMask;
5389 queueFamilyIndex, // deUint32 srcQueueFamilyIndex;
5390 queueFamilyIndex, // deUint32 destQueueFamilyIndex;
5391 *readImageBuffer, // VkBuffer buffer;
5392 0u, // VkDeviceSize offset;
5393 imageSizeBytes // VkDeviceSize size;
5395 vk.cmdPipelineBarrier(*cmdBuf, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 1, ©FinishBarrier, 0, (const VkImageMemoryBarrier*)DE_NULL);
5398 VK_CHECK(vk.endCommandBuffer(*cmdBuf));
5400 // Upload vertex data
5402 const VkMappedMemoryRange range =
5404 VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, // VkStructureType sType;
5405 DE_NULL, // const void* pNext;
5406 vertexBufferMemory->getMemory(), // VkDeviceMemory mem;
5407 0, // VkDeviceSize offset;
5408 (VkDeviceSize)sizeof(vertexData), // VkDeviceSize size;
5410 void* vertexBufPtr = vertexBufferMemory->getHostPtr();
5412 deMemcpy(vertexBufPtr, &vertexData[0], sizeof(vertexData));
5413 VK_CHECK(vk.flushMappedMemoryRanges(vkDevice, 1u, &range));
5416 // Submit & wait for completion
5418 const VkFenceCreateInfo fenceParams =
5420 VK_STRUCTURE_TYPE_FENCE_CREATE_INFO, // VkStructureType sType;
5421 DE_NULL, // const void* pNext;
5422 0u, // VkFenceCreateFlags flags;
5424 const Unique<VkFence> fence (createFence(vk, vkDevice, &fenceParams));
5425 const VkSubmitInfo submitInfo =
5427 VK_STRUCTURE_TYPE_SUBMIT_INFO,
5430 (const VkSemaphore*)DE_NULL,
5431 (const VkPipelineStageFlags*)DE_NULL,
5435 (const VkSemaphore*)DE_NULL,
5438 VK_CHECK(vk.queueSubmit(queue, 1u, &submitInfo, *fence));
5439 VK_CHECK(vk.waitForFences(vkDevice, 1u, &fence.get(), DE_TRUE, ~0ull));
5442 const void* imagePtr = readImageBufferMemory->getHostPtr();
5443 const tcu::ConstPixelBufferAccess pixelBuffer(tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNORM_INT8),
5444 renderSize.x(), renderSize.y(), 1, imagePtr);
5447 const VkMappedMemoryRange range =
5449 VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, // VkStructureType sType;
5450 DE_NULL, // const void* pNext;
5451 readImageBufferMemory->getMemory(), // VkDeviceMemory mem;
5452 0, // VkDeviceSize offset;
5453 imageSizeBytes, // VkDeviceSize size;
5456 VK_CHECK(vk.invalidateMappedMemoryRanges(vkDevice, 1u, &range));
5457 context.getTestContext().getLog() << TestLog::Image("Result", "Result", pixelBuffer);
5460 const RGBA threshold(1, 1, 1, 1);
5461 const RGBA upperLeft(pixelBuffer.getPixel(1, 1));
5462 if (!tcu::compareThreshold(upperLeft, instance.outputColors[0], threshold))
5463 return TestStatus::fail("Upper left corner mismatch");
5465 const RGBA upperRight(pixelBuffer.getPixel(pixelBuffer.getWidth() - 1, 1));
5466 if (!tcu::compareThreshold(upperRight, instance.outputColors[1], threshold))
5467 return TestStatus::fail("Upper right corner mismatch");
5469 const RGBA lowerLeft(pixelBuffer.getPixel(1, pixelBuffer.getHeight() - 1));
5470 if (!tcu::compareThreshold(lowerLeft, instance.outputColors[2], threshold))
5471 return TestStatus::fail("Lower left corner mismatch");
5473 const RGBA lowerRight(pixelBuffer.getPixel(pixelBuffer.getWidth() - 1, pixelBuffer.getHeight() - 1));
5474 if (!tcu::compareThreshold(lowerRight, instance.outputColors[3], threshold))
5475 return TestStatus::fail("Lower right corner mismatch");
5477 return TestStatus::pass("Rendered output matches input");
5480 void createTestsForAllStages (const std::string& name, const RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments, const vector<deInt32>& specConstants, tcu::TestCaseGroup* tests)
5482 const ShaderElement vertFragPipelineStages[] =
5484 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
5485 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
5488 const ShaderElement tessPipelineStages[] =
5490 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
5491 ShaderElement("tessc", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
5492 ShaderElement("tesse", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
5493 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
5496 const ShaderElement geomPipelineStages[] =
5498 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
5499 ShaderElement("geom", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
5500 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
5503 StageToSpecConstantMap specConstantMap;
5505 specConstantMap[VK_SHADER_STAGE_VERTEX_BIT] = specConstants;
5506 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_vert", "", addShaderCodeCustomVertex, runAndVerifyDefaultPipeline,
5507 createInstanceContext(vertFragPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5509 specConstantMap.clear();
5510 specConstantMap[VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT] = specConstants;
5511 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_tessc", "", addShaderCodeCustomTessControl, runAndVerifyDefaultPipeline,
5512 createInstanceContext(tessPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5514 specConstantMap.clear();
5515 specConstantMap[VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT] = specConstants;
5516 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_tesse", "", addShaderCodeCustomTessEval, runAndVerifyDefaultPipeline,
5517 createInstanceContext(tessPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5519 specConstantMap.clear();
5520 specConstantMap[VK_SHADER_STAGE_GEOMETRY_BIT] = specConstants;
5521 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_geom", "", addShaderCodeCustomGeometry, runAndVerifyDefaultPipeline,
5522 createInstanceContext(geomPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5524 specConstantMap.clear();
5525 specConstantMap[VK_SHADER_STAGE_FRAGMENT_BIT] = specConstants;
5526 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_frag", "", addShaderCodeCustomFragment, runAndVerifyDefaultPipeline,
5527 createInstanceContext(vertFragPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5530 inline void createTestsForAllStages (const std::string& name, const RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments, tcu::TestCaseGroup* tests)
5532 vector<deInt32> noSpecConstants;
5533 createTestsForAllStages(name, inputColors, outputColors, testCodeFragments, noSpecConstants, tests);
5538 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
5540 struct NameCodePair { string name, code; };
5541 RGBA defaultColors[4];
5542 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
5543 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
5544 map<string, string> fragments = passthruFragments();
5545 const NameCodePair tests[] =
5547 {"unknown", "OpSource Unknown 321"},
5548 {"essl", "OpSource ESSL 310"},
5549 {"glsl", "OpSource GLSL 450"},
5550 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
5551 {"opencl_c", "OpSource OpenCL_C 120"},
5552 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
5553 {"file", opsourceGLSLWithFile},
5554 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
5555 // Longest possible source string: SPIR-V limits instructions to 65535
5556 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
5557 // contain 65530 UTF8 characters (one word each) plus one last word
5558 // containing 3 ASCII characters and \0.
5559 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
5562 getDefaultColors(defaultColors);
5563 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5565 fragments["debug"] = tests[testNdx].code;
5566 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5569 return opSourceTests.release();
5572 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
5574 struct NameCodePair { string name, code; };
5575 RGBA defaultColors[4];
5576 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
5577 map<string, string> fragments = passthruFragments();
5578 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
5579 const NameCodePair tests[] =
5581 {"empty", opsource + "OpSourceContinued \"\""},
5582 {"short", opsource + "OpSourceContinued \"abcde\""},
5583 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
5584 // Longest possible source string: SPIR-V limits instructions to 65535
5585 // words, of which the first one is OpSourceContinued/length; the rest
5586 // will contain 65533 UTF8 characters (one word each) plus one last word
5587 // containing 3 ASCII characters and \0.
5588 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
5591 getDefaultColors(defaultColors);
5592 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5594 fragments["debug"] = tests[testNdx].code;
5595 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5598 return opSourceTests.release();
5601 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
5603 RGBA defaultColors[4];
5604 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
5605 map<string, string> fragments;
5606 getDefaultColors(defaultColors);
5607 fragments["debug"] =
5608 "%name = OpString \"name\"\n";
5610 fragments["pre_main"] =
5613 "OpLine %name 1 1\n"
5615 "OpLine %name 1 1\n"
5616 "OpLine %name 1 1\n"
5617 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5619 "OpLine %name 1 1\n"
5621 "OpLine %name 1 1\n"
5622 "OpLine %name 1 1\n"
5623 "%second_param1 = OpFunctionParameter %v4f32\n"
5626 "%label_secondfunction = OpLabel\n"
5628 "OpReturnValue %second_param1\n"
5633 fragments["testfun"] =
5634 // A %test_code function that returns its argument unchanged.
5637 "OpLine %name 1 1\n"
5638 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5640 "%param1 = OpFunctionParameter %v4f32\n"
5643 "%label_testfun = OpLabel\n"
5645 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5646 "OpReturnValue %val1\n"
5648 "OpLine %name 1 1\n"
5651 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
5653 return opLineTests.release();
5657 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
5659 RGBA defaultColors[4];
5660 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
5661 map<string, string> fragments;
5662 std::vector<std::pair<std::string, std::string> > problemStrings;
5664 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
5665 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
5666 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
5667 getDefaultColors(defaultColors);
5669 fragments["debug"] =
5670 "%other_name = OpString \"other_name\"\n";
5672 fragments["pre_main"] =
5673 "OpLine %file_name 32 0\n"
5674 "OpLine %file_name 32 32\n"
5675 "OpLine %file_name 32 40\n"
5676 "OpLine %other_name 32 40\n"
5677 "OpLine %other_name 0 100\n"
5678 "OpLine %other_name 0 4294967295\n"
5679 "OpLine %other_name 4294967295 0\n"
5680 "OpLine %other_name 32 40\n"
5681 "OpLine %file_name 0 0\n"
5682 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5683 "OpLine %file_name 1 0\n"
5684 "%second_param1 = OpFunctionParameter %v4f32\n"
5685 "OpLine %file_name 1 3\n"
5686 "OpLine %file_name 1 2\n"
5687 "%label_secondfunction = OpLabel\n"
5688 "OpLine %file_name 0 2\n"
5689 "OpReturnValue %second_param1\n"
5691 "OpLine %file_name 0 2\n"
5692 "OpLine %file_name 0 2\n";
5694 fragments["testfun"] =
5695 // A %test_code function that returns its argument unchanged.
5696 "OpLine %file_name 1 0\n"
5697 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5698 "OpLine %file_name 16 330\n"
5699 "%param1 = OpFunctionParameter %v4f32\n"
5700 "OpLine %file_name 14 442\n"
5701 "%label_testfun = OpLabel\n"
5702 "OpLine %file_name 11 1024\n"
5703 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5704 "OpLine %file_name 2 97\n"
5705 "OpReturnValue %val1\n"
5707 "OpLine %file_name 5 32\n";
5709 for (size_t i = 0; i < problemStrings.size(); ++i)
5711 map<string, string> testFragments = fragments;
5712 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
5713 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
5716 return opLineTests.release();
5719 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
5721 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
5725 const char functionStart[] =
5726 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5727 "%param1 = OpFunctionParameter %v4f32\n"
5730 const char functionEnd[] =
5731 "OpReturnValue %transformed_param\n"
5734 struct NameConstantsCode
5741 NameConstantsCode tests[] =
5745 "%cnull = OpConstantNull %v4f32\n",
5746 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
5750 "%cnull = OpConstantNull %f32\n",
5751 "%vp = OpVariable %fp_v4f32 Function\n"
5752 "%v = OpLoad %v4f32 %vp\n"
5753 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
5754 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
5755 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
5756 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
5757 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
5761 "%cnull = OpConstantNull %bool\n",
5762 "%v = OpVariable %fp_v4f32 Function\n"
5763 " OpStore %v %param1\n"
5764 " OpSelectionMerge %false_label None\n"
5765 " OpBranchConditional %cnull %true_label %false_label\n"
5766 "%true_label = OpLabel\n"
5767 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
5768 " OpBranch %false_label\n"
5769 "%false_label = OpLabel\n"
5770 "%transformed_param = OpLoad %v4f32 %v\n"
5774 "%cnull = OpConstantNull %i32\n",
5775 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
5776 "%b = OpIEqual %bool %cnull %c_i32_0\n"
5777 " OpSelectionMerge %false_label None\n"
5778 " OpBranchConditional %b %true_label %false_label\n"
5779 "%true_label = OpLabel\n"
5780 " OpStore %v %param1\n"
5781 " OpBranch %false_label\n"
5782 "%false_label = OpLabel\n"
5783 "%transformed_param = OpLoad %v4f32 %v\n"
5787 "%stype = OpTypeStruct %f32 %v4f32\n"
5788 "%fp_stype = OpTypePointer Function %stype\n"
5789 "%cnull = OpConstantNull %stype\n",
5790 "%v = OpVariable %fp_stype Function %cnull\n"
5791 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
5792 "%f_val = OpLoad %v4f32 %f\n"
5793 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
5797 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
5798 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
5799 "%cnull = OpConstantNull %a4_v4f32\n",
5800 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
5801 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5802 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
5803 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
5804 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
5805 "%f_val = OpLoad %v4f32 %f\n"
5806 "%f1_val = OpLoad %v4f32 %f1\n"
5807 "%f2_val = OpLoad %v4f32 %f2\n"
5808 "%f3_val = OpLoad %v4f32 %f3\n"
5809 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
5810 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
5811 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
5812 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
5816 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5817 "%cnull = OpConstantNull %mat4x4_f32\n",
5818 // Our null matrix * any vector should result in a zero vector.
5819 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
5820 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
5824 getHalfColorsFullAlpha(colors);
5826 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5828 map<string, string> fragments;
5829 fragments["pre_main"] = tests[testNdx].constants;
5830 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5831 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
5833 return opConstantNullTests.release();
5835 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
5837 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
5838 RGBA inputColors[4];
5839 RGBA outputColors[4];
5842 const char functionStart[] =
5843 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5844 "%param1 = OpFunctionParameter %v4f32\n"
5847 const char functionEnd[] =
5848 "OpReturnValue %transformed_param\n"
5851 struct NameConstantsCode
5858 NameConstantsCode tests[] =
5863 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
5864 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
5869 "%stype = OpTypeStruct %v4f32 %f32\n"
5870 "%fp_stype = OpTypePointer Function %stype\n"
5871 "%f32_n_1 = OpConstant %f32 -1.0\n"
5872 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5873 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
5874 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
5876 "%v = OpVariable %fp_stype Function %cval\n"
5877 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5878 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
5879 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
5880 "%f32_val = OpLoad %f32 %f32_ptr\n"
5881 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
5882 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
5883 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
5886 // [1|0|0|0.5] [x] = x + 0.5
5887 // [0|1|0|0.5] [y] = y + 0.5
5888 // [0|0|1|0.5] [z] = z + 0.5
5889 // [0|0|0|1 ] [1] = 1
5892 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5893 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
5894 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
5895 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
5896 "%v4f32_0_5_0_5_0_5_1 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_1\n"
5897 "%cval = OpConstantComposite %mat4x4_f32 %v4f32_1_0_0_0 %v4f32_0_1_0_0 %v4f32_0_0_1_0 %v4f32_0_5_0_5_0_5_1\n",
5899 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
5904 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5905 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5906 "%f32_n_1 = OpConstant %f32 -1.0\n"
5907 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5908 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
5910 "%v = OpVariable %fp_a4f32 Function %carr\n"
5911 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
5912 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
5913 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5914 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
5915 "%f_val = OpLoad %f32 %f\n"
5916 "%f1_val = OpLoad %f32 %f1\n"
5917 "%f2_val = OpLoad %f32 %f2\n"
5918 "%f3_val = OpLoad %f32 %f3\n"
5919 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
5920 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
5921 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
5922 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
5923 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5930 // [ 1.0, 1.0, 1.0, 1.0]
5934 // [ 0.0, 0.5, 0.0, 0.0]
5938 // [ 1.0, 1.0, 1.0, 1.0]
5941 "array_of_struct_of_array",
5943 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5944 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5945 "%stype = OpTypeStruct %f32 %a4f32\n"
5946 "%a3stype = OpTypeArray %stype %c_u32_3\n"
5947 "%fp_a3stype = OpTypePointer Function %a3stype\n"
5948 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
5949 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5950 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
5951 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
5952 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
5954 "%v = OpVariable %fp_a3stype Function %carr\n"
5955 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
5956 "%f_l = OpLoad %f32 %f\n"
5957 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
5958 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5962 getHalfColorsFullAlpha(inputColors);
5963 outputColors[0] = RGBA(255, 255, 255, 255);
5964 outputColors[1] = RGBA(255, 127, 127, 255);
5965 outputColors[2] = RGBA(127, 255, 127, 255);
5966 outputColors[3] = RGBA(127, 127, 255, 255);
5968 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5970 map<string, string> fragments;
5971 fragments["pre_main"] = tests[testNdx].constants;
5972 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5973 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
5975 return opConstantCompositeTests.release();
5978 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
5980 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
5981 RGBA inputColors[4];
5982 RGBA outputColors[4];
5983 map<string, string> fragments;
5985 // vec4 test_code(vec4 param) {
5986 // vec4 result = param;
5987 // for (int i = 0; i < 4; ++i) {
5988 // if (i == 0) result[i] = 0.;
5989 // else result[i] = 1. - result[i];
5993 const char function[] =
5994 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5995 "%param1 = OpFunctionParameter %v4f32\n"
5997 "%iptr = OpVariable %fp_i32 Function\n"
5998 "%result = OpVariable %fp_v4f32 Function\n"
5999 " OpStore %iptr %c_i32_0\n"
6000 " OpStore %result %param1\n"
6003 // Loop entry block.
6005 "%ival = OpLoad %i32 %iptr\n"
6006 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6007 " OpLoopMerge %exit %loop None\n"
6008 " OpBranchConditional %lt_4 %if_entry %exit\n"
6010 // Merge block for loop.
6012 "%ret = OpLoad %v4f32 %result\n"
6013 " OpReturnValue %ret\n"
6015 // If-statement entry block.
6016 "%if_entry = OpLabel\n"
6017 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6018 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
6019 " OpSelectionMerge %if_exit None\n"
6020 " OpBranchConditional %eq_0 %if_true %if_false\n"
6022 // False branch for if-statement.
6023 "%if_false = OpLabel\n"
6024 "%val = OpLoad %f32 %loc\n"
6025 "%sub = OpFSub %f32 %c_f32_1 %val\n"
6026 " OpStore %loc %sub\n"
6027 " OpBranch %if_exit\n"
6029 // Merge block for if-statement.
6030 "%if_exit = OpLabel\n"
6031 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6032 " OpStore %iptr %ival_next\n"
6035 // True branch for if-statement.
6036 "%if_true = OpLabel\n"
6037 " OpStore %loc %c_f32_0\n"
6038 " OpBranch %if_exit\n"
6042 fragments["testfun"] = function;
6044 inputColors[0] = RGBA(127, 127, 127, 0);
6045 inputColors[1] = RGBA(127, 0, 0, 0);
6046 inputColors[2] = RGBA(0, 127, 0, 0);
6047 inputColors[3] = RGBA(0, 0, 127, 0);
6049 outputColors[0] = RGBA(0, 128, 128, 255);
6050 outputColors[1] = RGBA(0, 255, 255, 255);
6051 outputColors[2] = RGBA(0, 128, 255, 255);
6052 outputColors[3] = RGBA(0, 255, 128, 255);
6054 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
6056 return group.release();
6059 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
6061 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
6062 RGBA inputColors[4];
6063 RGBA outputColors[4];
6064 map<string, string> fragments;
6066 const char typesAndConstants[] =
6067 "%c_f32_p2 = OpConstant %f32 0.2\n"
6068 "%c_f32_p4 = OpConstant %f32 0.4\n"
6069 "%c_f32_p6 = OpConstant %f32 0.6\n"
6070 "%c_f32_p8 = OpConstant %f32 0.8\n";
6072 // vec4 test_code(vec4 param) {
6073 // vec4 result = param;
6074 // for (int i = 0; i < 4; ++i) {
6076 // case 0: result[i] += .2; break;
6077 // case 1: result[i] += .6; break;
6078 // case 2: result[i] += .4; break;
6079 // case 3: result[i] += .8; break;
6080 // default: break; // unreachable
6085 const char function[] =
6086 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6087 "%param1 = OpFunctionParameter %v4f32\n"
6089 "%iptr = OpVariable %fp_i32 Function\n"
6090 "%result = OpVariable %fp_v4f32 Function\n"
6091 " OpStore %iptr %c_i32_0\n"
6092 " OpStore %result %param1\n"
6095 // Loop entry block.
6097 "%ival = OpLoad %i32 %iptr\n"
6098 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6099 " OpLoopMerge %exit %loop None\n"
6100 " OpBranchConditional %lt_4 %switch_entry %exit\n"
6102 // Merge block for loop.
6104 "%ret = OpLoad %v4f32 %result\n"
6105 " OpReturnValue %ret\n"
6107 // Switch-statement entry block.
6108 "%switch_entry = OpLabel\n"
6109 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6110 "%val = OpLoad %f32 %loc\n"
6111 " OpSelectionMerge %switch_exit None\n"
6112 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6114 "%case2 = OpLabel\n"
6115 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
6116 " OpStore %loc %addp4\n"
6117 " OpBranch %switch_exit\n"
6119 "%switch_default = OpLabel\n"
6122 "%case3 = OpLabel\n"
6123 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
6124 " OpStore %loc %addp8\n"
6125 " OpBranch %switch_exit\n"
6127 "%case0 = OpLabel\n"
6128 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
6129 " OpStore %loc %addp2\n"
6130 " OpBranch %switch_exit\n"
6132 // Merge block for switch-statement.
6133 "%switch_exit = OpLabel\n"
6134 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6135 " OpStore %iptr %ival_next\n"
6138 "%case1 = OpLabel\n"
6139 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
6140 " OpStore %loc %addp6\n"
6141 " OpBranch %switch_exit\n"
6145 fragments["pre_main"] = typesAndConstants;
6146 fragments["testfun"] = function;
6148 inputColors[0] = RGBA(127, 27, 127, 51);
6149 inputColors[1] = RGBA(127, 0, 0, 51);
6150 inputColors[2] = RGBA(0, 27, 0, 51);
6151 inputColors[3] = RGBA(0, 0, 127, 51);
6153 outputColors[0] = RGBA(178, 180, 229, 255);
6154 outputColors[1] = RGBA(178, 153, 102, 255);
6155 outputColors[2] = RGBA(51, 180, 102, 255);
6156 outputColors[3] = RGBA(51, 153, 229, 255);
6158 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
6160 return group.release();
6163 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
6165 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
6166 RGBA inputColors[4];
6167 RGBA outputColors[4];
6168 map<string, string> fragments;
6170 const char decorations[] =
6171 "OpDecorate %array_group ArrayStride 4\n"
6172 "OpDecorate %struct_member_group Offset 0\n"
6173 "%array_group = OpDecorationGroup\n"
6174 "%struct_member_group = OpDecorationGroup\n"
6176 "OpDecorate %group1 RelaxedPrecision\n"
6177 "OpDecorate %group3 RelaxedPrecision\n"
6178 "OpDecorate %group3 Invariant\n"
6179 "OpDecorate %group3 Restrict\n"
6180 "%group0 = OpDecorationGroup\n"
6181 "%group1 = OpDecorationGroup\n"
6182 "%group3 = OpDecorationGroup\n";
6184 const char typesAndConstants[] =
6185 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
6186 "%struct1 = OpTypeStruct %a3f32\n"
6187 "%struct2 = OpTypeStruct %a3f32\n"
6188 "%fp_struct1 = OpTypePointer Function %struct1\n"
6189 "%fp_struct2 = OpTypePointer Function %struct2\n"
6190 "%c_f32_2 = OpConstant %f32 2.\n"
6191 "%c_f32_n2 = OpConstant %f32 -2.\n"
6193 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
6194 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
6195 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
6196 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
6198 const char function[] =
6199 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6200 "%param = OpFunctionParameter %v4f32\n"
6201 "%entry = OpLabel\n"
6202 "%result = OpVariable %fp_v4f32 Function\n"
6203 "%v_struct1 = OpVariable %fp_struct1 Function\n"
6204 "%v_struct2 = OpVariable %fp_struct2 Function\n"
6205 " OpStore %result %param\n"
6206 " OpStore %v_struct1 %c_struct1\n"
6207 " OpStore %v_struct2 %c_struct2\n"
6208 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
6209 "%val1 = OpLoad %f32 %ptr1\n"
6210 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
6211 "%val2 = OpLoad %f32 %ptr2\n"
6212 "%addvalues = OpFAdd %f32 %val1 %val2\n"
6213 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
6214 "%val = OpLoad %f32 %ptr\n"
6215 "%addresult = OpFAdd %f32 %addvalues %val\n"
6216 " OpStore %ptr %addresult\n"
6217 "%ret = OpLoad %v4f32 %result\n"
6218 " OpReturnValue %ret\n"
6221 struct CaseNameDecoration
6227 CaseNameDecoration tests[] =
6230 "same_decoration_group_on_multiple_types",
6231 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
6234 "empty_decoration_group",
6235 "OpGroupDecorate %group0 %a3f32\n"
6236 "OpGroupDecorate %group0 %result\n"
6239 "one_element_decoration_group",
6240 "OpGroupDecorate %array_group %a3f32\n"
6243 "multiple_elements_decoration_group",
6244 "OpGroupDecorate %group3 %v_struct1\n"
6247 "multiple_decoration_groups_on_same_variable",
6248 "OpGroupDecorate %group0 %v_struct2\n"
6249 "OpGroupDecorate %group1 %v_struct2\n"
6250 "OpGroupDecorate %group3 %v_struct2\n"
6253 "same_decoration_group_multiple_times",
6254 "OpGroupDecorate %group1 %addvalues\n"
6255 "OpGroupDecorate %group1 %addvalues\n"
6256 "OpGroupDecorate %group1 %addvalues\n"
6261 getHalfColorsFullAlpha(inputColors);
6262 getHalfColorsFullAlpha(outputColors);
6264 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
6266 fragments["decoration"] = decorations + tests[idx].decoration;
6267 fragments["pre_main"] = typesAndConstants;
6268 fragments["testfun"] = function;
6270 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
6273 return group.release();
6276 struct SpecConstantTwoIntGraphicsCase
6278 const char* caseName;
6279 const char* scDefinition0;
6280 const char* scDefinition1;
6281 const char* scResultType;
6282 const char* scOperation;
6283 deInt32 scActualValue0;
6284 deInt32 scActualValue1;
6285 const char* resultOperation;
6286 RGBA expectedColors[4];
6288 SpecConstantTwoIntGraphicsCase (const char* name,
6289 const char* definition0,
6290 const char* definition1,
6291 const char* resultType,
6292 const char* operation,
6295 const char* resultOp,
6296 const RGBA (&output)[4])
6298 , scDefinition0 (definition0)
6299 , scDefinition1 (definition1)
6300 , scResultType (resultType)
6301 , scOperation (operation)
6302 , scActualValue0 (value0)
6303 , scActualValue1 (value1)
6304 , resultOperation (resultOp)
6306 expectedColors[0] = output[0];
6307 expectedColors[1] = output[1];
6308 expectedColors[2] = output[2];
6309 expectedColors[3] = output[3];
6313 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
6315 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
6316 vector<SpecConstantTwoIntGraphicsCase> cases;
6317 RGBA inputColors[4];
6318 RGBA outputColors0[4];
6319 RGBA outputColors1[4];
6320 RGBA outputColors2[4];
6322 const char decorations1[] =
6323 "OpDecorate %sc_0 SpecId 0\n"
6324 "OpDecorate %sc_1 SpecId 1\n";
6326 const char typesAndConstants1[] =
6327 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
6328 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
6329 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
6331 const char function1[] =
6332 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6333 "%param = OpFunctionParameter %v4f32\n"
6334 "%label = OpLabel\n"
6335 "%result = OpVariable %fp_v4f32 Function\n"
6336 " OpStore %result %param\n"
6337 "%gen = ${GEN_RESULT}\n"
6338 "%index = OpIAdd %i32 %gen %c_i32_1\n"
6339 "%loc = OpAccessChain %fp_f32 %result %index\n"
6340 "%val = OpLoad %f32 %loc\n"
6341 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6342 " OpStore %loc %add\n"
6343 "%ret = OpLoad %v4f32 %result\n"
6344 " OpReturnValue %ret\n"
6347 inputColors[0] = RGBA(127, 127, 127, 255);
6348 inputColors[1] = RGBA(127, 0, 0, 255);
6349 inputColors[2] = RGBA(0, 127, 0, 255);
6350 inputColors[3] = RGBA(0, 0, 127, 255);
6352 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
6353 outputColors0[0] = RGBA(255, 127, 127, 255);
6354 outputColors0[1] = RGBA(255, 0, 0, 255);
6355 outputColors0[2] = RGBA(128, 127, 0, 255);
6356 outputColors0[3] = RGBA(128, 0, 127, 255);
6358 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
6359 outputColors1[0] = RGBA(127, 255, 127, 255);
6360 outputColors1[1] = RGBA(127, 128, 0, 255);
6361 outputColors1[2] = RGBA(0, 255, 0, 255);
6362 outputColors1[3] = RGBA(0, 128, 127, 255);
6364 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
6365 outputColors2[0] = RGBA(127, 127, 255, 255);
6366 outputColors2[1] = RGBA(127, 0, 128, 255);
6367 outputColors2[2] = RGBA(0, 127, 128, 255);
6368 outputColors2[3] = RGBA(0, 0, 255, 255);
6370 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
6371 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
6372 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
6374 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
6375 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
6376 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
6377 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
6378 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
6379 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6380 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6381 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
6382 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
6383 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
6384 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
6385 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
6386 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
6387 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
6388 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
6389 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
6390 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6391 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
6392 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
6393 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
6394 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6395 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
6396 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
6397 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6398 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6399 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6400 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6401 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
6402 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
6403 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
6404 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
6405 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
6406 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
6408 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
6410 map<string, string> specializations;
6411 map<string, string> fragments;
6412 vector<deInt32> specConstants;
6414 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
6415 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
6416 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
6417 specializations["SC_OP"] = cases[caseNdx].scOperation;
6418 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
6420 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
6421 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
6422 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
6424 specConstants.push_back(cases[caseNdx].scActualValue0);
6425 specConstants.push_back(cases[caseNdx].scActualValue1);
6427 createTestsForAllStages(cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants, group.get());
6430 const char decorations2[] =
6431 "OpDecorate %sc_0 SpecId 0\n"
6432 "OpDecorate %sc_1 SpecId 1\n"
6433 "OpDecorate %sc_2 SpecId 2\n";
6435 const char typesAndConstants2[] =
6436 "%v3i32 = OpTypeVector %i32 3\n"
6438 "%sc_0 = OpSpecConstant %i32 0\n"
6439 "%sc_1 = OpSpecConstant %i32 0\n"
6440 "%sc_2 = OpSpecConstant %i32 0\n"
6442 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
6443 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
6444 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
6445 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
6446 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %sc_vec3_1 1 0 4\n" // (0, sc_0, sc_1)
6447 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2 5 1 2\n" // (sc_2, sc_0, sc_1)
6448 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
6449 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
6450 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
6451 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
6452 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
6454 const char function2[] =
6455 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6456 "%param = OpFunctionParameter %v4f32\n"
6457 "%label = OpLabel\n"
6458 "%result = OpVariable %fp_v4f32 Function\n"
6459 " OpStore %result %param\n"
6460 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
6461 "%val = OpLoad %f32 %loc\n"
6462 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6463 " OpStore %loc %add\n"
6464 "%ret = OpLoad %v4f32 %result\n"
6465 " OpReturnValue %ret\n"
6468 map<string, string> fragments;
6469 vector<deInt32> specConstants;
6471 fragments["decoration"] = decorations2;
6472 fragments["pre_main"] = typesAndConstants2;
6473 fragments["testfun"] = function2;
6475 specConstants.push_back(56789);
6476 specConstants.push_back(-2);
6477 specConstants.push_back(56788);
6479 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
6481 return group.release();
6484 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
6486 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
6487 RGBA inputColors[4];
6488 RGBA outputColors1[4];
6489 RGBA outputColors2[4];
6490 RGBA outputColors3[4];
6491 map<string, string> fragments1;
6492 map<string, string> fragments2;
6493 map<string, string> fragments3;
6495 const char typesAndConstants1[] =
6496 "%c_f32_p2 = OpConstant %f32 0.2\n"
6497 "%c_f32_p4 = OpConstant %f32 0.4\n"
6498 "%c_f32_p5 = OpConstant %f32 0.5\n"
6499 "%c_f32_p8 = OpConstant %f32 0.8\n";
6501 // vec4 test_code(vec4 param) {
6502 // vec4 result = param;
6503 // for (int i = 0; i < 4; ++i) {
6506 // case 0: operand = .2; break;
6507 // case 1: operand = .5; break;
6508 // case 2: operand = .4; break;
6509 // case 3: operand = .0; break;
6510 // default: break; // unreachable
6512 // result[i] += operand;
6516 const char function1[] =
6517 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6518 "%param1 = OpFunctionParameter %v4f32\n"
6520 "%iptr = OpVariable %fp_i32 Function\n"
6521 "%result = OpVariable %fp_v4f32 Function\n"
6522 " OpStore %iptr %c_i32_0\n"
6523 " OpStore %result %param1\n"
6527 "%ival = OpLoad %i32 %iptr\n"
6528 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6529 " OpLoopMerge %exit %loop None\n"
6530 " OpBranchConditional %lt_4 %entry %exit\n"
6532 "%entry = OpLabel\n"
6533 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6534 "%val = OpLoad %f32 %loc\n"
6535 " OpSelectionMerge %phi None\n"
6536 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6538 "%case0 = OpLabel\n"
6540 "%case1 = OpLabel\n"
6542 "%case2 = OpLabel\n"
6544 "%case3 = OpLabel\n"
6547 "%default = OpLabel\n"
6551 "%operand = OpPhi %f32 %c_f32_p4 %case2 %c_f32_p5 %case1 %c_f32_p2 %case0 %c_f32_0 %case3\n" // not in the order of blocks
6552 "%add = OpFAdd %f32 %val %operand\n"
6553 " OpStore %loc %add\n"
6554 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6555 " OpStore %iptr %ival_next\n"
6559 "%ret = OpLoad %v4f32 %result\n"
6560 " OpReturnValue %ret\n"
6564 fragments1["pre_main"] = typesAndConstants1;
6565 fragments1["testfun"] = function1;
6567 getHalfColorsFullAlpha(inputColors);
6569 outputColors1[0] = RGBA(178, 255, 229, 255);
6570 outputColors1[1] = RGBA(178, 127, 102, 255);
6571 outputColors1[2] = RGBA(51, 255, 102, 255);
6572 outputColors1[3] = RGBA(51, 127, 229, 255);
6574 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
6576 const char typesAndConstants2[] =
6577 "%c_f32_p2 = OpConstant %f32 0.2\n";
6579 // Add .4 to the second element of the given parameter.
6580 const char function2[] =
6581 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6582 "%param = OpFunctionParameter %v4f32\n"
6583 "%entry = OpLabel\n"
6584 "%result = OpVariable %fp_v4f32 Function\n"
6585 " OpStore %result %param\n"
6586 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6587 "%val = OpLoad %f32 %loc\n"
6591 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
6592 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
6593 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
6594 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
6595 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
6596 " OpLoopMerge %exit %phi None\n"
6597 " OpBranchConditional %still_loop %phi %exit\n"
6600 " OpStore %loc %accum\n"
6601 "%ret = OpLoad %v4f32 %result\n"
6602 " OpReturnValue %ret\n"
6606 fragments2["pre_main"] = typesAndConstants2;
6607 fragments2["testfun"] = function2;
6609 outputColors2[0] = RGBA(127, 229, 127, 255);
6610 outputColors2[1] = RGBA(127, 102, 0, 255);
6611 outputColors2[2] = RGBA(0, 229, 0, 255);
6612 outputColors2[3] = RGBA(0, 102, 127, 255);
6614 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
6616 const char typesAndConstants3[] =
6617 "%true = OpConstantTrue %bool\n"
6618 "%false = OpConstantFalse %bool\n"
6619 "%c_f32_p2 = OpConstant %f32 0.2\n";
6621 // Swap the second and the third element of the given parameter.
6622 const char function3[] =
6623 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6624 "%param = OpFunctionParameter %v4f32\n"
6625 "%entry = OpLabel\n"
6626 "%result = OpVariable %fp_v4f32 Function\n"
6627 " OpStore %result %param\n"
6628 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6629 "%a_init = OpLoad %f32 %a_loc\n"
6630 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
6631 "%b_init = OpLoad %f32 %b_loc\n"
6635 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
6636 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
6637 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
6638 " OpLoopMerge %exit %phi None\n"
6639 " OpBranchConditional %still_loop %phi %exit\n"
6642 " OpStore %a_loc %a_next\n"
6643 " OpStore %b_loc %b_next\n"
6644 "%ret = OpLoad %v4f32 %result\n"
6645 " OpReturnValue %ret\n"
6649 fragments3["pre_main"] = typesAndConstants3;
6650 fragments3["testfun"] = function3;
6652 outputColors3[0] = RGBA(127, 127, 127, 255);
6653 outputColors3[1] = RGBA(127, 0, 0, 255);
6654 outputColors3[2] = RGBA(0, 0, 127, 255);
6655 outputColors3[3] = RGBA(0, 127, 0, 255);
6657 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
6659 return group.release();
6662 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
6664 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
6665 RGBA inputColors[4];
6666 RGBA outputColors[4];
6668 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
6669 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
6670 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
6671 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
6672 const char constantsAndTypes[] =
6673 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
6674 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
6675 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
6676 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
6677 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n"
6680 const char function[] =
6681 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6682 "%param = OpFunctionParameter %v4f32\n"
6683 "%label = OpLabel\n"
6684 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
6685 "%var2 = OpVariable %fp_f32 Function\n"
6686 "%red = OpCompositeExtract %f32 %param 0\n"
6687 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
6688 " OpStore %var2 %plus_red\n"
6689 "%val1 = OpLoad %f32 %var1\n"
6690 "%val2 = OpLoad %f32 %var2\n"
6691 "%mul = OpFMul %f32 %val1 %val2\n"
6692 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
6693 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
6694 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
6695 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
6696 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
6697 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
6698 " OpReturnValue %ret\n"
6701 struct CaseNameDecoration
6708 CaseNameDecoration tests[] = {
6709 {"multiplication", "OpDecorate %mul NoContraction"},
6710 {"addition", "OpDecorate %add NoContraction"},
6711 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
6714 getHalfColorsFullAlpha(inputColors);
6716 for (deUint8 idx = 0; idx < 4; ++idx)
6718 inputColors[idx].setRed(0);
6719 outputColors[idx] = RGBA(0, 0, 0, 255);
6722 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
6724 map<string, string> fragments;
6726 fragments["decoration"] = tests[testNdx].decoration;
6727 fragments["pre_main"] = constantsAndTypes;
6728 fragments["testfun"] = function;
6730 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
6733 return group.release();
6736 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
6738 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
6741 const char constantsAndTypes[] =
6742 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
6743 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
6744 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
6745 "%fp_stype = OpTypePointer Function %stype\n";
6747 const char function[] =
6748 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6749 "%param1 = OpFunctionParameter %v4f32\n"
6751 "%v1 = OpVariable %fp_v4f32 Function\n"
6752 "%v2 = OpVariable %fp_a2f32 Function\n"
6753 "%v3 = OpVariable %fp_f32 Function\n"
6754 "%v = OpVariable %fp_stype Function\n"
6755 "%vv = OpVariable %fp_stype Function\n"
6756 "%vvv = OpVariable %fp_f32 Function\n"
6758 " OpStore %v1 %c_v4f32_1_1_1_1\n"
6759 " OpStore %v2 %c_a2f32_1\n"
6760 " OpStore %v3 %c_f32_1\n"
6762 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
6763 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
6764 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
6765 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
6766 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
6767 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
6769 " OpStore %p_v4f32 %v1_v ${access_type}\n"
6770 " OpStore %p_a2f32 %v2_v ${access_type}\n"
6771 " OpStore %p_f32 %v3_v ${access_type}\n"
6773 " OpCopyMemory %vv %v ${access_type}\n"
6774 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
6776 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
6777 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
6778 "%v_f32_3 = OpLoad %f32 %vvv\n"
6780 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
6781 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
6782 " OpReturnValue %ret2\n"
6785 struct NameMemoryAccess
6792 NameMemoryAccess tests[] =
6795 { "volatile", "Volatile" },
6796 { "aligned", "Aligned 1" },
6797 { "volatile_aligned", "Volatile|Aligned 1" },
6798 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
6799 { "volatile_nontemporal", "Volatile|Nontemporal" },
6800 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
6803 getHalfColorsFullAlpha(colors);
6805 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
6807 map<string, string> fragments;
6808 map<string, string> memoryAccess;
6809 memoryAccess["access_type"] = tests[testNdx].accessType;
6811 fragments["pre_main"] = constantsAndTypes;
6812 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
6813 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
6815 return memoryAccessTests.release();
6817 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
6819 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
6820 RGBA defaultColors[4];
6821 map<string, string> fragments;
6822 getDefaultColors(defaultColors);
6824 // First, simple cases that don't do anything with the OpUndef result.
6825 fragments["testfun"] =
6826 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6827 "%param1 = OpFunctionParameter %v4f32\n"
6828 "%label_testfun = OpLabel\n"
6829 "%undef = OpUndef %type\n"
6830 "OpReturnValue %param1\n"
6833 struct NameCodePair { string name, code; };
6834 const NameCodePair tests[] =
6836 {"bool", "%type = OpTypeBool"},
6837 {"vec2uint32", "%type = OpTypeVector %u32 2"},
6838 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown"},
6839 {"sampler", "%type = OpTypeSampler"},
6840 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img"},
6841 {"pointer", "%type = OpTypePointer Function %i32"},
6842 {"runtimearray", "%type = OpTypeRuntimeArray %f32"},
6843 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100"},
6844 {"struct", "%type = OpTypeStruct %f32 %i32 %u32"}};
6845 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
6847 fragments["pre_main"] = tests[testNdx].code;
6848 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
6852 fragments["testfun"] =
6853 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6854 "%param1 = OpFunctionParameter %v4f32\n"
6855 "%label_testfun = OpLabel\n"
6856 "%undef = OpUndef %f32\n"
6857 "%zero = OpFMul %f32 %undef %c_f32_0\n"
6858 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
6859 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
6860 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6861 "%b = OpFAdd %f32 %a %actually_zero\n"
6862 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
6863 "OpReturnValue %ret\n"
6866 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6868 fragments["testfun"] =
6869 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6870 "%param1 = OpFunctionParameter %v4f32\n"
6871 "%label_testfun = OpLabel\n"
6872 "%undef = OpUndef %i32\n"
6873 "%zero = OpIMul %i32 %undef %c_i32_0\n"
6874 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6875 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6876 "OpReturnValue %ret\n"
6879 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6881 fragments["testfun"] =
6882 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6883 "%param1 = OpFunctionParameter %v4f32\n"
6884 "%label_testfun = OpLabel\n"
6885 "%undef = OpUndef %u32\n"
6886 "%zero = OpIMul %u32 %undef %c_i32_0\n"
6887 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6888 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6889 "OpReturnValue %ret\n"
6892 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6894 fragments["testfun"] =
6895 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6896 "%param1 = OpFunctionParameter %v4f32\n"
6897 "%label_testfun = OpLabel\n"
6898 "%undef = OpUndef %v4f32\n"
6899 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
6900 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
6901 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
6902 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
6903 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
6904 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6905 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6906 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6907 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6908 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6909 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
6910 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
6911 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
6912 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6913 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6914 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6915 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6916 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6917 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6918 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6919 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6920 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6921 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6922 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6923 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6924 "OpReturnValue %ret\n"
6927 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6929 fragments["pre_main"] =
6930 "%v2f32 = OpTypeVector %f32 2\n"
6931 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
6932 fragments["testfun"] =
6933 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6934 "%param1 = OpFunctionParameter %v4f32\n"
6935 "%label_testfun = OpLabel\n"
6936 "%undef = OpUndef %m2x2f32\n"
6937 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
6938 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
6939 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
6940 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
6941 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
6942 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6943 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6944 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6945 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6946 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6947 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
6948 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
6949 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
6950 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6951 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6952 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6953 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6954 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6955 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6956 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6957 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6958 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6959 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6960 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6961 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6962 "OpReturnValue %ret\n"
6965 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
6967 return opUndefTests.release();
6970 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
6972 const RGBA inputColors[4] =
6975 RGBA(0, 0, 255, 255),
6976 RGBA(0, 255, 0, 255),
6977 RGBA(0, 255, 255, 255)
6980 const RGBA expectedColors[4] =
6982 RGBA(255, 0, 0, 255),
6983 RGBA(255, 0, 0, 255),
6984 RGBA(255, 0, 0, 255),
6985 RGBA(255, 0, 0, 255)
6988 const struct SingleFP16Possibility
6991 const char* constant; // Value to assign to %test_constant.
6993 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
6999 -constructNormalizedFloat(1, 0x300000),
7000 "%cond = OpFOrdEqual %bool %c %test_constant\n"
7005 constructNormalizedFloat(7, 0x000000),
7006 "%cond = OpFOrdEqual %bool %c %test_constant\n"
7008 // SPIR-V requires that OpQuantizeToF16 flushes
7009 // any numbers that would end up denormalized in F16 to zero.
7013 std::ldexp(1.5f, -140),
7014 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
7019 -std::ldexp(1.5f, -140),
7020 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
7025 std::ldexp(1.0f, -16),
7026 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
7027 }, // too small positive
7029 "negative_too_small",
7031 -std::ldexp(1.0f, -32),
7032 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
7033 }, // too small negative
7037 -std::ldexp(1.0f, 128),
7039 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
7040 "%inf = OpIsInf %bool %c\n"
7041 "%cond = OpLogicalAnd %bool %gz %inf\n"
7046 std::ldexp(1.0f, 128),
7048 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
7049 "%inf = OpIsInf %bool %c\n"
7050 "%cond = OpLogicalAnd %bool %gz %inf\n"
7053 "round_to_negative_inf",
7055 -std::ldexp(1.0f, 32),
7057 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
7058 "%inf = OpIsInf %bool %c\n"
7059 "%cond = OpLogicalAnd %bool %gz %inf\n"
7064 std::ldexp(1.0f, 16),
7066 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
7067 "%inf = OpIsInf %bool %c\n"
7068 "%cond = OpLogicalAnd %bool %gz %inf\n"
7073 std::numeric_limits<float>::quiet_NaN(),
7075 // Test for any NaN value, as NaNs are not preserved
7076 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
7077 "%cond = OpIsNan %bool %direct_quant\n"
7082 std::numeric_limits<float>::quiet_NaN(),
7084 // Test for any NaN value, as NaNs are not preserved
7085 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
7086 "%cond = OpIsNan %bool %direct_quant\n"
7089 const char* constants =
7090 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
7092 StringTemplate function (
7093 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7094 "%param1 = OpFunctionParameter %v4f32\n"
7095 "%label_testfun = OpLabel\n"
7096 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7097 "%b = OpFAdd %f32 %test_constant %a\n"
7098 "%c = OpQuantizeToF16 %f32 %b\n"
7100 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7101 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
7102 " OpReturnValue %retval\n"
7106 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
7107 const char* specConstants =
7108 "%test_constant = OpSpecConstant %f32 0.\n"
7109 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
7111 StringTemplate specConstantFunction(
7112 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7113 "%param1 = OpFunctionParameter %v4f32\n"
7114 "%label_testfun = OpLabel\n"
7116 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7117 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
7118 " OpReturnValue %retval\n"
7122 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
7124 map<string, string> codeSpecialization;
7125 map<string, string> fragments;
7126 codeSpecialization["condition"] = tests[idx].condition;
7127 fragments["testfun"] = function.specialize(codeSpecialization);
7128 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
7129 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7132 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
7134 map<string, string> codeSpecialization;
7135 map<string, string> fragments;
7136 vector<deInt32> passConstants;
7137 deInt32 specConstant;
7139 codeSpecialization["condition"] = tests[idx].condition;
7140 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
7141 fragments["decoration"] = specDecorations;
7142 fragments["pre_main"] = specConstants;
7144 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
7145 passConstants.push_back(specConstant);
7147 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7151 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
7153 RGBA inputColors[4] = {
7155 RGBA(0, 0, 255, 255),
7156 RGBA(0, 255, 0, 255),
7157 RGBA(0, 255, 255, 255)
7160 RGBA expectedColors[4] =
7162 RGBA(255, 0, 0, 255),
7163 RGBA(255, 0, 0, 255),
7164 RGBA(255, 0, 0, 255),
7165 RGBA(255, 0, 0, 255)
7168 struct DualFP16Possibility
7173 const char* possibleOutput1;
7174 const char* possibleOutput2;
7177 "positive_round_up_or_round_down",
7179 constructNormalizedFloat(8, 0x300300),
7184 "negative_round_up_or_round_down",
7186 -constructNormalizedFloat(-7, 0x600800),
7193 constructNormalizedFloat(2, 0x01e000),
7198 "carry_to_exponent",
7200 constructNormalizedFloat(1, 0xffe000),
7205 StringTemplate constants (
7206 "%input_const = OpConstant %f32 ${input}\n"
7207 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7208 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7211 StringTemplate specConstants (
7212 "%input_const = OpSpecConstant %f32 0.\n"
7213 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7214 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7217 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
7219 const char* function =
7220 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7221 "%param1 = OpFunctionParameter %v4f32\n"
7222 "%label_testfun = OpLabel\n"
7223 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7224 // For the purposes of this test we assume that 0.f will always get
7225 // faithfully passed through the pipeline stages.
7226 "%b = OpFAdd %f32 %input_const %a\n"
7227 "%c = OpQuantizeToF16 %f32 %b\n"
7228 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
7229 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
7230 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
7231 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7232 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
7233 " OpReturnValue %retval\n"
7236 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7237 map<string, string> fragments;
7238 map<string, string> constantSpecialization;
7240 constantSpecialization["input"] = tests[idx].input;
7241 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7242 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7243 fragments["testfun"] = function;
7244 fragments["pre_main"] = constants.specialize(constantSpecialization);
7245 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7248 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7249 map<string, string> fragments;
7250 map<string, string> constantSpecialization;
7251 vector<deInt32> passConstants;
7252 deInt32 specConstant;
7254 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7255 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7256 fragments["testfun"] = function;
7257 fragments["decoration"] = specDecorations;
7258 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
7260 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
7261 passConstants.push_back(specConstant);
7263 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7267 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
7269 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
7270 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
7271 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
7272 return opQuantizeTests.release();
7275 struct ShaderPermutation
7277 deUint8 vertexPermutation;
7278 deUint8 geometryPermutation;
7279 deUint8 tesscPermutation;
7280 deUint8 tessePermutation;
7281 deUint8 fragmentPermutation;
7284 ShaderPermutation getShaderPermutation(deUint8 inputValue)
7286 ShaderPermutation permutation =
7288 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
7289 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
7290 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
7291 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
7292 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
7297 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
7299 RGBA defaultColors[4];
7300 RGBA invertedColors[4];
7301 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
7303 const ShaderElement combinedPipeline[] =
7305 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
7306 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
7307 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7308 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7309 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
7312 getDefaultColors(defaultColors);
7313 getInvertedDefaultColors(invertedColors);
7314 addFunctionCaseWithPrograms<InstanceContext>(moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline, createInstanceContext(combinedPipeline, map<string, string>()));
7316 const char* numbers[] =
7321 for (deInt8 idx = 0; idx < 32; ++idx)
7323 ShaderPermutation permutation = getShaderPermutation(idx);
7324 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
7325 const ShaderElement pipeline[] =
7327 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
7328 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
7329 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7330 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7331 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
7334 // If there are an even number of swaps, then it should be no-op.
7335 // If there are an odd number, the color should be flipped.
7336 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
7338 addFunctionCaseWithPrograms<InstanceContext>(moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline, createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
7342 addFunctionCaseWithPrograms<InstanceContext>(moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline, createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
7345 return moduleTests.release();
7348 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
7350 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
7351 RGBA defaultColors[4];
7352 getDefaultColors(defaultColors);
7353 map<string, string> fragments;
7354 fragments["pre_main"] =
7355 "%c_f32_5 = OpConstant %f32 5.\n";
7357 // A loop with a single block. The Continue Target is the loop block
7358 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
7359 // -- the "continue construct" forms the entire loop.
7360 fragments["testfun"] =
7361 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7362 "%param1 = OpFunctionParameter %v4f32\n"
7364 "%entry = OpLabel\n"
7365 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7368 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7370 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7371 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
7372 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7373 "%val = OpFAdd %f32 %val1 %delta\n"
7374 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
7375 "%count__ = OpISub %i32 %count %c_i32_1\n"
7376 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7377 "OpLoopMerge %exit %loop None\n"
7378 "OpBranchConditional %again %loop %exit\n"
7381 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7382 "OpReturnValue %result\n"
7386 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
7388 // Body comprised of multiple basic blocks.
7389 const StringTemplate multiBlock(
7390 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7391 "%param1 = OpFunctionParameter %v4f32\n"
7393 "%entry = OpLabel\n"
7394 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7397 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7399 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
7400 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
7401 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
7402 // There are several possibilities for the Continue Target below. Each
7403 // will be specialized into a separate test case.
7404 "OpLoopMerge %exit ${continue_target} None\n"
7408 ";delta_next = (delta > 0) ? -1 : 1;\n"
7409 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
7410 "OpSelectionMerge %gather DontFlatten\n"
7411 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
7414 "OpBranch %gather\n"
7417 "OpBranch %gather\n"
7419 "%gather = OpLabel\n"
7420 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
7421 "%val = OpFAdd %f32 %val1 %delta\n"
7422 "%count__ = OpISub %i32 %count %c_i32_1\n"
7423 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7424 "OpBranchConditional %again %loop %exit\n"
7427 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7428 "OpReturnValue %result\n"
7432 map<string, string> continue_target;
7434 // The Continue Target is the loop block itself.
7435 continue_target["continue_target"] = "%loop";
7436 fragments["testfun"] = multiBlock.specialize(continue_target);
7437 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
7439 // The Continue Target is at the end of the loop.
7440 continue_target["continue_target"] = "%gather";
7441 fragments["testfun"] = multiBlock.specialize(continue_target);
7442 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
7444 // A loop with continue statement.
7445 fragments["testfun"] =
7446 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7447 "%param1 = OpFunctionParameter %v4f32\n"
7449 "%entry = OpLabel\n"
7450 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7453 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
7455 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7456 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
7457 "OpLoopMerge %exit %continue None\n"
7461 ";skip if %count==2\n"
7462 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
7463 "OpSelectionMerge %continue DontFlatten\n"
7464 "OpBranchConditional %eq2 %continue %body\n"
7467 "%fcount = OpConvertSToF %f32 %count\n"
7468 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7469 "OpBranch %continue\n"
7471 "%continue = OpLabel\n"
7472 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
7473 "%count__ = OpISub %i32 %count %c_i32_1\n"
7474 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7475 "OpBranchConditional %again %loop %exit\n"
7478 "%same = OpFSub %f32 %val %c_f32_8\n"
7479 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7480 "OpReturnValue %result\n"
7482 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
7484 // A loop with break.
7485 fragments["testfun"] =
7486 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7487 "%param1 = OpFunctionParameter %v4f32\n"
7489 "%entry = OpLabel\n"
7490 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7491 "%dot = OpDot %f32 %param1 %param1\n"
7492 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7493 "%zero = OpConvertFToU %u32 %div\n"
7494 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7495 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7498 ";adds 4 and 3 to %val0 (exits early)\n"
7500 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7501 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7502 "OpLoopMerge %exit %continue None\n"
7506 ";end loop if %count==%two\n"
7507 "%above2 = OpSGreaterThan %bool %count %two\n"
7508 "OpSelectionMerge %continue DontFlatten\n"
7509 "OpBranchConditional %above2 %body %exit\n"
7512 "%fcount = OpConvertSToF %f32 %count\n"
7513 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7514 "OpBranch %continue\n"
7516 "%continue = OpLabel\n"
7517 "%count__ = OpISub %i32 %count %c_i32_1\n"
7518 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7519 "OpBranchConditional %again %loop %exit\n"
7522 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
7523 "%same = OpFSub %f32 %val_post %c_f32_7\n"
7524 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7525 "OpReturnValue %result\n"
7527 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
7529 // A loop with return.
7530 fragments["testfun"] =
7531 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7532 "%param1 = OpFunctionParameter %v4f32\n"
7534 "%entry = OpLabel\n"
7535 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7536 "%dot = OpDot %f32 %param1 %param1\n"
7537 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7538 "%zero = OpConvertFToU %u32 %div\n"
7539 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7540 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7543 ";returns early without modifying %param1\n"
7545 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7546 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7547 "OpLoopMerge %exit %continue None\n"
7551 ";return if %count==%two\n"
7552 "%above2 = OpSGreaterThan %bool %count %two\n"
7553 "OpSelectionMerge %continue DontFlatten\n"
7554 "OpBranchConditional %above2 %body %early_exit\n"
7556 "%early_exit = OpLabel\n"
7557 "OpReturnValue %param1\n"
7560 "%fcount = OpConvertSToF %f32 %count\n"
7561 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7562 "OpBranch %continue\n"
7564 "%continue = OpLabel\n"
7565 "%count__ = OpISub %i32 %count %c_i32_1\n"
7566 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7567 "OpBranchConditional %again %loop %exit\n"
7570 ";should never get here, so return an incorrect result\n"
7571 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
7572 "OpReturnValue %result\n"
7574 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
7576 return testGroup.release();
7579 // Adds a new test to group using custom fragments for the tessellation-control
7580 // stage and passthrough fragments for all other stages. Uses default colors
7581 // for input and expected output.
7582 void addTessCtrlTest(tcu::TestCaseGroup* group, const char* name, const map<string, string>& fragments)
7584 RGBA defaultColors[4];
7585 getDefaultColors(defaultColors);
7586 const ShaderElement pipelineStages[] =
7588 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
7589 ShaderElement("tessc", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7590 ShaderElement("tesse", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7591 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
7594 addFunctionCaseWithPrograms<InstanceContext>(group, name, "", addShaderCodeCustomTessControl,
7595 runAndVerifyDefaultPipeline, createInstanceContext(
7596 pipelineStages, defaultColors, defaultColors, fragments, StageToSpecConstantMap()));
7599 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
7600 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
7602 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
7603 map<string, string> fragments;
7605 // A barrier inside a function body.
7606 fragments["pre_main"] =
7607 "%Workgroup = OpConstant %i32 2\n"
7608 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
7609 fragments["testfun"] =
7610 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7611 "%param1 = OpFunctionParameter %v4f32\n"
7612 "%label_testfun = OpLabel\n"
7613 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7614 "OpReturnValue %param1\n"
7616 addTessCtrlTest(testGroup.get(), "in_function", fragments);
7618 // Common setup code for the following tests.
7619 fragments["pre_main"] =
7620 "%Workgroup = OpConstant %i32 2\n"
7621 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7622 "%c_f32_5 = OpConstant %f32 5.\n";
7623 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
7624 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7625 "%param1 = OpFunctionParameter %v4f32\n"
7626 "%entry = OpLabel\n"
7627 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7628 "%dot = OpDot %f32 %param1 %param1\n"
7629 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7630 "%zero = OpConvertFToU %u32 %div\n";
7632 // Barriers inside OpSwitch branches.
7633 fragments["testfun"] =
7635 "OpSelectionMerge %switch_exit None\n"
7636 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
7638 "%case1 = OpLabel\n"
7639 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7640 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7641 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7642 "OpBranch %switch_exit\n"
7644 "%switch_default = OpLabel\n"
7645 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7646 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7647 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7648 "OpBranch %switch_exit\n"
7650 "%case0 = OpLabel\n"
7651 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7652 "OpBranch %switch_exit\n"
7654 "%switch_exit = OpLabel\n"
7655 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
7656 "OpReturnValue %ret\n"
7658 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
7660 // Barriers inside if-then-else.
7661 fragments["testfun"] =
7663 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
7664 "OpSelectionMerge %exit DontFlatten\n"
7665 "OpBranchConditional %eq0 %then %else\n"
7668 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7669 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7670 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7674 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7678 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
7679 "OpReturnValue %ret\n"
7681 addTessCtrlTest(testGroup.get(), "in_if", fragments);
7683 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
7684 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
7685 fragments["testfun"] =
7687 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
7688 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
7689 "OpSelectionMerge %exit DontFlatten\n"
7690 "OpBranchConditional %thread0 %then %else\n"
7693 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7697 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
7701 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
7702 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7703 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
7704 "OpReturnValue %ret\n"
7706 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
7708 // A barrier inside a loop.
7709 fragments["pre_main"] =
7710 "%Workgroup = OpConstant %i32 2\n"
7711 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7712 "%c_f32_10 = OpConstant %f32 10.\n";
7713 fragments["testfun"] =
7714 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7715 "%param1 = OpFunctionParameter %v4f32\n"
7716 "%entry = OpLabel\n"
7717 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7720 ";adds 4, 3, 2, and 1 to %val0\n"
7722 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7723 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7724 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7725 "%fcount = OpConvertSToF %f32 %count\n"
7726 "%val = OpFAdd %f32 %val1 %fcount\n"
7727 "%count__ = OpISub %i32 %count %c_i32_1\n"
7728 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7729 "OpLoopMerge %exit %loop None\n"
7730 "OpBranchConditional %again %loop %exit\n"
7733 "%same = OpFSub %f32 %val %c_f32_10\n"
7734 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7735 "OpReturnValue %ret\n"
7737 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
7739 return testGroup.release();
7742 // Test for the OpFRem instruction.
7743 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
7745 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
7746 map<string, string> fragments;
7747 RGBA inputColors[4];
7748 RGBA outputColors[4];
7750 fragments["pre_main"] =
7751 "%c_f32_3 = OpConstant %f32 3.0\n"
7752 "%c_f32_n3 = OpConstant %f32 -3.0\n"
7753 "%c_f32_4 = OpConstant %f32 4.0\n"
7754 "%c_f32_p75 = OpConstant %f32 0.75\n"
7755 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
7756 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
7757 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
7759 // The test does the following.
7760 // vec4 result = (param1 * 8.0) - 4.0;
7761 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
7762 fragments["testfun"] =
7763 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7764 "%param1 = OpFunctionParameter %v4f32\n"
7765 "%label_testfun = OpLabel\n"
7766 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
7767 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
7768 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
7769 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
7770 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
7771 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
7772 "OpReturnValue %xy_0_1\n"
7776 inputColors[0] = RGBA(16, 16, 0, 255);
7777 inputColors[1] = RGBA(232, 232, 0, 255);
7778 inputColors[2] = RGBA(232, 16, 0, 255);
7779 inputColors[3] = RGBA(16, 232, 0, 255);
7781 outputColors[0] = RGBA(64, 64, 0, 255);
7782 outputColors[1] = RGBA(255, 255, 0, 255);
7783 outputColors[2] = RGBA(255, 64, 0, 255);
7784 outputColors[3] = RGBA(64, 255, 0, 255);
7786 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
7787 return testGroup.release();
7792 INTEGER_TYPE_SIGNED_16,
7793 INTEGER_TYPE_SIGNED_32,
7794 INTEGER_TYPE_SIGNED_64,
7796 INTEGER_TYPE_UNSIGNED_16,
7797 INTEGER_TYPE_UNSIGNED_32,
7798 INTEGER_TYPE_UNSIGNED_64,
7801 const string getBitWidthStr (IntegerType type)
7805 case INTEGER_TYPE_SIGNED_16:
7806 case INTEGER_TYPE_UNSIGNED_16: return "16";
7808 case INTEGER_TYPE_SIGNED_32:
7809 case INTEGER_TYPE_UNSIGNED_32: return "32";
7811 case INTEGER_TYPE_SIGNED_64:
7812 case INTEGER_TYPE_UNSIGNED_64: return "64";
7814 default: DE_ASSERT(false);
7819 bool isSigned (IntegerType type)
7821 return (type <= INTEGER_TYPE_SIGNED_64);
7824 const string getTypeName (IntegerType type)
7826 string prefix = isSigned(type) ? "" : "u";
7827 return prefix + "int" + getBitWidthStr(type);
7830 const string getTestName (IntegerType from, IntegerType to)
7832 return getTypeName(from) + "_to_" + getTypeName(to);
7835 const string getAsmTypeDeclaration (IntegerType type)
7837 string sign = isSigned(type) ? " 1" : " 0";
7838 return "OpTypeInt " + getBitWidthStr(type) + sign;
7841 const string getConvertCaseShaderStr (const string& instruction, map<string, string> types)
7843 const StringTemplate shader (
7844 "OpCapability Shader\n"
7845 "${int_capabilities}"
7846 "OpMemoryModel Logical GLSL450\n"
7847 "OpEntryPoint GLCompute %main \"main\" %id\n"
7848 "OpExecutionMode %main LocalSize 1 1 1\n"
7849 "OpSource GLSL 430\n"
7850 "OpName %main \"main\"\n"
7851 "OpName %id \"gl_GlobalInvocationID\"\n"
7853 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7854 "OpDecorate %indata DescriptorSet 0\n"
7855 "OpDecorate %indata Binding 0\n"
7856 "OpDecorate %outdata DescriptorSet 0\n"
7857 "OpDecorate %outdata Binding 1\n"
7858 "OpDecorate %in_buf BufferBlock\n"
7859 "OpDecorate %out_buf BufferBlock\n"
7860 "OpMemberDecorate %in_buf 0 Offset 0\n"
7861 "OpMemberDecorate %out_buf 0 Offset 0\n"
7863 "%void = OpTypeVoid\n"
7864 "%voidf = OpTypeFunction %void\n"
7865 "%u32 = OpTypeInt 32 0\n"
7866 "%i32 = OpTypeInt 32 1\n"
7867 "%uvec3 = OpTypeVector %u32 3\n"
7868 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7870 "%in_type = ${inputType}\n"
7871 "%out_type = ${outputType}\n"
7873 "%in_ptr = OpTypePointer Uniform %in_type\n"
7874 "%out_ptr = OpTypePointer Uniform %out_type\n"
7875 "%in_arr = OpTypeRuntimeArray %in_type\n"
7876 "%out_arr = OpTypeRuntimeArray %out_type\n"
7877 "%in_buf = OpTypeStruct %in_arr\n"
7878 "%out_buf = OpTypeStruct %out_arr\n"
7879 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
7880 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
7881 "%indata = OpVariable %in_bufptr Uniform\n"
7882 "%outdata = OpVariable %out_bufptr Uniform\n"
7883 "%inputptr = OpTypePointer Input %in_type\n"
7884 "%id = OpVariable %uvec3ptr Input\n"
7886 "%zero = OpConstant %i32 0\n"
7888 "%main = OpFunction %void None %voidf\n"
7889 "%label = OpLabel\n"
7890 "%idval = OpLoad %uvec3 %id\n"
7891 "%x = OpCompositeExtract %u32 %idval 0\n"
7892 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
7893 "%outloc = OpAccessChain %out_ptr %outdata %zero %x\n"
7894 "%inval = OpLoad %in_type %inloc\n"
7895 "%conv = ${instruction} %out_type %inval\n"
7896 " OpStore %outloc %conv\n"
7901 types["instruction"] = instruction;
7903 return shader.specialize(types);
7906 template<typename T>
7907 BufferSp getSpecializedBuffer (deInt64 number)
7909 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
7912 BufferSp getBuffer (IntegerType type, deInt64 number)
7916 case INTEGER_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
7917 case INTEGER_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
7918 case INTEGER_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
7920 case INTEGER_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
7921 case INTEGER_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
7922 case INTEGER_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
7924 default: DE_ASSERT(false);
7925 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
7929 bool usesInt16 (IntegerType from, IntegerType to)
7931 return (from == INTEGER_TYPE_SIGNED_16 || from == INTEGER_TYPE_UNSIGNED_16
7932 || to == INTEGER_TYPE_SIGNED_16 || to == INTEGER_TYPE_UNSIGNED_16);
7935 bool usesInt64 (IntegerType from, IntegerType to)
7937 return (from == INTEGER_TYPE_SIGNED_64 || from == INTEGER_TYPE_UNSIGNED_64
7938 || to == INTEGER_TYPE_SIGNED_64 || to == INTEGER_TYPE_UNSIGNED_64);
7941 ConvertTestFeatures getUsedFeatures (IntegerType from, IntegerType to)
7943 if (usesInt16(from, to))
7945 if (usesInt64(from, to))
7947 return CONVERT_TEST_USES_INT16_INT64;
7951 return CONVERT_TEST_USES_INT16;
7956 return CONVERT_TEST_USES_INT64;
7962 ConvertCase (IntegerType from, IntegerType to, deInt64 number)
7965 , m_features (getUsedFeatures(from, to))
7966 , m_name (getTestName(from, to))
7967 , m_inputBuffer (getBuffer(from, number))
7968 , m_outputBuffer (getBuffer(to, number))
7970 m_asmTypes["inputType"] = getAsmTypeDeclaration(from);
7971 m_asmTypes["outputType"] = getAsmTypeDeclaration(to);
7973 if (m_features == CONVERT_TEST_USES_INT16)
7975 m_asmTypes["int_capabilities"] = "OpCapability Int16\n";
7977 else if (m_features == CONVERT_TEST_USES_INT64)
7979 m_asmTypes["int_capabilities"] = "OpCapability Int64\n";
7981 else if (m_features == CONVERT_TEST_USES_INT16_INT64)
7983 m_asmTypes["int_capabilities"] = "OpCapability Int16\n \
7984 OpCapability Int64\n";
7992 IntegerType m_fromType;
7993 IntegerType m_toType;
7994 ConvertTestFeatures m_features;
7996 map<string, string> m_asmTypes;
7997 BufferSp m_inputBuffer;
7998 BufferSp m_outputBuffer;
8001 void createSConvertCases (vector<ConvertCase>& testCases)
8003 // Convert int to int
8004 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_32, 14669));
8005 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_64, 3341));
8007 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_SIGNED_64, 973610259));
8009 // Convert int to unsigned int
8010 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_32, 9288));
8011 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_64, 15460));
8013 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_UNSIGNED_64, 346213461));
8016 // Test for the OpSConvert instruction.
8017 tcu::TestCaseGroup* createSConvertTests (tcu::TestContext& testCtx)
8019 const string instruction ("OpSConvert");
8020 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "sconvert", "OpSConvert"));
8021 vector<ConvertCase> testCases;
8022 createSConvertCases(testCases);
8024 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8026 ComputeShaderSpec spec;
8028 spec.assembly = getConvertCaseShaderStr(instruction, test->m_asmTypes);
8029 spec.inputs.push_back(test->m_inputBuffer);
8030 spec.outputs.push_back(test->m_inputBuffer);
8031 spec.numWorkGroups = IVec3(1, 1, 1);
8033 group->addChild(new ConvertTestCase(testCtx, test->m_name.c_str(), "Convert integers with OpSConvert.", spec, test->m_features));
8036 return group.release();
8039 void createUConvertCases (vector<ConvertCase>& testCases)
8041 // Convert unsigned int to unsigned int
8042 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_32, 60653));
8043 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_64, 17991));
8045 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_UNSIGNED_64, 904256275));
8047 // Convert unsigned int to int
8048 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_32, 38002));
8049 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_64, 64921));
8051 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_SIGNED_64, 4294956295ll));
8054 // Test for the OpUConvert instruction.
8055 tcu::TestCaseGroup* createUConvertTests (tcu::TestContext& testCtx)
8057 const string instruction ("OpUConvert");
8058 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "uconvert", "OpUConvert"));
8059 vector<ConvertCase> testCases;
8060 createUConvertCases(testCases);
8062 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8064 ComputeShaderSpec spec;
8066 spec.assembly = getConvertCaseShaderStr(instruction, test->m_asmTypes);
8067 spec.inputs.push_back(test->m_inputBuffer);
8068 spec.outputs.push_back(test->m_inputBuffer);
8069 spec.numWorkGroups = IVec3(1, 1, 1);
8071 group->addChild(new ConvertTestCase(testCtx, test->m_name.c_str(), "Convert integers with OpUConvert.", spec, test->m_features));
8073 return group.release();
8084 const string getNumberTypeName (const NumberType type)
8086 if (type == TYPE_INT)
8090 else if (type == TYPE_UINT)
8094 else if (type == TYPE_FLOAT)
8105 deInt32 getInt(de::Random& rnd)
8107 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
8110 template <typename T>
8111 const string numberToString (T number)
8113 std::stringstream ss;
8118 const string repeatString (const string& str, int times)
8121 for (int i = 0; i < times; ++i)
8128 const string getRandomConstantString (const NumberType type, de::Random& rnd)
8130 if (type == TYPE_INT)
8132 return numberToString<deInt32>(getInt(rnd));
8134 else if (type == TYPE_UINT)
8136 return numberToString<deUint32>(rnd.getUint32());
8138 else if (type == TYPE_FLOAT)
8140 return numberToString<float>(rnd.getFloat());
8149 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8151 map<string, string> params;
8154 for (int width = 2; width <= 4; ++width)
8156 string randomConst = numberToString(getInt(rnd));
8157 string widthStr = numberToString(width);
8158 int index = rnd.getInt(0, width-1);
8160 params["name"] = "vec_" + widthStr;
8161 params["compositeType"] = "%composite = OpTypeVector %custom " + widthStr +"\n";
8162 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
8163 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8164 params["indexes"] = numberToString(index);
8165 testCases.push_back(params);
8169 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8171 const int limit = 10;
8172 map<string, string> params;
8174 for (int width = 2; width <= limit; ++width)
8176 string randomConst = numberToString(getInt(rnd));
8177 string widthStr = numberToString(width);
8178 int index = rnd.getInt(0, width-1);
8180 params["name"] = "array_" + widthStr;
8181 params["compositeType"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
8182 + "%composite = OpTypeArray %custom %arraywidth\n";
8184 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
8185 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8186 params["indexes"] = numberToString(index);
8187 testCases.push_back(params);
8191 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8193 const int limit = 10;
8194 map<string, string> params;
8196 for (int width = 2; width <= limit; ++width)
8198 string randomConst = numberToString(getInt(rnd));
8199 int index = rnd.getInt(0, width-1);
8201 params["name"] = "struct_" + numberToString(width);
8202 params["compositeType"] = "%composite = OpTypeStruct" + repeatString(" %custom", width) + "\n";
8203 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
8204 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8205 params["indexes"] = numberToString(index);
8206 testCases.push_back(params);
8210 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8212 map<string, string> params;
8215 for (int width = 2; width <= 4; ++width)
8217 string widthStr = numberToString(width);
8219 for (int column = 2 ; column <= 4; ++column)
8221 int index_0 = rnd.getInt(0, column-1);
8222 int index_1 = rnd.getInt(0, width-1);
8223 string columnStr = numberToString(column);
8225 params["name"] = "matrix_" + widthStr + "x" + columnStr;
8226 params["compositeType"] = string("%vectype = OpTypeVector %custom " + widthStr + "\n")
8227 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
8229 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n"
8230 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
8232 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
8233 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
8234 testCases.push_back(params);
8239 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8241 createVectorCompositeCases(testCases, rnd, type);
8242 createArrayCompositeCases(testCases, rnd, type);
8243 createStructCompositeCases(testCases, rnd, type);
8244 // Matrix only supports float types
8245 if (type == TYPE_FLOAT)
8247 createMatrixCompositeCases(testCases, rnd, type);
8251 const string getAssemblyTypeDeclaration (const NumberType type)
8255 case TYPE_INT: return "OpTypeInt 32 1";
8256 case TYPE_UINT: return "OpTypeInt 32 0";
8257 case TYPE_FLOAT: return "OpTypeFloat 32";
8258 default: DE_ASSERT(false); return "";
8262 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
8264 map<string, string> parameters(params);
8266 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
8268 return StringTemplate (
8269 "OpCapability Shader\n"
8270 "OpCapability Matrix\n"
8271 "OpMemoryModel Logical GLSL450\n"
8272 "OpEntryPoint GLCompute %main \"main\" %id\n"
8273 "OpExecutionMode %main LocalSize 1 1 1\n"
8275 "OpSource GLSL 430\n"
8276 "OpName %main \"main\"\n"
8277 "OpName %id \"gl_GlobalInvocationID\"\n"
8280 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8281 "OpDecorate %buf BufferBlock\n"
8282 "OpDecorate %indata DescriptorSet 0\n"
8283 "OpDecorate %indata Binding 0\n"
8284 "OpDecorate %outdata DescriptorSet 0\n"
8285 "OpDecorate %outdata Binding 1\n"
8286 "OpDecorate %customarr ArrayStride 4\n"
8287 "OpMemberDecorate %buf 0 Offset 0\n"
8290 "%void = OpTypeVoid\n"
8291 "%voidf = OpTypeFunction %void\n"
8292 "%u32 = OpTypeInt 32 0\n"
8293 "%i32 = OpTypeInt 32 1\n"
8294 "%uvec3 = OpTypeVector %u32 3\n"
8295 "%uvec3ptr = OpTypePointer Input %uvec3\n"
8298 "%custom = ${typeDeclaration}\n"
8304 // Inherited from custom
8305 "%customptr = OpTypePointer Uniform %custom\n"
8306 "%customarr = OpTypeRuntimeArray %custom\n"
8307 "%buf = OpTypeStruct %customarr\n"
8308 "%bufptr = OpTypePointer Uniform %buf\n"
8310 "%indata = OpVariable %bufptr Uniform\n"
8311 "%outdata = OpVariable %bufptr Uniform\n"
8313 "%id = OpVariable %uvec3ptr Input\n"
8314 "%zero = OpConstant %i32 0\n"
8316 "%main = OpFunction %void None %voidf\n"
8317 "%label = OpLabel\n"
8318 "%idval = OpLoad %uvec3 %id\n"
8319 "%x = OpCompositeExtract %u32 %idval 0\n"
8321 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
8322 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
8323 // Read the input value
8324 "%inval = OpLoad %custom %inloc\n"
8325 // Create the composite and fill it
8326 "${compositeConstruct}"
8327 // Insert the input value to a place
8328 "%instance2 = OpCompositeInsert %composite %inval %instance ${indexes}\n"
8329 // Read back the value from the position
8330 "%out_val = OpCompositeExtract %custom %instance2 ${indexes}\n"
8331 // Store it in the output position
8332 " OpStore %outloc %out_val\n"
8335 ).specialize(parameters);
8338 template<typename T>
8339 BufferSp createCompositeBuffer(T number)
8341 return BufferSp(new Buffer<T>(vector<T>(1, number)));
8344 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
8346 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
8347 de::Random rnd (deStringHash(group->getName()));
8349 for (int type = TYPE_INT; type != TYPE_END; ++type)
8351 NumberType numberType = NumberType(type);
8352 const string typeName = getNumberTypeName(numberType);
8353 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
8354 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8355 vector<map<string, string> > testCases;
8357 createCompositeCases(testCases, rnd, numberType);
8359 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8361 ComputeShaderSpec spec;
8363 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
8369 deInt32 number = getInt(rnd);
8370 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8371 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8376 deUint32 number = rnd.getUint32();
8377 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8378 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8383 float number = rnd.getFloat();
8384 spec.inputs.push_back(createCompositeBuffer<float>(number));
8385 spec.outputs.push_back(createCompositeBuffer<float>(number));
8392 spec.numWorkGroups = IVec3(1, 1, 1);
8393 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
8395 group->addChild(subGroup.release());
8397 return group.release();
8400 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
8402 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
8403 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
8404 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
8406 computeTests->addChild(createOpNopGroup(testCtx));
8407 computeTests->addChild(createOpFUnordGroup(testCtx));
8408 computeTests->addChild(createOpLineGroup(testCtx));
8409 computeTests->addChild(createOpNoLineGroup(testCtx));
8410 computeTests->addChild(createOpConstantNullGroup(testCtx));
8411 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
8412 computeTests->addChild(createOpConstantUsageGroup(testCtx));
8413 computeTests->addChild(createSpecConstantGroup(testCtx));
8414 computeTests->addChild(createOpSourceGroup(testCtx));
8415 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
8416 computeTests->addChild(createDecorationGroupGroup(testCtx));
8417 computeTests->addChild(createOpPhiGroup(testCtx));
8418 computeTests->addChild(createLoopControlGroup(testCtx));
8419 computeTests->addChild(createFunctionControlGroup(testCtx));
8420 computeTests->addChild(createSelectionControlGroup(testCtx));
8421 computeTests->addChild(createBlockOrderGroup(testCtx));
8422 computeTests->addChild(createMultipleShaderGroup(testCtx));
8423 computeTests->addChild(createMemoryAccessGroup(testCtx));
8424 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
8425 computeTests->addChild(createOpCopyObjectGroup(testCtx));
8426 computeTests->addChild(createNoContractionGroup(testCtx));
8427 computeTests->addChild(createOpUndefGroup(testCtx));
8428 computeTests->addChild(createOpUnreachableGroup(testCtx));
8429 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
8430 computeTests ->addChild(createOpFRemGroup(testCtx));
8431 computeTests->addChild(createSConvertTests(testCtx));
8432 computeTests->addChild(createUConvertTests(testCtx));
8433 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
8435 RGBA defaultColors[4];
8436 getDefaultColors(defaultColors);
8438 de::MovePtr<tcu::TestCaseGroup> opnopTests (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
8439 map<string, string> opNopFragments;
8440 opNopFragments["testfun"] =
8441 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
8442 "%param1 = OpFunctionParameter %v4f32\n"
8443 "%label_testfun = OpLabel\n"
8452 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
8453 "%b = OpFAdd %f32 %a %a\n"
8455 "%c = OpFSub %f32 %b %a\n"
8456 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
8459 "OpReturnValue %ret\n"
8462 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, opnopTests.get());
8465 graphicsTests->addChild(opnopTests.release());
8466 graphicsTests->addChild(createOpSourceTests(testCtx));
8467 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
8468 graphicsTests->addChild(createOpLineTests(testCtx));
8469 graphicsTests->addChild(createOpNoLineTests(testCtx));
8470 graphicsTests->addChild(createOpConstantNullTests(testCtx));
8471 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
8472 graphicsTests->addChild(createMemoryAccessTests(testCtx));
8473 graphicsTests->addChild(createOpUndefTests(testCtx));
8474 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
8475 graphicsTests->addChild(createModuleTests(testCtx));
8476 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
8477 graphicsTests->addChild(createOpPhiTests(testCtx));
8478 graphicsTests->addChild(createNoContractionTests(testCtx));
8479 graphicsTests->addChild(createOpQuantizeTests(testCtx));
8480 graphicsTests->addChild(createLoopTests(testCtx));
8481 graphicsTests->addChild(createSpecConstantTests(testCtx));
8482 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
8483 graphicsTests->addChild(createBarrierTests(testCtx));
8484 graphicsTests->addChild(createDecorationGroupTests(testCtx));
8485 graphicsTests->addChild(createFRemTests(testCtx));
8487 instructionTests->addChild(computeTests.release());
8488 instructionTests->addChild(graphicsTests.release());
8490 return instructionTests.release();