1 /*-------------------------------------------------------------------------
2 * Vulkan Conformance Tests
3 * ------------------------
5 * Copyright (c) 2015 Google Inc.
6 * Copyright (c) 2016 The Khronos Group Inc.
8 * Licensed under the Apache License, Version 2.0 (the "License");
9 * you may not use this file except in compliance with the License.
10 * You may obtain a copy of the License at
12 * http://www.apache.org/licenses/LICENSE-2.0
14 * Unless required by applicable law or agreed to in writing, software
15 * distributed under the License is distributed on an "AS IS" BASIS,
16 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17 * See the License for the specific language governing permissions and
18 * limitations under the License.
22 * \brief SPIR-V Assembly Tests for Instructions (special opcode/operand)
23 *//*--------------------------------------------------------------------*/
25 #include "vktSpvAsmInstructionTests.hpp"
27 #include "tcuCommandLine.hpp"
28 #include "tcuFormatUtil.hpp"
29 #include "tcuFloat.hpp"
30 #include "tcuRGBA.hpp"
31 #include "tcuStringTemplate.hpp"
32 #include "tcuTestLog.hpp"
33 #include "tcuVectorUtil.hpp"
34 #include "tcuInterval.hpp"
37 #include "vkDeviceUtil.hpp"
38 #include "vkMemUtil.hpp"
39 #include "vkPlatform.hpp"
40 #include "vkPrograms.hpp"
41 #include "vkQueryUtil.hpp"
43 #include "vkRefUtil.hpp"
44 #include "vkStrUtil.hpp"
45 #include "vkTypeUtil.hpp"
47 #include "deStringUtil.hpp"
48 #include "deUniquePtr.hpp"
50 #include "tcuStringTemplate.hpp"
52 #include "vktSpvAsm8bitStorageTests.hpp"
53 #include "vktSpvAsm16bitStorageTests.hpp"
54 #include "vktSpvAsmUboMatrixPaddingTests.hpp"
55 #include "vktSpvAsmConditionalBranchTests.hpp"
56 #include "vktSpvAsmIndexingTests.hpp"
57 #include "vktSpvAsmImageSamplerTests.hpp"
58 #include "vktSpvAsmComputeShaderCase.hpp"
59 #include "vktSpvAsmComputeShaderTestUtil.hpp"
60 #include "vktSpvAsmGraphicsShaderTestUtil.hpp"
61 #include "vktSpvAsmVariablePointersTests.hpp"
62 #include "vktSpvAsmSpirvVersionTests.hpp"
63 #include "vktTestCaseUtil.hpp"
64 #include "vktSpvAsmLoopDepLenTests.hpp"
65 #include "vktSpvAsmLoopDepInfTests.hpp"
77 namespace SpirVAssembly
91 using tcu::TestStatus;
94 using tcu::StringTemplate;
98 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0)
100 T* const typedPtr = (T*)dst;
101 for (int ndx = 0; ndx < numValues; ndx++)
102 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue);
105 // Filter is a function that returns true if a value should pass, false otherwise.
106 template<typename T, typename FilterT>
107 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, FilterT filter, int offset = 0)
109 T* const typedPtr = (T*)dst;
111 for (int ndx = 0; ndx < numValues; ndx++)
114 value = randomScalar<T>(rnd, minValue, maxValue);
115 while (!filter(value));
117 typedPtr[offset + ndx] = value;
121 // Gets a 64-bit integer with a more logarithmic distribution
122 deInt64 randomInt64LogDistributed (de::Random& rnd)
124 deInt64 val = rnd.getUint64();
125 val &= (1ull << rnd.getInt(1, 63)) - 1;
131 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues)
133 for (int ndx = 0; ndx < numValues; ndx++)
134 dst[ndx] = randomInt64LogDistributed(rnd);
137 template<typename FilterT>
138 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues, FilterT filter)
140 for (int ndx = 0; ndx < numValues; ndx++)
144 value = randomInt64LogDistributed(rnd);
145 } while (!filter(value));
150 inline bool filterNonNegative (const deInt64 value)
155 inline bool filterPositive (const deInt64 value)
160 inline bool filterNotZero (const deInt64 value)
165 static void floorAll (vector<float>& values)
167 for (size_t i = 0; i < values.size(); i++)
168 values[i] = deFloatFloor(values[i]);
171 static void floorAll (vector<Vec4>& values)
173 for (size_t i = 0; i < values.size(); i++)
174 values[i] = floor(values[i]);
182 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {}
185 // Assembly code used for testing LocalSize, OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code:
189 // layout(std140, set = 0, binding = 0) readonly buffer Input {
192 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
196 // layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
199 // uint x = gl_GlobalInvocationID.x;
200 // output_data.elements[x] = -input_data.elements[x];
203 static string getAsmForLocalSizeTest(bool useLiteralLocalSize, bool useSpecConstantWorkgroupSize, IVec3 workGroupSize, deUint32 ndx)
205 std::ostringstream out;
206 out << getComputeAsmShaderPreambleWithoutLocalSize();
208 if (useLiteralLocalSize)
210 out << "OpExecutionMode %main LocalSize "
211 << workGroupSize.x() << " " << workGroupSize.y() << " " << workGroupSize.z() << "\n";
214 out << "OpSource GLSL 430\n"
215 "OpName %main \"main\"\n"
216 "OpName %id \"gl_GlobalInvocationID\"\n"
217 "OpDecorate %id BuiltIn GlobalInvocationId\n";
219 if (useSpecConstantWorkgroupSize)
221 out << "OpDecorate %spec_0 SpecId 100\n"
222 << "OpDecorate %spec_1 SpecId 101\n"
223 << "OpDecorate %spec_2 SpecId 102\n"
224 << "OpDecorate %gl_WorkGroupSize BuiltIn WorkgroupSize\n";
227 out << getComputeAsmInputOutputBufferTraits()
228 << getComputeAsmCommonTypes()
229 << getComputeAsmInputOutputBuffer()
230 << "%id = OpVariable %uvec3ptr Input\n"
231 << "%zero = OpConstant %i32 0 \n";
233 if (useSpecConstantWorkgroupSize)
235 out << "%spec_0 = OpSpecConstant %u32 "<< workGroupSize.x() << "\n"
236 << "%spec_1 = OpSpecConstant %u32 "<< workGroupSize.y() << "\n"
237 << "%spec_2 = OpSpecConstant %u32 "<< workGroupSize.z() << "\n"
238 << "%gl_WorkGroupSize = OpSpecConstantComposite %uvec3 %spec_0 %spec_1 %spec_2\n";
241 out << "%main = OpFunction %void None %voidf\n"
242 << "%label = OpLabel\n"
243 << "%idval = OpLoad %uvec3 %id\n"
244 << "%ndx = OpCompositeExtract %u32 %idval " << ndx << "\n"
246 "%inloc = OpAccessChain %f32ptr %indata %zero %ndx\n"
247 "%inval = OpLoad %f32 %inloc\n"
248 "%neg = OpFNegate %f32 %inval\n"
249 "%outloc = OpAccessChain %f32ptr %outdata %zero %ndx\n"
250 " OpStore %outloc %neg\n"
256 tcu::TestCaseGroup* createLocalSizeGroup (tcu::TestContext& testCtx)
258 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "localsize", ""));
259 ComputeShaderSpec spec;
260 de::Random rnd (deStringHash(group->getName()));
261 const deUint32 numElements = 64u;
262 vector<float> positiveFloats (numElements, 0);
263 vector<float> negativeFloats (numElements, 0);
265 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
267 for (size_t ndx = 0; ndx < numElements; ++ndx)
268 negativeFloats[ndx] = -positiveFloats[ndx];
270 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
271 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
273 spec.numWorkGroups = IVec3(numElements, 1, 1);
275 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, 1, 1), 0u);
276 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize", "", spec));
278 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, 1, 1), 0u);
279 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize", "", spec));
281 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, 1, 1), 0u);
282 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize", "", spec));
284 spec.numWorkGroups = IVec3(1, 1, 1);
286 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(numElements, 1, 1), 0u);
287 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_x", "", spec));
289 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(numElements, 1, 1), 0u);
290 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_x", "", spec));
292 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(numElements, 1, 1), 0u);
293 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_x", "", spec));
295 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, numElements, 1), 1u);
296 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_y", "", spec));
298 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, numElements, 1), 1u);
299 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_y", "", spec));
301 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, numElements, 1), 1u);
302 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_y", "", spec));
304 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, 1, numElements), 2u);
305 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_z", "", spec));
307 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, 1, numElements), 2u);
308 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_z", "", spec));
310 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, 1, numElements), 2u);
311 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_z", "", spec));
313 return group.release();
316 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
318 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
319 ComputeShaderSpec spec;
320 de::Random rnd (deStringHash(group->getName()));
321 const int numElements = 100;
322 vector<float> positiveFloats (numElements, 0);
323 vector<float> negativeFloats (numElements, 0);
325 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
327 for (size_t ndx = 0; ndx < numElements; ++ndx)
328 negativeFloats[ndx] = -positiveFloats[ndx];
331 string(getComputeAsmShaderPreamble()) +
333 "OpSource GLSL 430\n"
334 "OpName %main \"main\"\n"
335 "OpName %id \"gl_GlobalInvocationID\"\n"
337 "OpDecorate %id BuiltIn GlobalInvocationId\n"
339 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes())
341 + string(getComputeAsmInputOutputBuffer()) +
343 "%id = OpVariable %uvec3ptr Input\n"
344 "%zero = OpConstant %i32 0\n"
346 "%main = OpFunction %void None %voidf\n"
348 "%idval = OpLoad %uvec3 %id\n"
349 "%x = OpCompositeExtract %u32 %idval 0\n"
351 " OpNop\n" // Inside a function body
353 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
354 "%inval = OpLoad %f32 %inloc\n"
355 "%neg = OpFNegate %f32 %inval\n"
356 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
357 " OpStore %outloc %neg\n"
360 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
361 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
362 spec.numWorkGroups = IVec3(numElements, 1, 1);
364 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
366 return group.release();
369 bool compareFUnord (const std::vector<BufferSp>& inputs, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
371 if (outputAllocs.size() != 1)
374 vector<deUint8> input1Bytes;
375 vector<deUint8> input2Bytes;
376 vector<deUint8> expectedBytes;
378 inputs[0]->getBytes(input1Bytes);
379 inputs[1]->getBytes(input2Bytes);
380 expectedOutputs[0]->getBytes(expectedBytes);
382 const deInt32* const expectedOutputAsInt = reinterpret_cast<const deInt32* const>(&expectedBytes.front());
383 const deInt32* const outputAsInt = static_cast<const deInt32* const>(outputAllocs[0]->getHostPtr());
384 const float* const input1AsFloat = reinterpret_cast<const float* const>(&input1Bytes.front());
385 const float* const input2AsFloat = reinterpret_cast<const float* const>(&input2Bytes.front());
386 bool returnValue = true;
388 for (size_t idx = 0; idx < expectedBytes.size() / sizeof(deInt32); ++idx)
390 if (outputAsInt[idx] != expectedOutputAsInt[idx])
392 log << TestLog::Message << "ERROR: Sub-case failed. inputs: " << input1AsFloat[idx] << "," << input2AsFloat[idx] << " output: " << outputAsInt[idx]<< " expected output: " << expectedOutputAsInt[idx] << TestLog::EndMessage;
399 typedef VkBool32 (*compareFuncType) (float, float);
405 compareFuncType compareFunc;
407 OpFUnordCase (const char* _name, const char* _opCode, compareFuncType _compareFunc)
410 , compareFunc (_compareFunc) {}
413 #define ADD_OPFUNORD_CASE(NAME, OPCODE, OPERATOR) \
415 struct compare_##NAME { static VkBool32 compare(float x, float y) { return (x OPERATOR y) ? VK_TRUE : VK_FALSE; } }; \
416 cases.push_back(OpFUnordCase(#NAME, OPCODE, compare_##NAME::compare)); \
417 } while (deGetFalse())
419 tcu::TestCaseGroup* createOpFUnordGroup (tcu::TestContext& testCtx)
421 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfunord", "Test the OpFUnord* opcodes"));
422 de::Random rnd (deStringHash(group->getName()));
423 const int numElements = 100;
424 vector<OpFUnordCase> cases;
426 const StringTemplate shaderTemplate (
428 string(getComputeAsmShaderPreamble()) +
430 "OpSource GLSL 430\n"
431 "OpName %main \"main\"\n"
432 "OpName %id \"gl_GlobalInvocationID\"\n"
434 "OpDecorate %id BuiltIn GlobalInvocationId\n"
436 "OpDecorate %buf BufferBlock\n"
437 "OpDecorate %buf2 BufferBlock\n"
438 "OpDecorate %indata1 DescriptorSet 0\n"
439 "OpDecorate %indata1 Binding 0\n"
440 "OpDecorate %indata2 DescriptorSet 0\n"
441 "OpDecorate %indata2 Binding 1\n"
442 "OpDecorate %outdata DescriptorSet 0\n"
443 "OpDecorate %outdata Binding 2\n"
444 "OpDecorate %f32arr ArrayStride 4\n"
445 "OpDecorate %i32arr ArrayStride 4\n"
446 "OpMemberDecorate %buf 0 Offset 0\n"
447 "OpMemberDecorate %buf2 0 Offset 0\n"
449 + string(getComputeAsmCommonTypes()) +
451 "%buf = OpTypeStruct %f32arr\n"
452 "%bufptr = OpTypePointer Uniform %buf\n"
453 "%indata1 = OpVariable %bufptr Uniform\n"
454 "%indata2 = OpVariable %bufptr Uniform\n"
456 "%buf2 = OpTypeStruct %i32arr\n"
457 "%buf2ptr = OpTypePointer Uniform %buf2\n"
458 "%outdata = OpVariable %buf2ptr Uniform\n"
460 "%id = OpVariable %uvec3ptr Input\n"
461 "%zero = OpConstant %i32 0\n"
462 "%consti1 = OpConstant %i32 1\n"
463 "%constf1 = OpConstant %f32 1.0\n"
465 "%main = OpFunction %void None %voidf\n"
467 "%idval = OpLoad %uvec3 %id\n"
468 "%x = OpCompositeExtract %u32 %idval 0\n"
470 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
471 "%inval1 = OpLoad %f32 %inloc1\n"
472 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
473 "%inval2 = OpLoad %f32 %inloc2\n"
474 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
476 "%result = ${OPCODE} %bool %inval1 %inval2\n"
477 "%int_res = OpSelect %i32 %result %consti1 %zero\n"
478 " OpStore %outloc %int_res\n"
483 ADD_OPFUNORD_CASE(equal, "OpFUnordEqual", ==);
484 ADD_OPFUNORD_CASE(less, "OpFUnordLessThan", <);
485 ADD_OPFUNORD_CASE(lessequal, "OpFUnordLessThanEqual", <=);
486 ADD_OPFUNORD_CASE(greater, "OpFUnordGreaterThan", >);
487 ADD_OPFUNORD_CASE(greaterequal, "OpFUnordGreaterThanEqual", >=);
488 ADD_OPFUNORD_CASE(notequal, "OpFUnordNotEqual", !=);
490 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
492 map<string, string> specializations;
493 ComputeShaderSpec spec;
494 const float NaN = std::numeric_limits<float>::quiet_NaN();
495 vector<float> inputFloats1 (numElements, 0);
496 vector<float> inputFloats2 (numElements, 0);
497 vector<deInt32> expectedInts (numElements, 0);
499 specializations["OPCODE"] = cases[caseNdx].opCode;
500 spec.assembly = shaderTemplate.specialize(specializations);
502 fillRandomScalars(rnd, 1.f, 100.f, &inputFloats1[0], numElements);
503 for (size_t ndx = 0; ndx < numElements; ++ndx)
507 case 0: inputFloats2[ndx] = inputFloats1[ndx] + 1.0f; break;
508 case 1: inputFloats2[ndx] = inputFloats1[ndx] - 1.0f; break;
509 case 2: inputFloats2[ndx] = inputFloats1[ndx]; break;
510 case 3: inputFloats2[ndx] = NaN; break;
511 case 4: inputFloats2[ndx] = inputFloats1[ndx]; inputFloats1[ndx] = NaN; break;
512 case 5: inputFloats2[ndx] = NaN; inputFloats1[ndx] = NaN; break;
514 expectedInts[ndx] = tcu::Float32(inputFloats1[ndx]).isNaN() || tcu::Float32(inputFloats2[ndx]).isNaN() || cases[caseNdx].compareFunc(inputFloats1[ndx], inputFloats2[ndx]);
517 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
518 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
519 spec.outputs.push_back(BufferSp(new Int32Buffer(expectedInts)));
520 spec.numWorkGroups = IVec3(numElements, 1, 1);
521 spec.verifyIO = &compareFUnord;
522 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
525 return group.release();
531 const char* assembly;
532 OpAtomicType opAtomic;
533 deInt32 numOutputElements;
535 OpAtomicCase (const char* _name, const char* _assembly, OpAtomicType _opAtomic, deInt32 _numOutputElements)
537 , assembly (_assembly)
538 , opAtomic (_opAtomic)
539 , numOutputElements (_numOutputElements) {}
542 tcu::TestCaseGroup* createOpAtomicGroup (tcu::TestContext& testCtx, bool useStorageBuffer)
544 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx,
545 useStorageBuffer ? "opatomic_storage_buffer" : "opatomic",
546 "Test the OpAtomic* opcodes"));
547 const int numElements = 65535;
548 vector<OpAtomicCase> cases;
550 const StringTemplate shaderTemplate (
552 string("OpCapability Shader\n") +
553 (useStorageBuffer ? "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" : "") +
554 "OpMemoryModel Logical GLSL450\n"
555 "OpEntryPoint GLCompute %main \"main\" %id\n"
556 "OpExecutionMode %main LocalSize 1 1 1\n" +
558 "OpSource GLSL 430\n"
559 "OpName %main \"main\"\n"
560 "OpName %id \"gl_GlobalInvocationID\"\n"
562 "OpDecorate %id BuiltIn GlobalInvocationId\n"
564 "OpDecorate %buf ${BLOCK_DECORATION}\n"
565 "OpDecorate %indata DescriptorSet 0\n"
566 "OpDecorate %indata Binding 0\n"
567 "OpDecorate %i32arr ArrayStride 4\n"
568 "OpMemberDecorate %buf 0 Offset 0\n"
570 "OpDecorate %sumbuf ${BLOCK_DECORATION}\n"
571 "OpDecorate %sum DescriptorSet 0\n"
572 "OpDecorate %sum Binding 1\n"
573 "OpMemberDecorate %sumbuf 0 Coherent\n"
574 "OpMemberDecorate %sumbuf 0 Offset 0\n"
576 + getComputeAsmCommonTypes("${BLOCK_POINTER_TYPE}") +
578 "%buf = OpTypeStruct %i32arr\n"
579 "%bufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %buf\n"
580 "%indata = OpVariable %bufptr ${BLOCK_POINTER_TYPE}\n"
582 "%sumbuf = OpTypeStruct %i32arr\n"
583 "%sumbufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %sumbuf\n"
584 "%sum = OpVariable %sumbufptr ${BLOCK_POINTER_TYPE}\n"
586 "%id = OpVariable %uvec3ptr Input\n"
587 "%minusone = OpConstant %i32 -1\n"
588 "%zero = OpConstant %i32 0\n"
589 "%one = OpConstant %u32 1\n"
590 "%two = OpConstant %i32 2\n"
592 "%main = OpFunction %void None %voidf\n"
594 "%idval = OpLoad %uvec3 %id\n"
595 "%x = OpCompositeExtract %u32 %idval 0\n"
597 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
598 "%inval = OpLoad %i32 %inloc\n"
600 "%outloc = OpAccessChain %i32ptr %sum %zero ${INDEX}\n"
606 #define ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, NUM_OUTPUT_ELEMENTS) \
608 DE_STATIC_ASSERT((NUM_OUTPUT_ELEMENTS) == 1 || (NUM_OUTPUT_ELEMENTS) == numElements); \
609 cases.push_back(OpAtomicCase(#NAME, ASSEMBLY, OPATOMIC, NUM_OUTPUT_ELEMENTS)); \
610 } while (deGetFalse())
611 #define ADD_OPATOMIC_CASE_1(NAME, ASSEMBLY, OPATOMIC) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, 1)
612 #define ADD_OPATOMIC_CASE_N(NAME, ASSEMBLY, OPATOMIC) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, numElements)
614 ADD_OPATOMIC_CASE_1(iadd, "%unused = OpAtomicIAdd %i32 %outloc %one %zero %inval\n", OPATOMIC_IADD );
615 ADD_OPATOMIC_CASE_1(isub, "%unused = OpAtomicISub %i32 %outloc %one %zero %inval\n", OPATOMIC_ISUB );
616 ADD_OPATOMIC_CASE_1(iinc, "%unused = OpAtomicIIncrement %i32 %outloc %one %zero\n", OPATOMIC_IINC );
617 ADD_OPATOMIC_CASE_1(idec, "%unused = OpAtomicIDecrement %i32 %outloc %one %zero\n", OPATOMIC_IDEC );
618 ADD_OPATOMIC_CASE_N(load, "%inval2 = OpAtomicLoad %i32 %inloc %zero %zero\n"
619 " OpStore %outloc %inval2\n", OPATOMIC_LOAD );
620 ADD_OPATOMIC_CASE_N(store, " OpAtomicStore %outloc %zero %zero %inval\n", OPATOMIC_STORE );
621 ADD_OPATOMIC_CASE_N(compex, "%even = OpSMod %i32 %inval %two\n"
622 " OpStore %outloc %even\n"
623 "%unused = OpAtomicCompareExchange %i32 %outloc %one %zero %zero %minusone %zero\n", OPATOMIC_COMPEX );
625 #undef ADD_OPATOMIC_CASE
626 #undef ADD_OPATOMIC_CASE_1
627 #undef ADD_OPATOMIC_CASE_N
629 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
631 map<string, string> specializations;
632 ComputeShaderSpec spec;
633 vector<deInt32> inputInts (numElements, 0);
634 vector<deInt32> expected (cases[caseNdx].numOutputElements, -1);
636 specializations["INDEX"] = (cases[caseNdx].numOutputElements == 1) ? "%zero" : "%x";
637 specializations["INSTRUCTION"] = cases[caseNdx].assembly;
638 specializations["BLOCK_DECORATION"] = useStorageBuffer ? "Block" : "BufferBlock";
639 specializations["BLOCK_POINTER_TYPE"] = useStorageBuffer ? "StorageBuffer" : "Uniform";
640 spec.assembly = shaderTemplate.specialize(specializations);
642 if (useStorageBuffer)
643 spec.extensions.push_back("VK_KHR_storage_buffer_storage_class");
645 spec.inputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_INPUT)));
646 spec.outputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_EXPECTED)));
647 spec.numWorkGroups = IVec3(numElements, 1, 1);
648 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
651 return group.release();
654 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
656 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
657 ComputeShaderSpec spec;
658 de::Random rnd (deStringHash(group->getName()));
659 const int numElements = 100;
660 vector<float> positiveFloats (numElements, 0);
661 vector<float> negativeFloats (numElements, 0);
663 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
665 for (size_t ndx = 0; ndx < numElements; ++ndx)
666 negativeFloats[ndx] = -positiveFloats[ndx];
669 string(getComputeAsmShaderPreamble()) +
671 "%fname1 = OpString \"negateInputs.comp\"\n"
672 "%fname2 = OpString \"negateInputs\"\n"
674 "OpSource GLSL 430\n"
675 "OpName %main \"main\"\n"
676 "OpName %id \"gl_GlobalInvocationID\"\n"
678 "OpDecorate %id BuiltIn GlobalInvocationId\n"
680 + string(getComputeAsmInputOutputBufferTraits()) +
682 "OpLine %fname1 0 0\n" // At the earliest possible position
684 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
686 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
687 "OpLine %fname2 1 0\n" // Different filenames
688 "OpLine %fname1 1000 100000\n"
690 "%id = OpVariable %uvec3ptr Input\n"
691 "%zero = OpConstant %i32 0\n"
693 "OpLine %fname1 1 1\n" // Before a function
695 "%main = OpFunction %void None %voidf\n"
698 "OpLine %fname1 1 1\n" // In a function
700 "%idval = OpLoad %uvec3 %id\n"
701 "%x = OpCompositeExtract %u32 %idval 0\n"
702 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
703 "%inval = OpLoad %f32 %inloc\n"
704 "%neg = OpFNegate %f32 %inval\n"
705 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
706 " OpStore %outloc %neg\n"
709 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
710 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
711 spec.numWorkGroups = IVec3(numElements, 1, 1);
713 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
715 return group.release();
718 bool veryfiBinaryShader (const ProgramBinary& binary)
720 const size_t paternCount = 3u;
721 bool paternsCheck[paternCount] =
725 const string patersns[paternCount] =
731 size_t paternNdx = 0u;
733 for (size_t ndx = 0u; ndx < binary.getSize(); ++ndx)
735 if (false == paternsCheck[paternNdx] &&
736 patersns[paternNdx][0] == static_cast<char>(binary.getBinary()[ndx]) &&
737 deMemoryEqual((const char*)&binary.getBinary()[ndx], &patersns[paternNdx][0], patersns[paternNdx].length()))
739 paternsCheck[paternNdx]= true;
741 if (paternNdx == paternCount)
746 for (size_t ndx = 0u; ndx < paternCount; ++ndx)
748 if (!paternsCheck[ndx])
755 tcu::TestCaseGroup* createOpModuleProcessedGroup (tcu::TestContext& testCtx)
757 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opmoduleprocessed", "Test the OpModuleProcessed instruction"));
758 ComputeShaderSpec spec;
759 de::Random rnd (deStringHash(group->getName()));
760 const int numElements = 10;
761 vector<float> positiveFloats (numElements, 0);
762 vector<float> negativeFloats (numElements, 0);
764 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
766 for (size_t ndx = 0; ndx < numElements; ++ndx)
767 negativeFloats[ndx] = -positiveFloats[ndx];
770 string(getComputeAsmShaderPreamble()) +
771 "%fname = OpString \"negateInputs.comp\"\n"
773 "OpSource GLSL 430\n"
774 "OpName %main \"main\"\n"
775 "OpName %id \"gl_GlobalInvocationID\"\n"
776 "OpModuleProcessed \"VULKAN CTS\"\n" //OpModuleProcessed;
777 "OpModuleProcessed \"Negative values\"\n"
778 "OpModuleProcessed \"Date: 2017/09/21\"\n"
779 "OpDecorate %id BuiltIn GlobalInvocationId\n"
781 + string(getComputeAsmInputOutputBufferTraits())
783 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
785 "OpLine %fname 0 1\n"
787 "OpLine %fname 1000 1\n"
789 "%id = OpVariable %uvec3ptr Input\n"
790 "%zero = OpConstant %i32 0\n"
791 "%main = OpFunction %void None %voidf\n"
794 "%idval = OpLoad %uvec3 %id\n"
795 "%x = OpCompositeExtract %u32 %idval 0\n"
797 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
798 "%inval = OpLoad %f32 %inloc\n"
799 "%neg = OpFNegate %f32 %inval\n"
800 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
801 " OpStore %outloc %neg\n"
804 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
805 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
806 spec.numWorkGroups = IVec3(numElements, 1, 1);
807 spec.verifyBinary = veryfiBinaryShader;
808 spec.spirvVersion = SPIRV_VERSION_1_3;
810 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpModuleProcessed Tests", spec));
812 return group.release();
815 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
817 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
818 ComputeShaderSpec spec;
819 de::Random rnd (deStringHash(group->getName()));
820 const int numElements = 100;
821 vector<float> positiveFloats (numElements, 0);
822 vector<float> negativeFloats (numElements, 0);
824 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
826 for (size_t ndx = 0; ndx < numElements; ++ndx)
827 negativeFloats[ndx] = -positiveFloats[ndx];
830 string(getComputeAsmShaderPreamble()) +
832 "%fname = OpString \"negateInputs.comp\"\n"
834 "OpSource GLSL 430\n"
835 "OpName %main \"main\"\n"
836 "OpName %id \"gl_GlobalInvocationID\"\n"
838 "OpDecorate %id BuiltIn GlobalInvocationId\n"
840 + string(getComputeAsmInputOutputBufferTraits()) +
842 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
844 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
846 "OpLine %fname 0 1\n"
847 "OpNoLine\n" // Immediately following a preceding OpLine
849 "OpLine %fname 1000 1\n"
851 "%id = OpVariable %uvec3ptr Input\n"
852 "%zero = OpConstant %i32 0\n"
854 "OpNoLine\n" // Contents after the previous OpLine
856 "%main = OpFunction %void None %voidf\n"
858 "%idval = OpLoad %uvec3 %id\n"
859 "%x = OpCompositeExtract %u32 %idval 0\n"
861 "OpNoLine\n" // Multiple OpNoLine
865 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
866 "%inval = OpLoad %f32 %inloc\n"
867 "%neg = OpFNegate %f32 %inval\n"
868 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
869 " OpStore %outloc %neg\n"
872 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
873 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
874 spec.numWorkGroups = IVec3(numElements, 1, 1);
876 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
878 return group.release();
881 // Compare instruction for the contraction compute case.
882 // Returns true if the output is what is expected from the test case.
883 bool compareNoContractCase(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
885 if (outputAllocs.size() != 1)
888 // Only size is needed because we are not comparing the exact values.
889 size_t byteSize = expectedOutputs[0]->getByteSize();
891 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
893 for(size_t i = 0; i < byteSize / sizeof(float); ++i) {
894 if (outputAsFloat[i] != 0.f &&
895 outputAsFloat[i] != -ldexp(1, -24)) {
903 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
905 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
906 vector<CaseParameter> cases;
907 const int numElements = 100;
908 vector<float> inputFloats1 (numElements, 0);
909 vector<float> inputFloats2 (numElements, 0);
910 vector<float> outputFloats (numElements, 0);
911 const StringTemplate shaderTemplate (
912 string(getComputeAsmShaderPreamble()) +
914 "OpName %main \"main\"\n"
915 "OpName %id \"gl_GlobalInvocationID\"\n"
917 "OpDecorate %id BuiltIn GlobalInvocationId\n"
921 "OpDecorate %buf BufferBlock\n"
922 "OpDecorate %indata1 DescriptorSet 0\n"
923 "OpDecorate %indata1 Binding 0\n"
924 "OpDecorate %indata2 DescriptorSet 0\n"
925 "OpDecorate %indata2 Binding 1\n"
926 "OpDecorate %outdata DescriptorSet 0\n"
927 "OpDecorate %outdata Binding 2\n"
928 "OpDecorate %f32arr ArrayStride 4\n"
929 "OpMemberDecorate %buf 0 Offset 0\n"
931 + string(getComputeAsmCommonTypes()) +
933 "%buf = OpTypeStruct %f32arr\n"
934 "%bufptr = OpTypePointer Uniform %buf\n"
935 "%indata1 = OpVariable %bufptr Uniform\n"
936 "%indata2 = OpVariable %bufptr Uniform\n"
937 "%outdata = OpVariable %bufptr Uniform\n"
939 "%id = OpVariable %uvec3ptr Input\n"
940 "%zero = OpConstant %i32 0\n"
941 "%c_f_m1 = OpConstant %f32 -1.\n"
943 "%main = OpFunction %void None %voidf\n"
945 "%idval = OpLoad %uvec3 %id\n"
946 "%x = OpCompositeExtract %u32 %idval 0\n"
947 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
948 "%inval1 = OpLoad %f32 %inloc1\n"
949 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
950 "%inval2 = OpLoad %f32 %inloc2\n"
951 "%mul = OpFMul %f32 %inval1 %inval2\n"
952 "%add = OpFAdd %f32 %mul %c_f_m1\n"
953 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
954 " OpStore %outloc %add\n"
958 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
959 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
960 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
962 for (size_t ndx = 0; ndx < numElements; ++ndx)
964 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
965 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
966 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
967 // conducted separately and the result is rounded to 1, or 0x1.fffffcp-1
968 // So the final result will be 0.f or 0x1p-24.
969 // If the operation is combined into a precise fused multiply-add, then the result would be
970 // 2^-46 (0xa8800000).
971 outputFloats[ndx] = 0.f;
974 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
976 map<string, string> specializations;
977 ComputeShaderSpec spec;
979 specializations["DECORATION"] = cases[caseNdx].param;
980 spec.assembly = shaderTemplate.specialize(specializations);
981 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
982 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
983 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
984 spec.numWorkGroups = IVec3(numElements, 1, 1);
985 // Check against the two possible answers based on rounding mode.
986 spec.verifyIO = &compareNoContractCase;
988 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
990 return group.release();
993 bool compareFRem(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
995 if (outputAllocs.size() != 1)
998 vector<deUint8> expectedBytes;
999 expectedOutputs[0]->getBytes(expectedBytes);
1001 const float* expectedOutputAsFloat = reinterpret_cast<const float*>(&expectedBytes.front());
1002 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1004 for (size_t idx = 0; idx < expectedBytes.size() / sizeof(float); ++idx)
1006 const float f0 = expectedOutputAsFloat[idx];
1007 const float f1 = outputAsFloat[idx];
1008 // \todo relative error needs to be fairly high because FRem may be implemented as
1009 // (roughly) frac(a/b)*b, so LSB errors can be magnified. But this should be fine for now.
1010 if (deFloatAbs((f1 - f0) / f0) > 0.02)
1017 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
1019 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
1020 ComputeShaderSpec spec;
1021 de::Random rnd (deStringHash(group->getName()));
1022 const int numElements = 200;
1023 vector<float> inputFloats1 (numElements, 0);
1024 vector<float> inputFloats2 (numElements, 0);
1025 vector<float> outputFloats (numElements, 0);
1027 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1028 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
1030 for (size_t ndx = 0; ndx < numElements; ++ndx)
1032 // Guard against divisors near zero.
1033 if (std::fabs(inputFloats2[ndx]) < 1e-3)
1034 inputFloats2[ndx] = 8.f;
1036 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1037 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
1041 string(getComputeAsmShaderPreamble()) +
1043 "OpName %main \"main\"\n"
1044 "OpName %id \"gl_GlobalInvocationID\"\n"
1046 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1048 "OpDecorate %buf BufferBlock\n"
1049 "OpDecorate %indata1 DescriptorSet 0\n"
1050 "OpDecorate %indata1 Binding 0\n"
1051 "OpDecorate %indata2 DescriptorSet 0\n"
1052 "OpDecorate %indata2 Binding 1\n"
1053 "OpDecorate %outdata DescriptorSet 0\n"
1054 "OpDecorate %outdata Binding 2\n"
1055 "OpDecorate %f32arr ArrayStride 4\n"
1056 "OpMemberDecorate %buf 0 Offset 0\n"
1058 + string(getComputeAsmCommonTypes()) +
1060 "%buf = OpTypeStruct %f32arr\n"
1061 "%bufptr = OpTypePointer Uniform %buf\n"
1062 "%indata1 = OpVariable %bufptr Uniform\n"
1063 "%indata2 = OpVariable %bufptr Uniform\n"
1064 "%outdata = OpVariable %bufptr Uniform\n"
1066 "%id = OpVariable %uvec3ptr Input\n"
1067 "%zero = OpConstant %i32 0\n"
1069 "%main = OpFunction %void None %voidf\n"
1070 "%label = OpLabel\n"
1071 "%idval = OpLoad %uvec3 %id\n"
1072 "%x = OpCompositeExtract %u32 %idval 0\n"
1073 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1074 "%inval1 = OpLoad %f32 %inloc1\n"
1075 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1076 "%inval2 = OpLoad %f32 %inloc2\n"
1077 "%rem = OpFRem %f32 %inval1 %inval2\n"
1078 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1079 " OpStore %outloc %rem\n"
1083 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1084 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1085 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1086 spec.numWorkGroups = IVec3(numElements, 1, 1);
1087 spec.verifyIO = &compareFRem;
1089 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1091 return group.release();
1094 bool compareNMin (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1096 if (outputAllocs.size() != 1)
1099 const BufferSp& expectedOutput (expectedOutputs[0]);
1100 std::vector<deUint8> data;
1101 expectedOutput->getBytes(data);
1103 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1104 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1106 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float); ++idx)
1108 const float f0 = expectedOutputAsFloat[idx];
1109 const float f1 = outputAsFloat[idx];
1111 // For NMin, we accept NaN as output if both inputs were NaN.
1112 // Otherwise the NaN is the wrong choise, as on architectures that
1113 // do not handle NaN, those are huge values.
1114 if (!(tcu::Float32(f1).isNaN() && tcu::Float32(f0).isNaN()) && deFloatAbs(f1 - f0) > 0.00001f)
1121 tcu::TestCaseGroup* createOpNMinGroup (tcu::TestContext& testCtx)
1123 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnmin", "Test the OpNMin instruction"));
1124 ComputeShaderSpec spec;
1125 de::Random rnd (deStringHash(group->getName()));
1126 const int numElements = 200;
1127 vector<float> inputFloats1 (numElements, 0);
1128 vector<float> inputFloats2 (numElements, 0);
1129 vector<float> outputFloats (numElements, 0);
1131 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1132 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1134 // Make the first case a full-NAN case.
1135 inputFloats1[0] = TCU_NAN;
1136 inputFloats2[0] = TCU_NAN;
1138 for (size_t ndx = 0; ndx < numElements; ++ndx)
1140 // By default, pick the smallest
1141 outputFloats[ndx] = std::min(inputFloats1[ndx], inputFloats2[ndx]);
1143 // Make half of the cases NaN cases
1146 // Alternate between the NaN operand
1149 outputFloats[ndx] = inputFloats2[ndx];
1150 inputFloats1[ndx] = TCU_NAN;
1154 outputFloats[ndx] = inputFloats1[ndx];
1155 inputFloats2[ndx] = TCU_NAN;
1161 "OpCapability Shader\n"
1162 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1163 "OpMemoryModel Logical GLSL450\n"
1164 "OpEntryPoint GLCompute %main \"main\" %id\n"
1165 "OpExecutionMode %main LocalSize 1 1 1\n"
1167 "OpName %main \"main\"\n"
1168 "OpName %id \"gl_GlobalInvocationID\"\n"
1170 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1172 "OpDecorate %buf BufferBlock\n"
1173 "OpDecorate %indata1 DescriptorSet 0\n"
1174 "OpDecorate %indata1 Binding 0\n"
1175 "OpDecorate %indata2 DescriptorSet 0\n"
1176 "OpDecorate %indata2 Binding 1\n"
1177 "OpDecorate %outdata DescriptorSet 0\n"
1178 "OpDecorate %outdata Binding 2\n"
1179 "OpDecorate %f32arr ArrayStride 4\n"
1180 "OpMemberDecorate %buf 0 Offset 0\n"
1182 + string(getComputeAsmCommonTypes()) +
1184 "%buf = OpTypeStruct %f32arr\n"
1185 "%bufptr = OpTypePointer Uniform %buf\n"
1186 "%indata1 = OpVariable %bufptr Uniform\n"
1187 "%indata2 = OpVariable %bufptr Uniform\n"
1188 "%outdata = OpVariable %bufptr Uniform\n"
1190 "%id = OpVariable %uvec3ptr Input\n"
1191 "%zero = OpConstant %i32 0\n"
1193 "%main = OpFunction %void None %voidf\n"
1194 "%label = OpLabel\n"
1195 "%idval = OpLoad %uvec3 %id\n"
1196 "%x = OpCompositeExtract %u32 %idval 0\n"
1197 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1198 "%inval1 = OpLoad %f32 %inloc1\n"
1199 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1200 "%inval2 = OpLoad %f32 %inloc2\n"
1201 "%rem = OpExtInst %f32 %std450 NMin %inval1 %inval2\n"
1202 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1203 " OpStore %outloc %rem\n"
1207 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1208 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1209 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1210 spec.numWorkGroups = IVec3(numElements, 1, 1);
1211 spec.verifyIO = &compareNMin;
1213 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1215 return group.release();
1218 bool compareNMax (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1220 if (outputAllocs.size() != 1)
1223 const BufferSp& expectedOutput = expectedOutputs[0];
1224 std::vector<deUint8> data;
1225 expectedOutput->getBytes(data);
1227 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1228 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1230 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float); ++idx)
1232 const float f0 = expectedOutputAsFloat[idx];
1233 const float f1 = outputAsFloat[idx];
1235 // For NMax, NaN is considered acceptable result, since in
1236 // architectures that do not handle NaNs, those are huge values.
1237 if (!tcu::Float32(f1).isNaN() && deFloatAbs(f1 - f0) > 0.00001f)
1244 tcu::TestCaseGroup* createOpNMaxGroup (tcu::TestContext& testCtx)
1246 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, "opnmax", "Test the OpNMax instruction"));
1247 ComputeShaderSpec spec;
1248 de::Random rnd (deStringHash(group->getName()));
1249 const int numElements = 200;
1250 vector<float> inputFloats1 (numElements, 0);
1251 vector<float> inputFloats2 (numElements, 0);
1252 vector<float> outputFloats (numElements, 0);
1254 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1255 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1257 // Make the first case a full-NAN case.
1258 inputFloats1[0] = TCU_NAN;
1259 inputFloats2[0] = TCU_NAN;
1261 for (size_t ndx = 0; ndx < numElements; ++ndx)
1263 // By default, pick the biggest
1264 outputFloats[ndx] = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1266 // Make half of the cases NaN cases
1269 // Alternate between the NaN operand
1272 outputFloats[ndx] = inputFloats2[ndx];
1273 inputFloats1[ndx] = TCU_NAN;
1277 outputFloats[ndx] = inputFloats1[ndx];
1278 inputFloats2[ndx] = TCU_NAN;
1284 "OpCapability Shader\n"
1285 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1286 "OpMemoryModel Logical GLSL450\n"
1287 "OpEntryPoint GLCompute %main \"main\" %id\n"
1288 "OpExecutionMode %main LocalSize 1 1 1\n"
1290 "OpName %main \"main\"\n"
1291 "OpName %id \"gl_GlobalInvocationID\"\n"
1293 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1295 "OpDecorate %buf BufferBlock\n"
1296 "OpDecorate %indata1 DescriptorSet 0\n"
1297 "OpDecorate %indata1 Binding 0\n"
1298 "OpDecorate %indata2 DescriptorSet 0\n"
1299 "OpDecorate %indata2 Binding 1\n"
1300 "OpDecorate %outdata DescriptorSet 0\n"
1301 "OpDecorate %outdata Binding 2\n"
1302 "OpDecorate %f32arr ArrayStride 4\n"
1303 "OpMemberDecorate %buf 0 Offset 0\n"
1305 + string(getComputeAsmCommonTypes()) +
1307 "%buf = OpTypeStruct %f32arr\n"
1308 "%bufptr = OpTypePointer Uniform %buf\n"
1309 "%indata1 = OpVariable %bufptr Uniform\n"
1310 "%indata2 = OpVariable %bufptr Uniform\n"
1311 "%outdata = OpVariable %bufptr Uniform\n"
1313 "%id = OpVariable %uvec3ptr Input\n"
1314 "%zero = OpConstant %i32 0\n"
1316 "%main = OpFunction %void None %voidf\n"
1317 "%label = OpLabel\n"
1318 "%idval = OpLoad %uvec3 %id\n"
1319 "%x = OpCompositeExtract %u32 %idval 0\n"
1320 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1321 "%inval1 = OpLoad %f32 %inloc1\n"
1322 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1323 "%inval2 = OpLoad %f32 %inloc2\n"
1324 "%rem = OpExtInst %f32 %std450 NMax %inval1 %inval2\n"
1325 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1326 " OpStore %outloc %rem\n"
1330 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1331 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1332 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1333 spec.numWorkGroups = IVec3(numElements, 1, 1);
1334 spec.verifyIO = &compareNMax;
1336 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1338 return group.release();
1341 bool compareNClamp (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1343 if (outputAllocs.size() != 1)
1346 const BufferSp& expectedOutput = expectedOutputs[0];
1347 std::vector<deUint8> data;
1348 expectedOutput->getBytes(data);
1350 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1351 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1353 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float) / 2; ++idx)
1355 const float e0 = expectedOutputAsFloat[idx * 2];
1356 const float e1 = expectedOutputAsFloat[idx * 2 + 1];
1357 const float res = outputAsFloat[idx];
1359 // For NClamp, we have two possible outcomes based on
1360 // whether NaNs are handled or not.
1361 // If either min or max value is NaN, the result is undefined,
1362 // so this test doesn't stress those. If the clamped value is
1363 // NaN, and NaNs are handled, the result is min; if NaNs are not
1364 // handled, they are big values that result in max.
1365 // If all three parameters are NaN, the result should be NaN.
1366 if (!((tcu::Float32(e0).isNaN() && tcu::Float32(res).isNaN()) ||
1367 (deFloatAbs(e0 - res) < 0.00001f) ||
1368 (deFloatAbs(e1 - res) < 0.00001f)))
1375 tcu::TestCaseGroup* createOpNClampGroup (tcu::TestContext& testCtx)
1377 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnclamp", "Test the OpNClamp instruction"));
1378 ComputeShaderSpec spec;
1379 de::Random rnd (deStringHash(group->getName()));
1380 const int numElements = 200;
1381 vector<float> inputFloats1 (numElements, 0);
1382 vector<float> inputFloats2 (numElements, 0);
1383 vector<float> inputFloats3 (numElements, 0);
1384 vector<float> outputFloats (numElements * 2, 0);
1386 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1387 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1388 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats3[0], numElements);
1390 for (size_t ndx = 0; ndx < numElements; ++ndx)
1392 // Results are only defined if max value is bigger than min value.
1393 if (inputFloats2[ndx] > inputFloats3[ndx])
1395 float t = inputFloats2[ndx];
1396 inputFloats2[ndx] = inputFloats3[ndx];
1397 inputFloats3[ndx] = t;
1400 // By default, do the clamp, setting both possible answers
1401 float defaultRes = std::min(std::max(inputFloats1[ndx], inputFloats2[ndx]), inputFloats3[ndx]);
1403 float maxResA = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1404 float maxResB = maxResA;
1406 // Alternate between the NaN cases
1409 inputFloats1[ndx] = TCU_NAN;
1410 // If NaN is handled, the result should be same as the clamp minimum.
1411 // If NaN is not handled, the result should clamp to the clamp maximum.
1412 maxResA = inputFloats2[ndx];
1413 maxResB = inputFloats3[ndx];
1417 // Not a NaN case - only one legal result.
1418 maxResA = defaultRes;
1419 maxResB = defaultRes;
1422 outputFloats[ndx * 2] = maxResA;
1423 outputFloats[ndx * 2 + 1] = maxResB;
1426 // Make the first case a full-NAN case.
1427 inputFloats1[0] = TCU_NAN;
1428 inputFloats2[0] = TCU_NAN;
1429 inputFloats3[0] = TCU_NAN;
1430 outputFloats[0] = TCU_NAN;
1431 outputFloats[1] = TCU_NAN;
1434 "OpCapability Shader\n"
1435 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1436 "OpMemoryModel Logical GLSL450\n"
1437 "OpEntryPoint GLCompute %main \"main\" %id\n"
1438 "OpExecutionMode %main LocalSize 1 1 1\n"
1440 "OpName %main \"main\"\n"
1441 "OpName %id \"gl_GlobalInvocationID\"\n"
1443 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1445 "OpDecorate %buf BufferBlock\n"
1446 "OpDecorate %indata1 DescriptorSet 0\n"
1447 "OpDecorate %indata1 Binding 0\n"
1448 "OpDecorate %indata2 DescriptorSet 0\n"
1449 "OpDecorate %indata2 Binding 1\n"
1450 "OpDecorate %indata3 DescriptorSet 0\n"
1451 "OpDecorate %indata3 Binding 2\n"
1452 "OpDecorate %outdata DescriptorSet 0\n"
1453 "OpDecorate %outdata Binding 3\n"
1454 "OpDecorate %f32arr ArrayStride 4\n"
1455 "OpMemberDecorate %buf 0 Offset 0\n"
1457 + string(getComputeAsmCommonTypes()) +
1459 "%buf = OpTypeStruct %f32arr\n"
1460 "%bufptr = OpTypePointer Uniform %buf\n"
1461 "%indata1 = OpVariable %bufptr Uniform\n"
1462 "%indata2 = OpVariable %bufptr Uniform\n"
1463 "%indata3 = OpVariable %bufptr Uniform\n"
1464 "%outdata = OpVariable %bufptr Uniform\n"
1466 "%id = OpVariable %uvec3ptr Input\n"
1467 "%zero = OpConstant %i32 0\n"
1469 "%main = OpFunction %void None %voidf\n"
1470 "%label = OpLabel\n"
1471 "%idval = OpLoad %uvec3 %id\n"
1472 "%x = OpCompositeExtract %u32 %idval 0\n"
1473 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1474 "%inval1 = OpLoad %f32 %inloc1\n"
1475 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1476 "%inval2 = OpLoad %f32 %inloc2\n"
1477 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1478 "%inval3 = OpLoad %f32 %inloc3\n"
1479 "%rem = OpExtInst %f32 %std450 NClamp %inval1 %inval2 %inval3\n"
1480 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1481 " OpStore %outloc %rem\n"
1485 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1486 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1487 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1488 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1489 spec.numWorkGroups = IVec3(numElements, 1, 1);
1490 spec.verifyIO = &compareNClamp;
1492 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1494 return group.release();
1497 tcu::TestCaseGroup* createOpSRemComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1499 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem", "Test the OpSRem instruction"));
1500 de::Random rnd (deStringHash(group->getName()));
1501 const int numElements = 200;
1503 const struct CaseParams
1506 const char* failMessage; // customized status message
1507 qpTestResult failResult; // override status on failure
1508 int op1Min, op1Max; // operand ranges
1512 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1513 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1515 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1517 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1519 const CaseParams& params = cases[caseNdx];
1520 ComputeShaderSpec spec;
1521 vector<deInt32> inputInts1 (numElements, 0);
1522 vector<deInt32> inputInts2 (numElements, 0);
1523 vector<deInt32> outputInts (numElements, 0);
1525 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1526 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1528 for (int ndx = 0; ndx < numElements; ++ndx)
1530 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1531 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1535 string(getComputeAsmShaderPreamble()) +
1537 "OpName %main \"main\"\n"
1538 "OpName %id \"gl_GlobalInvocationID\"\n"
1540 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1542 "OpDecorate %buf BufferBlock\n"
1543 "OpDecorate %indata1 DescriptorSet 0\n"
1544 "OpDecorate %indata1 Binding 0\n"
1545 "OpDecorate %indata2 DescriptorSet 0\n"
1546 "OpDecorate %indata2 Binding 1\n"
1547 "OpDecorate %outdata DescriptorSet 0\n"
1548 "OpDecorate %outdata Binding 2\n"
1549 "OpDecorate %i32arr ArrayStride 4\n"
1550 "OpMemberDecorate %buf 0 Offset 0\n"
1552 + string(getComputeAsmCommonTypes()) +
1554 "%buf = OpTypeStruct %i32arr\n"
1555 "%bufptr = OpTypePointer Uniform %buf\n"
1556 "%indata1 = OpVariable %bufptr Uniform\n"
1557 "%indata2 = OpVariable %bufptr Uniform\n"
1558 "%outdata = OpVariable %bufptr Uniform\n"
1560 "%id = OpVariable %uvec3ptr Input\n"
1561 "%zero = OpConstant %i32 0\n"
1563 "%main = OpFunction %void None %voidf\n"
1564 "%label = OpLabel\n"
1565 "%idval = OpLoad %uvec3 %id\n"
1566 "%x = OpCompositeExtract %u32 %idval 0\n"
1567 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1568 "%inval1 = OpLoad %i32 %inloc1\n"
1569 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1570 "%inval2 = OpLoad %i32 %inloc2\n"
1571 "%rem = OpSRem %i32 %inval1 %inval2\n"
1572 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1573 " OpStore %outloc %rem\n"
1577 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1578 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1579 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1580 spec.numWorkGroups = IVec3(numElements, 1, 1);
1581 spec.failResult = params.failResult;
1582 spec.failMessage = params.failMessage;
1584 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1587 return group.release();
1590 tcu::TestCaseGroup* createOpSRemComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1592 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem64", "Test the 64-bit OpSRem instruction"));
1593 de::Random rnd (deStringHash(group->getName()));
1594 const int numElements = 200;
1596 const struct CaseParams
1599 const char* failMessage; // customized status message
1600 qpTestResult failResult; // override status on failure
1604 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1605 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1607 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1609 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1611 const CaseParams& params = cases[caseNdx];
1612 ComputeShaderSpec spec;
1613 vector<deInt64> inputInts1 (numElements, 0);
1614 vector<deInt64> inputInts2 (numElements, 0);
1615 vector<deInt64> outputInts (numElements, 0);
1617 if (params.positive)
1619 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1620 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1624 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1625 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1628 for (int ndx = 0; ndx < numElements; ++ndx)
1630 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1631 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1635 "OpCapability Int64\n"
1637 + string(getComputeAsmShaderPreamble()) +
1639 "OpName %main \"main\"\n"
1640 "OpName %id \"gl_GlobalInvocationID\"\n"
1642 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1644 "OpDecorate %buf BufferBlock\n"
1645 "OpDecorate %indata1 DescriptorSet 0\n"
1646 "OpDecorate %indata1 Binding 0\n"
1647 "OpDecorate %indata2 DescriptorSet 0\n"
1648 "OpDecorate %indata2 Binding 1\n"
1649 "OpDecorate %outdata DescriptorSet 0\n"
1650 "OpDecorate %outdata Binding 2\n"
1651 "OpDecorate %i64arr ArrayStride 8\n"
1652 "OpMemberDecorate %buf 0 Offset 0\n"
1654 + string(getComputeAsmCommonTypes())
1655 + string(getComputeAsmCommonInt64Types()) +
1657 "%buf = OpTypeStruct %i64arr\n"
1658 "%bufptr = OpTypePointer Uniform %buf\n"
1659 "%indata1 = OpVariable %bufptr Uniform\n"
1660 "%indata2 = OpVariable %bufptr Uniform\n"
1661 "%outdata = OpVariable %bufptr Uniform\n"
1663 "%id = OpVariable %uvec3ptr Input\n"
1664 "%zero = OpConstant %i64 0\n"
1666 "%main = OpFunction %void None %voidf\n"
1667 "%label = OpLabel\n"
1668 "%idval = OpLoad %uvec3 %id\n"
1669 "%x = OpCompositeExtract %u32 %idval 0\n"
1670 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1671 "%inval1 = OpLoad %i64 %inloc1\n"
1672 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1673 "%inval2 = OpLoad %i64 %inloc2\n"
1674 "%rem = OpSRem %i64 %inval1 %inval2\n"
1675 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1676 " OpStore %outloc %rem\n"
1680 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1681 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1682 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1683 spec.numWorkGroups = IVec3(numElements, 1, 1);
1684 spec.failResult = params.failResult;
1685 spec.failMessage = params.failMessage;
1687 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1690 return group.release();
1693 tcu::TestCaseGroup* createOpSModComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1695 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod", "Test the OpSMod instruction"));
1696 de::Random rnd (deStringHash(group->getName()));
1697 const int numElements = 200;
1699 const struct CaseParams
1702 const char* failMessage; // customized status message
1703 qpTestResult failResult; // override status on failure
1704 int op1Min, op1Max; // operand ranges
1708 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1709 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1711 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1713 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1715 const CaseParams& params = cases[caseNdx];
1717 ComputeShaderSpec spec;
1718 vector<deInt32> inputInts1 (numElements, 0);
1719 vector<deInt32> inputInts2 (numElements, 0);
1720 vector<deInt32> outputInts (numElements, 0);
1722 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1723 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1725 for (int ndx = 0; ndx < numElements; ++ndx)
1727 deInt32 rem = inputInts1[ndx] % inputInts2[ndx];
1730 outputInts[ndx] = 0;
1732 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1734 // They have the same sign
1735 outputInts[ndx] = rem;
1739 // They have opposite sign. The remainder operation takes the
1740 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1741 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1742 // the result has the correct sign and that it is still
1743 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1745 // See also http://mathforum.org/library/drmath/view/52343.html
1746 outputInts[ndx] = rem + inputInts2[ndx];
1751 string(getComputeAsmShaderPreamble()) +
1753 "OpName %main \"main\"\n"
1754 "OpName %id \"gl_GlobalInvocationID\"\n"
1756 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1758 "OpDecorate %buf BufferBlock\n"
1759 "OpDecorate %indata1 DescriptorSet 0\n"
1760 "OpDecorate %indata1 Binding 0\n"
1761 "OpDecorate %indata2 DescriptorSet 0\n"
1762 "OpDecorate %indata2 Binding 1\n"
1763 "OpDecorate %outdata DescriptorSet 0\n"
1764 "OpDecorate %outdata Binding 2\n"
1765 "OpDecorate %i32arr ArrayStride 4\n"
1766 "OpMemberDecorate %buf 0 Offset 0\n"
1768 + string(getComputeAsmCommonTypes()) +
1770 "%buf = OpTypeStruct %i32arr\n"
1771 "%bufptr = OpTypePointer Uniform %buf\n"
1772 "%indata1 = OpVariable %bufptr Uniform\n"
1773 "%indata2 = OpVariable %bufptr Uniform\n"
1774 "%outdata = OpVariable %bufptr Uniform\n"
1776 "%id = OpVariable %uvec3ptr Input\n"
1777 "%zero = OpConstant %i32 0\n"
1779 "%main = OpFunction %void None %voidf\n"
1780 "%label = OpLabel\n"
1781 "%idval = OpLoad %uvec3 %id\n"
1782 "%x = OpCompositeExtract %u32 %idval 0\n"
1783 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1784 "%inval1 = OpLoad %i32 %inloc1\n"
1785 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1786 "%inval2 = OpLoad %i32 %inloc2\n"
1787 "%rem = OpSMod %i32 %inval1 %inval2\n"
1788 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1789 " OpStore %outloc %rem\n"
1793 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1794 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1795 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1796 spec.numWorkGroups = IVec3(numElements, 1, 1);
1797 spec.failResult = params.failResult;
1798 spec.failMessage = params.failMessage;
1800 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1803 return group.release();
1806 tcu::TestCaseGroup* createOpSModComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1808 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod64", "Test the OpSMod instruction"));
1809 de::Random rnd (deStringHash(group->getName()));
1810 const int numElements = 200;
1812 const struct CaseParams
1815 const char* failMessage; // customized status message
1816 qpTestResult failResult; // override status on failure
1820 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1821 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1823 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1825 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1827 const CaseParams& params = cases[caseNdx];
1829 ComputeShaderSpec spec;
1830 vector<deInt64> inputInts1 (numElements, 0);
1831 vector<deInt64> inputInts2 (numElements, 0);
1832 vector<deInt64> outputInts (numElements, 0);
1835 if (params.positive)
1837 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1838 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1842 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1843 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1846 for (int ndx = 0; ndx < numElements; ++ndx)
1848 deInt64 rem = inputInts1[ndx] % inputInts2[ndx];
1851 outputInts[ndx] = 0;
1853 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1855 // They have the same sign
1856 outputInts[ndx] = rem;
1860 // They have opposite sign. The remainder operation takes the
1861 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1862 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1863 // the result has the correct sign and that it is still
1864 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1866 // See also http://mathforum.org/library/drmath/view/52343.html
1867 outputInts[ndx] = rem + inputInts2[ndx];
1872 "OpCapability Int64\n"
1874 + string(getComputeAsmShaderPreamble()) +
1876 "OpName %main \"main\"\n"
1877 "OpName %id \"gl_GlobalInvocationID\"\n"
1879 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1881 "OpDecorate %buf BufferBlock\n"
1882 "OpDecorate %indata1 DescriptorSet 0\n"
1883 "OpDecorate %indata1 Binding 0\n"
1884 "OpDecorate %indata2 DescriptorSet 0\n"
1885 "OpDecorate %indata2 Binding 1\n"
1886 "OpDecorate %outdata DescriptorSet 0\n"
1887 "OpDecorate %outdata Binding 2\n"
1888 "OpDecorate %i64arr ArrayStride 8\n"
1889 "OpMemberDecorate %buf 0 Offset 0\n"
1891 + string(getComputeAsmCommonTypes())
1892 + string(getComputeAsmCommonInt64Types()) +
1894 "%buf = OpTypeStruct %i64arr\n"
1895 "%bufptr = OpTypePointer Uniform %buf\n"
1896 "%indata1 = OpVariable %bufptr Uniform\n"
1897 "%indata2 = OpVariable %bufptr Uniform\n"
1898 "%outdata = OpVariable %bufptr Uniform\n"
1900 "%id = OpVariable %uvec3ptr Input\n"
1901 "%zero = OpConstant %i64 0\n"
1903 "%main = OpFunction %void None %voidf\n"
1904 "%label = OpLabel\n"
1905 "%idval = OpLoad %uvec3 %id\n"
1906 "%x = OpCompositeExtract %u32 %idval 0\n"
1907 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1908 "%inval1 = OpLoad %i64 %inloc1\n"
1909 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1910 "%inval2 = OpLoad %i64 %inloc2\n"
1911 "%rem = OpSMod %i64 %inval1 %inval2\n"
1912 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1913 " OpStore %outloc %rem\n"
1917 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1918 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1919 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1920 spec.numWorkGroups = IVec3(numElements, 1, 1);
1921 spec.failResult = params.failResult;
1922 spec.failMessage = params.failMessage;
1924 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1927 return group.release();
1930 // Copy contents in the input buffer to the output buffer.
1931 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
1933 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
1934 de::Random rnd (deStringHash(group->getName()));
1935 const int numElements = 100;
1937 // 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.
1938 ComputeShaderSpec spec1;
1939 vector<Vec4> inputFloats1 (numElements);
1940 vector<Vec4> outputFloats1 (numElements);
1942 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
1944 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1945 floorAll(inputFloats1);
1947 for (size_t ndx = 0; ndx < numElements; ++ndx)
1948 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
1951 string(getComputeAsmShaderPreamble()) +
1953 "OpName %main \"main\"\n"
1954 "OpName %id \"gl_GlobalInvocationID\"\n"
1956 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1957 "OpDecorate %vec4arr ArrayStride 16\n"
1959 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1961 "%vec4 = OpTypeVector %f32 4\n"
1962 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
1963 "%vec4ptr_f = OpTypePointer Function %vec4\n"
1964 "%vec4arr = OpTypeRuntimeArray %vec4\n"
1965 "%buf = OpTypeStruct %vec4arr\n"
1966 "%bufptr = OpTypePointer Uniform %buf\n"
1967 "%indata = OpVariable %bufptr Uniform\n"
1968 "%outdata = OpVariable %bufptr Uniform\n"
1970 "%id = OpVariable %uvec3ptr Input\n"
1971 "%zero = OpConstant %i32 0\n"
1972 "%c_f_0 = OpConstant %f32 0.\n"
1973 "%c_f_0_5 = OpConstant %f32 0.5\n"
1974 "%c_f_1_5 = OpConstant %f32 1.5\n"
1975 "%c_f_2_5 = OpConstant %f32 2.5\n"
1976 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
1978 "%main = OpFunction %void None %voidf\n"
1979 "%label = OpLabel\n"
1980 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
1981 "%idval = OpLoad %uvec3 %id\n"
1982 "%x = OpCompositeExtract %u32 %idval 0\n"
1983 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
1984 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
1985 " OpCopyMemory %v_vec4 %inloc\n"
1986 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
1987 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
1988 " OpStore %outloc %add\n"
1992 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
1993 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
1994 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1996 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
1998 // The following case copies a float[100] variable from the input buffer to the output buffer.
1999 ComputeShaderSpec spec2;
2000 vector<float> inputFloats2 (numElements);
2001 vector<float> outputFloats2 (numElements);
2003 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
2005 for (size_t ndx = 0; ndx < numElements; ++ndx)
2006 outputFloats2[ndx] = inputFloats2[ndx];
2009 string(getComputeAsmShaderPreamble()) +
2011 "OpName %main \"main\"\n"
2012 "OpName %id \"gl_GlobalInvocationID\"\n"
2014 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2015 "OpDecorate %f32arr100 ArrayStride 4\n"
2017 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2019 "%hundred = OpConstant %u32 100\n"
2020 "%f32arr100 = OpTypeArray %f32 %hundred\n"
2021 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
2022 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
2023 "%buf = OpTypeStruct %f32arr100\n"
2024 "%bufptr = OpTypePointer Uniform %buf\n"
2025 "%indata = OpVariable %bufptr Uniform\n"
2026 "%outdata = OpVariable %bufptr Uniform\n"
2028 "%id = OpVariable %uvec3ptr Input\n"
2029 "%zero = OpConstant %i32 0\n"
2031 "%main = OpFunction %void None %voidf\n"
2032 "%label = OpLabel\n"
2033 "%var = OpVariable %f32arr100ptr_f Function\n"
2034 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
2035 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
2036 " OpCopyMemory %var %inarr\n"
2037 " OpCopyMemory %outarr %var\n"
2041 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2042 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2043 spec2.numWorkGroups = IVec3(1, 1, 1);
2045 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
2047 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
2048 ComputeShaderSpec spec3;
2049 vector<float> inputFloats3 (16);
2050 vector<float> outputFloats3 (16);
2052 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
2054 for (size_t ndx = 0; ndx < 16; ++ndx)
2055 outputFloats3[ndx] = inputFloats3[ndx];
2058 string(getComputeAsmShaderPreamble()) +
2060 "OpName %main \"main\"\n"
2061 "OpName %id \"gl_GlobalInvocationID\"\n"
2063 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2064 "OpMemberDecorate %buf 0 Offset 0\n"
2065 "OpMemberDecorate %buf 1 Offset 16\n"
2066 "OpMemberDecorate %buf 2 Offset 32\n"
2067 "OpMemberDecorate %buf 3 Offset 48\n"
2069 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2071 "%vec4 = OpTypeVector %f32 4\n"
2072 "%buf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
2073 "%bufptr = OpTypePointer Uniform %buf\n"
2074 "%indata = OpVariable %bufptr Uniform\n"
2075 "%outdata = OpVariable %bufptr Uniform\n"
2076 "%vec4stptr = OpTypePointer Function %buf\n"
2078 "%id = OpVariable %uvec3ptr Input\n"
2079 "%zero = OpConstant %i32 0\n"
2081 "%main = OpFunction %void None %voidf\n"
2082 "%label = OpLabel\n"
2083 "%var = OpVariable %vec4stptr Function\n"
2084 " OpCopyMemory %var %indata\n"
2085 " OpCopyMemory %outdata %var\n"
2089 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2090 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
2091 spec3.numWorkGroups = IVec3(1, 1, 1);
2093 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
2095 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
2096 ComputeShaderSpec spec4;
2097 vector<float> inputFloats4 (numElements);
2098 vector<float> outputFloats4 (numElements);
2100 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
2102 for (size_t ndx = 0; ndx < numElements; ++ndx)
2103 outputFloats4[ndx] = -inputFloats4[ndx];
2106 string(getComputeAsmShaderPreamble()) +
2108 "OpName %main \"main\"\n"
2109 "OpName %id \"gl_GlobalInvocationID\"\n"
2111 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2113 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2115 "%f32ptr_f = OpTypePointer Function %f32\n"
2116 "%id = OpVariable %uvec3ptr Input\n"
2117 "%zero = OpConstant %i32 0\n"
2119 "%main = OpFunction %void None %voidf\n"
2120 "%label = OpLabel\n"
2121 "%var = OpVariable %f32ptr_f Function\n"
2122 "%idval = OpLoad %uvec3 %id\n"
2123 "%x = OpCompositeExtract %u32 %idval 0\n"
2124 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2125 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2126 " OpCopyMemory %var %inloc\n"
2127 "%val = OpLoad %f32 %var\n"
2128 "%neg = OpFNegate %f32 %val\n"
2129 " OpStore %outloc %neg\n"
2133 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2134 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
2135 spec4.numWorkGroups = IVec3(numElements, 1, 1);
2137 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
2139 return group.release();
2142 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
2144 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
2145 ComputeShaderSpec spec;
2146 de::Random rnd (deStringHash(group->getName()));
2147 const int numElements = 100;
2148 vector<float> inputFloats (numElements, 0);
2149 vector<float> outputFloats (numElements, 0);
2151 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
2153 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2154 floorAll(inputFloats);
2156 for (size_t ndx = 0; ndx < numElements; ++ndx)
2157 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
2160 string(getComputeAsmShaderPreamble()) +
2162 "OpName %main \"main\"\n"
2163 "OpName %id \"gl_GlobalInvocationID\"\n"
2165 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2167 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2169 "%fmat = OpTypeMatrix %fvec3 3\n"
2170 "%three = OpConstant %u32 3\n"
2171 "%farr = OpTypeArray %f32 %three\n"
2172 "%fst = OpTypeStruct %f32 %f32\n"
2174 + string(getComputeAsmInputOutputBuffer()) +
2176 "%id = OpVariable %uvec3ptr Input\n"
2177 "%zero = OpConstant %i32 0\n"
2178 "%c_f = OpConstant %f32 1.5\n"
2179 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
2180 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
2181 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
2182 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
2184 "%main = OpFunction %void None %voidf\n"
2185 "%label = OpLabel\n"
2186 "%c_f_copy = OpCopyObject %f32 %c_f\n"
2187 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
2188 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
2189 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
2190 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
2191 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
2192 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
2193 "%farr_elem = OpCompositeExtract %f32 %c_farr_copy 2\n"
2194 "%fst_elem = OpCompositeExtract %f32 %c_fst_copy 1\n"
2195 // Add up. 1.5 * 5 = 7.5.
2196 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
2197 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
2198 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
2199 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
2201 "%idval = OpLoad %uvec3 %id\n"
2202 "%x = OpCompositeExtract %u32 %idval 0\n"
2203 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2204 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2205 "%inval = OpLoad %f32 %inloc\n"
2206 "%add = OpFAdd %f32 %add4 %inval\n"
2207 " OpStore %outloc %add\n"
2210 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2211 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2212 spec.numWorkGroups = IVec3(numElements, 1, 1);
2214 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
2216 return group.release();
2218 // Assembly code used for testing OpUnreachable is based on GLSL source code:
2222 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2223 // float elements[];
2225 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2226 // float elements[];
2229 // void not_called_func() {
2230 // // place OpUnreachable here
2233 // uint modulo4(uint val) {
2234 // switch (val % uint(4)) {
2235 // case 0: return 3;
2236 // case 1: return 2;
2237 // case 2: return 1;
2238 // case 3: return 0;
2239 // default: return 100; // place OpUnreachable here
2245 // // place OpUnreachable here
2249 // uint x = gl_GlobalInvocationID.x;
2250 // if (const5() > modulo4(1000)) {
2251 // output_data.elements[x] = -input_data.elements[x];
2253 // // place OpUnreachable here
2254 // output_data.elements[x] = input_data.elements[x];
2258 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
2260 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
2261 ComputeShaderSpec spec;
2262 de::Random rnd (deStringHash(group->getName()));
2263 const int numElements = 100;
2264 vector<float> positiveFloats (numElements, 0);
2265 vector<float> negativeFloats (numElements, 0);
2267 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2269 for (size_t ndx = 0; ndx < numElements; ++ndx)
2270 negativeFloats[ndx] = -positiveFloats[ndx];
2273 string(getComputeAsmShaderPreamble()) +
2275 "OpSource GLSL 430\n"
2276 "OpName %main \"main\"\n"
2277 "OpName %func_not_called_func \"not_called_func(\"\n"
2278 "OpName %func_modulo4 \"modulo4(u1;\"\n"
2279 "OpName %func_const5 \"const5(\"\n"
2280 "OpName %id \"gl_GlobalInvocationID\"\n"
2282 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2284 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2286 "%u32ptr = OpTypePointer Function %u32\n"
2287 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
2288 "%unitf = OpTypeFunction %u32\n"
2290 "%id = OpVariable %uvec3ptr Input\n"
2291 "%zero = OpConstant %u32 0\n"
2292 "%one = OpConstant %u32 1\n"
2293 "%two = OpConstant %u32 2\n"
2294 "%three = OpConstant %u32 3\n"
2295 "%four = OpConstant %u32 4\n"
2296 "%five = OpConstant %u32 5\n"
2297 "%hundred = OpConstant %u32 100\n"
2298 "%thousand = OpConstant %u32 1000\n"
2300 + string(getComputeAsmInputOutputBuffer()) +
2303 "%main = OpFunction %void None %voidf\n"
2304 "%main_entry = OpLabel\n"
2305 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
2306 "%idval = OpLoad %uvec3 %id\n"
2307 "%x = OpCompositeExtract %u32 %idval 0\n"
2308 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2309 "%inval = OpLoad %f32 %inloc\n"
2310 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2311 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
2312 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
2313 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
2314 " OpSelectionMerge %if_end None\n"
2315 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2316 "%if_true = OpLabel\n"
2317 "%negate = OpFNegate %f32 %inval\n"
2318 " OpStore %outloc %negate\n"
2319 " OpBranch %if_end\n"
2320 "%if_false = OpLabel\n"
2321 " OpUnreachable\n" // Unreachable else branch for if statement
2322 "%if_end = OpLabel\n"
2326 // not_called_function()
2327 "%func_not_called_func = OpFunction %void None %voidf\n"
2328 "%not_called_func_entry = OpLabel\n"
2329 " OpUnreachable\n" // Unreachable entry block in not called static function
2333 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
2334 "%valptr = OpFunctionParameter %u32ptr\n"
2335 "%modulo4_entry = OpLabel\n"
2336 "%val = OpLoad %u32 %valptr\n"
2337 "%modulo = OpUMod %u32 %val %four\n"
2338 " OpSelectionMerge %switch_merge None\n"
2339 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
2340 "%case0 = OpLabel\n"
2341 " OpReturnValue %three\n"
2342 "%case1 = OpLabel\n"
2343 " OpReturnValue %two\n"
2344 "%case2 = OpLabel\n"
2345 " OpReturnValue %one\n"
2346 "%case3 = OpLabel\n"
2347 " OpReturnValue %zero\n"
2348 "%default = OpLabel\n"
2349 " OpUnreachable\n" // Unreachable default case for switch statement
2350 "%switch_merge = OpLabel\n"
2351 " OpUnreachable\n" // Unreachable merge block for switch statement
2355 "%func_const5 = OpFunction %u32 None %unitf\n"
2356 "%const5_entry = OpLabel\n"
2357 " OpReturnValue %five\n"
2358 "%unreachable = OpLabel\n"
2359 " OpUnreachable\n" // Unreachable block in function
2361 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2362 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2363 spec.numWorkGroups = IVec3(numElements, 1, 1);
2365 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
2367 return group.release();
2370 // Assembly code used for testing decoration group is based on GLSL source code:
2374 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
2375 // float elements[];
2377 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
2378 // float elements[];
2380 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
2381 // float elements[];
2383 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
2384 // float elements[];
2386 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
2387 // float elements[];
2389 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
2390 // float elements[];
2394 // uint x = gl_GlobalInvocationID.x;
2395 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
2397 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
2399 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
2400 ComputeShaderSpec spec;
2401 de::Random rnd (deStringHash(group->getName()));
2402 const int numElements = 100;
2403 vector<float> inputFloats0 (numElements, 0);
2404 vector<float> inputFloats1 (numElements, 0);
2405 vector<float> inputFloats2 (numElements, 0);
2406 vector<float> inputFloats3 (numElements, 0);
2407 vector<float> inputFloats4 (numElements, 0);
2408 vector<float> outputFloats (numElements, 0);
2410 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
2411 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
2412 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
2413 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
2414 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
2416 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2417 floorAll(inputFloats0);
2418 floorAll(inputFloats1);
2419 floorAll(inputFloats2);
2420 floorAll(inputFloats3);
2421 floorAll(inputFloats4);
2423 for (size_t ndx = 0; ndx < numElements; ++ndx)
2424 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
2427 string(getComputeAsmShaderPreamble()) +
2429 "OpSource GLSL 430\n"
2430 "OpName %main \"main\"\n"
2431 "OpName %id \"gl_GlobalInvocationID\"\n"
2433 // Not using group decoration on variable.
2434 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2435 // Not using group decoration on type.
2436 "OpDecorate %f32arr ArrayStride 4\n"
2438 "OpDecorate %groups BufferBlock\n"
2439 "OpDecorate %groupm Offset 0\n"
2440 "%groups = OpDecorationGroup\n"
2441 "%groupm = OpDecorationGroup\n"
2443 // Group decoration on multiple structs.
2444 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
2445 // Group decoration on multiple struct members.
2446 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
2448 "OpDecorate %group1 DescriptorSet 0\n"
2449 "OpDecorate %group3 DescriptorSet 0\n"
2450 "OpDecorate %group3 NonWritable\n"
2451 "OpDecorate %group3 Restrict\n"
2452 "%group0 = OpDecorationGroup\n"
2453 "%group1 = OpDecorationGroup\n"
2454 "%group3 = OpDecorationGroup\n"
2456 // Applying the same decoration group multiple times.
2457 "OpGroupDecorate %group1 %outdata\n"
2458 "OpGroupDecorate %group1 %outdata\n"
2459 "OpGroupDecorate %group1 %outdata\n"
2460 "OpDecorate %outdata DescriptorSet 0\n"
2461 "OpDecorate %outdata Binding 5\n"
2462 // Applying decoration group containing nothing.
2463 "OpGroupDecorate %group0 %indata0\n"
2464 "OpDecorate %indata0 DescriptorSet 0\n"
2465 "OpDecorate %indata0 Binding 0\n"
2466 // Applying decoration group containing one decoration.
2467 "OpGroupDecorate %group1 %indata1\n"
2468 "OpDecorate %indata1 Binding 1\n"
2469 // Applying decoration group containing multiple decorations.
2470 "OpGroupDecorate %group3 %indata2 %indata3\n"
2471 "OpDecorate %indata2 Binding 2\n"
2472 "OpDecorate %indata3 Binding 3\n"
2473 // Applying multiple decoration groups (with overlapping).
2474 "OpGroupDecorate %group0 %indata4\n"
2475 "OpGroupDecorate %group1 %indata4\n"
2476 "OpGroupDecorate %group3 %indata4\n"
2477 "OpDecorate %indata4 Binding 4\n"
2479 + string(getComputeAsmCommonTypes()) +
2481 "%id = OpVariable %uvec3ptr Input\n"
2482 "%zero = OpConstant %i32 0\n"
2484 "%outbuf = OpTypeStruct %f32arr\n"
2485 "%outbufptr = OpTypePointer Uniform %outbuf\n"
2486 "%outdata = OpVariable %outbufptr Uniform\n"
2487 "%inbuf0 = OpTypeStruct %f32arr\n"
2488 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
2489 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
2490 "%inbuf1 = OpTypeStruct %f32arr\n"
2491 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
2492 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
2493 "%inbuf2 = OpTypeStruct %f32arr\n"
2494 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
2495 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
2496 "%inbuf3 = OpTypeStruct %f32arr\n"
2497 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
2498 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
2499 "%inbuf4 = OpTypeStruct %f32arr\n"
2500 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
2501 "%indata4 = OpVariable %inbufptr Uniform\n"
2503 "%main = OpFunction %void None %voidf\n"
2504 "%label = OpLabel\n"
2505 "%idval = OpLoad %uvec3 %id\n"
2506 "%x = OpCompositeExtract %u32 %idval 0\n"
2507 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
2508 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
2509 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
2510 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
2511 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
2512 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2513 "%inval0 = OpLoad %f32 %inloc0\n"
2514 "%inval1 = OpLoad %f32 %inloc1\n"
2515 "%inval2 = OpLoad %f32 %inloc2\n"
2516 "%inval3 = OpLoad %f32 %inloc3\n"
2517 "%inval4 = OpLoad %f32 %inloc4\n"
2518 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
2519 "%add1 = OpFAdd %f32 %add0 %inval2\n"
2520 "%add2 = OpFAdd %f32 %add1 %inval3\n"
2521 "%add = OpFAdd %f32 %add2 %inval4\n"
2522 " OpStore %outloc %add\n"
2525 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
2526 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
2527 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2528 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2529 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2530 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2531 spec.numWorkGroups = IVec3(numElements, 1, 1);
2533 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
2535 return group.release();
2538 struct SpecConstantTwoIntCase
2540 const char* caseName;
2541 const char* scDefinition0;
2542 const char* scDefinition1;
2543 const char* scResultType;
2544 const char* scOperation;
2545 deInt32 scActualValue0;
2546 deInt32 scActualValue1;
2547 const char* resultOperation;
2548 vector<deInt32> expectedOutput;
2550 SpecConstantTwoIntCase (const char* name,
2551 const char* definition0,
2552 const char* definition1,
2553 const char* resultType,
2554 const char* operation,
2557 const char* resultOp,
2558 const vector<deInt32>& output)
2560 , scDefinition0 (definition0)
2561 , scDefinition1 (definition1)
2562 , scResultType (resultType)
2563 , scOperation (operation)
2564 , scActualValue0 (value0)
2565 , scActualValue1 (value1)
2566 , resultOperation (resultOp)
2567 , expectedOutput (output) {}
2570 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
2572 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
2573 vector<SpecConstantTwoIntCase> cases;
2574 de::Random rnd (deStringHash(group->getName()));
2575 const int numElements = 100;
2576 vector<deInt32> inputInts (numElements, 0);
2577 vector<deInt32> outputInts1 (numElements, 0);
2578 vector<deInt32> outputInts2 (numElements, 0);
2579 vector<deInt32> outputInts3 (numElements, 0);
2580 vector<deInt32> outputInts4 (numElements, 0);
2581 const StringTemplate shaderTemplate (
2582 "${CAPABILITIES:opt}"
2583 + string(getComputeAsmShaderPreamble()) +
2585 "OpName %main \"main\"\n"
2586 "OpName %id \"gl_GlobalInvocationID\"\n"
2588 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2589 "OpDecorate %sc_0 SpecId 0\n"
2590 "OpDecorate %sc_1 SpecId 1\n"
2591 "OpDecorate %i32arr ArrayStride 4\n"
2593 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2595 "${OPTYPE_DEFINITIONS:opt}"
2596 "%buf = OpTypeStruct %i32arr\n"
2597 "%bufptr = OpTypePointer Uniform %buf\n"
2598 "%indata = OpVariable %bufptr Uniform\n"
2599 "%outdata = OpVariable %bufptr Uniform\n"
2601 "%id = OpVariable %uvec3ptr Input\n"
2602 "%zero = OpConstant %i32 0\n"
2604 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
2605 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
2606 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
2608 "%main = OpFunction %void None %voidf\n"
2609 "%label = OpLabel\n"
2610 "${TYPE_CONVERT:opt}"
2611 "%idval = OpLoad %uvec3 %id\n"
2612 "%x = OpCompositeExtract %u32 %idval 0\n"
2613 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2614 "%inval = OpLoad %i32 %inloc\n"
2615 "%final = ${GEN_RESULT}\n"
2616 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2617 " OpStore %outloc %final\n"
2619 " OpFunctionEnd\n");
2621 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
2623 for (size_t ndx = 0; ndx < numElements; ++ndx)
2625 outputInts1[ndx] = inputInts[ndx] + 42;
2626 outputInts2[ndx] = inputInts[ndx];
2627 outputInts3[ndx] = inputInts[ndx] - 11200;
2628 outputInts4[ndx] = inputInts[ndx] + 1;
2631 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
2632 const char addSc32ToInput[] = "OpIAdd %i32 %inval %sc_final32";
2633 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
2634 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
2636 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
2637 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
2638 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
2639 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
2640 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
2641 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2642 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2643 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
2644 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
2645 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
2646 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
2647 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2648 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2649 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
2650 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
2651 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
2652 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2653 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
2654 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
2655 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
2656 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2657 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
2658 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
2659 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2660 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2661 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2662 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2663 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
2664 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
2665 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
2666 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
2667 cases.push_back(SpecConstantTwoIntCase("sconvert", " %i32 0", " %i32 0", "%i16", "SConvert %sc_0", -11200, 0, addSc32ToInput, outputInts3));
2668 // -969998336 stored as 32-bit two's complement is the binary representation of -11200 as IEEE-754 Float
2669 cases.push_back(SpecConstantTwoIntCase("fconvert", " %f32 0", " %f32 0", "%f64", "FConvert %sc_0", -969998336, 0, addSc32ToInput, outputInts3));
2671 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2673 map<string, string> specializations;
2674 ComputeShaderSpec spec;
2675 ComputeTestFeatures features = COMPUTE_TEST_USES_NONE;
2677 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
2678 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
2679 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
2680 specializations["SC_OP"] = cases[caseNdx].scOperation;
2681 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
2683 // Special SPIR-V code for SConvert-case
2684 if (strcmp(cases[caseNdx].caseName, "sconvert") == 0)
2686 features = COMPUTE_TEST_USES_INT16;
2687 specializations["CAPABILITIES"] = "OpCapability Int16\n"; // Adds 16-bit integer capability
2688 specializations["OPTYPE_DEFINITIONS"] = "%i16 = OpTypeInt 16 1\n"; // Adds 16-bit integer type
2689 specializations["TYPE_CONVERT"] = "%sc_final32 = OpSConvert %i32 %sc_final\n"; // Converts 16-bit integer to 32-bit integer
2692 // Special SPIR-V code for FConvert-case
2693 if (strcmp(cases[caseNdx].caseName, "fconvert") == 0)
2695 features = COMPUTE_TEST_USES_FLOAT64;
2696 specializations["CAPABILITIES"] = "OpCapability Float64\n"; // Adds 64-bit float capability
2697 specializations["OPTYPE_DEFINITIONS"] = "%f64 = OpTypeFloat 64\n"; // Adds 64-bit float type
2698 specializations["TYPE_CONVERT"] = "%sc_final32 = OpConvertFToS %i32 %sc_final\n"; // Converts 64-bit float to 32-bit integer
2701 spec.assembly = shaderTemplate.specialize(specializations);
2702 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2703 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
2704 spec.numWorkGroups = IVec3(numElements, 1, 1);
2705 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
2706 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
2708 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec, features));
2711 ComputeShaderSpec spec;
2714 string(getComputeAsmShaderPreamble()) +
2716 "OpName %main \"main\"\n"
2717 "OpName %id \"gl_GlobalInvocationID\"\n"
2719 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2720 "OpDecorate %sc_0 SpecId 0\n"
2721 "OpDecorate %sc_1 SpecId 1\n"
2722 "OpDecorate %sc_2 SpecId 2\n"
2723 "OpDecorate %i32arr ArrayStride 4\n"
2725 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2727 "%ivec3 = OpTypeVector %i32 3\n"
2728 "%buf = OpTypeStruct %i32arr\n"
2729 "%bufptr = OpTypePointer Uniform %buf\n"
2730 "%indata = OpVariable %bufptr Uniform\n"
2731 "%outdata = OpVariable %bufptr Uniform\n"
2733 "%id = OpVariable %uvec3ptr Input\n"
2734 "%zero = OpConstant %i32 0\n"
2735 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
2736 "%vec3_undef = OpUndef %ivec3\n"
2738 "%sc_0 = OpSpecConstant %i32 0\n"
2739 "%sc_1 = OpSpecConstant %i32 0\n"
2740 "%sc_2 = OpSpecConstant %i32 0\n"
2741 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
2742 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
2743 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
2744 "%sc_vec3_0_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
2745 "%sc_vec3_1_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
2746 "%sc_vec3_2_s = OpSpecConstantOp %ivec3 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
2747 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
2748 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
2749 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
2750 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
2751 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
2752 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
2753 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
2755 "%main = OpFunction %void None %voidf\n"
2756 "%label = OpLabel\n"
2757 "%idval = OpLoad %uvec3 %id\n"
2758 "%x = OpCompositeExtract %u32 %idval 0\n"
2759 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2760 "%inval = OpLoad %i32 %inloc\n"
2761 "%final = OpIAdd %i32 %inval %sc_final\n"
2762 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2763 " OpStore %outloc %final\n"
2766 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2767 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
2768 spec.numWorkGroups = IVec3(numElements, 1, 1);
2769 spec.specConstants.push_back(123);
2770 spec.specConstants.push_back(56);
2771 spec.specConstants.push_back(-77);
2773 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
2775 return group.release();
2778 void createOpPhiVartypeTests (de::MovePtr<tcu::TestCaseGroup>& group, tcu::TestContext& testCtx)
2780 ComputeShaderSpec specInt;
2781 ComputeShaderSpec specFloat;
2782 ComputeShaderSpec specVec3;
2783 ComputeShaderSpec specMat4;
2784 ComputeShaderSpec specArray;
2785 ComputeShaderSpec specStruct;
2786 de::Random rnd (deStringHash(group->getName()));
2787 const int numElements = 100;
2788 vector<float> inputFloats (numElements, 0);
2789 vector<float> outputFloats (numElements, 0);
2791 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
2793 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2794 floorAll(inputFloats);
2796 for (size_t ndx = 0; ndx < numElements; ++ndx)
2798 // Just check if the value is positive or not
2799 outputFloats[ndx] = (inputFloats[ndx] > 0) ? 1.0f : -1.0f;
2802 // All of the tests are of the form:
2806 // if (inputdata > 0)
2813 specFloat.assembly =
2814 string(getComputeAsmShaderPreamble()) +
2816 "OpSource GLSL 430\n"
2817 "OpName %main \"main\"\n"
2818 "OpName %id \"gl_GlobalInvocationID\"\n"
2820 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2822 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2824 "%id = OpVariable %uvec3ptr Input\n"
2825 "%zero = OpConstant %i32 0\n"
2826 "%float_0 = OpConstant %f32 0.0\n"
2827 "%float_1 = OpConstant %f32 1.0\n"
2828 "%float_n1 = OpConstant %f32 -1.0\n"
2830 "%main = OpFunction %void None %voidf\n"
2831 "%entry = OpLabel\n"
2832 "%idval = OpLoad %uvec3 %id\n"
2833 "%x = OpCompositeExtract %u32 %idval 0\n"
2834 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2835 "%inval = OpLoad %f32 %inloc\n"
2837 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2838 " OpSelectionMerge %cm None\n"
2839 " OpBranchConditional %comp %tb %fb\n"
2845 "%res = OpPhi %f32 %float_1 %tb %float_n1 %fb\n"
2847 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2848 " OpStore %outloc %res\n"
2852 specFloat.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2853 specFloat.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2854 specFloat.numWorkGroups = IVec3(numElements, 1, 1);
2857 string(getComputeAsmShaderPreamble()) +
2859 "OpSource GLSL 430\n"
2860 "OpName %main \"main\"\n"
2861 "OpName %id \"gl_GlobalInvocationID\"\n"
2863 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2865 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2867 "%id = OpVariable %uvec3ptr Input\n"
2868 "%v4f32 = OpTypeVector %f32 4\n"
2869 "%mat4v4f32 = OpTypeMatrix %v4f32 4\n"
2870 "%zero = OpConstant %i32 0\n"
2871 "%float_0 = OpConstant %f32 0.0\n"
2872 "%float_1 = OpConstant %f32 1.0\n"
2873 "%float_n1 = OpConstant %f32 -1.0\n"
2874 "%m11 = OpConstantComposite %v4f32 %float_1 %float_0 %float_0 %float_0\n"
2875 "%m12 = OpConstantComposite %v4f32 %float_0 %float_1 %float_0 %float_0\n"
2876 "%m13 = OpConstantComposite %v4f32 %float_0 %float_0 %float_1 %float_0\n"
2877 "%m14 = OpConstantComposite %v4f32 %float_0 %float_0 %float_0 %float_1\n"
2878 "%m1 = OpConstantComposite %mat4v4f32 %m11 %m12 %m13 %m14\n"
2879 "%m21 = OpConstantComposite %v4f32 %float_n1 %float_0 %float_0 %float_0\n"
2880 "%m22 = OpConstantComposite %v4f32 %float_0 %float_n1 %float_0 %float_0\n"
2881 "%m23 = OpConstantComposite %v4f32 %float_0 %float_0 %float_n1 %float_0\n"
2882 "%m24 = OpConstantComposite %v4f32 %float_0 %float_0 %float_0 %float_n1\n"
2883 "%m2 = OpConstantComposite %mat4v4f32 %m21 %m22 %m23 %m24\n"
2885 "%main = OpFunction %void None %voidf\n"
2886 "%entry = OpLabel\n"
2887 "%idval = OpLoad %uvec3 %id\n"
2888 "%x = OpCompositeExtract %u32 %idval 0\n"
2889 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2890 "%inval = OpLoad %f32 %inloc\n"
2892 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2893 " OpSelectionMerge %cm None\n"
2894 " OpBranchConditional %comp %tb %fb\n"
2900 "%mres = OpPhi %mat4v4f32 %m1 %tb %m2 %fb\n"
2901 "%res = OpCompositeExtract %f32 %mres 2 2\n"
2903 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2904 " OpStore %outloc %res\n"
2908 specMat4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2909 specMat4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2910 specMat4.numWorkGroups = IVec3(numElements, 1, 1);
2913 string(getComputeAsmShaderPreamble()) +
2915 "OpSource GLSL 430\n"
2916 "OpName %main \"main\"\n"
2917 "OpName %id \"gl_GlobalInvocationID\"\n"
2919 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2921 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2923 "%id = OpVariable %uvec3ptr Input\n"
2924 "%zero = OpConstant %i32 0\n"
2925 "%float_0 = OpConstant %f32 0.0\n"
2926 "%float_1 = OpConstant %f32 1.0\n"
2927 "%float_n1 = OpConstant %f32 -1.0\n"
2928 "%v1 = OpConstantComposite %fvec3 %float_1 %float_1 %float_1\n"
2929 "%v2 = OpConstantComposite %fvec3 %float_n1 %float_n1 %float_n1\n"
2931 "%main = OpFunction %void None %voidf\n"
2932 "%entry = OpLabel\n"
2933 "%idval = OpLoad %uvec3 %id\n"
2934 "%x = OpCompositeExtract %u32 %idval 0\n"
2935 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2936 "%inval = OpLoad %f32 %inloc\n"
2938 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2939 " OpSelectionMerge %cm None\n"
2940 " OpBranchConditional %comp %tb %fb\n"
2946 "%vres = OpPhi %fvec3 %v1 %tb %v2 %fb\n"
2947 "%res = OpCompositeExtract %f32 %vres 2\n"
2949 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2950 " OpStore %outloc %res\n"
2954 specVec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2955 specVec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2956 specVec3.numWorkGroups = IVec3(numElements, 1, 1);
2959 string(getComputeAsmShaderPreamble()) +
2961 "OpSource GLSL 430\n"
2962 "OpName %main \"main\"\n"
2963 "OpName %id \"gl_GlobalInvocationID\"\n"
2965 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2967 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2969 "%id = OpVariable %uvec3ptr Input\n"
2970 "%zero = OpConstant %i32 0\n"
2971 "%float_0 = OpConstant %f32 0.0\n"
2972 "%i1 = OpConstant %i32 1\n"
2973 "%i2 = OpConstant %i32 -1\n"
2975 "%main = OpFunction %void None %voidf\n"
2976 "%entry = OpLabel\n"
2977 "%idval = OpLoad %uvec3 %id\n"
2978 "%x = OpCompositeExtract %u32 %idval 0\n"
2979 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2980 "%inval = OpLoad %f32 %inloc\n"
2982 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2983 " OpSelectionMerge %cm None\n"
2984 " OpBranchConditional %comp %tb %fb\n"
2990 "%ires = OpPhi %i32 %i1 %tb %i2 %fb\n"
2991 "%res = OpConvertSToF %f32 %ires\n"
2993 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2994 " OpStore %outloc %res\n"
2998 specInt.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2999 specInt.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3000 specInt.numWorkGroups = IVec3(numElements, 1, 1);
3002 specArray.assembly =
3003 string(getComputeAsmShaderPreamble()) +
3005 "OpSource GLSL 430\n"
3006 "OpName %main \"main\"\n"
3007 "OpName %id \"gl_GlobalInvocationID\"\n"
3009 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3011 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3013 "%id = OpVariable %uvec3ptr Input\n"
3014 "%zero = OpConstant %i32 0\n"
3015 "%u7 = OpConstant %u32 7\n"
3016 "%float_0 = OpConstant %f32 0.0\n"
3017 "%float_1 = OpConstant %f32 1.0\n"
3018 "%float_n1 = OpConstant %f32 -1.0\n"
3019 "%f32a7 = OpTypeArray %f32 %u7\n"
3020 "%a1 = OpConstantComposite %f32a7 %float_1 %float_1 %float_1 %float_1 %float_1 %float_1 %float_1\n"
3021 "%a2 = OpConstantComposite %f32a7 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1\n"
3022 "%main = OpFunction %void None %voidf\n"
3023 "%entry = OpLabel\n"
3024 "%idval = OpLoad %uvec3 %id\n"
3025 "%x = OpCompositeExtract %u32 %idval 0\n"
3026 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3027 "%inval = OpLoad %f32 %inloc\n"
3029 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3030 " OpSelectionMerge %cm None\n"
3031 " OpBranchConditional %comp %tb %fb\n"
3037 "%ares = OpPhi %f32a7 %a1 %tb %a2 %fb\n"
3038 "%res = OpCompositeExtract %f32 %ares 5\n"
3040 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3041 " OpStore %outloc %res\n"
3045 specArray.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3046 specArray.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3047 specArray.numWorkGroups = IVec3(numElements, 1, 1);
3049 specStruct.assembly =
3050 string(getComputeAsmShaderPreamble()) +
3052 "OpSource GLSL 430\n"
3053 "OpName %main \"main\"\n"
3054 "OpName %id \"gl_GlobalInvocationID\"\n"
3056 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3058 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3060 "%id = OpVariable %uvec3ptr Input\n"
3061 "%zero = OpConstant %i32 0\n"
3062 "%float_0 = OpConstant %f32 0.0\n"
3063 "%float_1 = OpConstant %f32 1.0\n"
3064 "%float_n1 = OpConstant %f32 -1.0\n"
3066 "%v2f32 = OpTypeVector %f32 2\n"
3067 "%Data2 = OpTypeStruct %f32 %v2f32\n"
3068 "%Data = OpTypeStruct %Data2 %f32\n"
3070 "%in1a = OpConstantComposite %v2f32 %float_1 %float_1\n"
3071 "%in1b = OpConstantComposite %Data2 %float_1 %in1a\n"
3072 "%s1 = OpConstantComposite %Data %in1b %float_1\n"
3073 "%in2a = OpConstantComposite %v2f32 %float_n1 %float_n1\n"
3074 "%in2b = OpConstantComposite %Data2 %float_n1 %in2a\n"
3075 "%s2 = OpConstantComposite %Data %in2b %float_n1\n"
3077 "%main = OpFunction %void None %voidf\n"
3078 "%entry = OpLabel\n"
3079 "%idval = OpLoad %uvec3 %id\n"
3080 "%x = OpCompositeExtract %u32 %idval 0\n"
3081 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3082 "%inval = OpLoad %f32 %inloc\n"
3084 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3085 " OpSelectionMerge %cm None\n"
3086 " OpBranchConditional %comp %tb %fb\n"
3092 "%sres = OpPhi %Data %s1 %tb %s2 %fb\n"
3093 "%res = OpCompositeExtract %f32 %sres 0 0\n"
3095 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3096 " OpStore %outloc %res\n"
3100 specStruct.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3101 specStruct.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3102 specStruct.numWorkGroups = IVec3(numElements, 1, 1);
3104 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_int", "OpPhi with int variables", specInt));
3105 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_float", "OpPhi with float variables", specFloat));
3106 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_vec3", "OpPhi with vec3 variables", specVec3));
3107 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_mat4", "OpPhi with mat4 variables", specMat4));
3108 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_array", "OpPhi with array variables", specArray));
3109 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_struct", "OpPhi with struct variables", specStruct));
3112 string generateConstantDefinitions (int count)
3114 std::ostringstream r;
3115 for (int i = 0; i < count; i++)
3116 r << "%cf" << (i * 10 + 5) << " = OpConstant %f32 " <<(i * 10 + 5) << ".0\n";
3121 string generateSwitchCases (int count)
3123 std::ostringstream r;
3124 for (int i = 0; i < count; i++)
3125 r << " " << i << " %case" << i;
3130 string generateSwitchTargets (int count)
3132 std::ostringstream r;
3133 for (int i = 0; i < count; i++)
3134 r << "%case" << i << " = OpLabel\n OpBranch %phi\n";
3139 string generateOpPhiParams (int count)
3141 std::ostringstream r;
3142 for (int i = 0; i < count; i++)
3143 r << " %cf" << (i * 10 + 5) << " %case" << i;
3148 string generateIntWidth (int value)
3150 std::ostringstream r;
3155 // Expand input string by injecting "ABC" between the input
3156 // string characters. The acc/add/treshold parameters are used
3157 // to skip some of the injections to make the result less
3158 // uniform (and a lot shorter).
3159 string expandOpPhiCase5 (const string& s, int &acc, int add, int treshold)
3161 std::ostringstream res;
3162 const char* p = s.c_str();
3178 // Calculate expected result based on the code string
3179 float calcOpPhiCase5 (float val, const string& s)
3181 const char* p = s.c_str();
3184 const float tv[8] = { 0.5f, 1.5f, 3.5f, 7.5f, 15.5f, 31.5f, 63.5f, 127.5f };
3185 const float v = deFloatAbs(val);
3190 for (int i = 7; i >= 0; --i)
3191 x[i] = std::fmod((float)v, (float)(2 << i));
3192 for (int i = 7; i >= 0; --i)
3193 b[i] = x[i] > tv[i];
3200 if (skip == 0 && b[depth])
3211 if (b[depth] || skip)
3225 // In the code string, the letters represent the following:
3228 // if (certain bit is set)
3239 // AABCBC leads to if(){r++;if(){r++;}else{}}else{}
3240 // ABABCC leads to if(){r++;}else{if(){r++;}else{}}
3241 // ABCABC leads to if(){r++;}else{}if(){r++;}else{}
3243 // Code generation gets a bit complicated due to the else-branches,
3244 // which do not generate new values. Thus, the generator needs to
3245 // keep track of the previous variable change seen by the else
3247 string generateOpPhiCase5 (const string& s)
3249 std::stack<int> idStack;
3250 std::stack<std::string> value;
3251 std::stack<std::string> valueLabel;
3252 std::stack<std::string> mergeLeft;
3253 std::stack<std::string> mergeRight;
3254 std::ostringstream res;
3255 const char* p = s.c_str();
3261 value.push("%f32_0");
3262 valueLabel.push("%f32_0 %entry");
3270 idStack.push(currId);
3271 res << "\tOpSelectionMerge %m" << currId << " None\n";
3272 res << "\tOpBranchConditional %b" << depth << " %t" << currId << " %f" << currId << "\n";
3273 res << "%t" << currId << " = OpLabel\n";
3274 res << "%rt" << currId << " = OpFAdd %f32 " << value.top() << " %f32_1\n";
3275 std::ostringstream tag;
3276 tag << "%rt" << currId;
3277 value.push(tag.str());
3278 tag << " %t" << currId;
3279 valueLabel.push(tag.str());
3284 mergeLeft.push(valueLabel.top());
3287 res << "\tOpBranch %m" << currId << "\n";
3288 res << "%f" << currId << " = OpLabel\n";
3289 std::ostringstream tag;
3290 tag << value.top() << " %f" << currId;
3292 valueLabel.push(tag.str());
3297 mergeRight.push(valueLabel.top());
3298 res << "\tOpBranch %m" << currId << "\n";
3299 res << "%m" << currId << " = OpLabel\n";
3301 res << "%res"; // last result goes to %res
3303 res << "%rm" << currId;
3304 res << " = OpPhi %f32 " << mergeLeft.top() << " " << mergeRight.top() << "\n";
3305 std::ostringstream tag;
3306 tag << "%rm" << currId;
3308 value.push(tag.str());
3309 tag << " %m" << currId;
3311 valueLabel.push(tag.str());
3316 currId = idStack.top();
3324 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
3326 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
3327 ComputeShaderSpec spec1;
3328 ComputeShaderSpec spec2;
3329 ComputeShaderSpec spec3;
3330 ComputeShaderSpec spec4;
3331 ComputeShaderSpec spec5;
3332 de::Random rnd (deStringHash(group->getName()));
3333 const int numElements = 100;
3334 vector<float> inputFloats (numElements, 0);
3335 vector<float> outputFloats1 (numElements, 0);
3336 vector<float> outputFloats2 (numElements, 0);
3337 vector<float> outputFloats3 (numElements, 0);
3338 vector<float> outputFloats4 (numElements, 0);
3339 vector<float> outputFloats5 (numElements, 0);
3340 std::string codestring = "ABC";
3341 const int test4Width = 1024;
3343 // Build case 5 code string. Each iteration makes the hierarchy more complicated.
3344 // 9 iterations with (7, 24) parameters makes the hierarchy 8 deep with about 1500 lines of
3346 for (int i = 0, acc = 0; i < 9; i++)
3347 codestring = expandOpPhiCase5(codestring, acc, 7, 24);
3349 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
3351 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3352 floorAll(inputFloats);
3354 for (size_t ndx = 0; ndx < numElements; ++ndx)
3358 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
3359 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
3360 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
3363 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
3364 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
3366 int index4 = (int)deFloor(deAbs((float)ndx * inputFloats[ndx]));
3367 outputFloats4[ndx] = (float)(index4 % test4Width) * 10.0f + 5.0f;
3369 outputFloats5[ndx] = calcOpPhiCase5(inputFloats[ndx], codestring);
3373 string(getComputeAsmShaderPreamble()) +
3375 "OpSource GLSL 430\n"
3376 "OpName %main \"main\"\n"
3377 "OpName %id \"gl_GlobalInvocationID\"\n"
3379 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3381 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3383 "%id = OpVariable %uvec3ptr Input\n"
3384 "%zero = OpConstant %i32 0\n"
3385 "%three = OpConstant %u32 3\n"
3386 "%constf5p5 = OpConstant %f32 5.5\n"
3387 "%constf20p5 = OpConstant %f32 20.5\n"
3388 "%constf1p75 = OpConstant %f32 1.75\n"
3389 "%constf8p5 = OpConstant %f32 8.5\n"
3390 "%constf6p5 = OpConstant %f32 6.5\n"
3392 "%main = OpFunction %void None %voidf\n"
3393 "%entry = OpLabel\n"
3394 "%idval = OpLoad %uvec3 %id\n"
3395 "%x = OpCompositeExtract %u32 %idval 0\n"
3396 "%selector = OpUMod %u32 %x %three\n"
3397 " OpSelectionMerge %phi None\n"
3398 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
3400 // Case 1 before OpPhi.
3401 "%case1 = OpLabel\n"
3404 "%default = OpLabel\n"
3408 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
3409 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3410 "%inval = OpLoad %f32 %inloc\n"
3411 "%add = OpFAdd %f32 %inval %operand\n"
3412 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3413 " OpStore %outloc %add\n"
3416 // Case 0 after OpPhi.
3417 "%case0 = OpLabel\n"
3421 // Case 2 after OpPhi.
3422 "%case2 = OpLabel\n"
3426 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3427 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
3428 spec1.numWorkGroups = IVec3(numElements, 1, 1);
3430 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
3433 string(getComputeAsmShaderPreamble()) +
3435 "OpName %main \"main\"\n"
3436 "OpName %id \"gl_GlobalInvocationID\"\n"
3438 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3440 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3442 "%id = OpVariable %uvec3ptr Input\n"
3443 "%zero = OpConstant %i32 0\n"
3444 "%one = OpConstant %i32 1\n"
3445 "%three = OpConstant %i32 3\n"
3446 "%constf6p5 = OpConstant %f32 6.5\n"
3448 "%main = OpFunction %void None %voidf\n"
3449 "%entry = OpLabel\n"
3450 "%idval = OpLoad %uvec3 %id\n"
3451 "%x = OpCompositeExtract %u32 %idval 0\n"
3452 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3453 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3454 "%inval = OpLoad %f32 %inloc\n"
3458 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
3459 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
3460 "%step_next = OpIAdd %i32 %step %one\n"
3461 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
3462 "%still_loop = OpSLessThan %bool %step %three\n"
3463 " OpLoopMerge %exit %phi None\n"
3464 " OpBranchConditional %still_loop %phi %exit\n"
3467 " OpStore %outloc %accum\n"
3470 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3471 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
3472 spec2.numWorkGroups = IVec3(numElements, 1, 1);
3474 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
3477 string(getComputeAsmShaderPreamble()) +
3479 "OpName %main \"main\"\n"
3480 "OpName %id \"gl_GlobalInvocationID\"\n"
3482 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3484 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3486 "%f32ptr_f = OpTypePointer Function %f32\n"
3487 "%id = OpVariable %uvec3ptr Input\n"
3488 "%true = OpConstantTrue %bool\n"
3489 "%false = OpConstantFalse %bool\n"
3490 "%zero = OpConstant %i32 0\n"
3491 "%constf8p5 = OpConstant %f32 8.5\n"
3493 "%main = OpFunction %void None %voidf\n"
3494 "%entry = OpLabel\n"
3495 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
3496 "%idval = OpLoad %uvec3 %id\n"
3497 "%x = OpCompositeExtract %u32 %idval 0\n"
3498 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3499 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3500 "%a_init = OpLoad %f32 %inloc\n"
3501 "%b_init = OpLoad %f32 %b\n"
3505 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
3506 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
3507 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
3508 " OpLoopMerge %exit %phi None\n"
3509 " OpBranchConditional %still_loop %phi %exit\n"
3512 "%sub = OpFSub %f32 %a_next %b_next\n"
3513 " OpStore %outloc %sub\n"
3516 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3517 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
3518 spec3.numWorkGroups = IVec3(numElements, 1, 1);
3520 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
3523 "OpCapability Shader\n"
3524 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
3525 "OpMemoryModel Logical GLSL450\n"
3526 "OpEntryPoint GLCompute %main \"main\" %id\n"
3527 "OpExecutionMode %main LocalSize 1 1 1\n"
3529 "OpSource GLSL 430\n"
3530 "OpName %main \"main\"\n"
3531 "OpName %id \"gl_GlobalInvocationID\"\n"
3533 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3535 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3537 "%id = OpVariable %uvec3ptr Input\n"
3538 "%zero = OpConstant %i32 0\n"
3539 "%cimod = OpConstant %u32 " + generateIntWidth(test4Width) + "\n"
3541 + generateConstantDefinitions(test4Width) +
3543 "%main = OpFunction %void None %voidf\n"
3544 "%entry = OpLabel\n"
3545 "%idval = OpLoad %uvec3 %id\n"
3546 "%x = OpCompositeExtract %u32 %idval 0\n"
3547 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3548 "%inval = OpLoad %f32 %inloc\n"
3549 "%xf = OpConvertUToF %f32 %x\n"
3550 "%xm = OpFMul %f32 %xf %inval\n"
3551 "%xa = OpExtInst %f32 %ext FAbs %xm\n"
3552 "%xi = OpConvertFToU %u32 %xa\n"
3553 "%selector = OpUMod %u32 %xi %cimod\n"
3554 " OpSelectionMerge %phi None\n"
3555 " OpSwitch %selector %default "
3557 + generateSwitchCases(test4Width) +
3559 "%default = OpLabel\n"
3562 + generateSwitchTargets(test4Width) +
3565 "%result = OpPhi %f32"
3567 + generateOpPhiParams(test4Width) +
3569 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3570 " OpStore %outloc %result\n"
3574 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3575 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
3576 spec4.numWorkGroups = IVec3(numElements, 1, 1);
3578 group->addChild(new SpvAsmComputeShaderCase(testCtx, "wide", "OpPhi with a lot of parameters", spec4));
3581 "OpCapability Shader\n"
3582 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
3583 "OpMemoryModel Logical GLSL450\n"
3584 "OpEntryPoint GLCompute %main \"main\" %id\n"
3585 "OpExecutionMode %main LocalSize 1 1 1\n"
3586 "%code = OpString \"" + codestring + "\"\n"
3588 "OpSource GLSL 430\n"
3589 "OpName %main \"main\"\n"
3590 "OpName %id \"gl_GlobalInvocationID\"\n"
3592 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3594 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3596 "%id = OpVariable %uvec3ptr Input\n"
3597 "%zero = OpConstant %i32 0\n"
3598 "%f32_0 = OpConstant %f32 0.0\n"
3599 "%f32_0_5 = OpConstant %f32 0.5\n"
3600 "%f32_1 = OpConstant %f32 1.0\n"
3601 "%f32_1_5 = OpConstant %f32 1.5\n"
3602 "%f32_2 = OpConstant %f32 2.0\n"
3603 "%f32_3_5 = OpConstant %f32 3.5\n"
3604 "%f32_4 = OpConstant %f32 4.0\n"
3605 "%f32_7_5 = OpConstant %f32 7.5\n"
3606 "%f32_8 = OpConstant %f32 8.0\n"
3607 "%f32_15_5 = OpConstant %f32 15.5\n"
3608 "%f32_16 = OpConstant %f32 16.0\n"
3609 "%f32_31_5 = OpConstant %f32 31.5\n"
3610 "%f32_32 = OpConstant %f32 32.0\n"
3611 "%f32_63_5 = OpConstant %f32 63.5\n"
3612 "%f32_64 = OpConstant %f32 64.0\n"
3613 "%f32_127_5 = OpConstant %f32 127.5\n"
3614 "%f32_128 = OpConstant %f32 128.0\n"
3615 "%f32_256 = OpConstant %f32 256.0\n"
3617 "%main = OpFunction %void None %voidf\n"
3618 "%entry = OpLabel\n"
3619 "%idval = OpLoad %uvec3 %id\n"
3620 "%x = OpCompositeExtract %u32 %idval 0\n"
3621 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3622 "%inval = OpLoad %f32 %inloc\n"
3624 "%xabs = OpExtInst %f32 %ext FAbs %inval\n"
3625 "%x8 = OpFMod %f32 %xabs %f32_256\n"
3626 "%x7 = OpFMod %f32 %xabs %f32_128\n"
3627 "%x6 = OpFMod %f32 %xabs %f32_64\n"
3628 "%x5 = OpFMod %f32 %xabs %f32_32\n"
3629 "%x4 = OpFMod %f32 %xabs %f32_16\n"
3630 "%x3 = OpFMod %f32 %xabs %f32_8\n"
3631 "%x2 = OpFMod %f32 %xabs %f32_4\n"
3632 "%x1 = OpFMod %f32 %xabs %f32_2\n"
3634 "%b7 = OpFOrdGreaterThanEqual %bool %x8 %f32_127_5\n"
3635 "%b6 = OpFOrdGreaterThanEqual %bool %x7 %f32_63_5\n"
3636 "%b5 = OpFOrdGreaterThanEqual %bool %x6 %f32_31_5\n"
3637 "%b4 = OpFOrdGreaterThanEqual %bool %x5 %f32_15_5\n"
3638 "%b3 = OpFOrdGreaterThanEqual %bool %x4 %f32_7_5\n"
3639 "%b2 = OpFOrdGreaterThanEqual %bool %x3 %f32_3_5\n"
3640 "%b1 = OpFOrdGreaterThanEqual %bool %x2 %f32_1_5\n"
3641 "%b0 = OpFOrdGreaterThanEqual %bool %x1 %f32_0_5\n"
3643 + generateOpPhiCase5(codestring) +
3645 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3646 " OpStore %outloc %res\n"
3650 spec5.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3651 spec5.outputs.push_back(BufferSp(new Float32Buffer(outputFloats5)));
3652 spec5.numWorkGroups = IVec3(numElements, 1, 1);
3654 group->addChild(new SpvAsmComputeShaderCase(testCtx, "nested", "Stress OpPhi with a lot of nesting", spec5));
3656 createOpPhiVartypeTests(group, testCtx);
3658 return group.release();
3661 // Assembly code used for testing block order is based on GLSL source code:
3665 // layout(std140, set = 0, binding = 0) readonly buffer Input {
3666 // float elements[];
3668 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
3669 // float elements[];
3673 // uint x = gl_GlobalInvocationID.x;
3674 // output_data.elements[x] = input_data.elements[x];
3675 // if (x > uint(50)) {
3676 // switch (x % uint(3)) {
3677 // case 0: output_data.elements[x] += 1.5f; break;
3678 // case 1: output_data.elements[x] += 42.f; break;
3679 // case 2: output_data.elements[x] -= 27.f; break;
3683 // output_data.elements[x] = -input_data.elements[x];
3686 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
3688 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
3689 ComputeShaderSpec spec;
3690 de::Random rnd (deStringHash(group->getName()));
3691 const int numElements = 100;
3692 vector<float> inputFloats (numElements, 0);
3693 vector<float> outputFloats (numElements, 0);
3695 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3697 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3698 floorAll(inputFloats);
3700 for (size_t ndx = 0; ndx <= 50; ++ndx)
3701 outputFloats[ndx] = -inputFloats[ndx];
3703 for (size_t ndx = 51; ndx < numElements; ++ndx)
3707 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
3708 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
3709 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
3715 string(getComputeAsmShaderPreamble()) +
3717 "OpSource GLSL 430\n"
3718 "OpName %main \"main\"\n"
3719 "OpName %id \"gl_GlobalInvocationID\"\n"
3721 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3723 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3725 "%u32ptr = OpTypePointer Function %u32\n"
3726 "%u32ptr_input = OpTypePointer Input %u32\n"
3728 + string(getComputeAsmInputOutputBuffer()) +
3730 "%id = OpVariable %uvec3ptr Input\n"
3731 "%zero = OpConstant %i32 0\n"
3732 "%const3 = OpConstant %u32 3\n"
3733 "%const50 = OpConstant %u32 50\n"
3734 "%constf1p5 = OpConstant %f32 1.5\n"
3735 "%constf27 = OpConstant %f32 27.0\n"
3736 "%constf42 = OpConstant %f32 42.0\n"
3738 "%main = OpFunction %void None %voidf\n"
3741 "%entry = OpLabel\n"
3743 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
3744 "%xvar = OpVariable %u32ptr Function\n"
3745 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
3746 "%x = OpLoad %u32 %xptr\n"
3747 " OpStore %xvar %x\n"
3749 "%cmp = OpUGreaterThan %bool %x %const50\n"
3750 " OpSelectionMerge %if_merge None\n"
3751 " OpBranchConditional %cmp %if_true %if_false\n"
3753 // False branch for if-statement: placed in the middle of switch cases and before true branch.
3754 "%if_false = OpLabel\n"
3755 "%x_f = OpLoad %u32 %xvar\n"
3756 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
3757 "%inval_f = OpLoad %f32 %inloc_f\n"
3758 "%negate = OpFNegate %f32 %inval_f\n"
3759 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
3760 " OpStore %outloc_f %negate\n"
3761 " OpBranch %if_merge\n"
3763 // Merge block for if-statement: placed in the middle of true and false branch.
3764 "%if_merge = OpLabel\n"
3767 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
3768 "%if_true = OpLabel\n"
3769 "%xval_t = OpLoad %u32 %xvar\n"
3770 "%mod = OpUMod %u32 %xval_t %const3\n"
3771 " OpSelectionMerge %switch_merge None\n"
3772 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
3774 // Merge block for switch-statement: placed before the case
3775 // bodies. But it must follow OpSwitch which dominates it.
3776 "%switch_merge = OpLabel\n"
3777 " OpBranch %if_merge\n"
3779 // Case 1 for switch-statement: placed before case 0.
3780 // It must follow the OpSwitch that dominates it.
3781 "%case1 = OpLabel\n"
3782 "%x_1 = OpLoad %u32 %xvar\n"
3783 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
3784 "%inval_1 = OpLoad %f32 %inloc_1\n"
3785 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
3786 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
3787 " OpStore %outloc_1 %addf42\n"
3788 " OpBranch %switch_merge\n"
3790 // Case 2 for switch-statement.
3791 "%case2 = OpLabel\n"
3792 "%x_2 = OpLoad %u32 %xvar\n"
3793 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
3794 "%inval_2 = OpLoad %f32 %inloc_2\n"
3795 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
3796 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
3797 " OpStore %outloc_2 %subf27\n"
3798 " OpBranch %switch_merge\n"
3800 // Default case for switch-statement: placed in the middle of normal cases.
3801 "%default = OpLabel\n"
3802 " OpBranch %switch_merge\n"
3804 // Case 0 for switch-statement: out of order.
3805 "%case0 = OpLabel\n"
3806 "%x_0 = OpLoad %u32 %xvar\n"
3807 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
3808 "%inval_0 = OpLoad %f32 %inloc_0\n"
3809 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
3810 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
3811 " OpStore %outloc_0 %addf1p5\n"
3812 " OpBranch %switch_merge\n"
3815 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3816 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3817 spec.numWorkGroups = IVec3(numElements, 1, 1);
3819 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
3821 return group.release();
3824 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
3826 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
3827 ComputeShaderSpec spec1;
3828 ComputeShaderSpec spec2;
3829 de::Random rnd (deStringHash(group->getName()));
3830 const int numElements = 100;
3831 vector<float> inputFloats (numElements, 0);
3832 vector<float> outputFloats1 (numElements, 0);
3833 vector<float> outputFloats2 (numElements, 0);
3834 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
3836 for (size_t ndx = 0; ndx < numElements; ++ndx)
3838 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
3839 outputFloats2[ndx] = -inputFloats[ndx];
3842 const string assembly(
3843 "OpCapability Shader\n"
3844 "OpCapability ClipDistance\n"
3845 "OpMemoryModel Logical GLSL450\n"
3846 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
3847 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
3848 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
3849 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
3850 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
3851 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
3853 "OpName %comp_main1 \"entrypoint1\"\n"
3854 "OpName %comp_main2 \"entrypoint2\"\n"
3855 "OpName %vert_main \"entrypoint2\"\n"
3856 "OpName %id \"gl_GlobalInvocationID\"\n"
3857 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
3858 "OpName %vertexIndex \"gl_VertexIndex\"\n"
3859 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
3860 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
3861 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
3862 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
3864 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3865 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
3866 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
3867 "OpDecorate %vert_builtin_st Block\n"
3868 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
3869 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
3870 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
3872 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3874 "%zero = OpConstant %i32 0\n"
3875 "%one = OpConstant %u32 1\n"
3876 "%c_f32_1 = OpConstant %f32 1\n"
3878 "%i32inputptr = OpTypePointer Input %i32\n"
3879 "%vec4 = OpTypeVector %f32 4\n"
3880 "%vec4ptr = OpTypePointer Output %vec4\n"
3881 "%f32arr1 = OpTypeArray %f32 %one\n"
3882 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
3883 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
3884 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
3886 "%id = OpVariable %uvec3ptr Input\n"
3887 "%vertexIndex = OpVariable %i32inputptr Input\n"
3888 "%instanceIndex = OpVariable %i32inputptr Input\n"
3889 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
3891 // gl_Position = vec4(1.);
3892 "%vert_main = OpFunction %void None %voidf\n"
3893 "%vert_entry = OpLabel\n"
3894 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
3895 " OpStore %position %c_vec4_1\n"
3900 "%comp_main1 = OpFunction %void None %voidf\n"
3901 "%comp1_entry = OpLabel\n"
3902 "%idval1 = OpLoad %uvec3 %id\n"
3903 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
3904 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
3905 "%inval1 = OpLoad %f32 %inloc1\n"
3906 "%add = OpFAdd %f32 %inval1 %inval1\n"
3907 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
3908 " OpStore %outloc1 %add\n"
3913 "%comp_main2 = OpFunction %void None %voidf\n"
3914 "%comp2_entry = OpLabel\n"
3915 "%idval2 = OpLoad %uvec3 %id\n"
3916 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
3917 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
3918 "%inval2 = OpLoad %f32 %inloc2\n"
3919 "%neg = OpFNegate %f32 %inval2\n"
3920 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
3921 " OpStore %outloc2 %neg\n"
3923 " OpFunctionEnd\n");
3925 spec1.assembly = assembly;
3926 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3927 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
3928 spec1.numWorkGroups = IVec3(numElements, 1, 1);
3929 spec1.entryPoint = "entrypoint1";
3931 spec2.assembly = assembly;
3932 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3933 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
3934 spec2.numWorkGroups = IVec3(numElements, 1, 1);
3935 spec2.entryPoint = "entrypoint2";
3937 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
3938 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
3940 return group.release();
3943 inline std::string makeLongUTF8String (size_t num4ByteChars)
3945 // An example of a longest valid UTF-8 character. Be explicit about the
3946 // character type because Microsoft compilers can otherwise interpret the
3947 // character string as being over wide (16-bit) characters. Ideally, we
3948 // would just use a C++11 UTF-8 string literal, but we want to support older
3949 // Microsoft compilers.
3950 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
3951 std::string longString;
3952 longString.reserve(num4ByteChars * 4);
3953 for (size_t count = 0; count < num4ByteChars; count++)
3955 longString += earthAfrica;
3960 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
3962 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
3963 vector<CaseParameter> cases;
3964 de::Random rnd (deStringHash(group->getName()));
3965 const int numElements = 100;
3966 vector<float> positiveFloats (numElements, 0);
3967 vector<float> negativeFloats (numElements, 0);
3968 const StringTemplate shaderTemplate (
3969 "OpCapability Shader\n"
3970 "OpMemoryModel Logical GLSL450\n"
3972 "OpEntryPoint GLCompute %main \"main\" %id\n"
3973 "OpExecutionMode %main LocalSize 1 1 1\n"
3977 "OpName %main \"main\"\n"
3978 "OpName %id \"gl_GlobalInvocationID\"\n"
3980 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3982 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3984 "%id = OpVariable %uvec3ptr Input\n"
3985 "%zero = OpConstant %i32 0\n"
3987 "%main = OpFunction %void None %voidf\n"
3988 "%label = OpLabel\n"
3989 "%idval = OpLoad %uvec3 %id\n"
3990 "%x = OpCompositeExtract %u32 %idval 0\n"
3991 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3992 "%inval = OpLoad %f32 %inloc\n"
3993 "%neg = OpFNegate %f32 %inval\n"
3994 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3995 " OpStore %outloc %neg\n"
3997 " OpFunctionEnd\n");
3999 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
4000 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
4001 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
4002 "OpSource GLSL 430 %fname"));
4003 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
4004 "OpSource GLSL 430 %fname"));
4005 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
4006 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
4007 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
4008 "OpSource GLSL 430 %fname \"\""));
4009 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
4010 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
4011 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
4012 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
4013 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
4014 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
4015 "OpSourceContinued \"id main() {}\""));
4016 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
4017 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4018 "OpSourceContinued \"\""));
4019 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
4020 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4021 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
4022 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
4023 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4024 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
4025 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
4026 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
4027 "OpSourceContinued \"void\"\n"
4028 "OpSourceContinued \"main()\"\n"
4029 "OpSourceContinued \"{}\""));
4030 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
4031 "OpSource GLSL 430 %fname \"\"\n"
4032 "OpSourceContinued \"#version 430\nvoid main() {}\""));
4034 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4036 for (size_t ndx = 0; ndx < numElements; ++ndx)
4037 negativeFloats[ndx] = -positiveFloats[ndx];
4039 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4041 map<string, string> specializations;
4042 ComputeShaderSpec spec;
4044 specializations["SOURCE"] = cases[caseNdx].param;
4045 spec.assembly = shaderTemplate.specialize(specializations);
4046 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4047 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4048 spec.numWorkGroups = IVec3(numElements, 1, 1);
4050 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4053 return group.release();
4056 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
4058 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
4059 vector<CaseParameter> cases;
4060 de::Random rnd (deStringHash(group->getName()));
4061 const int numElements = 100;
4062 vector<float> inputFloats (numElements, 0);
4063 vector<float> outputFloats (numElements, 0);
4064 const StringTemplate shaderTemplate (
4065 string(getComputeAsmShaderPreamble()) +
4067 "OpSourceExtension \"${EXTENSION}\"\n"
4069 "OpName %main \"main\"\n"
4070 "OpName %id \"gl_GlobalInvocationID\"\n"
4072 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4074 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4076 "%id = OpVariable %uvec3ptr Input\n"
4077 "%zero = OpConstant %i32 0\n"
4079 "%main = OpFunction %void None %voidf\n"
4080 "%label = OpLabel\n"
4081 "%idval = OpLoad %uvec3 %id\n"
4082 "%x = OpCompositeExtract %u32 %idval 0\n"
4083 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4084 "%inval = OpLoad %f32 %inloc\n"
4085 "%neg = OpFNegate %f32 %inval\n"
4086 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4087 " OpStore %outloc %neg\n"
4089 " OpFunctionEnd\n");
4091 cases.push_back(CaseParameter("empty_extension", ""));
4092 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
4093 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
4094 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
4095 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
4097 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
4099 for (size_t ndx = 0; ndx < numElements; ++ndx)
4100 outputFloats[ndx] = -inputFloats[ndx];
4102 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4104 map<string, string> specializations;
4105 ComputeShaderSpec spec;
4107 specializations["EXTENSION"] = cases[caseNdx].param;
4108 spec.assembly = shaderTemplate.specialize(specializations);
4109 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4110 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4111 spec.numWorkGroups = IVec3(numElements, 1, 1);
4113 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4116 return group.release();
4119 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
4120 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
4122 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
4123 vector<CaseParameter> cases;
4124 de::Random rnd (deStringHash(group->getName()));
4125 const int numElements = 100;
4126 vector<float> positiveFloats (numElements, 0);
4127 vector<float> negativeFloats (numElements, 0);
4128 const StringTemplate shaderTemplate (
4129 string(getComputeAsmShaderPreamble()) +
4131 "OpSource GLSL 430\n"
4132 "OpName %main \"main\"\n"
4133 "OpName %id \"gl_GlobalInvocationID\"\n"
4135 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4137 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4138 "%uvec2 = OpTypeVector %u32 2\n"
4139 "%bvec3 = OpTypeVector %bool 3\n"
4140 "%fvec4 = OpTypeVector %f32 4\n"
4141 "%fmat33 = OpTypeMatrix %fvec3 3\n"
4142 "%const100 = OpConstant %u32 100\n"
4143 "%uarr100 = OpTypeArray %i32 %const100\n"
4144 "%struct = OpTypeStruct %f32 %i32 %u32\n"
4145 "%pointer = OpTypePointer Function %i32\n"
4146 + string(getComputeAsmInputOutputBuffer()) +
4148 "%null = OpConstantNull ${TYPE}\n"
4150 "%id = OpVariable %uvec3ptr Input\n"
4151 "%zero = OpConstant %i32 0\n"
4153 "%main = OpFunction %void None %voidf\n"
4154 "%label = OpLabel\n"
4155 "%idval = OpLoad %uvec3 %id\n"
4156 "%x = OpCompositeExtract %u32 %idval 0\n"
4157 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4158 "%inval = OpLoad %f32 %inloc\n"
4159 "%neg = OpFNegate %f32 %inval\n"
4160 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4161 " OpStore %outloc %neg\n"
4163 " OpFunctionEnd\n");
4165 cases.push_back(CaseParameter("bool", "%bool"));
4166 cases.push_back(CaseParameter("sint32", "%i32"));
4167 cases.push_back(CaseParameter("uint32", "%u32"));
4168 cases.push_back(CaseParameter("float32", "%f32"));
4169 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
4170 cases.push_back(CaseParameter("vec3bool", "%bvec3"));
4171 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
4172 cases.push_back(CaseParameter("matrix", "%fmat33"));
4173 cases.push_back(CaseParameter("array", "%uarr100"));
4174 cases.push_back(CaseParameter("struct", "%struct"));
4175 cases.push_back(CaseParameter("pointer", "%pointer"));
4177 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4179 for (size_t ndx = 0; ndx < numElements; ++ndx)
4180 negativeFloats[ndx] = -positiveFloats[ndx];
4182 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4184 map<string, string> specializations;
4185 ComputeShaderSpec spec;
4187 specializations["TYPE"] = cases[caseNdx].param;
4188 spec.assembly = shaderTemplate.specialize(specializations);
4189 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4190 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4191 spec.numWorkGroups = IVec3(numElements, 1, 1);
4193 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4196 return group.release();
4199 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
4200 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
4202 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
4203 vector<CaseParameter> cases;
4204 de::Random rnd (deStringHash(group->getName()));
4205 const int numElements = 100;
4206 vector<float> positiveFloats (numElements, 0);
4207 vector<float> negativeFloats (numElements, 0);
4208 const StringTemplate shaderTemplate (
4209 string(getComputeAsmShaderPreamble()) +
4211 "OpSource GLSL 430\n"
4212 "OpName %main \"main\"\n"
4213 "OpName %id \"gl_GlobalInvocationID\"\n"
4215 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4217 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4219 "%id = OpVariable %uvec3ptr Input\n"
4220 "%zero = OpConstant %i32 0\n"
4224 "%main = OpFunction %void None %voidf\n"
4225 "%label = OpLabel\n"
4226 "%idval = OpLoad %uvec3 %id\n"
4227 "%x = OpCompositeExtract %u32 %idval 0\n"
4228 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4229 "%inval = OpLoad %f32 %inloc\n"
4230 "%neg = OpFNegate %f32 %inval\n"
4231 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4232 " OpStore %outloc %neg\n"
4234 " OpFunctionEnd\n");
4236 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
4237 "%const = OpConstantComposite %uvec3 %five %zero %five"));
4238 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
4239 "%ten = OpConstant %f32 10.\n"
4240 "%fzero = OpConstant %f32 0.\n"
4241 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
4242 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
4243 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
4244 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
4245 "%fzero = OpConstant %f32 0.\n"
4246 "%one = OpConstant %f32 1.\n"
4247 "%point5 = OpConstant %f32 0.5\n"
4248 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
4249 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
4250 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
4251 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
4252 "%st2 = OpTypeStruct %i32 %i32\n"
4253 "%struct = OpTypeStruct %st1 %st2\n"
4254 "%point5 = OpConstant %f32 0.5\n"
4255 "%one = OpConstant %u32 1\n"
4256 "%ten = OpConstant %i32 10\n"
4257 "%st1val = OpConstantComposite %st1 %one %point5\n"
4258 "%st2val = OpConstantComposite %st2 %ten %ten\n"
4259 "%const = OpConstantComposite %struct %st1val %st2val"));
4261 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4263 for (size_t ndx = 0; ndx < numElements; ++ndx)
4264 negativeFloats[ndx] = -positiveFloats[ndx];
4266 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4268 map<string, string> specializations;
4269 ComputeShaderSpec spec;
4271 specializations["CONSTANT"] = cases[caseNdx].param;
4272 spec.assembly = shaderTemplate.specialize(specializations);
4273 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4274 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4275 spec.numWorkGroups = IVec3(numElements, 1, 1);
4277 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4280 return group.release();
4283 // Creates a floating point number with the given exponent, and significand
4284 // bits set. It can only create normalized numbers. Only the least significant
4285 // 24 bits of the significand will be examined. The final bit of the
4286 // significand will also be ignored. This allows alignment to be written
4287 // similarly to C99 hex-floats.
4288 // For example if you wanted to write 0x1.7f34p-12 you would call
4289 // constructNormalizedFloat(-12, 0x7f3400)
4290 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
4294 for (deInt32 idx = 0; idx < 23; ++idx)
4296 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
4300 return std::ldexp(f, exponent);
4303 // Compare instruction for the OpQuantizeF16 compute exact case.
4304 // Returns true if the output is what is expected from the test case.
4305 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
4307 if (outputAllocs.size() != 1)
4310 // Only size is needed because we cannot compare Nans.
4311 size_t byteSize = expectedOutputs[0]->getByteSize();
4313 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
4315 if (byteSize != 4*sizeof(float)) {
4319 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
4320 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
4325 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
4326 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
4331 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
4332 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
4337 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
4338 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
4345 // Checks that every output from a test-case is a float NaN.
4346 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
4348 if (outputAllocs.size() != 1)
4351 // Only size is needed because we cannot compare Nans.
4352 size_t byteSize = expectedOutputs[0]->getByteSize();
4354 const float* const output_as_float = static_cast<const float* const>(outputAllocs[0]->getHostPtr());
4356 for (size_t idx = 0; idx < byteSize / sizeof(float); ++idx)
4358 if (!deFloatIsNaN(output_as_float[idx]))
4367 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
4368 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
4370 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
4372 const std::string shader (
4373 string(getComputeAsmShaderPreamble()) +
4375 "OpSource GLSL 430\n"
4376 "OpName %main \"main\"\n"
4377 "OpName %id \"gl_GlobalInvocationID\"\n"
4379 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4381 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4383 "%id = OpVariable %uvec3ptr Input\n"
4384 "%zero = OpConstant %i32 0\n"
4386 "%main = OpFunction %void None %voidf\n"
4387 "%label = OpLabel\n"
4388 "%idval = OpLoad %uvec3 %id\n"
4389 "%x = OpCompositeExtract %u32 %idval 0\n"
4390 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4391 "%inval = OpLoad %f32 %inloc\n"
4392 "%quant = OpQuantizeToF16 %f32 %inval\n"
4393 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4394 " OpStore %outloc %quant\n"
4396 " OpFunctionEnd\n");
4399 ComputeShaderSpec spec;
4400 const deUint32 numElements = 100;
4401 vector<float> infinities;
4402 vector<float> results;
4404 infinities.reserve(numElements);
4405 results.reserve(numElements);
4407 for (size_t idx = 0; idx < numElements; ++idx)
4412 infinities.push_back(std::numeric_limits<float>::infinity());
4413 results.push_back(std::numeric_limits<float>::infinity());
4416 infinities.push_back(-std::numeric_limits<float>::infinity());
4417 results.push_back(-std::numeric_limits<float>::infinity());
4420 infinities.push_back(std::ldexp(1.0f, 16));
4421 results.push_back(std::numeric_limits<float>::infinity());
4424 infinities.push_back(std::ldexp(-1.0f, 32));
4425 results.push_back(-std::numeric_limits<float>::infinity());
4430 spec.assembly = shader;
4431 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
4432 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
4433 spec.numWorkGroups = IVec3(numElements, 1, 1);
4435 group->addChild(new SpvAsmComputeShaderCase(
4436 testCtx, "infinities", "Check that infinities propagated and created", spec));
4440 ComputeShaderSpec spec;
4442 const deUint32 numElements = 100;
4444 nans.reserve(numElements);
4446 for (size_t idx = 0; idx < numElements; ++idx)
4450 nans.push_back(std::numeric_limits<float>::quiet_NaN());
4454 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
4458 spec.assembly = shader;
4459 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
4460 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
4461 spec.numWorkGroups = IVec3(numElements, 1, 1);
4462 spec.verifyIO = &compareNan;
4464 group->addChild(new SpvAsmComputeShaderCase(
4465 testCtx, "propagated_nans", "Check that nans are propagated", spec));
4469 ComputeShaderSpec spec;
4470 vector<float> small;
4471 vector<float> zeros;
4472 const deUint32 numElements = 100;
4474 small.reserve(numElements);
4475 zeros.reserve(numElements);
4477 for (size_t idx = 0; idx < numElements; ++idx)
4482 small.push_back(0.f);
4483 zeros.push_back(0.f);
4486 small.push_back(-0.f);
4487 zeros.push_back(-0.f);
4490 small.push_back(std::ldexp(1.0f, -16));
4491 zeros.push_back(0.f);
4494 small.push_back(std::ldexp(-1.0f, -32));
4495 zeros.push_back(-0.f);
4498 small.push_back(std::ldexp(1.0f, -127));
4499 zeros.push_back(0.f);
4502 small.push_back(-std::ldexp(1.0f, -128));
4503 zeros.push_back(-0.f);
4508 spec.assembly = shader;
4509 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
4510 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
4511 spec.numWorkGroups = IVec3(numElements, 1, 1);
4513 group->addChild(new SpvAsmComputeShaderCase(
4514 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
4518 ComputeShaderSpec spec;
4519 vector<float> exact;
4520 const deUint32 numElements = 200;
4522 exact.reserve(numElements);
4524 for (size_t idx = 0; idx < numElements; ++idx)
4525 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
4527 spec.assembly = shader;
4528 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
4529 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
4530 spec.numWorkGroups = IVec3(numElements, 1, 1);
4532 group->addChild(new SpvAsmComputeShaderCase(
4533 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
4537 ComputeShaderSpec spec;
4538 vector<float> inputs;
4539 const deUint32 numElements = 4;
4541 inputs.push_back(constructNormalizedFloat(8, 0x300300));
4542 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
4543 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
4544 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
4546 spec.assembly = shader;
4547 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
4548 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4549 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
4550 spec.numWorkGroups = IVec3(numElements, 1, 1);
4552 group->addChild(new SpvAsmComputeShaderCase(
4553 testCtx, "rounded", "Check that are rounded when needed", spec));
4556 return group.release();
4559 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
4561 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
4563 const std::string shader (
4564 string(getComputeAsmShaderPreamble()) +
4566 "OpName %main \"main\"\n"
4567 "OpName %id \"gl_GlobalInvocationID\"\n"
4569 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4571 "OpDecorate %sc_0 SpecId 0\n"
4572 "OpDecorate %sc_1 SpecId 1\n"
4573 "OpDecorate %sc_2 SpecId 2\n"
4574 "OpDecorate %sc_3 SpecId 3\n"
4575 "OpDecorate %sc_4 SpecId 4\n"
4576 "OpDecorate %sc_5 SpecId 5\n"
4578 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4580 "%id = OpVariable %uvec3ptr Input\n"
4581 "%zero = OpConstant %i32 0\n"
4582 "%c_u32_6 = OpConstant %u32 6\n"
4584 "%sc_0 = OpSpecConstant %f32 0.\n"
4585 "%sc_1 = OpSpecConstant %f32 0.\n"
4586 "%sc_2 = OpSpecConstant %f32 0.\n"
4587 "%sc_3 = OpSpecConstant %f32 0.\n"
4588 "%sc_4 = OpSpecConstant %f32 0.\n"
4589 "%sc_5 = OpSpecConstant %f32 0.\n"
4591 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
4592 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
4593 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
4594 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
4595 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
4596 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
4598 "%main = OpFunction %void None %voidf\n"
4599 "%label = OpLabel\n"
4600 "%idval = OpLoad %uvec3 %id\n"
4601 "%x = OpCompositeExtract %u32 %idval 0\n"
4602 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4603 "%selector = OpUMod %u32 %x %c_u32_6\n"
4604 " OpSelectionMerge %exit None\n"
4605 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
4607 "%case0 = OpLabel\n"
4608 " OpStore %outloc %sc_0_quant\n"
4611 "%case1 = OpLabel\n"
4612 " OpStore %outloc %sc_1_quant\n"
4615 "%case2 = OpLabel\n"
4616 " OpStore %outloc %sc_2_quant\n"
4619 "%case3 = OpLabel\n"
4620 " OpStore %outloc %sc_3_quant\n"
4623 "%case4 = OpLabel\n"
4624 " OpStore %outloc %sc_4_quant\n"
4627 "%case5 = OpLabel\n"
4628 " OpStore %outloc %sc_5_quant\n"
4634 " OpFunctionEnd\n");
4637 ComputeShaderSpec spec;
4638 const deUint8 numCases = 4;
4639 vector<float> inputs (numCases, 0.f);
4640 vector<float> outputs;
4642 spec.assembly = shader;
4643 spec.numWorkGroups = IVec3(numCases, 1, 1);
4645 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
4646 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
4647 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
4648 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
4650 outputs.push_back(std::numeric_limits<float>::infinity());
4651 outputs.push_back(-std::numeric_limits<float>::infinity());
4652 outputs.push_back(std::numeric_limits<float>::infinity());
4653 outputs.push_back(-std::numeric_limits<float>::infinity());
4655 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4656 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4658 group->addChild(new SpvAsmComputeShaderCase(
4659 testCtx, "infinities", "Check that infinities propagated and created", spec));
4663 ComputeShaderSpec spec;
4664 const deUint8 numCases = 2;
4665 vector<float> inputs (numCases, 0.f);
4666 vector<float> outputs;
4668 spec.assembly = shader;
4669 spec.numWorkGroups = IVec3(numCases, 1, 1);
4670 spec.verifyIO = &compareNan;
4672 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
4673 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
4675 for (deUint8 idx = 0; idx < numCases; ++idx)
4676 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
4678 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4679 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4681 group->addChild(new SpvAsmComputeShaderCase(
4682 testCtx, "propagated_nans", "Check that nans are propagated", spec));
4686 ComputeShaderSpec spec;
4687 const deUint8 numCases = 6;
4688 vector<float> inputs (numCases, 0.f);
4689 vector<float> outputs;
4691 spec.assembly = shader;
4692 spec.numWorkGroups = IVec3(numCases, 1, 1);
4694 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
4695 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
4696 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
4697 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
4698 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
4699 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
4701 outputs.push_back(0.f);
4702 outputs.push_back(-0.f);
4703 outputs.push_back(0.f);
4704 outputs.push_back(-0.f);
4705 outputs.push_back(0.f);
4706 outputs.push_back(-0.f);
4708 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4709 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4711 group->addChild(new SpvAsmComputeShaderCase(
4712 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
4716 ComputeShaderSpec spec;
4717 const deUint8 numCases = 6;
4718 vector<float> inputs (numCases, 0.f);
4719 vector<float> outputs;
4721 spec.assembly = shader;
4722 spec.numWorkGroups = IVec3(numCases, 1, 1);
4724 for (deUint8 idx = 0; idx < 6; ++idx)
4726 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
4727 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
4728 outputs.push_back(f);
4731 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4732 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4734 group->addChild(new SpvAsmComputeShaderCase(
4735 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
4739 ComputeShaderSpec spec;
4740 const deUint8 numCases = 4;
4741 vector<float> inputs (numCases, 0.f);
4742 vector<float> outputs;
4744 spec.assembly = shader;
4745 spec.numWorkGroups = IVec3(numCases, 1, 1);
4746 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
4748 outputs.push_back(constructNormalizedFloat(8, 0x300300));
4749 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
4750 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
4751 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
4753 for (deUint8 idx = 0; idx < numCases; ++idx)
4754 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
4756 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4757 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4759 group->addChild(new SpvAsmComputeShaderCase(
4760 testCtx, "rounded", "Check that are rounded when needed", spec));
4763 return group.release();
4766 // Checks that constant null/composite values can be used in computation.
4767 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
4769 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
4770 ComputeShaderSpec spec;
4771 de::Random rnd (deStringHash(group->getName()));
4772 const int numElements = 100;
4773 vector<float> positiveFloats (numElements, 0);
4774 vector<float> negativeFloats (numElements, 0);
4776 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4778 for (size_t ndx = 0; ndx < numElements; ++ndx)
4779 negativeFloats[ndx] = -positiveFloats[ndx];
4782 "OpCapability Shader\n"
4783 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
4784 "OpMemoryModel Logical GLSL450\n"
4785 "OpEntryPoint GLCompute %main \"main\" %id\n"
4786 "OpExecutionMode %main LocalSize 1 1 1\n"
4788 "OpSource GLSL 430\n"
4789 "OpName %main \"main\"\n"
4790 "OpName %id \"gl_GlobalInvocationID\"\n"
4792 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4794 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4796 "%fmat = OpTypeMatrix %fvec3 3\n"
4797 "%ten = OpConstant %u32 10\n"
4798 "%f32arr10 = OpTypeArray %f32 %ten\n"
4799 "%fst = OpTypeStruct %f32 %f32\n"
4801 + string(getComputeAsmInputOutputBuffer()) +
4803 "%id = OpVariable %uvec3ptr Input\n"
4804 "%zero = OpConstant %i32 0\n"
4806 // Create a bunch of null values
4807 "%unull = OpConstantNull %u32\n"
4808 "%fnull = OpConstantNull %f32\n"
4809 "%vnull = OpConstantNull %fvec3\n"
4810 "%mnull = OpConstantNull %fmat\n"
4811 "%anull = OpConstantNull %f32arr10\n"
4812 "%snull = OpConstantComposite %fst %fnull %fnull\n"
4814 "%main = OpFunction %void None %voidf\n"
4815 "%label = OpLabel\n"
4816 "%idval = OpLoad %uvec3 %id\n"
4817 "%x = OpCompositeExtract %u32 %idval 0\n"
4818 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4819 "%inval = OpLoad %f32 %inloc\n"
4820 "%neg = OpFNegate %f32 %inval\n"
4822 // Get the abs() of (a certain element of) those null values
4823 "%unull_cov = OpConvertUToF %f32 %unull\n"
4824 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
4825 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
4826 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
4827 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
4828 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
4829 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
4830 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
4831 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
4832 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
4833 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
4836 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
4837 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
4838 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
4839 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
4840 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
4841 "%final = OpFAdd %f32 %add5 %snull_abs\n"
4843 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4844 " OpStore %outloc %final\n" // write to output
4847 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4848 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4849 spec.numWorkGroups = IVec3(numElements, 1, 1);
4851 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
4853 return group.release();
4856 // Assembly code used for testing loop control is based on GLSL source code:
4859 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4860 // float elements[];
4862 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4863 // float elements[];
4867 // uint x = gl_GlobalInvocationID.x;
4868 // output_data.elements[x] = input_data.elements[x];
4869 // for (uint i = 0; i < 4; ++i)
4870 // output_data.elements[x] += 1.f;
4872 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
4874 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
4875 vector<CaseParameter> cases;
4876 de::Random rnd (deStringHash(group->getName()));
4877 const int numElements = 100;
4878 vector<float> inputFloats (numElements, 0);
4879 vector<float> outputFloats (numElements, 0);
4880 const StringTemplate shaderTemplate (
4881 string(getComputeAsmShaderPreamble()) +
4883 "OpSource GLSL 430\n"
4884 "OpName %main \"main\"\n"
4885 "OpName %id \"gl_GlobalInvocationID\"\n"
4887 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4889 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4891 "%u32ptr = OpTypePointer Function %u32\n"
4893 "%id = OpVariable %uvec3ptr Input\n"
4894 "%zero = OpConstant %i32 0\n"
4895 "%uzero = OpConstant %u32 0\n"
4896 "%one = OpConstant %i32 1\n"
4897 "%constf1 = OpConstant %f32 1.0\n"
4898 "%four = OpConstant %u32 4\n"
4900 "%main = OpFunction %void None %voidf\n"
4901 "%entry = OpLabel\n"
4902 "%i = OpVariable %u32ptr Function\n"
4903 " OpStore %i %uzero\n"
4905 "%idval = OpLoad %uvec3 %id\n"
4906 "%x = OpCompositeExtract %u32 %idval 0\n"
4907 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4908 "%inval = OpLoad %f32 %inloc\n"
4909 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4910 " OpStore %outloc %inval\n"
4911 " OpBranch %loop_entry\n"
4913 "%loop_entry = OpLabel\n"
4914 "%i_val = OpLoad %u32 %i\n"
4915 "%cmp_lt = OpULessThan %bool %i_val %four\n"
4916 " OpLoopMerge %loop_merge %loop_body ${CONTROL}\n"
4917 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
4918 "%loop_body = OpLabel\n"
4919 "%outval = OpLoad %f32 %outloc\n"
4920 "%addf1 = OpFAdd %f32 %outval %constf1\n"
4921 " OpStore %outloc %addf1\n"
4922 "%new_i = OpIAdd %u32 %i_val %one\n"
4923 " OpStore %i %new_i\n"
4924 " OpBranch %loop_entry\n"
4925 "%loop_merge = OpLabel\n"
4927 " OpFunctionEnd\n");
4929 cases.push_back(CaseParameter("none", "None"));
4930 cases.push_back(CaseParameter("unroll", "Unroll"));
4931 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
4932 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
4934 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4936 for (size_t ndx = 0; ndx < numElements; ++ndx)
4937 outputFloats[ndx] = inputFloats[ndx] + 4.f;
4939 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4941 map<string, string> specializations;
4942 ComputeShaderSpec spec;
4944 specializations["CONTROL"] = cases[caseNdx].param;
4945 spec.assembly = shaderTemplate.specialize(specializations);
4946 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4947 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4948 spec.numWorkGroups = IVec3(numElements, 1, 1);
4950 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4953 group->addChild(new SpvAsmLoopControlDependencyLengthCase(testCtx, "dependency_length", "dependency_length"));
4954 group->addChild(new SpvAsmLoopControlDependencyInfiniteCase(testCtx, "dependency_infinite", "dependency_infinite"));
4956 return group.release();
4959 // Assembly code used for testing selection control is based on GLSL source code:
4962 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4963 // float elements[];
4965 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4966 // float elements[];
4970 // uint x = gl_GlobalInvocationID.x;
4971 // float val = input_data.elements[x];
4973 // output_data.elements[x] = val + 1.f;
4975 // output_data.elements[x] = val - 1.f;
4977 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
4979 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
4980 vector<CaseParameter> cases;
4981 de::Random rnd (deStringHash(group->getName()));
4982 const int numElements = 100;
4983 vector<float> inputFloats (numElements, 0);
4984 vector<float> outputFloats (numElements, 0);
4985 const StringTemplate shaderTemplate (
4986 string(getComputeAsmShaderPreamble()) +
4988 "OpSource GLSL 430\n"
4989 "OpName %main \"main\"\n"
4990 "OpName %id \"gl_GlobalInvocationID\"\n"
4992 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4994 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4996 "%id = OpVariable %uvec3ptr Input\n"
4997 "%zero = OpConstant %i32 0\n"
4998 "%constf1 = OpConstant %f32 1.0\n"
4999 "%constf10 = OpConstant %f32 10.0\n"
5001 "%main = OpFunction %void None %voidf\n"
5002 "%entry = OpLabel\n"
5003 "%idval = OpLoad %uvec3 %id\n"
5004 "%x = OpCompositeExtract %u32 %idval 0\n"
5005 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5006 "%inval = OpLoad %f32 %inloc\n"
5007 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5008 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
5010 " OpSelectionMerge %if_end ${CONTROL}\n"
5011 " OpBranchConditional %cmp_gt %if_true %if_false\n"
5012 "%if_true = OpLabel\n"
5013 "%addf1 = OpFAdd %f32 %inval %constf1\n"
5014 " OpStore %outloc %addf1\n"
5015 " OpBranch %if_end\n"
5016 "%if_false = OpLabel\n"
5017 "%subf1 = OpFSub %f32 %inval %constf1\n"
5018 " OpStore %outloc %subf1\n"
5019 " OpBranch %if_end\n"
5020 "%if_end = OpLabel\n"
5022 " OpFunctionEnd\n");
5024 cases.push_back(CaseParameter("none", "None"));
5025 cases.push_back(CaseParameter("flatten", "Flatten"));
5026 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
5027 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
5029 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5031 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
5032 floorAll(inputFloats);
5034 for (size_t ndx = 0; ndx < numElements; ++ndx)
5035 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
5037 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5039 map<string, string> specializations;
5040 ComputeShaderSpec spec;
5042 specializations["CONTROL"] = cases[caseNdx].param;
5043 spec.assembly = shaderTemplate.specialize(specializations);
5044 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5045 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5046 spec.numWorkGroups = IVec3(numElements, 1, 1);
5048 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5051 return group.release();
5054 // Assembly code used for testing function control is based on GLSL source code:
5058 // layout(std140, set = 0, binding = 0) readonly buffer Input {
5059 // float elements[];
5061 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
5062 // float elements[];
5065 // float const10() { return 10.f; }
5068 // uint x = gl_GlobalInvocationID.x;
5069 // output_data.elements[x] = input_data.elements[x] + const10();
5071 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
5073 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
5074 vector<CaseParameter> cases;
5075 de::Random rnd (deStringHash(group->getName()));
5076 const int numElements = 100;
5077 vector<float> inputFloats (numElements, 0);
5078 vector<float> outputFloats (numElements, 0);
5079 const StringTemplate shaderTemplate (
5080 string(getComputeAsmShaderPreamble()) +
5082 "OpSource GLSL 430\n"
5083 "OpName %main \"main\"\n"
5084 "OpName %func_const10 \"const10(\"\n"
5085 "OpName %id \"gl_GlobalInvocationID\"\n"
5087 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5089 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5091 "%f32f = OpTypeFunction %f32\n"
5092 "%id = OpVariable %uvec3ptr Input\n"
5093 "%zero = OpConstant %i32 0\n"
5094 "%constf10 = OpConstant %f32 10.0\n"
5096 "%main = OpFunction %void None %voidf\n"
5097 "%entry = OpLabel\n"
5098 "%idval = OpLoad %uvec3 %id\n"
5099 "%x = OpCompositeExtract %u32 %idval 0\n"
5100 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5101 "%inval = OpLoad %f32 %inloc\n"
5102 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
5103 "%fadd = OpFAdd %f32 %inval %ret_10\n"
5104 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5105 " OpStore %outloc %fadd\n"
5109 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
5110 "%label = OpLabel\n"
5111 " OpReturnValue %constf10\n"
5112 " OpFunctionEnd\n");
5114 cases.push_back(CaseParameter("none", "None"));
5115 cases.push_back(CaseParameter("inline", "Inline"));
5116 cases.push_back(CaseParameter("dont_inline", "DontInline"));
5117 cases.push_back(CaseParameter("pure", "Pure"));
5118 cases.push_back(CaseParameter("const", "Const"));
5119 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
5120 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
5121 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
5122 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
5124 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5126 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
5127 floorAll(inputFloats);
5129 for (size_t ndx = 0; ndx < numElements; ++ndx)
5130 outputFloats[ndx] = inputFloats[ndx] + 10.f;
5132 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5134 map<string, string> specializations;
5135 ComputeShaderSpec spec;
5137 specializations["CONTROL"] = cases[caseNdx].param;
5138 spec.assembly = shaderTemplate.specialize(specializations);
5139 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5140 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5141 spec.numWorkGroups = IVec3(numElements, 1, 1);
5143 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5146 return group.release();
5149 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
5151 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
5152 vector<CaseParameter> cases;
5153 de::Random rnd (deStringHash(group->getName()));
5154 const int numElements = 100;
5155 vector<float> inputFloats (numElements, 0);
5156 vector<float> outputFloats (numElements, 0);
5157 const StringTemplate shaderTemplate (
5158 string(getComputeAsmShaderPreamble()) +
5160 "OpSource GLSL 430\n"
5161 "OpName %main \"main\"\n"
5162 "OpName %id \"gl_GlobalInvocationID\"\n"
5164 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5166 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5168 "%f32ptr_f = OpTypePointer Function %f32\n"
5170 "%id = OpVariable %uvec3ptr Input\n"
5171 "%zero = OpConstant %i32 0\n"
5172 "%four = OpConstant %i32 4\n"
5174 "%main = OpFunction %void None %voidf\n"
5175 "%label = OpLabel\n"
5176 "%copy = OpVariable %f32ptr_f Function\n"
5177 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
5178 "%x = OpCompositeExtract %u32 %idval 0\n"
5179 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5180 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5181 " OpCopyMemory %copy %inloc ${ACCESS}\n"
5182 "%val1 = OpLoad %f32 %copy\n"
5183 "%val2 = OpLoad %f32 %inloc\n"
5184 "%add = OpFAdd %f32 %val1 %val2\n"
5185 " OpStore %outloc %add ${ACCESS}\n"
5187 " OpFunctionEnd\n");
5189 cases.push_back(CaseParameter("null", ""));
5190 cases.push_back(CaseParameter("none", "None"));
5191 cases.push_back(CaseParameter("volatile", "Volatile"));
5192 cases.push_back(CaseParameter("aligned", "Aligned 4"));
5193 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
5194 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
5195 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
5197 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5199 for (size_t ndx = 0; ndx < numElements; ++ndx)
5200 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
5202 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5204 map<string, string> specializations;
5205 ComputeShaderSpec spec;
5207 specializations["ACCESS"] = cases[caseNdx].param;
5208 spec.assembly = shaderTemplate.specialize(specializations);
5209 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5210 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5211 spec.numWorkGroups = IVec3(numElements, 1, 1);
5213 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5216 return group.release();
5219 // Checks that we can get undefined values for various types, without exercising a computation with it.
5220 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
5222 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
5223 vector<CaseParameter> cases;
5224 de::Random rnd (deStringHash(group->getName()));
5225 const int numElements = 100;
5226 vector<float> positiveFloats (numElements, 0);
5227 vector<float> negativeFloats (numElements, 0);
5228 const StringTemplate shaderTemplate (
5229 string(getComputeAsmShaderPreamble()) +
5231 "OpSource GLSL 430\n"
5232 "OpName %main \"main\"\n"
5233 "OpName %id \"gl_GlobalInvocationID\"\n"
5235 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5237 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
5238 "%uvec2 = OpTypeVector %u32 2\n"
5239 "%fvec4 = OpTypeVector %f32 4\n"
5240 "%fmat33 = OpTypeMatrix %fvec3 3\n"
5241 "%image = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
5242 "%sampler = OpTypeSampler\n"
5243 "%simage = OpTypeSampledImage %image\n"
5244 "%const100 = OpConstant %u32 100\n"
5245 "%uarr100 = OpTypeArray %i32 %const100\n"
5246 "%struct = OpTypeStruct %f32 %i32 %u32\n"
5247 "%pointer = OpTypePointer Function %i32\n"
5248 + string(getComputeAsmInputOutputBuffer()) +
5250 "%id = OpVariable %uvec3ptr Input\n"
5251 "%zero = OpConstant %i32 0\n"
5253 "%main = OpFunction %void None %voidf\n"
5254 "%label = OpLabel\n"
5256 "%undef = OpUndef ${TYPE}\n"
5258 "%idval = OpLoad %uvec3 %id\n"
5259 "%x = OpCompositeExtract %u32 %idval 0\n"
5261 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5262 "%inval = OpLoad %f32 %inloc\n"
5263 "%neg = OpFNegate %f32 %inval\n"
5264 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5265 " OpStore %outloc %neg\n"
5267 " OpFunctionEnd\n");
5269 cases.push_back(CaseParameter("bool", "%bool"));
5270 cases.push_back(CaseParameter("sint32", "%i32"));
5271 cases.push_back(CaseParameter("uint32", "%u32"));
5272 cases.push_back(CaseParameter("float32", "%f32"));
5273 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
5274 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
5275 cases.push_back(CaseParameter("matrix", "%fmat33"));
5276 cases.push_back(CaseParameter("image", "%image"));
5277 cases.push_back(CaseParameter("sampler", "%sampler"));
5278 cases.push_back(CaseParameter("sampledimage", "%simage"));
5279 cases.push_back(CaseParameter("array", "%uarr100"));
5280 cases.push_back(CaseParameter("runtimearray", "%f32arr"));
5281 cases.push_back(CaseParameter("struct", "%struct"));
5282 cases.push_back(CaseParameter("pointer", "%pointer"));
5284 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
5286 for (size_t ndx = 0; ndx < numElements; ++ndx)
5287 negativeFloats[ndx] = -positiveFloats[ndx];
5289 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5291 map<string, string> specializations;
5292 ComputeShaderSpec spec;
5294 specializations["TYPE"] = cases[caseNdx].param;
5295 spec.assembly = shaderTemplate.specialize(specializations);
5296 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
5297 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
5298 spec.numWorkGroups = IVec3(numElements, 1, 1);
5300 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5303 return group.release();
5307 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
5309 struct NameCodePair { string name, code; };
5310 RGBA defaultColors[4];
5311 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
5312 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
5313 map<string, string> fragments = passthruFragments();
5314 const NameCodePair tests[] =
5316 {"unknown", "OpSource Unknown 321"},
5317 {"essl", "OpSource ESSL 310"},
5318 {"glsl", "OpSource GLSL 450"},
5319 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
5320 {"opencl_c", "OpSource OpenCL_C 120"},
5321 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
5322 {"file", opsourceGLSLWithFile},
5323 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
5324 // Longest possible source string: SPIR-V limits instructions to 65535
5325 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
5326 // contain 65530 UTF8 characters (one word each) plus one last word
5327 // containing 3 ASCII characters and \0.
5328 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
5331 getDefaultColors(defaultColors);
5332 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5334 fragments["debug"] = tests[testNdx].code;
5335 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5338 return opSourceTests.release();
5341 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
5343 struct NameCodePair { string name, code; };
5344 RGBA defaultColors[4];
5345 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
5346 map<string, string> fragments = passthruFragments();
5347 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
5348 const NameCodePair tests[] =
5350 {"empty", opsource + "OpSourceContinued \"\""},
5351 {"short", opsource + "OpSourceContinued \"abcde\""},
5352 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
5353 // Longest possible source string: SPIR-V limits instructions to 65535
5354 // words, of which the first one is OpSourceContinued/length; the rest
5355 // will contain 65533 UTF8 characters (one word each) plus one last word
5356 // containing 3 ASCII characters and \0.
5357 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
5360 getDefaultColors(defaultColors);
5361 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5363 fragments["debug"] = tests[testNdx].code;
5364 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5367 return opSourceTests.release();
5369 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
5371 RGBA defaultColors[4];
5372 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
5373 map<string, string> fragments;
5374 getDefaultColors(defaultColors);
5375 fragments["debug"] =
5376 "%name = OpString \"name\"\n";
5378 fragments["pre_main"] =
5381 "OpLine %name 1 1\n"
5383 "OpLine %name 1 1\n"
5384 "OpLine %name 1 1\n"
5385 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5387 "OpLine %name 1 1\n"
5389 "OpLine %name 1 1\n"
5390 "OpLine %name 1 1\n"
5391 "%second_param1 = OpFunctionParameter %v4f32\n"
5394 "%label_secondfunction = OpLabel\n"
5396 "OpReturnValue %second_param1\n"
5401 fragments["testfun"] =
5402 // A %test_code function that returns its argument unchanged.
5405 "OpLine %name 1 1\n"
5406 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5408 "%param1 = OpFunctionParameter %v4f32\n"
5411 "%label_testfun = OpLabel\n"
5413 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5414 "OpReturnValue %val1\n"
5416 "OpLine %name 1 1\n"
5419 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
5421 return opLineTests.release();
5424 tcu::TestCaseGroup* createOpModuleProcessedTests(tcu::TestContext& testCtx)
5426 RGBA defaultColors[4];
5427 de::MovePtr<tcu::TestCaseGroup> opModuleProcessedTests (new tcu::TestCaseGroup(testCtx, "opmoduleprocessed", "OpModuleProcessed instruction"));
5428 map<string, string> fragments;
5429 std::vector<std::string> noExtensions;
5430 GraphicsResources resources;
5432 getDefaultColors(defaultColors);
5433 resources.verifyBinary = veryfiBinaryShader;
5434 resources.spirvVersion = SPIRV_VERSION_1_3;
5436 fragments["moduleprocessed"] =
5437 "OpModuleProcessed \"VULKAN CTS\"\n"
5438 "OpModuleProcessed \"Negative values\"\n"
5439 "OpModuleProcessed \"Date: 2017/09/21\"\n";
5441 fragments["pre_main"] =
5442 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5443 "%second_param1 = OpFunctionParameter %v4f32\n"
5444 "%label_secondfunction = OpLabel\n"
5445 "OpReturnValue %second_param1\n"
5448 fragments["testfun"] =
5449 // A %test_code function that returns its argument unchanged.
5450 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5451 "%param1 = OpFunctionParameter %v4f32\n"
5452 "%label_testfun = OpLabel\n"
5453 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5454 "OpReturnValue %val1\n"
5457 createTestsForAllStages ("opmoduleprocessed", defaultColors, defaultColors, fragments, resources, noExtensions, opModuleProcessedTests.get());
5459 return opModuleProcessedTests.release();
5463 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
5465 RGBA defaultColors[4];
5466 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
5467 map<string, string> fragments;
5468 std::vector<std::pair<std::string, std::string> > problemStrings;
5470 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
5471 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
5472 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
5473 getDefaultColors(defaultColors);
5475 fragments["debug"] =
5476 "%other_name = OpString \"other_name\"\n";
5478 fragments["pre_main"] =
5479 "OpLine %file_name 32 0\n"
5480 "OpLine %file_name 32 32\n"
5481 "OpLine %file_name 32 40\n"
5482 "OpLine %other_name 32 40\n"
5483 "OpLine %other_name 0 100\n"
5484 "OpLine %other_name 0 4294967295\n"
5485 "OpLine %other_name 4294967295 0\n"
5486 "OpLine %other_name 32 40\n"
5487 "OpLine %file_name 0 0\n"
5488 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5489 "OpLine %file_name 1 0\n"
5490 "%second_param1 = OpFunctionParameter %v4f32\n"
5491 "OpLine %file_name 1 3\n"
5492 "OpLine %file_name 1 2\n"
5493 "%label_secondfunction = OpLabel\n"
5494 "OpLine %file_name 0 2\n"
5495 "OpReturnValue %second_param1\n"
5497 "OpLine %file_name 0 2\n"
5498 "OpLine %file_name 0 2\n";
5500 fragments["testfun"] =
5501 // A %test_code function that returns its argument unchanged.
5502 "OpLine %file_name 1 0\n"
5503 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5504 "OpLine %file_name 16 330\n"
5505 "%param1 = OpFunctionParameter %v4f32\n"
5506 "OpLine %file_name 14 442\n"
5507 "%label_testfun = OpLabel\n"
5508 "OpLine %file_name 11 1024\n"
5509 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5510 "OpLine %file_name 2 97\n"
5511 "OpReturnValue %val1\n"
5513 "OpLine %file_name 5 32\n";
5515 for (size_t i = 0; i < problemStrings.size(); ++i)
5517 map<string, string> testFragments = fragments;
5518 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
5519 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
5522 return opLineTests.release();
5525 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
5527 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
5531 const char functionStart[] =
5532 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5533 "%param1 = OpFunctionParameter %v4f32\n"
5536 const char functionEnd[] =
5537 "OpReturnValue %transformed_param\n"
5540 struct NameConstantsCode
5547 NameConstantsCode tests[] =
5551 "%cnull = OpConstantNull %v4f32\n",
5552 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
5556 "%cnull = OpConstantNull %f32\n",
5557 "%vp = OpVariable %fp_v4f32 Function\n"
5558 "%v = OpLoad %v4f32 %vp\n"
5559 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
5560 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
5561 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
5562 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
5563 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
5567 "%cnull = OpConstantNull %bool\n",
5568 "%v = OpVariable %fp_v4f32 Function\n"
5569 " OpStore %v %param1\n"
5570 " OpSelectionMerge %false_label None\n"
5571 " OpBranchConditional %cnull %true_label %false_label\n"
5572 "%true_label = OpLabel\n"
5573 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
5574 " OpBranch %false_label\n"
5575 "%false_label = OpLabel\n"
5576 "%transformed_param = OpLoad %v4f32 %v\n"
5580 "%cnull = OpConstantNull %i32\n",
5581 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
5582 "%b = OpIEqual %bool %cnull %c_i32_0\n"
5583 " OpSelectionMerge %false_label None\n"
5584 " OpBranchConditional %b %true_label %false_label\n"
5585 "%true_label = OpLabel\n"
5586 " OpStore %v %param1\n"
5587 " OpBranch %false_label\n"
5588 "%false_label = OpLabel\n"
5589 "%transformed_param = OpLoad %v4f32 %v\n"
5593 "%stype = OpTypeStruct %f32 %v4f32\n"
5594 "%fp_stype = OpTypePointer Function %stype\n"
5595 "%cnull = OpConstantNull %stype\n",
5596 "%v = OpVariable %fp_stype Function %cnull\n"
5597 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
5598 "%f_val = OpLoad %v4f32 %f\n"
5599 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
5603 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
5604 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
5605 "%cnull = OpConstantNull %a4_v4f32\n",
5606 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
5607 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5608 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
5609 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
5610 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
5611 "%f_val = OpLoad %v4f32 %f\n"
5612 "%f1_val = OpLoad %v4f32 %f1\n"
5613 "%f2_val = OpLoad %v4f32 %f2\n"
5614 "%f3_val = OpLoad %v4f32 %f3\n"
5615 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
5616 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
5617 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
5618 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
5622 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5623 "%cnull = OpConstantNull %mat4x4_f32\n",
5624 // Our null matrix * any vector should result in a zero vector.
5625 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
5626 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
5630 getHalfColorsFullAlpha(colors);
5632 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5634 map<string, string> fragments;
5635 fragments["pre_main"] = tests[testNdx].constants;
5636 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5637 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
5639 return opConstantNullTests.release();
5641 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
5643 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
5644 RGBA inputColors[4];
5645 RGBA outputColors[4];
5648 const char functionStart[] =
5649 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5650 "%param1 = OpFunctionParameter %v4f32\n"
5653 const char functionEnd[] =
5654 "OpReturnValue %transformed_param\n"
5657 struct NameConstantsCode
5664 NameConstantsCode tests[] =
5669 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
5670 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
5675 "%stype = OpTypeStruct %v4f32 %f32\n"
5676 "%fp_stype = OpTypePointer Function %stype\n"
5677 "%f32_n_1 = OpConstant %f32 -1.0\n"
5678 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5679 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
5680 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
5682 "%v = OpVariable %fp_stype Function %cval\n"
5683 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5684 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
5685 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
5686 "%f32_val = OpLoad %f32 %f32_ptr\n"
5687 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
5688 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
5689 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
5692 // [1|0|0|0.5] [x] = x + 0.5
5693 // [0|1|0|0.5] [y] = y + 0.5
5694 // [0|0|1|0.5] [z] = z + 0.5
5695 // [0|0|0|1 ] [1] = 1
5698 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5699 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
5700 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
5701 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
5702 "%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"
5703 "%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",
5705 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
5710 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5711 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5712 "%f32_n_1 = OpConstant %f32 -1.0\n"
5713 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5714 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
5716 "%v = OpVariable %fp_a4f32 Function %carr\n"
5717 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
5718 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
5719 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5720 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
5721 "%f_val = OpLoad %f32 %f\n"
5722 "%f1_val = OpLoad %f32 %f1\n"
5723 "%f2_val = OpLoad %f32 %f2\n"
5724 "%f3_val = OpLoad %f32 %f3\n"
5725 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
5726 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
5727 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
5728 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
5729 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5736 // [ 1.0, 1.0, 1.0, 1.0]
5740 // [ 0.0, 0.5, 0.0, 0.0]
5744 // [ 1.0, 1.0, 1.0, 1.0]
5747 "array_of_struct_of_array",
5749 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5750 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5751 "%stype = OpTypeStruct %f32 %a4f32\n"
5752 "%a3stype = OpTypeArray %stype %c_u32_3\n"
5753 "%fp_a3stype = OpTypePointer Function %a3stype\n"
5754 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
5755 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5756 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
5757 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
5758 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
5760 "%v = OpVariable %fp_a3stype Function %carr\n"
5761 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
5762 "%f_l = OpLoad %f32 %f\n"
5763 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
5764 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5768 getHalfColorsFullAlpha(inputColors);
5769 outputColors[0] = RGBA(255, 255, 255, 255);
5770 outputColors[1] = RGBA(255, 127, 127, 255);
5771 outputColors[2] = RGBA(127, 255, 127, 255);
5772 outputColors[3] = RGBA(127, 127, 255, 255);
5774 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5776 map<string, string> fragments;
5777 fragments["pre_main"] = tests[testNdx].constants;
5778 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5779 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
5781 return opConstantCompositeTests.release();
5784 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
5786 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
5787 RGBA inputColors[4];
5788 RGBA outputColors[4];
5789 map<string, string> fragments;
5791 // vec4 test_code(vec4 param) {
5792 // vec4 result = param;
5793 // for (int i = 0; i < 4; ++i) {
5794 // if (i == 0) result[i] = 0.;
5795 // else result[i] = 1. - result[i];
5799 const char function[] =
5800 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5801 "%param1 = OpFunctionParameter %v4f32\n"
5803 "%iptr = OpVariable %fp_i32 Function\n"
5804 "%result = OpVariable %fp_v4f32 Function\n"
5805 " OpStore %iptr %c_i32_0\n"
5806 " OpStore %result %param1\n"
5809 // Loop entry block.
5811 "%ival = OpLoad %i32 %iptr\n"
5812 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5813 " OpLoopMerge %exit %if_entry None\n"
5814 " OpBranchConditional %lt_4 %if_entry %exit\n"
5816 // Merge block for loop.
5818 "%ret = OpLoad %v4f32 %result\n"
5819 " OpReturnValue %ret\n"
5821 // If-statement entry block.
5822 "%if_entry = OpLabel\n"
5823 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5824 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
5825 " OpSelectionMerge %if_exit None\n"
5826 " OpBranchConditional %eq_0 %if_true %if_false\n"
5828 // False branch for if-statement.
5829 "%if_false = OpLabel\n"
5830 "%val = OpLoad %f32 %loc\n"
5831 "%sub = OpFSub %f32 %c_f32_1 %val\n"
5832 " OpStore %loc %sub\n"
5833 " OpBranch %if_exit\n"
5835 // Merge block for if-statement.
5836 "%if_exit = OpLabel\n"
5837 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5838 " OpStore %iptr %ival_next\n"
5841 // True branch for if-statement.
5842 "%if_true = OpLabel\n"
5843 " OpStore %loc %c_f32_0\n"
5844 " OpBranch %if_exit\n"
5848 fragments["testfun"] = function;
5850 inputColors[0] = RGBA(127, 127, 127, 0);
5851 inputColors[1] = RGBA(127, 0, 0, 0);
5852 inputColors[2] = RGBA(0, 127, 0, 0);
5853 inputColors[3] = RGBA(0, 0, 127, 0);
5855 outputColors[0] = RGBA(0, 128, 128, 255);
5856 outputColors[1] = RGBA(0, 255, 255, 255);
5857 outputColors[2] = RGBA(0, 128, 255, 255);
5858 outputColors[3] = RGBA(0, 255, 128, 255);
5860 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5862 return group.release();
5865 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
5867 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
5868 RGBA inputColors[4];
5869 RGBA outputColors[4];
5870 map<string, string> fragments;
5872 const char typesAndConstants[] =
5873 "%c_f32_p2 = OpConstant %f32 0.2\n"
5874 "%c_f32_p4 = OpConstant %f32 0.4\n"
5875 "%c_f32_p6 = OpConstant %f32 0.6\n"
5876 "%c_f32_p8 = OpConstant %f32 0.8\n";
5878 // vec4 test_code(vec4 param) {
5879 // vec4 result = param;
5880 // for (int i = 0; i < 4; ++i) {
5882 // case 0: result[i] += .2; break;
5883 // case 1: result[i] += .6; break;
5884 // case 2: result[i] += .4; break;
5885 // case 3: result[i] += .8; break;
5886 // default: break; // unreachable
5891 const char function[] =
5892 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5893 "%param1 = OpFunctionParameter %v4f32\n"
5895 "%iptr = OpVariable %fp_i32 Function\n"
5896 "%result = OpVariable %fp_v4f32 Function\n"
5897 " OpStore %iptr %c_i32_0\n"
5898 " OpStore %result %param1\n"
5901 // Loop entry block.
5903 "%ival = OpLoad %i32 %iptr\n"
5904 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5905 " OpLoopMerge %exit %switch_exit None\n"
5906 " OpBranchConditional %lt_4 %switch_entry %exit\n"
5908 // Merge block for loop.
5910 "%ret = OpLoad %v4f32 %result\n"
5911 " OpReturnValue %ret\n"
5913 // Switch-statement entry block.
5914 "%switch_entry = OpLabel\n"
5915 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5916 "%val = OpLoad %f32 %loc\n"
5917 " OpSelectionMerge %switch_exit None\n"
5918 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5920 "%case2 = OpLabel\n"
5921 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
5922 " OpStore %loc %addp4\n"
5923 " OpBranch %switch_exit\n"
5925 "%switch_default = OpLabel\n"
5928 "%case3 = OpLabel\n"
5929 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
5930 " OpStore %loc %addp8\n"
5931 " OpBranch %switch_exit\n"
5933 "%case0 = OpLabel\n"
5934 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
5935 " OpStore %loc %addp2\n"
5936 " OpBranch %switch_exit\n"
5938 // Merge block for switch-statement.
5939 "%switch_exit = OpLabel\n"
5940 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5941 " OpStore %iptr %ival_next\n"
5944 "%case1 = OpLabel\n"
5945 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
5946 " OpStore %loc %addp6\n"
5947 " OpBranch %switch_exit\n"
5951 fragments["pre_main"] = typesAndConstants;
5952 fragments["testfun"] = function;
5954 inputColors[0] = RGBA(127, 27, 127, 51);
5955 inputColors[1] = RGBA(127, 0, 0, 51);
5956 inputColors[2] = RGBA(0, 27, 0, 51);
5957 inputColors[3] = RGBA(0, 0, 127, 51);
5959 outputColors[0] = RGBA(178, 180, 229, 255);
5960 outputColors[1] = RGBA(178, 153, 102, 255);
5961 outputColors[2] = RGBA(51, 180, 102, 255);
5962 outputColors[3] = RGBA(51, 153, 229, 255);
5964 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5966 return group.release();
5969 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
5971 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
5972 RGBA inputColors[4];
5973 RGBA outputColors[4];
5974 map<string, string> fragments;
5976 const char decorations[] =
5977 "OpDecorate %array_group ArrayStride 4\n"
5978 "OpDecorate %struct_member_group Offset 0\n"
5979 "%array_group = OpDecorationGroup\n"
5980 "%struct_member_group = OpDecorationGroup\n"
5982 "OpDecorate %group1 RelaxedPrecision\n"
5983 "OpDecorate %group3 RelaxedPrecision\n"
5984 "OpDecorate %group3 Invariant\n"
5985 "OpDecorate %group3 Restrict\n"
5986 "%group0 = OpDecorationGroup\n"
5987 "%group1 = OpDecorationGroup\n"
5988 "%group3 = OpDecorationGroup\n";
5990 const char typesAndConstants[] =
5991 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
5992 "%struct1 = OpTypeStruct %a3f32\n"
5993 "%struct2 = OpTypeStruct %a3f32\n"
5994 "%fp_struct1 = OpTypePointer Function %struct1\n"
5995 "%fp_struct2 = OpTypePointer Function %struct2\n"
5996 "%c_f32_2 = OpConstant %f32 2.\n"
5997 "%c_f32_n2 = OpConstant %f32 -2.\n"
5999 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
6000 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
6001 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
6002 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
6004 const char function[] =
6005 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6006 "%param = OpFunctionParameter %v4f32\n"
6007 "%entry = OpLabel\n"
6008 "%result = OpVariable %fp_v4f32 Function\n"
6009 "%v_struct1 = OpVariable %fp_struct1 Function\n"
6010 "%v_struct2 = OpVariable %fp_struct2 Function\n"
6011 " OpStore %result %param\n"
6012 " OpStore %v_struct1 %c_struct1\n"
6013 " OpStore %v_struct2 %c_struct2\n"
6014 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
6015 "%val1 = OpLoad %f32 %ptr1\n"
6016 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
6017 "%val2 = OpLoad %f32 %ptr2\n"
6018 "%addvalues = OpFAdd %f32 %val1 %val2\n"
6019 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
6020 "%val = OpLoad %f32 %ptr\n"
6021 "%addresult = OpFAdd %f32 %addvalues %val\n"
6022 " OpStore %ptr %addresult\n"
6023 "%ret = OpLoad %v4f32 %result\n"
6024 " OpReturnValue %ret\n"
6027 struct CaseNameDecoration
6033 CaseNameDecoration tests[] =
6036 "same_decoration_group_on_multiple_types",
6037 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
6040 "empty_decoration_group",
6041 "OpGroupDecorate %group0 %a3f32\n"
6042 "OpGroupDecorate %group0 %result\n"
6045 "one_element_decoration_group",
6046 "OpGroupDecorate %array_group %a3f32\n"
6049 "multiple_elements_decoration_group",
6050 "OpGroupDecorate %group3 %v_struct1\n"
6053 "multiple_decoration_groups_on_same_variable",
6054 "OpGroupDecorate %group0 %v_struct2\n"
6055 "OpGroupDecorate %group1 %v_struct2\n"
6056 "OpGroupDecorate %group3 %v_struct2\n"
6059 "same_decoration_group_multiple_times",
6060 "OpGroupDecorate %group1 %addvalues\n"
6061 "OpGroupDecorate %group1 %addvalues\n"
6062 "OpGroupDecorate %group1 %addvalues\n"
6067 getHalfColorsFullAlpha(inputColors);
6068 getHalfColorsFullAlpha(outputColors);
6070 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
6072 fragments["decoration"] = decorations + tests[idx].decoration;
6073 fragments["pre_main"] = typesAndConstants;
6074 fragments["testfun"] = function;
6076 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
6079 return group.release();
6082 struct SpecConstantTwoIntGraphicsCase
6084 const char* caseName;
6085 const char* scDefinition0;
6086 const char* scDefinition1;
6087 const char* scResultType;
6088 const char* scOperation;
6089 deInt32 scActualValue0;
6090 deInt32 scActualValue1;
6091 const char* resultOperation;
6092 RGBA expectedColors[4];
6094 SpecConstantTwoIntGraphicsCase (const char* name,
6095 const char* definition0,
6096 const char* definition1,
6097 const char* resultType,
6098 const char* operation,
6101 const char* resultOp,
6102 const RGBA (&output)[4])
6104 , scDefinition0 (definition0)
6105 , scDefinition1 (definition1)
6106 , scResultType (resultType)
6107 , scOperation (operation)
6108 , scActualValue0 (value0)
6109 , scActualValue1 (value1)
6110 , resultOperation (resultOp)
6112 expectedColors[0] = output[0];
6113 expectedColors[1] = output[1];
6114 expectedColors[2] = output[2];
6115 expectedColors[3] = output[3];
6119 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
6121 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
6122 vector<SpecConstantTwoIntGraphicsCase> cases;
6123 RGBA inputColors[4];
6124 RGBA outputColors0[4];
6125 RGBA outputColors1[4];
6126 RGBA outputColors2[4];
6128 const char decorations1[] =
6129 "OpDecorate %sc_0 SpecId 0\n"
6130 "OpDecorate %sc_1 SpecId 1\n";
6132 const char typesAndConstants1[] =
6133 "${OPTYPE_DEFINITIONS:opt}"
6134 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
6135 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
6136 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
6138 const char function1[] =
6139 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6140 "%param = OpFunctionParameter %v4f32\n"
6141 "%label = OpLabel\n"
6142 "${TYPE_CONVERT:opt}"
6143 "%result = OpVariable %fp_v4f32 Function\n"
6144 " OpStore %result %param\n"
6145 "%gen = ${GEN_RESULT}\n"
6146 "%index = OpIAdd %i32 %gen %c_i32_1\n"
6147 "%loc = OpAccessChain %fp_f32 %result %index\n"
6148 "%val = OpLoad %f32 %loc\n"
6149 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6150 " OpStore %loc %add\n"
6151 "%ret = OpLoad %v4f32 %result\n"
6152 " OpReturnValue %ret\n"
6155 inputColors[0] = RGBA(127, 127, 127, 255);
6156 inputColors[1] = RGBA(127, 0, 0, 255);
6157 inputColors[2] = RGBA(0, 127, 0, 255);
6158 inputColors[3] = RGBA(0, 0, 127, 255);
6160 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
6161 outputColors0[0] = RGBA(255, 127, 127, 255);
6162 outputColors0[1] = RGBA(255, 0, 0, 255);
6163 outputColors0[2] = RGBA(128, 127, 0, 255);
6164 outputColors0[3] = RGBA(128, 0, 127, 255);
6166 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
6167 outputColors1[0] = RGBA(127, 255, 127, 255);
6168 outputColors1[1] = RGBA(127, 128, 0, 255);
6169 outputColors1[2] = RGBA(0, 255, 0, 255);
6170 outputColors1[3] = RGBA(0, 128, 127, 255);
6172 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
6173 outputColors2[0] = RGBA(127, 127, 255, 255);
6174 outputColors2[1] = RGBA(127, 0, 128, 255);
6175 outputColors2[2] = RGBA(0, 127, 128, 255);
6176 outputColors2[3] = RGBA(0, 0, 255, 255);
6178 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
6179 const char addZeroToSc32[] = "OpIAdd %i32 %c_i32_0 %sc_op32";
6180 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
6181 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
6183 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
6184 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
6185 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
6186 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
6187 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
6188 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6189 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6190 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
6191 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
6192 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
6193 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
6194 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
6195 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
6196 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
6197 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
6198 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
6199 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6200 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
6201 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
6202 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
6203 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6204 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
6205 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
6206 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6207 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6208 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6209 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6210 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
6211 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
6212 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
6213 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
6214 cases.push_back(SpecConstantTwoIntGraphicsCase("sconvert", " %i32 0", " %i32 0", "%i16", "SConvert %sc_0", -1, 0, addZeroToSc32, outputColors0));
6215 // -1082130432 stored as 32-bit two's complement is the binary representation of -1 as IEEE-754 Float
6216 cases.push_back(SpecConstantTwoIntGraphicsCase("fconvert", " %f32 0", " %f32 0", "%f64", "FConvert %sc_0", -1082130432, 0, addZeroToSc32, outputColors0));
6217 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
6219 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
6221 map<string, string> specializations;
6222 map<string, string> fragments;
6223 vector<deInt32> specConstants;
6224 vector<string> features;
6225 PushConstants noPushConstants;
6226 GraphicsResources noResources;
6227 GraphicsInterfaces noInterfaces;
6228 std::vector<std::string> noExtensions;
6230 // Special SPIR-V code for SConvert-case
6231 if (strcmp(cases[caseNdx].caseName, "sconvert") == 0)
6233 features.push_back("shaderInt16");
6234 fragments["capability"] = "OpCapability Int16\n"; // Adds 16-bit integer capability
6235 specializations["OPTYPE_DEFINITIONS"] = "%i16 = OpTypeInt 16 1\n"; // Adds 16-bit integer type
6236 specializations["TYPE_CONVERT"] = "%sc_op32 = OpSConvert %i32 %sc_op\n"; // Converts 16-bit integer to 32-bit integer
6239 // Special SPIR-V code for FConvert-case
6240 if (strcmp(cases[caseNdx].caseName, "fconvert") == 0)
6242 features.push_back("shaderFloat64");
6243 fragments["capability"] = "OpCapability Float64\n"; // Adds 64-bit float capability
6244 specializations["OPTYPE_DEFINITIONS"] = "%f64 = OpTypeFloat 64\n"; // Adds 64-bit float type
6245 specializations["TYPE_CONVERT"] = "%sc_op32 = OpConvertFToS %i32 %sc_op\n"; // Converts 64-bit float to 32-bit integer
6248 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
6249 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
6250 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
6251 specializations["SC_OP"] = cases[caseNdx].scOperation;
6252 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
6254 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
6255 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
6256 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
6258 specConstants.push_back(cases[caseNdx].scActualValue0);
6259 specConstants.push_back(cases[caseNdx].scActualValue1);
6261 createTestsForAllStages(
6262 cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants,
6263 noPushConstants, noResources, noInterfaces, noExtensions, features, VulkanFeatures(), group.get());
6266 const char decorations2[] =
6267 "OpDecorate %sc_0 SpecId 0\n"
6268 "OpDecorate %sc_1 SpecId 1\n"
6269 "OpDecorate %sc_2 SpecId 2\n";
6271 const char typesAndConstants2[] =
6272 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
6273 "%vec3_undef = OpUndef %v3i32\n"
6275 "%sc_0 = OpSpecConstant %i32 0\n"
6276 "%sc_1 = OpSpecConstant %i32 0\n"
6277 "%sc_2 = OpSpecConstant %i32 0\n"
6278 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
6279 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
6280 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
6281 "%sc_vec3_0_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
6282 "%sc_vec3_1_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
6283 "%sc_vec3_2_s = OpSpecConstantOp %v3i32 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
6284 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
6285 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
6286 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
6287 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
6288 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
6289 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
6290 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
6292 const char function2[] =
6293 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6294 "%param = OpFunctionParameter %v4f32\n"
6295 "%label = OpLabel\n"
6296 "%result = OpVariable %fp_v4f32 Function\n"
6297 " OpStore %result %param\n"
6298 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
6299 "%val = OpLoad %f32 %loc\n"
6300 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6301 " OpStore %loc %add\n"
6302 "%ret = OpLoad %v4f32 %result\n"
6303 " OpReturnValue %ret\n"
6306 map<string, string> fragments;
6307 vector<deInt32> specConstants;
6309 fragments["decoration"] = decorations2;
6310 fragments["pre_main"] = typesAndConstants2;
6311 fragments["testfun"] = function2;
6313 specConstants.push_back(56789);
6314 specConstants.push_back(-2);
6315 specConstants.push_back(56788);
6317 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
6319 return group.release();
6322 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
6324 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
6325 RGBA inputColors[4];
6326 RGBA outputColors1[4];
6327 RGBA outputColors2[4];
6328 RGBA outputColors3[4];
6329 map<string, string> fragments1;
6330 map<string, string> fragments2;
6331 map<string, string> fragments3;
6333 const char typesAndConstants1[] =
6334 "%c_f32_p2 = OpConstant %f32 0.2\n"
6335 "%c_f32_p4 = OpConstant %f32 0.4\n"
6336 "%c_f32_p5 = OpConstant %f32 0.5\n"
6337 "%c_f32_p8 = OpConstant %f32 0.8\n";
6339 // vec4 test_code(vec4 param) {
6340 // vec4 result = param;
6341 // for (int i = 0; i < 4; ++i) {
6344 // case 0: operand = .2; break;
6345 // case 1: operand = .5; break;
6346 // case 2: operand = .4; break;
6347 // case 3: operand = .0; break;
6348 // default: break; // unreachable
6350 // result[i] += operand;
6354 const char function1[] =
6355 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6356 "%param1 = OpFunctionParameter %v4f32\n"
6358 "%iptr = OpVariable %fp_i32 Function\n"
6359 "%result = OpVariable %fp_v4f32 Function\n"
6360 " OpStore %iptr %c_i32_0\n"
6361 " OpStore %result %param1\n"
6365 "%ival = OpLoad %i32 %iptr\n"
6366 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6367 " OpLoopMerge %exit %phi None\n"
6368 " OpBranchConditional %lt_4 %entry %exit\n"
6370 "%entry = OpLabel\n"
6371 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6372 "%val = OpLoad %f32 %loc\n"
6373 " OpSelectionMerge %phi None\n"
6374 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6376 "%case0 = OpLabel\n"
6378 "%case1 = OpLabel\n"
6380 "%case2 = OpLabel\n"
6382 "%case3 = OpLabel\n"
6385 "%default = OpLabel\n"
6389 "%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
6390 "%add = OpFAdd %f32 %val %operand\n"
6391 " OpStore %loc %add\n"
6392 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6393 " OpStore %iptr %ival_next\n"
6397 "%ret = OpLoad %v4f32 %result\n"
6398 " OpReturnValue %ret\n"
6402 fragments1["pre_main"] = typesAndConstants1;
6403 fragments1["testfun"] = function1;
6405 getHalfColorsFullAlpha(inputColors);
6407 outputColors1[0] = RGBA(178, 255, 229, 255);
6408 outputColors1[1] = RGBA(178, 127, 102, 255);
6409 outputColors1[2] = RGBA(51, 255, 102, 255);
6410 outputColors1[3] = RGBA(51, 127, 229, 255);
6412 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
6414 const char typesAndConstants2[] =
6415 "%c_f32_p2 = OpConstant %f32 0.2\n";
6417 // Add .4 to the second element of the given parameter.
6418 const char function2[] =
6419 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6420 "%param = OpFunctionParameter %v4f32\n"
6421 "%entry = OpLabel\n"
6422 "%result = OpVariable %fp_v4f32 Function\n"
6423 " OpStore %result %param\n"
6424 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6425 "%val = OpLoad %f32 %loc\n"
6429 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
6430 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
6431 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
6432 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
6433 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
6434 " OpLoopMerge %exit %phi None\n"
6435 " OpBranchConditional %still_loop %phi %exit\n"
6438 " OpStore %loc %accum\n"
6439 "%ret = OpLoad %v4f32 %result\n"
6440 " OpReturnValue %ret\n"
6444 fragments2["pre_main"] = typesAndConstants2;
6445 fragments2["testfun"] = function2;
6447 outputColors2[0] = RGBA(127, 229, 127, 255);
6448 outputColors2[1] = RGBA(127, 102, 0, 255);
6449 outputColors2[2] = RGBA(0, 229, 0, 255);
6450 outputColors2[3] = RGBA(0, 102, 127, 255);
6452 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
6454 const char typesAndConstants3[] =
6455 "%true = OpConstantTrue %bool\n"
6456 "%false = OpConstantFalse %bool\n"
6457 "%c_f32_p2 = OpConstant %f32 0.2\n";
6459 // Swap the second and the third element of the given parameter.
6460 const char function3[] =
6461 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6462 "%param = OpFunctionParameter %v4f32\n"
6463 "%entry = OpLabel\n"
6464 "%result = OpVariable %fp_v4f32 Function\n"
6465 " OpStore %result %param\n"
6466 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6467 "%a_init = OpLoad %f32 %a_loc\n"
6468 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
6469 "%b_init = OpLoad %f32 %b_loc\n"
6473 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
6474 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
6475 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
6476 " OpLoopMerge %exit %phi None\n"
6477 " OpBranchConditional %still_loop %phi %exit\n"
6480 " OpStore %a_loc %a_next\n"
6481 " OpStore %b_loc %b_next\n"
6482 "%ret = OpLoad %v4f32 %result\n"
6483 " OpReturnValue %ret\n"
6487 fragments3["pre_main"] = typesAndConstants3;
6488 fragments3["testfun"] = function3;
6490 outputColors3[0] = RGBA(127, 127, 127, 255);
6491 outputColors3[1] = RGBA(127, 0, 0, 255);
6492 outputColors3[2] = RGBA(0, 0, 127, 255);
6493 outputColors3[3] = RGBA(0, 127, 0, 255);
6495 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
6497 return group.release();
6500 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
6502 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
6503 RGBA inputColors[4];
6504 RGBA outputColors[4];
6506 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
6507 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
6508 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
6509 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
6510 const char constantsAndTypes[] =
6511 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
6512 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
6513 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
6514 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
6515 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n";
6517 const char function[] =
6518 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6519 "%param = OpFunctionParameter %v4f32\n"
6520 "%label = OpLabel\n"
6521 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
6522 "%var2 = OpVariable %fp_f32 Function\n"
6523 "%red = OpCompositeExtract %f32 %param 0\n"
6524 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
6525 " OpStore %var2 %plus_red\n"
6526 "%val1 = OpLoad %f32 %var1\n"
6527 "%val2 = OpLoad %f32 %var2\n"
6528 "%mul = OpFMul %f32 %val1 %val2\n"
6529 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
6530 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
6531 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
6532 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
6533 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
6534 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
6535 " OpReturnValue %ret\n"
6538 struct CaseNameDecoration
6545 CaseNameDecoration tests[] = {
6546 {"multiplication", "OpDecorate %mul NoContraction"},
6547 {"addition", "OpDecorate %add NoContraction"},
6548 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
6551 getHalfColorsFullAlpha(inputColors);
6553 for (deUint8 idx = 0; idx < 4; ++idx)
6555 inputColors[idx].setRed(0);
6556 outputColors[idx] = RGBA(0, 0, 0, 255);
6559 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
6561 map<string, string> fragments;
6563 fragments["decoration"] = tests[testNdx].decoration;
6564 fragments["pre_main"] = constantsAndTypes;
6565 fragments["testfun"] = function;
6567 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
6570 return group.release();
6573 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
6575 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
6578 const char constantsAndTypes[] =
6579 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
6580 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
6581 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
6582 "%fp_stype = OpTypePointer Function %stype\n";
6584 const char function[] =
6585 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6586 "%param1 = OpFunctionParameter %v4f32\n"
6588 "%v1 = OpVariable %fp_v4f32 Function\n"
6589 "%v2 = OpVariable %fp_a2f32 Function\n"
6590 "%v3 = OpVariable %fp_f32 Function\n"
6591 "%v = OpVariable %fp_stype Function\n"
6592 "%vv = OpVariable %fp_stype Function\n"
6593 "%vvv = OpVariable %fp_f32 Function\n"
6595 " OpStore %v1 %c_v4f32_1_1_1_1\n"
6596 " OpStore %v2 %c_a2f32_1\n"
6597 " OpStore %v3 %c_f32_1\n"
6599 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
6600 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
6601 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
6602 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
6603 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
6604 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
6606 " OpStore %p_v4f32 %v1_v ${access_type}\n"
6607 " OpStore %p_a2f32 %v2_v ${access_type}\n"
6608 " OpStore %p_f32 %v3_v ${access_type}\n"
6610 " OpCopyMemory %vv %v ${access_type}\n"
6611 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
6613 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
6614 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
6615 "%v_f32_3 = OpLoad %f32 %vvv\n"
6617 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
6618 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
6619 " OpReturnValue %ret2\n"
6622 struct NameMemoryAccess
6629 NameMemoryAccess tests[] =
6632 { "volatile", "Volatile" },
6633 { "aligned", "Aligned 1" },
6634 { "volatile_aligned", "Volatile|Aligned 1" },
6635 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
6636 { "volatile_nontemporal", "Volatile|Nontemporal" },
6637 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
6640 getHalfColorsFullAlpha(colors);
6642 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
6644 map<string, string> fragments;
6645 map<string, string> memoryAccess;
6646 memoryAccess["access_type"] = tests[testNdx].accessType;
6648 fragments["pre_main"] = constantsAndTypes;
6649 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
6650 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
6652 return memoryAccessTests.release();
6654 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
6656 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
6657 RGBA defaultColors[4];
6658 map<string, string> fragments;
6659 getDefaultColors(defaultColors);
6661 // First, simple cases that don't do anything with the OpUndef result.
6662 struct NameCodePair { string name, decl, type; };
6663 const NameCodePair tests[] =
6665 {"bool", "", "%bool"},
6666 {"vec2uint32", "", "%v2u32"},
6667 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown", "%type"},
6668 {"sampler", "%type = OpTypeSampler", "%type"},
6669 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img", "%type"},
6670 {"pointer", "", "%fp_i32"},
6671 {"runtimearray", "%type = OpTypeRuntimeArray %f32", "%type"},
6672 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100", "%type"},
6673 {"struct", "%type = OpTypeStruct %f32 %i32 %u32", "%type"}};
6674 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
6676 fragments["undef_type"] = tests[testNdx].type;
6677 fragments["testfun"] = StringTemplate(
6678 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6679 "%param1 = OpFunctionParameter %v4f32\n"
6680 "%label_testfun = OpLabel\n"
6681 "%undef = OpUndef ${undef_type}\n"
6682 "OpReturnValue %param1\n"
6683 "OpFunctionEnd\n").specialize(fragments);
6684 fragments["pre_main"] = tests[testNdx].decl;
6685 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
6689 fragments["testfun"] =
6690 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6691 "%param1 = OpFunctionParameter %v4f32\n"
6692 "%label_testfun = OpLabel\n"
6693 "%undef = OpUndef %f32\n"
6694 "%zero = OpFMul %f32 %undef %c_f32_0\n"
6695 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
6696 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
6697 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6698 "%b = OpFAdd %f32 %a %actually_zero\n"
6699 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
6700 "OpReturnValue %ret\n"
6703 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6705 fragments["testfun"] =
6706 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6707 "%param1 = OpFunctionParameter %v4f32\n"
6708 "%label_testfun = OpLabel\n"
6709 "%undef = OpUndef %i32\n"
6710 "%zero = OpIMul %i32 %undef %c_i32_0\n"
6711 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6712 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6713 "OpReturnValue %ret\n"
6716 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6718 fragments["testfun"] =
6719 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6720 "%param1 = OpFunctionParameter %v4f32\n"
6721 "%label_testfun = OpLabel\n"
6722 "%undef = OpUndef %u32\n"
6723 "%zero = OpIMul %u32 %undef %c_i32_0\n"
6724 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6725 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6726 "OpReturnValue %ret\n"
6729 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6731 fragments["testfun"] =
6732 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6733 "%param1 = OpFunctionParameter %v4f32\n"
6734 "%label_testfun = OpLabel\n"
6735 "%undef = OpUndef %v4f32\n"
6736 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
6737 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
6738 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
6739 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
6740 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
6741 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6742 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6743 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6744 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6745 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6746 "%actually_zero_1 = OpSelect %f32 %is_nan_1 %c_f32_0 %zero_1\n"
6747 "%actually_zero_2 = OpSelect %f32 %is_nan_2 %c_f32_0 %zero_2\n"
6748 "%actually_zero_3 = OpSelect %f32 %is_nan_3 %c_f32_0 %zero_3\n"
6749 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6750 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6751 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6752 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6753 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6754 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6755 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6756 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6757 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6758 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6759 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6760 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6761 "OpReturnValue %ret\n"
6764 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6766 fragments["pre_main"] =
6767 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
6768 fragments["testfun"] =
6769 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6770 "%param1 = OpFunctionParameter %v4f32\n"
6771 "%label_testfun = OpLabel\n"
6772 "%undef = OpUndef %m2x2f32\n"
6773 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
6774 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
6775 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
6776 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
6777 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
6778 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6779 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6780 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6781 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6782 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6783 "%actually_zero_1 = OpSelect %f32 %is_nan_1 %c_f32_0 %zero_1\n"
6784 "%actually_zero_2 = OpSelect %f32 %is_nan_2 %c_f32_0 %zero_2\n"
6785 "%actually_zero_3 = OpSelect %f32 %is_nan_3 %c_f32_0 %zero_3\n"
6786 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6787 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6788 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6789 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6790 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6791 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6792 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6793 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6794 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6795 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6796 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6797 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6798 "OpReturnValue %ret\n"
6801 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
6803 return opUndefTests.release();
6806 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
6808 const RGBA inputColors[4] =
6811 RGBA(0, 0, 255, 255),
6812 RGBA(0, 255, 0, 255),
6813 RGBA(0, 255, 255, 255)
6816 const RGBA expectedColors[4] =
6818 RGBA(255, 0, 0, 255),
6819 RGBA(255, 0, 0, 255),
6820 RGBA(255, 0, 0, 255),
6821 RGBA(255, 0, 0, 255)
6824 const struct SingleFP16Possibility
6827 const char* constant; // Value to assign to %test_constant.
6829 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
6835 -constructNormalizedFloat(1, 0x300000),
6836 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6841 constructNormalizedFloat(7, 0x000000),
6842 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6844 // SPIR-V requires that OpQuantizeToF16 flushes
6845 // any numbers that would end up denormalized in F16 to zero.
6849 std::ldexp(1.5f, -140),
6850 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6855 -std::ldexp(1.5f, -140),
6856 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6861 std::ldexp(1.0f, -16),
6862 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6863 }, // too small positive
6865 "negative_too_small",
6867 -std::ldexp(1.0f, -32),
6868 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6869 }, // too small negative
6873 -std::ldexp(1.0f, 128),
6875 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6876 "%inf = OpIsInf %bool %c\n"
6877 "%cond = OpLogicalAnd %bool %gz %inf\n"
6882 std::ldexp(1.0f, 128),
6884 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6885 "%inf = OpIsInf %bool %c\n"
6886 "%cond = OpLogicalAnd %bool %gz %inf\n"
6889 "round_to_negative_inf",
6891 -std::ldexp(1.0f, 32),
6893 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6894 "%inf = OpIsInf %bool %c\n"
6895 "%cond = OpLogicalAnd %bool %gz %inf\n"
6900 std::ldexp(1.0f, 16),
6902 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6903 "%inf = OpIsInf %bool %c\n"
6904 "%cond = OpLogicalAnd %bool %gz %inf\n"
6909 std::numeric_limits<float>::quiet_NaN(),
6911 // Test for any NaN value, as NaNs are not preserved
6912 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6913 "%cond = OpIsNan %bool %direct_quant\n"
6918 std::numeric_limits<float>::quiet_NaN(),
6920 // Test for any NaN value, as NaNs are not preserved
6921 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6922 "%cond = OpIsNan %bool %direct_quant\n"
6925 const char* constants =
6926 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
6928 StringTemplate function (
6929 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6930 "%param1 = OpFunctionParameter %v4f32\n"
6931 "%label_testfun = OpLabel\n"
6932 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6933 "%b = OpFAdd %f32 %test_constant %a\n"
6934 "%c = OpQuantizeToF16 %f32 %b\n"
6936 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6937 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
6938 " OpReturnValue %retval\n"
6942 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
6943 const char* specConstants =
6944 "%test_constant = OpSpecConstant %f32 0.\n"
6945 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
6947 StringTemplate specConstantFunction(
6948 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6949 "%param1 = OpFunctionParameter %v4f32\n"
6950 "%label_testfun = OpLabel\n"
6952 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6953 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
6954 " OpReturnValue %retval\n"
6958 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6960 map<string, string> codeSpecialization;
6961 map<string, string> fragments;
6962 codeSpecialization["condition"] = tests[idx].condition;
6963 fragments["testfun"] = function.specialize(codeSpecialization);
6964 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
6965 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6968 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6970 map<string, string> codeSpecialization;
6971 map<string, string> fragments;
6972 vector<deInt32> passConstants;
6973 deInt32 specConstant;
6975 codeSpecialization["condition"] = tests[idx].condition;
6976 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
6977 fragments["decoration"] = specDecorations;
6978 fragments["pre_main"] = specConstants;
6980 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
6981 passConstants.push_back(specConstant);
6983 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6987 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
6989 RGBA inputColors[4] = {
6991 RGBA(0, 0, 255, 255),
6992 RGBA(0, 255, 0, 255),
6993 RGBA(0, 255, 255, 255)
6996 RGBA expectedColors[4] =
6998 RGBA(255, 0, 0, 255),
6999 RGBA(255, 0, 0, 255),
7000 RGBA(255, 0, 0, 255),
7001 RGBA(255, 0, 0, 255)
7004 struct DualFP16Possibility
7009 const char* possibleOutput1;
7010 const char* possibleOutput2;
7013 "positive_round_up_or_round_down",
7015 constructNormalizedFloat(8, 0x300300),
7020 "negative_round_up_or_round_down",
7022 -constructNormalizedFloat(-7, 0x600800),
7029 constructNormalizedFloat(2, 0x01e000),
7034 "carry_to_exponent",
7036 constructNormalizedFloat(1, 0xffe000),
7041 StringTemplate constants (
7042 "%input_const = OpConstant %f32 ${input}\n"
7043 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7044 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7047 StringTemplate specConstants (
7048 "%input_const = OpSpecConstant %f32 0.\n"
7049 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7050 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7053 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
7055 const char* function =
7056 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7057 "%param1 = OpFunctionParameter %v4f32\n"
7058 "%label_testfun = OpLabel\n"
7059 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7060 // For the purposes of this test we assume that 0.f will always get
7061 // faithfully passed through the pipeline stages.
7062 "%b = OpFAdd %f32 %input_const %a\n"
7063 "%c = OpQuantizeToF16 %f32 %b\n"
7064 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
7065 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
7066 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
7067 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7068 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
7069 " OpReturnValue %retval\n"
7072 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7073 map<string, string> fragments;
7074 map<string, string> constantSpecialization;
7076 constantSpecialization["input"] = tests[idx].input;
7077 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7078 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7079 fragments["testfun"] = function;
7080 fragments["pre_main"] = constants.specialize(constantSpecialization);
7081 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7084 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7085 map<string, string> fragments;
7086 map<string, string> constantSpecialization;
7087 vector<deInt32> passConstants;
7088 deInt32 specConstant;
7090 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7091 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7092 fragments["testfun"] = function;
7093 fragments["decoration"] = specDecorations;
7094 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
7096 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
7097 passConstants.push_back(specConstant);
7099 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7103 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
7105 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
7106 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
7107 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
7108 return opQuantizeTests.release();
7111 struct ShaderPermutation
7113 deUint8 vertexPermutation;
7114 deUint8 geometryPermutation;
7115 deUint8 tesscPermutation;
7116 deUint8 tessePermutation;
7117 deUint8 fragmentPermutation;
7120 ShaderPermutation getShaderPermutation(deUint8 inputValue)
7122 ShaderPermutation permutation =
7124 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
7125 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
7126 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
7127 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
7128 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
7133 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
7135 RGBA defaultColors[4];
7136 RGBA invertedColors[4];
7137 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
7139 const ShaderElement combinedPipeline[] =
7141 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
7142 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
7143 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7144 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7145 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
7148 getDefaultColors(defaultColors);
7149 getInvertedDefaultColors(invertedColors);
7150 addFunctionCaseWithPrograms<InstanceContext>(
7151 moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline,
7152 createInstanceContext(combinedPipeline, map<string, string>()));
7154 const char* numbers[] =
7159 for (deInt8 idx = 0; idx < 32; ++idx)
7161 ShaderPermutation permutation = getShaderPermutation(idx);
7162 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
7163 const ShaderElement pipeline[] =
7165 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
7166 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
7167 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7168 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7169 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
7172 // If there are an even number of swaps, then it should be no-op.
7173 // If there are an odd number, the color should be flipped.
7174 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
7176 addFunctionCaseWithPrograms<InstanceContext>(
7177 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
7178 createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
7182 addFunctionCaseWithPrograms<InstanceContext>(
7183 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
7184 createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
7187 return moduleTests.release();
7190 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
7192 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
7193 RGBA defaultColors[4];
7194 getDefaultColors(defaultColors);
7195 map<string, string> fragments;
7196 fragments["pre_main"] =
7197 "%c_f32_5 = OpConstant %f32 5.\n";
7199 // A loop with a single block. The Continue Target is the loop block
7200 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
7201 // -- the "continue construct" forms the entire loop.
7202 fragments["testfun"] =
7203 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7204 "%param1 = OpFunctionParameter %v4f32\n"
7206 "%entry = OpLabel\n"
7207 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7210 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7212 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7213 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
7214 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7215 "%val = OpFAdd %f32 %val1 %delta\n"
7216 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
7217 "%count__ = OpISub %i32 %count %c_i32_1\n"
7218 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7219 "OpLoopMerge %exit %loop None\n"
7220 "OpBranchConditional %again %loop %exit\n"
7223 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7224 "OpReturnValue %result\n"
7228 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
7230 // Body comprised of multiple basic blocks.
7231 const StringTemplate multiBlock(
7232 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7233 "%param1 = OpFunctionParameter %v4f32\n"
7235 "%entry = OpLabel\n"
7236 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7239 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7241 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
7242 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
7243 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
7244 // There are several possibilities for the Continue Target below. Each
7245 // will be specialized into a separate test case.
7246 "OpLoopMerge %exit ${continue_target} None\n"
7250 ";delta_next = (delta > 0) ? -1 : 1;\n"
7251 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
7252 "OpSelectionMerge %gather DontFlatten\n"
7253 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
7256 "OpBranch %gather\n"
7259 "OpBranch %gather\n"
7261 "%gather = OpLabel\n"
7262 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
7263 "%val = OpFAdd %f32 %val1 %delta\n"
7264 "%count__ = OpISub %i32 %count %c_i32_1\n"
7265 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7266 "OpBranchConditional %again %loop %exit\n"
7269 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7270 "OpReturnValue %result\n"
7274 map<string, string> continue_target;
7276 // The Continue Target is the loop block itself.
7277 continue_target["continue_target"] = "%loop";
7278 fragments["testfun"] = multiBlock.specialize(continue_target);
7279 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
7281 // The Continue Target is at the end of the loop.
7282 continue_target["continue_target"] = "%gather";
7283 fragments["testfun"] = multiBlock.specialize(continue_target);
7284 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
7286 // A loop with continue statement.
7287 fragments["testfun"] =
7288 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7289 "%param1 = OpFunctionParameter %v4f32\n"
7291 "%entry = OpLabel\n"
7292 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7295 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
7297 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7298 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
7299 "OpLoopMerge %exit %continue None\n"
7303 ";skip if %count==2\n"
7304 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
7305 "OpSelectionMerge %continue DontFlatten\n"
7306 "OpBranchConditional %eq2 %continue %body\n"
7309 "%fcount = OpConvertSToF %f32 %count\n"
7310 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7311 "OpBranch %continue\n"
7313 "%continue = OpLabel\n"
7314 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
7315 "%count__ = OpISub %i32 %count %c_i32_1\n"
7316 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7317 "OpBranchConditional %again %loop %exit\n"
7320 "%same = OpFSub %f32 %val %c_f32_8\n"
7321 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7322 "OpReturnValue %result\n"
7324 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
7326 // A loop with break.
7327 fragments["testfun"] =
7328 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7329 "%param1 = OpFunctionParameter %v4f32\n"
7331 "%entry = OpLabel\n"
7332 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7333 "%dot = OpDot %f32 %param1 %param1\n"
7334 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7335 "%zero = OpConvertFToU %u32 %div\n"
7336 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7337 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7340 ";adds 4 and 3 to %val0 (exits early)\n"
7342 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7343 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7344 "OpLoopMerge %exit %continue None\n"
7348 ";end loop if %count==%two\n"
7349 "%above2 = OpSGreaterThan %bool %count %two\n"
7350 "OpSelectionMerge %continue DontFlatten\n"
7351 "OpBranchConditional %above2 %body %exit\n"
7354 "%fcount = OpConvertSToF %f32 %count\n"
7355 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7356 "OpBranch %continue\n"
7358 "%continue = OpLabel\n"
7359 "%count__ = OpISub %i32 %count %c_i32_1\n"
7360 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7361 "OpBranchConditional %again %loop %exit\n"
7364 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
7365 "%same = OpFSub %f32 %val_post %c_f32_7\n"
7366 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7367 "OpReturnValue %result\n"
7369 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
7371 // A loop with return.
7372 fragments["testfun"] =
7373 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7374 "%param1 = OpFunctionParameter %v4f32\n"
7376 "%entry = OpLabel\n"
7377 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7378 "%dot = OpDot %f32 %param1 %param1\n"
7379 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7380 "%zero = OpConvertFToU %u32 %div\n"
7381 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7382 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7385 ";returns early without modifying %param1\n"
7387 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7388 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7389 "OpLoopMerge %exit %continue None\n"
7393 ";return if %count==%two\n"
7394 "%above2 = OpSGreaterThan %bool %count %two\n"
7395 "OpSelectionMerge %continue DontFlatten\n"
7396 "OpBranchConditional %above2 %body %early_exit\n"
7398 "%early_exit = OpLabel\n"
7399 "OpReturnValue %param1\n"
7402 "%fcount = OpConvertSToF %f32 %count\n"
7403 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7404 "OpBranch %continue\n"
7406 "%continue = OpLabel\n"
7407 "%count__ = OpISub %i32 %count %c_i32_1\n"
7408 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7409 "OpBranchConditional %again %loop %exit\n"
7412 ";should never get here, so return an incorrect result\n"
7413 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
7414 "OpReturnValue %result\n"
7416 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
7418 // Continue inside a switch block to break to enclosing loop's merge block.
7419 // Matches roughly the following GLSL code:
7420 // for (; keep_going; keep_going = false)
7422 // switch (int(param1.x))
7424 // case 0: continue;
7425 // case 1: continue;
7426 // default: continue;
7428 // dead code: modify return value to invalid result.
7430 fragments["pre_main"] =
7431 "%fp_bool = OpTypePointer Function %bool\n"
7432 "%true = OpConstantTrue %bool\n"
7433 "%false = OpConstantFalse %bool\n";
7435 fragments["testfun"] =
7436 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7437 "%param1 = OpFunctionParameter %v4f32\n"
7439 "%entry = OpLabel\n"
7440 "%keep_going = OpVariable %fp_bool Function\n"
7441 "%val_ptr = OpVariable %fp_f32 Function\n"
7442 "%param1_x = OpCompositeExtract %f32 %param1 0\n"
7443 "OpStore %keep_going %true\n"
7444 "OpBranch %forloop_begin\n"
7446 "%forloop_begin = OpLabel\n"
7447 "OpLoopMerge %forloop_merge %forloop_continue None\n"
7448 "OpBranch %forloop\n"
7450 "%forloop = OpLabel\n"
7451 "%for_condition = OpLoad %bool %keep_going\n"
7452 "OpBranchConditional %for_condition %forloop_body %forloop_merge\n"
7454 "%forloop_body = OpLabel\n"
7455 "OpStore %val_ptr %param1_x\n"
7456 "%param1_x_int = OpConvertFToS %i32 %param1_x\n"
7458 "OpSelectionMerge %switch_merge None\n"
7459 "OpSwitch %param1_x_int %default 0 %case_0 1 %case_1\n"
7460 "%case_0 = OpLabel\n"
7461 "OpBranch %forloop_continue\n"
7462 "%case_1 = OpLabel\n"
7463 "OpBranch %forloop_continue\n"
7464 "%default = OpLabel\n"
7465 "OpBranch %forloop_continue\n"
7466 "%switch_merge = OpLabel\n"
7467 ";should never get here, so change the return value to invalid result\n"
7468 "OpStore %val_ptr %c_f32_1\n"
7469 "OpBranch %forloop_continue\n"
7471 "%forloop_continue = OpLabel\n"
7472 "OpStore %keep_going %false\n"
7473 "OpBranch %forloop_begin\n"
7474 "%forloop_merge = OpLabel\n"
7476 "%val = OpLoad %f32 %val_ptr\n"
7477 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7478 "OpReturnValue %result\n"
7480 createTestsForAllStages("switch_continue", defaultColors, defaultColors, fragments, testGroup.get());
7482 return testGroup.release();
7485 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
7486 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
7488 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
7489 map<string, string> fragments;
7491 // A barrier inside a function body.
7492 fragments["pre_main"] =
7493 "%Workgroup = OpConstant %i32 2\n"
7494 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
7495 fragments["testfun"] =
7496 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7497 "%param1 = OpFunctionParameter %v4f32\n"
7498 "%label_testfun = OpLabel\n"
7499 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7500 "OpReturnValue %param1\n"
7502 addTessCtrlTest(testGroup.get(), "in_function", fragments);
7504 // Common setup code for the following tests.
7505 fragments["pre_main"] =
7506 "%Workgroup = OpConstant %i32 2\n"
7507 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7508 "%c_f32_5 = OpConstant %f32 5.\n";
7509 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
7510 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7511 "%param1 = OpFunctionParameter %v4f32\n"
7512 "%entry = OpLabel\n"
7513 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7514 "%dot = OpDot %f32 %param1 %param1\n"
7515 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7516 "%zero = OpConvertFToU %u32 %div\n";
7518 // Barriers inside OpSwitch branches.
7519 fragments["testfun"] =
7521 "OpSelectionMerge %switch_exit None\n"
7522 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
7524 "%case1 = OpLabel\n"
7525 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7526 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7527 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7528 "OpBranch %switch_exit\n"
7530 "%switch_default = OpLabel\n"
7531 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7532 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7533 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7534 "OpBranch %switch_exit\n"
7536 "%case0 = OpLabel\n"
7537 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7538 "OpBranch %switch_exit\n"
7540 "%switch_exit = OpLabel\n"
7541 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
7542 "OpReturnValue %ret\n"
7544 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
7546 // Barriers inside if-then-else.
7547 fragments["testfun"] =
7549 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
7550 "OpSelectionMerge %exit DontFlatten\n"
7551 "OpBranchConditional %eq0 %then %else\n"
7554 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7555 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7556 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7560 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7564 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
7565 "OpReturnValue %ret\n"
7567 addTessCtrlTest(testGroup.get(), "in_if", fragments);
7569 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
7570 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
7571 fragments["testfun"] =
7573 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
7574 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
7575 "OpSelectionMerge %exit DontFlatten\n"
7576 "OpBranchConditional %thread0 %then %else\n"
7579 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7583 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
7587 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
7588 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7589 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
7590 "OpReturnValue %ret\n"
7592 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
7594 // A barrier inside a loop.
7595 fragments["pre_main"] =
7596 "%Workgroup = OpConstant %i32 2\n"
7597 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7598 "%c_f32_10 = OpConstant %f32 10.\n";
7599 fragments["testfun"] =
7600 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7601 "%param1 = OpFunctionParameter %v4f32\n"
7602 "%entry = OpLabel\n"
7603 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7606 ";adds 4, 3, 2, and 1 to %val0\n"
7608 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7609 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7610 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7611 "%fcount = OpConvertSToF %f32 %count\n"
7612 "%val = OpFAdd %f32 %val1 %fcount\n"
7613 "%count__ = OpISub %i32 %count %c_i32_1\n"
7614 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7615 "OpLoopMerge %exit %loop None\n"
7616 "OpBranchConditional %again %loop %exit\n"
7619 "%same = OpFSub %f32 %val %c_f32_10\n"
7620 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7621 "OpReturnValue %ret\n"
7623 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
7625 return testGroup.release();
7628 // Test for the OpFRem instruction.
7629 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
7631 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
7632 map<string, string> fragments;
7633 RGBA inputColors[4];
7634 RGBA outputColors[4];
7636 fragments["pre_main"] =
7637 "%c_f32_3 = OpConstant %f32 3.0\n"
7638 "%c_f32_n3 = OpConstant %f32 -3.0\n"
7639 "%c_f32_4 = OpConstant %f32 4.0\n"
7640 "%c_f32_p75 = OpConstant %f32 0.75\n"
7641 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
7642 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
7643 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
7645 // The test does the following.
7646 // vec4 result = (param1 * 8.0) - 4.0;
7647 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
7648 fragments["testfun"] =
7649 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7650 "%param1 = OpFunctionParameter %v4f32\n"
7651 "%label_testfun = OpLabel\n"
7652 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
7653 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
7654 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
7655 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
7656 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
7657 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
7658 "OpReturnValue %xy_0_1\n"
7662 inputColors[0] = RGBA(16, 16, 0, 255);
7663 inputColors[1] = RGBA(232, 232, 0, 255);
7664 inputColors[2] = RGBA(232, 16, 0, 255);
7665 inputColors[3] = RGBA(16, 232, 0, 255);
7667 outputColors[0] = RGBA(64, 64, 0, 255);
7668 outputColors[1] = RGBA(255, 255, 0, 255);
7669 outputColors[2] = RGBA(255, 64, 0, 255);
7670 outputColors[3] = RGBA(64, 255, 0, 255);
7672 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
7673 return testGroup.release();
7676 // Test for the OpSRem instruction.
7677 tcu::TestCaseGroup* createOpSRemGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
7679 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "srem", "OpSRem"));
7680 map<string, string> fragments;
7682 fragments["pre_main"] =
7683 "%c_f32_255 = OpConstant %f32 255.0\n"
7684 "%c_i32_128 = OpConstant %i32 128\n"
7685 "%c_i32_255 = OpConstant %i32 255\n"
7686 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
7687 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
7688 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
7690 // The test does the following.
7691 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
7692 // ivec4 result = ivec4(srem(ints.x, ints.y), srem(ints.y, ints.z), srem(ints.z, ints.x), 255);
7693 // return float(result + 128) / 255.0;
7694 fragments["testfun"] =
7695 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7696 "%param1 = OpFunctionParameter %v4f32\n"
7697 "%label_testfun = OpLabel\n"
7698 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
7699 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
7700 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
7701 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
7702 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
7703 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
7704 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
7705 "%x_out = OpSRem %i32 %x_in %y_in\n"
7706 "%y_out = OpSRem %i32 %y_in %z_in\n"
7707 "%z_out = OpSRem %i32 %z_in %x_in\n"
7708 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
7709 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
7710 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
7711 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
7712 "OpReturnValue %float_out\n"
7715 const struct CaseParams
7718 const char* failMessageTemplate; // customized status message
7719 qpTestResult failResult; // override status on failure
7720 int operands[4][3]; // four (x, y, z) vectors of operands
7721 int results[4][3]; // four (x, y, z) vectors of results
7727 QP_TEST_RESULT_FAIL,
7728 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
7729 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
7733 "Inconsistent results, but within specification: ${reason}",
7734 negFailResult, // negative operands, not required by the spec
7735 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
7736 { { 5, 12, -2 }, { 0, -5, 2 }, { 3, 8, -6 }, { 25, -60, 0 } }, // results
7739 // If either operand is negative the result is undefined. Some implementations may still return correct values.
7741 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
7743 const CaseParams& params = cases[caseNdx];
7744 RGBA inputColors[4];
7745 RGBA outputColors[4];
7747 for (int i = 0; i < 4; ++i)
7749 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
7750 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
7753 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
7756 return testGroup.release();
7759 // Test for the OpSMod instruction.
7760 tcu::TestCaseGroup* createOpSModGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
7762 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "smod", "OpSMod"));
7763 map<string, string> fragments;
7765 fragments["pre_main"] =
7766 "%c_f32_255 = OpConstant %f32 255.0\n"
7767 "%c_i32_128 = OpConstant %i32 128\n"
7768 "%c_i32_255 = OpConstant %i32 255\n"
7769 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
7770 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
7771 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
7773 // The test does the following.
7774 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
7775 // ivec4 result = ivec4(smod(ints.x, ints.y), smod(ints.y, ints.z), smod(ints.z, ints.x), 255);
7776 // return float(result + 128) / 255.0;
7777 fragments["testfun"] =
7778 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7779 "%param1 = OpFunctionParameter %v4f32\n"
7780 "%label_testfun = OpLabel\n"
7781 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
7782 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
7783 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
7784 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
7785 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
7786 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
7787 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
7788 "%x_out = OpSMod %i32 %x_in %y_in\n"
7789 "%y_out = OpSMod %i32 %y_in %z_in\n"
7790 "%z_out = OpSMod %i32 %z_in %x_in\n"
7791 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
7792 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
7793 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
7794 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
7795 "OpReturnValue %float_out\n"
7798 const struct CaseParams
7801 const char* failMessageTemplate; // customized status message
7802 qpTestResult failResult; // override status on failure
7803 int operands[4][3]; // four (x, y, z) vectors of operands
7804 int results[4][3]; // four (x, y, z) vectors of results
7810 QP_TEST_RESULT_FAIL,
7811 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
7812 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
7816 "Inconsistent results, but within specification: ${reason}",
7817 negFailResult, // negative operands, not required by the spec
7818 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
7819 { { 5, -5, 3 }, { 0, 2, -3 }, { 3, -73, 69 }, { -35, 40, 0 } }, // results
7822 // If either operand is negative the result is undefined. Some implementations may still return correct values.
7824 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
7826 const CaseParams& params = cases[caseNdx];
7827 RGBA inputColors[4];
7828 RGBA outputColors[4];
7830 for (int i = 0; i < 4; ++i)
7832 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
7833 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
7836 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
7838 return testGroup.release();
7844 INTEGER_TYPE_SIGNED_16,
7845 INTEGER_TYPE_SIGNED_32,
7846 INTEGER_TYPE_SIGNED_64,
7848 INTEGER_TYPE_UNSIGNED_16,
7849 INTEGER_TYPE_UNSIGNED_32,
7850 INTEGER_TYPE_UNSIGNED_64,
7853 const string getBitWidthStr (IntegerType type)
7857 case INTEGER_TYPE_SIGNED_16:
7858 case INTEGER_TYPE_UNSIGNED_16: return "16";
7860 case INTEGER_TYPE_SIGNED_32:
7861 case INTEGER_TYPE_UNSIGNED_32: return "32";
7863 case INTEGER_TYPE_SIGNED_64:
7864 case INTEGER_TYPE_UNSIGNED_64: return "64";
7866 default: DE_ASSERT(false);
7871 const string getByteWidthStr (IntegerType type)
7875 case INTEGER_TYPE_SIGNED_16:
7876 case INTEGER_TYPE_UNSIGNED_16: return "2";
7878 case INTEGER_TYPE_SIGNED_32:
7879 case INTEGER_TYPE_UNSIGNED_32: return "4";
7881 case INTEGER_TYPE_SIGNED_64:
7882 case INTEGER_TYPE_UNSIGNED_64: return "8";
7884 default: DE_ASSERT(false);
7889 bool isSigned (IntegerType type)
7891 return (type <= INTEGER_TYPE_SIGNED_64);
7894 const string getTypeName (IntegerType type)
7896 string prefix = isSigned(type) ? "" : "u";
7897 return prefix + "int" + getBitWidthStr(type);
7900 const string getTestName (IntegerType from, IntegerType to)
7902 return getTypeName(from) + "_to_" + getTypeName(to);
7905 const string getAsmTypeDeclaration (IntegerType type)
7907 string sign = isSigned(type) ? " 1" : " 0";
7908 return "OpTypeInt " + getBitWidthStr(type) + sign;
7911 const string getAsmTypeName (IntegerType type)
7913 const string prefix = isSigned(type) ? "%i" : "%u";
7914 return prefix + getBitWidthStr(type);
7917 template<typename T>
7918 BufferSp getSpecializedBuffer (deInt64 number)
7920 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
7923 BufferSp getBuffer (IntegerType type, deInt64 number)
7927 case INTEGER_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
7928 case INTEGER_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
7929 case INTEGER_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
7931 case INTEGER_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
7932 case INTEGER_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
7933 case INTEGER_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
7935 default: DE_ASSERT(false);
7936 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
7940 bool usesInt16 (IntegerType from, IntegerType to)
7942 return (from == INTEGER_TYPE_SIGNED_16 || from == INTEGER_TYPE_UNSIGNED_16
7943 || to == INTEGER_TYPE_SIGNED_16 || to == INTEGER_TYPE_UNSIGNED_16);
7946 bool usesInt64 (IntegerType from, IntegerType to)
7948 return (from == INTEGER_TYPE_SIGNED_64 || from == INTEGER_TYPE_UNSIGNED_64
7949 || to == INTEGER_TYPE_SIGNED_64 || to == INTEGER_TYPE_UNSIGNED_64);
7952 ComputeTestFeatures getConversionUsedFeatures (IntegerType from, IntegerType to)
7954 if (usesInt16(from, to))
7956 if (usesInt64(from, to))
7958 return COMPUTE_TEST_USES_INT16_INT64;
7962 return COMPUTE_TEST_USES_INT16;
7967 return COMPUTE_TEST_USES_INT64;
7973 ConvertCase (IntegerType from, IntegerType to, deInt64 number)
7976 , m_features (getConversionUsedFeatures(from, to))
7977 , m_name (getTestName(from, to))
7978 , m_inputBuffer (getBuffer(from, number))
7979 , m_outputBuffer (getBuffer(to, number))
7981 m_asmTypes["inputType"] = getAsmTypeName(from);
7982 m_asmTypes["outputType"] = getAsmTypeName(to);
7984 if (m_features == COMPUTE_TEST_USES_INT16)
7986 m_asmTypes["int_capabilities"] = "OpCapability Int16\n"
7987 "OpCapability StorageUniformBufferBlock16\n";
7988 m_asmTypes["int_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
7989 "%u16 = OpTypeInt 16 0\n";
7990 m_asmTypes["int_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
7992 else if (m_features == COMPUTE_TEST_USES_INT64)
7994 m_asmTypes["int_capabilities"] = "OpCapability Int64\n";
7995 m_asmTypes["int_additional_decl"] = "%i64 = OpTypeInt 64 1\n"
7996 "%u64 = OpTypeInt 64 0\n";
7997 m_asmTypes["int_extensions"] = "";
7999 else if (m_features == COMPUTE_TEST_USES_INT16_INT64)
8001 m_asmTypes["int_capabilities"] = "OpCapability Int16\n"
8002 "OpCapability StorageUniformBufferBlock16\n"
8003 "OpCapability Int64\n";
8004 m_asmTypes["int_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
8005 "%u16 = OpTypeInt 16 0\n"
8006 "%i64 = OpTypeInt 64 1\n"
8007 "%u64 = OpTypeInt 64 0\n";
8008 m_asmTypes["int_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
8016 IntegerType m_fromType;
8017 IntegerType m_toType;
8018 ComputeTestFeatures m_features;
8020 map<string, string> m_asmTypes;
8021 BufferSp m_inputBuffer;
8022 BufferSp m_outputBuffer;
8025 const string getConvertCaseShaderStr (const string& instruction, const ConvertCase& convertCase)
8027 map<string, string> params = convertCase.m_asmTypes;
8029 params["instruction"] = instruction;
8031 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
8032 params["outDecorator"] = getByteWidthStr(convertCase.m_toType);
8034 const StringTemplate shader (
8035 "OpCapability Shader\n"
8036 "${int_capabilities}"
8038 "OpMemoryModel Logical GLSL450\n"
8039 "OpEntryPoint GLCompute %main \"main\" %id\n"
8040 "OpExecutionMode %main LocalSize 1 1 1\n"
8041 "OpSource GLSL 430\n"
8042 "OpName %main \"main\"\n"
8043 "OpName %id \"gl_GlobalInvocationID\"\n"
8045 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8046 "OpDecorate %indata DescriptorSet 0\n"
8047 "OpDecorate %indata Binding 0\n"
8048 "OpDecorate %outdata DescriptorSet 0\n"
8049 "OpDecorate %outdata Binding 1\n"
8050 "OpDecorate %in_arr ArrayStride ${inDecorator}\n"
8051 "OpDecorate %out_arr ArrayStride ${outDecorator}\n"
8052 "OpDecorate %in_buf BufferBlock\n"
8053 "OpDecorate %out_buf BufferBlock\n"
8054 "OpMemberDecorate %in_buf 0 Offset 0\n"
8055 "OpMemberDecorate %out_buf 0 Offset 0\n"
8057 "%void = OpTypeVoid\n"
8058 "%voidf = OpTypeFunction %void\n"
8059 "%u32 = OpTypeInt 32 0\n"
8060 "%i32 = OpTypeInt 32 1\n"
8061 "${int_additional_decl}"
8062 "%uvec3 = OpTypeVector %u32 3\n"
8063 "%uvec3ptr = OpTypePointer Input %uvec3\n"
8065 "%in_ptr = OpTypePointer Uniform ${inputType}\n"
8066 "%out_ptr = OpTypePointer Uniform ${outputType}\n"
8067 "%in_arr = OpTypeRuntimeArray ${inputType}\n"
8068 "%out_arr = OpTypeRuntimeArray ${outputType}\n"
8069 "%in_buf = OpTypeStruct %in_arr\n"
8070 "%out_buf = OpTypeStruct %out_arr\n"
8071 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
8072 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
8073 "%indata = OpVariable %in_bufptr Uniform\n"
8074 "%outdata = OpVariable %out_bufptr Uniform\n"
8075 "%inputptr = OpTypePointer Input ${inputType}\n"
8076 "%id = OpVariable %uvec3ptr Input\n"
8078 "%zero = OpConstant %i32 0\n"
8080 "%main = OpFunction %void None %voidf\n"
8081 "%label = OpLabel\n"
8082 "%idval = OpLoad %uvec3 %id\n"
8083 "%x = OpCompositeExtract %u32 %idval 0\n"
8084 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
8085 "%outloc = OpAccessChain %out_ptr %outdata %zero %x\n"
8086 "%inval = OpLoad ${inputType} %inloc\n"
8087 "%conv = ${instruction} ${outputType} %inval\n"
8088 " OpStore %outloc %conv\n"
8093 return shader.specialize(params);
8096 void createSConvertCases (vector<ConvertCase>& testCases)
8098 // Convert int to int
8099 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_32, 14669));
8100 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_64, 3341));
8102 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_SIGNED_64, 973610259));
8104 // Convert int to unsigned int
8105 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_32, 9288));
8106 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_64, 15460));
8108 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_UNSIGNED_64, 346213461));
8111 // Test for the OpSConvert instruction.
8112 tcu::TestCaseGroup* createSConvertTests (tcu::TestContext& testCtx)
8114 const string instruction ("OpSConvert");
8115 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "sconvert", "OpSConvert"));
8116 vector<ConvertCase> testCases;
8117 createSConvertCases(testCases);
8119 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8121 ComputeShaderSpec spec;
8123 spec.assembly = getConvertCaseShaderStr(instruction, *test);
8124 spec.inputs.push_back(test->m_inputBuffer);
8125 spec.outputs.push_back(test->m_outputBuffer);
8126 spec.numWorkGroups = IVec3(1, 1, 1);
8128 if (test->m_features == COMPUTE_TEST_USES_INT16 || test->m_features == COMPUTE_TEST_USES_INT16_INT64)
8130 spec.extensions.push_back("VK_KHR_16bit_storage");
8131 spec.requestedVulkanFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_UNIFORM_BUFFER_BLOCK;
8134 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpSConvert.", spec, test->m_features));
8137 return group.release();
8140 void createUConvertCases (vector<ConvertCase>& testCases)
8142 // Convert unsigned int to unsigned int
8143 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_32, 60653));
8144 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_64, 17991));
8146 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_UNSIGNED_64, 904256275));
8149 // Test for the OpUConvert instruction.
8150 tcu::TestCaseGroup* createUConvertTests (tcu::TestContext& testCtx)
8152 const string instruction ("OpUConvert");
8153 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "uconvert", "OpUConvert"));
8154 vector<ConvertCase> testCases;
8155 createUConvertCases(testCases);
8157 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8159 ComputeShaderSpec spec;
8161 spec.assembly = getConvertCaseShaderStr(instruction, *test);
8162 spec.inputs.push_back(test->m_inputBuffer);
8163 spec.outputs.push_back(test->m_outputBuffer);
8164 spec.numWorkGroups = IVec3(1, 1, 1);
8166 if (test->m_features == COMPUTE_TEST_USES_INT16 || test->m_features == COMPUTE_TEST_USES_INT16_INT64)
8168 spec.extensions.push_back("VK_KHR_16bit_storage");
8169 spec.requestedVulkanFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_UNIFORM_BUFFER_BLOCK;
8172 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpUConvert.", spec, test->m_features));
8174 return group.release();
8177 const string getNumberTypeName (const NumberType type)
8179 if (type == NUMBERTYPE_INT32)
8183 else if (type == NUMBERTYPE_UINT32)
8187 else if (type == NUMBERTYPE_FLOAT32)
8198 deInt32 getInt(de::Random& rnd)
8200 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
8203 const string repeatString (const string& str, int times)
8206 for (int i = 0; i < times; ++i)
8213 const string getRandomConstantString (const NumberType type, de::Random& rnd)
8215 if (type == NUMBERTYPE_INT32)
8217 return numberToString<deInt32>(getInt(rnd));
8219 else if (type == NUMBERTYPE_UINT32)
8221 return numberToString<deUint32>(rnd.getUint32());
8223 else if (type == NUMBERTYPE_FLOAT32)
8225 return numberToString<float>(rnd.getFloat());
8234 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8236 map<string, string> params;
8239 for (int width = 2; width <= 4; ++width)
8241 const string randomConst = numberToString(getInt(rnd));
8242 const string widthStr = numberToString(width);
8243 const string composite_type = "${customType}vec" + widthStr;
8244 const int index = rnd.getInt(0, width-1);
8246 params["type"] = "vec";
8247 params["name"] = params["type"] + "_" + widthStr;
8248 params["compositeDecl"] = composite_type + " = OpTypeVector ${customType} " + widthStr +"\n";
8249 params["compositeType"] = composite_type;
8250 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8251 params["compositeConstruct"] = "%instance = OpCompositeConstruct " + composite_type + repeatString(" %filler", width) + "\n";
8252 params["indexes"] = numberToString(index);
8253 testCases.push_back(params);
8257 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8259 const int limit = 10;
8260 map<string, string> params;
8262 for (int width = 2; width <= limit; ++width)
8264 string randomConst = numberToString(getInt(rnd));
8265 string widthStr = numberToString(width);
8266 int index = rnd.getInt(0, width-1);
8268 params["type"] = "array";
8269 params["name"] = params["type"] + "_" + widthStr;
8270 params["compositeDecl"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
8271 + "%composite = OpTypeArray ${customType} %arraywidth\n";
8272 params["compositeType"] = "%composite";
8273 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8274 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8275 params["indexes"] = numberToString(index);
8276 testCases.push_back(params);
8280 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8282 const int limit = 10;
8283 map<string, string> params;
8285 for (int width = 2; width <= limit; ++width)
8287 string randomConst = numberToString(getInt(rnd));
8288 int index = rnd.getInt(0, width-1);
8290 params["type"] = "struct";
8291 params["name"] = params["type"] + "_" + numberToString(width);
8292 params["compositeDecl"] = "%composite = OpTypeStruct" + repeatString(" ${customType}", width) + "\n";
8293 params["compositeType"] = "%composite";
8294 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8295 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8296 params["indexes"] = numberToString(index);
8297 testCases.push_back(params);
8301 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8303 map<string, string> params;
8306 for (int width = 2; width <= 4; ++width)
8308 string widthStr = numberToString(width);
8310 for (int column = 2 ; column <= 4; ++column)
8312 int index_0 = rnd.getInt(0, column-1);
8313 int index_1 = rnd.getInt(0, width-1);
8314 string columnStr = numberToString(column);
8316 params["type"] = "matrix";
8317 params["name"] = params["type"] + "_" + widthStr + "x" + columnStr;
8318 params["compositeDecl"] = string("%vectype = OpTypeVector ${customType} " + widthStr + "\n")
8319 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
8320 params["compositeType"] = "%composite";
8322 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n"
8323 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
8325 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
8326 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
8327 testCases.push_back(params);
8332 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8334 createVectorCompositeCases(testCases, rnd, type);
8335 createArrayCompositeCases(testCases, rnd, type);
8336 createStructCompositeCases(testCases, rnd, type);
8337 // Matrix only supports float types
8338 if (type == NUMBERTYPE_FLOAT32)
8340 createMatrixCompositeCases(testCases, rnd, type);
8344 const string getAssemblyTypeDeclaration (const NumberType type)
8348 case NUMBERTYPE_INT32: return "OpTypeInt 32 1";
8349 case NUMBERTYPE_UINT32: return "OpTypeInt 32 0";
8350 case NUMBERTYPE_FLOAT32: return "OpTypeFloat 32";
8351 default: DE_ASSERT(false); return "";
8355 const string getAssemblyTypeName (const NumberType type)
8359 case NUMBERTYPE_INT32: return "%i32";
8360 case NUMBERTYPE_UINT32: return "%u32";
8361 case NUMBERTYPE_FLOAT32: return "%f32";
8362 default: DE_ASSERT(false); return "";
8366 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
8368 map<string, string> parameters(params);
8370 const string customType = getAssemblyTypeName(type);
8371 map<string, string> substCustomType;
8372 substCustomType["customType"] = customType;
8373 parameters["compositeDecl"] = StringTemplate(parameters.at("compositeDecl")).specialize(substCustomType);
8374 parameters["compositeType"] = StringTemplate(parameters.at("compositeType")).specialize(substCustomType);
8375 parameters["compositeConstruct"] = StringTemplate(parameters.at("compositeConstruct")).specialize(substCustomType);
8376 parameters["filler"] = StringTemplate(parameters.at("filler")).specialize(substCustomType);
8377 parameters["customType"] = customType;
8378 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
8380 if (parameters.at("compositeType") != "%u32vec3")
8382 parameters["u32vec3Decl"] = "%u32vec3 = OpTypeVector %u32 3\n";
8385 return StringTemplate(
8386 "OpCapability Shader\n"
8387 "OpCapability Matrix\n"
8388 "OpMemoryModel Logical GLSL450\n"
8389 "OpEntryPoint GLCompute %main \"main\" %id\n"
8390 "OpExecutionMode %main LocalSize 1 1 1\n"
8392 "OpSource GLSL 430\n"
8393 "OpName %main \"main\"\n"
8394 "OpName %id \"gl_GlobalInvocationID\"\n"
8397 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8398 "OpDecorate %buf BufferBlock\n"
8399 "OpDecorate %indata DescriptorSet 0\n"
8400 "OpDecorate %indata Binding 0\n"
8401 "OpDecorate %outdata DescriptorSet 0\n"
8402 "OpDecorate %outdata Binding 1\n"
8403 "OpDecorate %customarr ArrayStride 4\n"
8404 "${compositeDecorator}"
8405 "OpMemberDecorate %buf 0 Offset 0\n"
8408 "%void = OpTypeVoid\n"
8409 "%voidf = OpTypeFunction %void\n"
8410 "%u32 = OpTypeInt 32 0\n"
8411 "%i32 = OpTypeInt 32 1\n"
8412 "%f32 = OpTypeFloat 32\n"
8414 // Composite declaration
8420 "${u32vec3Decl:opt}"
8421 "%uvec3ptr = OpTypePointer Input %u32vec3\n"
8423 // Inherited from custom
8424 "%customptr = OpTypePointer Uniform ${customType}\n"
8425 "%customarr = OpTypeRuntimeArray ${customType}\n"
8426 "%buf = OpTypeStruct %customarr\n"
8427 "%bufptr = OpTypePointer Uniform %buf\n"
8429 "%indata = OpVariable %bufptr Uniform\n"
8430 "%outdata = OpVariable %bufptr Uniform\n"
8432 "%id = OpVariable %uvec3ptr Input\n"
8433 "%zero = OpConstant %i32 0\n"
8435 "%main = OpFunction %void None %voidf\n"
8436 "%label = OpLabel\n"
8437 "%idval = OpLoad %u32vec3 %id\n"
8438 "%x = OpCompositeExtract %u32 %idval 0\n"
8440 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
8441 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
8442 // Read the input value
8443 "%inval = OpLoad ${customType} %inloc\n"
8444 // Create the composite and fill it
8445 "${compositeConstruct}"
8446 // Insert the input value to a place
8447 "%instance2 = OpCompositeInsert ${compositeType} %inval %instance ${indexes}\n"
8448 // Read back the value from the position
8449 "%out_val = OpCompositeExtract ${customType} %instance2 ${indexes}\n"
8450 // Store it in the output position
8451 " OpStore %outloc %out_val\n"
8454 ).specialize(parameters);
8457 template<typename T>
8458 BufferSp createCompositeBuffer(T number)
8460 return BufferSp(new Buffer<T>(vector<T>(1, number)));
8463 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
8465 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
8466 de::Random rnd (deStringHash(group->getName()));
8468 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8470 NumberType numberType = NumberType(type);
8471 const string typeName = getNumberTypeName(numberType);
8472 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
8473 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8474 vector<map<string, string> > testCases;
8476 createCompositeCases(testCases, rnd, numberType);
8478 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8480 ComputeShaderSpec spec;
8482 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
8486 case NUMBERTYPE_INT32:
8488 deInt32 number = getInt(rnd);
8489 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8490 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8493 case NUMBERTYPE_UINT32:
8495 deUint32 number = rnd.getUint32();
8496 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8497 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8500 case NUMBERTYPE_FLOAT32:
8502 float number = rnd.getFloat();
8503 spec.inputs.push_back(createCompositeBuffer<float>(number));
8504 spec.outputs.push_back(createCompositeBuffer<float>(number));
8511 spec.numWorkGroups = IVec3(1, 1, 1);
8512 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
8514 group->addChild(subGroup.release());
8516 return group.release();
8519 struct AssemblyStructInfo
8521 AssemblyStructInfo (const deUint32 comp, const deUint32 idx)
8526 deUint32 components;
8530 const string specializeInBoundsShaderTemplate (const NumberType type, const AssemblyStructInfo& structInfo, const map<string, string>& params)
8532 // Create the full index string
8533 string fullIndex = numberToString(structInfo.index) + " " + params.at("indexes");
8534 // Convert it to list of indexes
8535 vector<string> indexes = de::splitString(fullIndex, ' ');
8537 map<string, string> parameters (params);
8538 parameters["structType"] = repeatString(" ${compositeType}", structInfo.components);
8539 parameters["structConstruct"] = repeatString(" %instance", structInfo.components);
8540 parameters["insertIndexes"] = fullIndex;
8542 // In matrix cases the last two index is the CompositeExtract indexes
8543 const deUint32 extractIndexes = (parameters["type"] == "matrix") ? 2 : 1;
8545 // Construct the extractIndex
8546 for (vector<string>::const_iterator index = indexes.end() - extractIndexes; index != indexes.end(); ++index)
8548 parameters["extractIndexes"] += " " + *index;
8551 // Remove the last 1 or 2 element depends on matrix case or not
8552 indexes.erase(indexes.end() - extractIndexes, indexes.end());
8555 // Generate AccessChain index expressions (except for the last one, because we use ptr to the composite)
8556 for (vector<string>::const_iterator index = indexes.begin(); index != indexes.end(); ++index)
8558 string indexId = "%index_" + numberToString(id++);
8559 parameters["accessChainConstDeclaration"] += indexId + " = OpConstant %u32 " + *index + "\n";
8560 parameters["accessChainIndexes"] += " " + indexId;
8563 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
8565 const string customType = getAssemblyTypeName(type);
8566 map<string, string> substCustomType;
8567 substCustomType["customType"] = customType;
8568 parameters["compositeDecl"] = StringTemplate(parameters.at("compositeDecl")).specialize(substCustomType);
8569 parameters["compositeType"] = StringTemplate(parameters.at("compositeType")).specialize(substCustomType);
8570 parameters["compositeConstruct"] = StringTemplate(parameters.at("compositeConstruct")).specialize(substCustomType);
8571 parameters["filler"] = StringTemplate(parameters.at("filler")).specialize(substCustomType);
8572 parameters["customType"] = customType;
8574 const string compositeType = parameters.at("compositeType");
8575 map<string, string> substCompositeType;
8576 substCompositeType["compositeType"] = compositeType;
8577 parameters["structType"] = StringTemplate(parameters.at("structType")).specialize(substCompositeType);
8578 if (compositeType != "%u32vec3")
8580 parameters["u32vec3Decl"] = "%u32vec3 = OpTypeVector %u32 3\n";
8583 return StringTemplate(
8584 "OpCapability Shader\n"
8585 "OpCapability Matrix\n"
8586 "OpMemoryModel Logical GLSL450\n"
8587 "OpEntryPoint GLCompute %main \"main\" %id\n"
8588 "OpExecutionMode %main LocalSize 1 1 1\n"
8590 "OpSource GLSL 430\n"
8591 "OpName %main \"main\"\n"
8592 "OpName %id \"gl_GlobalInvocationID\"\n"
8594 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8595 "OpDecorate %buf BufferBlock\n"
8596 "OpDecorate %indata DescriptorSet 0\n"
8597 "OpDecorate %indata Binding 0\n"
8598 "OpDecorate %outdata DescriptorSet 0\n"
8599 "OpDecorate %outdata Binding 1\n"
8600 "OpDecorate %customarr ArrayStride 4\n"
8601 "${compositeDecorator}"
8602 "OpMemberDecorate %buf 0 Offset 0\n"
8604 "%void = OpTypeVoid\n"
8605 "%voidf = OpTypeFunction %void\n"
8606 "%i32 = OpTypeInt 32 1\n"
8607 "%u32 = OpTypeInt 32 0\n"
8608 "%f32 = OpTypeFloat 32\n"
8611 // %u32vec3 if not already declared in ${compositeDecl}
8612 "${u32vec3Decl:opt}"
8613 "%uvec3ptr = OpTypePointer Input %u32vec3\n"
8614 // Inherited from composite
8615 "%composite_p = OpTypePointer Function ${compositeType}\n"
8616 "%struct_t = OpTypeStruct${structType}\n"
8617 "%struct_p = OpTypePointer Function %struct_t\n"
8620 "${accessChainConstDeclaration}"
8621 // Inherited from custom
8622 "%customptr = OpTypePointer Uniform ${customType}\n"
8623 "%customarr = OpTypeRuntimeArray ${customType}\n"
8624 "%buf = OpTypeStruct %customarr\n"
8625 "%bufptr = OpTypePointer Uniform %buf\n"
8626 "%indata = OpVariable %bufptr Uniform\n"
8627 "%outdata = OpVariable %bufptr Uniform\n"
8629 "%id = OpVariable %uvec3ptr Input\n"
8630 "%zero = OpConstant %u32 0\n"
8631 "%main = OpFunction %void None %voidf\n"
8632 "%label = OpLabel\n"
8633 "%struct_v = OpVariable %struct_p Function\n"
8634 "%idval = OpLoad %u32vec3 %id\n"
8635 "%x = OpCompositeExtract %u32 %idval 0\n"
8636 // Create the input/output type
8637 "%inloc = OpInBoundsAccessChain %customptr %indata %zero %x\n"
8638 "%outloc = OpInBoundsAccessChain %customptr %outdata %zero %x\n"
8639 // Read the input value
8640 "%inval = OpLoad ${customType} %inloc\n"
8641 // Create the composite and fill it
8642 "${compositeConstruct}"
8643 // Create the struct and fill it with the composite
8644 "%struct = OpCompositeConstruct %struct_t${structConstruct}\n"
8646 "%comp_obj = OpCompositeInsert %struct_t %inval %struct ${insertIndexes}\n"
8648 " OpStore %struct_v %comp_obj\n"
8649 // Get deepest possible composite pointer
8650 "%inner_ptr = OpInBoundsAccessChain %composite_p %struct_v${accessChainIndexes}\n"
8651 "%read_obj = OpLoad ${compositeType} %inner_ptr\n"
8652 // Read back the stored value
8653 "%read_val = OpCompositeExtract ${customType} %read_obj${extractIndexes}\n"
8654 " OpStore %outloc %read_val\n"
8657 ).specialize(parameters);
8660 tcu::TestCaseGroup* createOpInBoundsAccessChainGroup (tcu::TestContext& testCtx)
8662 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opinboundsaccesschain", "Test the OpInBoundsAccessChain instruction"));
8663 de::Random rnd (deStringHash(group->getName()));
8665 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8667 NumberType numberType = NumberType(type);
8668 const string typeName = getNumberTypeName(numberType);
8669 const string description = "Test the OpInBoundsAccessChain instruction with " + typeName + "s";
8670 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8672 vector<map<string, string> > testCases;
8673 createCompositeCases(testCases, rnd, numberType);
8675 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8677 ComputeShaderSpec spec;
8679 // Number of components inside of a struct
8680 deUint32 structComponents = rnd.getInt(2, 8);
8681 // Component index value
8682 deUint32 structIndex = rnd.getInt(0, structComponents - 1);
8683 AssemblyStructInfo structInfo(structComponents, structIndex);
8685 spec.assembly = specializeInBoundsShaderTemplate(numberType, structInfo, *test);
8689 case NUMBERTYPE_INT32:
8691 deInt32 number = getInt(rnd);
8692 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8693 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8696 case NUMBERTYPE_UINT32:
8698 deUint32 number = rnd.getUint32();
8699 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8700 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8703 case NUMBERTYPE_FLOAT32:
8705 float number = rnd.getFloat();
8706 spec.inputs.push_back(createCompositeBuffer<float>(number));
8707 spec.outputs.push_back(createCompositeBuffer<float>(number));
8713 spec.numWorkGroups = IVec3(1, 1, 1);
8714 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpInBoundsAccessChain test", spec));
8716 group->addChild(subGroup.release());
8718 return group.release();
8721 // If the params missing, uninitialized case
8722 const string specializeDefaultOutputShaderTemplate (const NumberType type, const map<string, string>& params = map<string, string>())
8724 map<string, string> parameters(params);
8726 parameters["customType"] = getAssemblyTypeName(type);
8728 // Declare the const value, and use it in the initializer
8729 if (params.find("constValue") != params.end())
8731 parameters["variableInitializer"] = " %const";
8733 // Uninitialized case
8736 parameters["commentDecl"] = ";";
8739 return StringTemplate(
8740 "OpCapability Shader\n"
8741 "OpMemoryModel Logical GLSL450\n"
8742 "OpEntryPoint GLCompute %main \"main\" %id\n"
8743 "OpExecutionMode %main LocalSize 1 1 1\n"
8744 "OpSource GLSL 430\n"
8745 "OpName %main \"main\"\n"
8746 "OpName %id \"gl_GlobalInvocationID\"\n"
8748 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8749 "OpDecorate %indata DescriptorSet 0\n"
8750 "OpDecorate %indata Binding 0\n"
8751 "OpDecorate %outdata DescriptorSet 0\n"
8752 "OpDecorate %outdata Binding 1\n"
8753 "OpDecorate %in_arr ArrayStride 4\n"
8754 "OpDecorate %in_buf BufferBlock\n"
8755 "OpMemberDecorate %in_buf 0 Offset 0\n"
8757 "%void = OpTypeVoid\n"
8758 "%voidf = OpTypeFunction %void\n"
8759 "%u32 = OpTypeInt 32 0\n"
8760 "%i32 = OpTypeInt 32 1\n"
8761 "%f32 = OpTypeFloat 32\n"
8762 "%uvec3 = OpTypeVector %u32 3\n"
8763 "%uvec3ptr = OpTypePointer Input %uvec3\n"
8764 "${commentDecl:opt}%const = OpConstant ${customType} ${constValue:opt}\n"
8766 "%in_ptr = OpTypePointer Uniform ${customType}\n"
8767 "%in_arr = OpTypeRuntimeArray ${customType}\n"
8768 "%in_buf = OpTypeStruct %in_arr\n"
8769 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
8770 "%indata = OpVariable %in_bufptr Uniform\n"
8771 "%outdata = OpVariable %in_bufptr Uniform\n"
8772 "%id = OpVariable %uvec3ptr Input\n"
8773 "%var_ptr = OpTypePointer Function ${customType}\n"
8775 "%zero = OpConstant %i32 0\n"
8777 "%main = OpFunction %void None %voidf\n"
8778 "%label = OpLabel\n"
8779 "%out_var = OpVariable %var_ptr Function${variableInitializer:opt}\n"
8780 "%idval = OpLoad %uvec3 %id\n"
8781 "%x = OpCompositeExtract %u32 %idval 0\n"
8782 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
8783 "%outloc = OpAccessChain %in_ptr %outdata %zero %x\n"
8785 "%outval = OpLoad ${customType} %out_var\n"
8786 " OpStore %outloc %outval\n"
8789 ).specialize(parameters);
8792 bool compareFloats (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
8794 DE_ASSERT(outputAllocs.size() != 0);
8795 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
8797 // Use custom epsilon because of the float->string conversion
8798 const float epsilon = 0.00001f;
8800 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
8802 vector<deUint8> expectedBytes;
8806 expectedOutputs[outputNdx]->getBytes(expectedBytes);
8807 memcpy(&expected, &expectedBytes.front(), expectedBytes.size());
8808 memcpy(&actual, outputAllocs[outputNdx]->getHostPtr(), expectedBytes.size());
8810 // Test with epsilon
8811 if (fabs(expected - actual) > epsilon)
8813 log << TestLog::Message << "Error: The actual and expected values not matching."
8814 << " Expected: " << expected << " Actual: " << actual << " Epsilon: " << epsilon << TestLog::EndMessage;
8821 // Checks if the driver crash with uninitialized cases
8822 bool passthruVerify (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
8824 DE_ASSERT(outputAllocs.size() != 0);
8825 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
8827 // Copy and discard the result.
8828 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
8830 vector<deUint8> expectedBytes;
8831 expectedOutputs[outputNdx]->getBytes(expectedBytes);
8833 const size_t width = expectedBytes.size();
8834 vector<char> data (width);
8836 memcpy(&data[0], outputAllocs[outputNdx]->getHostPtr(), width);
8841 tcu::TestCaseGroup* createShaderDefaultOutputGroup (tcu::TestContext& testCtx)
8843 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "shader_default_output", "Test shader default output."));
8844 de::Random rnd (deStringHash(group->getName()));
8846 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8848 NumberType numberType = NumberType(type);
8849 const string typeName = getNumberTypeName(numberType);
8850 const string description = "Test the OpVariable initializer with " + typeName + ".";
8851 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8853 // 2 similar subcases (initialized and uninitialized)
8854 for (int subCase = 0; subCase < 2; ++subCase)
8856 ComputeShaderSpec spec;
8857 spec.numWorkGroups = IVec3(1, 1, 1);
8859 map<string, string> params;
8863 case NUMBERTYPE_INT32:
8865 deInt32 number = getInt(rnd);
8866 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8867 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8868 params["constValue"] = numberToString(number);
8871 case NUMBERTYPE_UINT32:
8873 deUint32 number = rnd.getUint32();
8874 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8875 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8876 params["constValue"] = numberToString(number);
8879 case NUMBERTYPE_FLOAT32:
8881 float number = rnd.getFloat();
8882 spec.inputs.push_back(createCompositeBuffer<float>(number));
8883 spec.outputs.push_back(createCompositeBuffer<float>(number));
8884 spec.verifyIO = &compareFloats;
8885 params["constValue"] = numberToString(number);
8892 // Initialized subcase
8895 spec.assembly = specializeDefaultOutputShaderTemplate(numberType, params);
8896 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "initialized", "OpVariable initializer tests.", spec));
8898 // Uninitialized subcase
8901 spec.assembly = specializeDefaultOutputShaderTemplate(numberType);
8902 spec.verifyIO = &passthruVerify;
8903 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "uninitialized", "OpVariable initializer tests.", spec));
8906 group->addChild(subGroup.release());
8908 return group.release();
8911 tcu::TestCaseGroup* createOpNopTests (tcu::TestContext& testCtx)
8913 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
8914 RGBA defaultColors[4];
8915 map<string, string> opNopFragments;
8917 getDefaultColors(defaultColors);
8919 opNopFragments["testfun"] =
8920 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
8921 "%param1 = OpFunctionParameter %v4f32\n"
8922 "%label_testfun = OpLabel\n"
8931 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
8932 "%b = OpFAdd %f32 %a %a\n"
8934 "%c = OpFSub %f32 %b %a\n"
8935 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
8938 "OpReturnValue %ret\n"
8941 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, testGroup.get());
8943 return testGroup.release();
8946 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
8948 const bool testComputePipeline = true;
8950 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
8951 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
8952 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
8954 computeTests->addChild(createSpivVersionCheckTests(testCtx, testComputePipeline));
8955 computeTests->addChild(createLocalSizeGroup(testCtx));
8956 computeTests->addChild(createOpNopGroup(testCtx));
8957 computeTests->addChild(createOpFUnordGroup(testCtx));
8958 computeTests->addChild(createOpAtomicGroup(testCtx, false));
8959 computeTests->addChild(createOpAtomicGroup(testCtx, true)); // Using new StorageBuffer decoration
8960 computeTests->addChild(createOpLineGroup(testCtx));
8961 computeTests->addChild(createOpModuleProcessedGroup(testCtx));
8962 computeTests->addChild(createOpNoLineGroup(testCtx));
8963 computeTests->addChild(createOpConstantNullGroup(testCtx));
8964 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
8965 computeTests->addChild(createOpConstantUsageGroup(testCtx));
8966 computeTests->addChild(createSpecConstantGroup(testCtx));
8967 computeTests->addChild(createOpSourceGroup(testCtx));
8968 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
8969 computeTests->addChild(createDecorationGroupGroup(testCtx));
8970 computeTests->addChild(createOpPhiGroup(testCtx));
8971 computeTests->addChild(createLoopControlGroup(testCtx));
8972 computeTests->addChild(createFunctionControlGroup(testCtx));
8973 computeTests->addChild(createSelectionControlGroup(testCtx));
8974 computeTests->addChild(createBlockOrderGroup(testCtx));
8975 computeTests->addChild(createMultipleShaderGroup(testCtx));
8976 computeTests->addChild(createMemoryAccessGroup(testCtx));
8977 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
8978 computeTests->addChild(createOpCopyObjectGroup(testCtx));
8979 computeTests->addChild(createNoContractionGroup(testCtx));
8980 computeTests->addChild(createOpUndefGroup(testCtx));
8981 computeTests->addChild(createOpUnreachableGroup(testCtx));
8982 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
8983 computeTests ->addChild(createOpFRemGroup(testCtx));
8984 computeTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_PASS));
8985 computeTests->addChild(createOpSRemComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
8986 computeTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_PASS));
8987 computeTests->addChild(createOpSModComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
8988 computeTests->addChild(createSConvertTests(testCtx));
8989 computeTests->addChild(createUConvertTests(testCtx));
8990 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
8991 computeTests->addChild(createOpInBoundsAccessChainGroup(testCtx));
8992 computeTests->addChild(createShaderDefaultOutputGroup(testCtx));
8993 computeTests->addChild(createOpNMinGroup(testCtx));
8994 computeTests->addChild(createOpNMaxGroup(testCtx));
8995 computeTests->addChild(createOpNClampGroup(testCtx));
8997 de::MovePtr<tcu::TestCaseGroup> computeAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
8999 computeAndroidTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9000 computeAndroidTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9002 computeTests->addChild(computeAndroidTests.release());
9005 computeTests->addChild(create8BitStorageComputeGroup(testCtx));
9006 computeTests->addChild(create16BitStorageComputeGroup(testCtx));
9007 computeTests->addChild(createUboMatrixPaddingComputeGroup(testCtx));
9008 computeTests->addChild(createConditionalBranchComputeGroup(testCtx));
9009 computeTests->addChild(createIndexingComputeGroup(testCtx));
9010 computeTests->addChild(createVariablePointersComputeGroup(testCtx));
9011 computeTests->addChild(createImageSamplerComputeGroup(testCtx));
9012 graphicsTests->addChild(createSpivVersionCheckTests(testCtx, !testComputePipeline));
9013 graphicsTests->addChild(createOpNopTests(testCtx));
9014 graphicsTests->addChild(createOpSourceTests(testCtx));
9015 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
9016 graphicsTests->addChild(createOpModuleProcessedTests(testCtx));
9017 graphicsTests->addChild(createOpLineTests(testCtx));
9018 graphicsTests->addChild(createOpNoLineTests(testCtx));
9019 graphicsTests->addChild(createOpConstantNullTests(testCtx));
9020 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
9021 graphicsTests->addChild(createMemoryAccessTests(testCtx));
9022 graphicsTests->addChild(createOpUndefTests(testCtx));
9023 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
9024 graphicsTests->addChild(createModuleTests(testCtx));
9025 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
9026 graphicsTests->addChild(createOpPhiTests(testCtx));
9027 graphicsTests->addChild(createNoContractionTests(testCtx));
9028 graphicsTests->addChild(createOpQuantizeTests(testCtx));
9029 graphicsTests->addChild(createLoopTests(testCtx));
9030 graphicsTests->addChild(createSpecConstantTests(testCtx));
9031 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
9032 graphicsTests->addChild(createBarrierTests(testCtx));
9033 graphicsTests->addChild(createDecorationGroupTests(testCtx));
9034 graphicsTests->addChild(createFRemTests(testCtx));
9035 graphicsTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
9036 graphicsTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
9039 de::MovePtr<tcu::TestCaseGroup> graphicsAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
9041 graphicsAndroidTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9042 graphicsAndroidTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9044 graphicsTests->addChild(graphicsAndroidTests.release());
9047 graphicsTests->addChild(create8BitStorageGraphicsGroup(testCtx));
9048 graphicsTests->addChild(create16BitStorageGraphicsGroup(testCtx));
9049 graphicsTests->addChild(createUboMatrixPaddingGraphicsGroup(testCtx));
9050 graphicsTests->addChild(createConditionalBranchGraphicsGroup(testCtx));
9051 graphicsTests->addChild(createIndexingGraphicsGroup(testCtx));
9052 graphicsTests->addChild(createVariablePointersGraphicsGroup(testCtx));
9053 graphicsTests->addChild(createImageSamplerGraphicsGroup(testCtx));
9055 instructionTests->addChild(computeTests.release());
9056 instructionTests->addChild(graphicsTests.release());
9058 return instructionTests.release();