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 "vktSpvAsmCrossStageInterfaceTests.hpp"
53 #include "vktSpvAsm8bitStorageTests.hpp"
54 #include "vktSpvAsm16bitStorageTests.hpp"
55 #include "vktSpvAsmUboMatrixPaddingTests.hpp"
56 #include "vktSpvAsmConditionalBranchTests.hpp"
57 #include "vktSpvAsmIndexingTests.hpp"
58 #include "vktSpvAsmImageSamplerTests.hpp"
59 #include "vktSpvAsmComputeShaderCase.hpp"
60 #include "vktSpvAsmComputeShaderTestUtil.hpp"
61 #include "vktSpvAsmGraphicsShaderTestUtil.hpp"
62 #include "vktSpvAsmVariablePointersTests.hpp"
63 #include "vktSpvAsmVariableInitTests.hpp"
64 #include "vktSpvAsmPointerParameterTests.hpp"
65 #include "vktSpvAsmSpirvVersionTests.hpp"
66 #include "vktTestCaseUtil.hpp"
67 #include "vktSpvAsmLoopDepLenTests.hpp"
68 #include "vktSpvAsmLoopDepInfTests.hpp"
80 namespace SpirVAssembly
94 using tcu::TestStatus;
97 using tcu::StringTemplate;
101 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0)
103 T* const typedPtr = (T*)dst;
104 for (int ndx = 0; ndx < numValues; ndx++)
105 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue);
108 // Filter is a function that returns true if a value should pass, false otherwise.
109 template<typename T, typename FilterT>
110 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, FilterT filter, int offset = 0)
112 T* const typedPtr = (T*)dst;
114 for (int ndx = 0; ndx < numValues; ndx++)
117 value = randomScalar<T>(rnd, minValue, maxValue);
118 while (!filter(value));
120 typedPtr[offset + ndx] = value;
124 // Gets a 64-bit integer with a more logarithmic distribution
125 deInt64 randomInt64LogDistributed (de::Random& rnd)
127 deInt64 val = rnd.getUint64();
128 val &= (1ull << rnd.getInt(1, 63)) - 1;
134 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues)
136 for (int ndx = 0; ndx < numValues; ndx++)
137 dst[ndx] = randomInt64LogDistributed(rnd);
140 template<typename FilterT>
141 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues, FilterT filter)
143 for (int ndx = 0; ndx < numValues; ndx++)
147 value = randomInt64LogDistributed(rnd);
148 } while (!filter(value));
153 inline bool filterNonNegative (const deInt64 value)
158 inline bool filterPositive (const deInt64 value)
163 inline bool filterNotZero (const deInt64 value)
168 static void floorAll (vector<float>& values)
170 for (size_t i = 0; i < values.size(); i++)
171 values[i] = deFloatFloor(values[i]);
174 static void floorAll (vector<Vec4>& values)
176 for (size_t i = 0; i < values.size(); i++)
177 values[i] = floor(values[i]);
185 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {}
188 // Assembly code used for testing LocalSize, OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code:
192 // layout(std140, set = 0, binding = 0) readonly buffer Input {
195 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
199 // layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
202 // uint x = gl_GlobalInvocationID.x;
203 // output_data.elements[x] = -input_data.elements[x];
206 static string getAsmForLocalSizeTest(bool useLiteralLocalSize, bool useSpecConstantWorkgroupSize, IVec3 workGroupSize, deUint32 ndx)
208 std::ostringstream out;
209 out << getComputeAsmShaderPreambleWithoutLocalSize();
211 if (useLiteralLocalSize)
213 out << "OpExecutionMode %main LocalSize "
214 << workGroupSize.x() << " " << workGroupSize.y() << " " << workGroupSize.z() << "\n";
217 out << "OpSource GLSL 430\n"
218 "OpName %main \"main\"\n"
219 "OpName %id \"gl_GlobalInvocationID\"\n"
220 "OpDecorate %id BuiltIn GlobalInvocationId\n";
222 if (useSpecConstantWorkgroupSize)
224 out << "OpDecorate %spec_0 SpecId 100\n"
225 << "OpDecorate %spec_1 SpecId 101\n"
226 << "OpDecorate %spec_2 SpecId 102\n"
227 << "OpDecorate %gl_WorkGroupSize BuiltIn WorkgroupSize\n";
230 out << getComputeAsmInputOutputBufferTraits()
231 << getComputeAsmCommonTypes()
232 << getComputeAsmInputOutputBuffer()
233 << "%id = OpVariable %uvec3ptr Input\n"
234 << "%zero = OpConstant %i32 0 \n";
236 if (useSpecConstantWorkgroupSize)
238 out << "%spec_0 = OpSpecConstant %u32 "<< workGroupSize.x() << "\n"
239 << "%spec_1 = OpSpecConstant %u32 "<< workGroupSize.y() << "\n"
240 << "%spec_2 = OpSpecConstant %u32 "<< workGroupSize.z() << "\n"
241 << "%gl_WorkGroupSize = OpSpecConstantComposite %uvec3 %spec_0 %spec_1 %spec_2\n";
244 out << "%main = OpFunction %void None %voidf\n"
245 << "%label = OpLabel\n"
246 << "%idval = OpLoad %uvec3 %id\n"
247 << "%ndx = OpCompositeExtract %u32 %idval " << ndx << "\n"
249 "%inloc = OpAccessChain %f32ptr %indata %zero %ndx\n"
250 "%inval = OpLoad %f32 %inloc\n"
251 "%neg = OpFNegate %f32 %inval\n"
252 "%outloc = OpAccessChain %f32ptr %outdata %zero %ndx\n"
253 " OpStore %outloc %neg\n"
259 tcu::TestCaseGroup* createLocalSizeGroup (tcu::TestContext& testCtx)
261 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "localsize", ""));
262 ComputeShaderSpec spec;
263 de::Random rnd (deStringHash(group->getName()));
264 const deUint32 numElements = 64u;
265 vector<float> positiveFloats (numElements, 0);
266 vector<float> negativeFloats (numElements, 0);
268 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
270 for (size_t ndx = 0; ndx < numElements; ++ndx)
271 negativeFloats[ndx] = -positiveFloats[ndx];
273 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
274 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
276 spec.numWorkGroups = IVec3(numElements, 1, 1);
278 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, 1, 1), 0u);
279 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize", "", spec));
281 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, 1, 1), 0u);
282 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize", "", spec));
284 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, 1, 1), 0u);
285 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize", "", spec));
287 spec.numWorkGroups = IVec3(1, 1, 1);
289 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(numElements, 1, 1), 0u);
290 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_x", "", spec));
292 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(numElements, 1, 1), 0u);
293 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_x", "", spec));
295 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(numElements, 1, 1), 0u);
296 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_x", "", spec));
298 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, numElements, 1), 1u);
299 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_y", "", spec));
301 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, numElements, 1), 1u);
302 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_y", "", spec));
304 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, numElements, 1), 1u);
305 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_y", "", spec));
307 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, 1, numElements), 2u);
308 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_z", "", spec));
310 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, 1, numElements), 2u);
311 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_z", "", spec));
313 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, 1, numElements), 2u);
314 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_z", "", spec));
316 return group.release();
319 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
321 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
322 ComputeShaderSpec spec;
323 de::Random rnd (deStringHash(group->getName()));
324 const int numElements = 100;
325 vector<float> positiveFloats (numElements, 0);
326 vector<float> negativeFloats (numElements, 0);
328 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
330 for (size_t ndx = 0; ndx < numElements; ++ndx)
331 negativeFloats[ndx] = -positiveFloats[ndx];
334 string(getComputeAsmShaderPreamble()) +
336 "OpSource GLSL 430\n"
337 "OpName %main \"main\"\n"
338 "OpName %id \"gl_GlobalInvocationID\"\n"
340 "OpDecorate %id BuiltIn GlobalInvocationId\n"
342 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes())
344 + string(getComputeAsmInputOutputBuffer()) +
346 "%id = OpVariable %uvec3ptr Input\n"
347 "%zero = OpConstant %i32 0\n"
349 "%main = OpFunction %void None %voidf\n"
351 "%idval = OpLoad %uvec3 %id\n"
352 "%x = OpCompositeExtract %u32 %idval 0\n"
354 " OpNop\n" // Inside a function body
356 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
357 "%inval = OpLoad %f32 %inloc\n"
358 "%neg = OpFNegate %f32 %inval\n"
359 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
360 " OpStore %outloc %neg\n"
363 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
364 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
365 spec.numWorkGroups = IVec3(numElements, 1, 1);
367 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
369 return group.release();
372 bool compareFUnord (const std::vector<Resource>& inputs, const vector<AllocationSp>& outputAllocs, const std::vector<Resource>& expectedOutputs, TestLog& log)
374 if (outputAllocs.size() != 1)
377 vector<deUint8> input1Bytes;
378 vector<deUint8> input2Bytes;
379 vector<deUint8> expectedBytes;
381 inputs[0].getBytes(input1Bytes);
382 inputs[1].getBytes(input2Bytes);
383 expectedOutputs[0].getBytes(expectedBytes);
385 const deInt32* const expectedOutputAsInt = reinterpret_cast<const deInt32*>(&expectedBytes.front());
386 const deInt32* const outputAsInt = static_cast<const deInt32*>(outputAllocs[0]->getHostPtr());
387 const float* const input1AsFloat = reinterpret_cast<const float*>(&input1Bytes.front());
388 const float* const input2AsFloat = reinterpret_cast<const float*>(&input2Bytes.front());
389 bool returnValue = true;
391 for (size_t idx = 0; idx < expectedBytes.size() / sizeof(deInt32); ++idx)
393 if (outputAsInt[idx] != expectedOutputAsInt[idx])
395 log << TestLog::Message << "ERROR: Sub-case failed. inputs: " << input1AsFloat[idx] << "," << input2AsFloat[idx] << " output: " << outputAsInt[idx]<< " expected output: " << expectedOutputAsInt[idx] << TestLog::EndMessage;
402 typedef VkBool32 (*compareFuncType) (float, float);
408 compareFuncType compareFunc;
410 OpFUnordCase (const char* _name, const char* _opCode, compareFuncType _compareFunc)
413 , compareFunc (_compareFunc) {}
416 #define ADD_OPFUNORD_CASE(NAME, OPCODE, OPERATOR) \
418 struct compare_##NAME { static VkBool32 compare(float x, float y) { return (x OPERATOR y) ? VK_TRUE : VK_FALSE; } }; \
419 cases.push_back(OpFUnordCase(#NAME, OPCODE, compare_##NAME::compare)); \
420 } while (deGetFalse())
422 tcu::TestCaseGroup* createOpFUnordGroup (tcu::TestContext& testCtx)
424 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfunord", "Test the OpFUnord* opcodes"));
425 de::Random rnd (deStringHash(group->getName()));
426 const int numElements = 100;
427 vector<OpFUnordCase> cases;
429 const StringTemplate shaderTemplate (
431 string(getComputeAsmShaderPreamble()) +
433 "OpSource GLSL 430\n"
434 "OpName %main \"main\"\n"
435 "OpName %id \"gl_GlobalInvocationID\"\n"
437 "OpDecorate %id BuiltIn GlobalInvocationId\n"
439 "OpDecorate %buf BufferBlock\n"
440 "OpDecorate %buf2 BufferBlock\n"
441 "OpDecorate %indata1 DescriptorSet 0\n"
442 "OpDecorate %indata1 Binding 0\n"
443 "OpDecorate %indata2 DescriptorSet 0\n"
444 "OpDecorate %indata2 Binding 1\n"
445 "OpDecorate %outdata DescriptorSet 0\n"
446 "OpDecorate %outdata Binding 2\n"
447 "OpDecorate %f32arr ArrayStride 4\n"
448 "OpDecorate %i32arr ArrayStride 4\n"
449 "OpMemberDecorate %buf 0 Offset 0\n"
450 "OpMemberDecorate %buf2 0 Offset 0\n"
452 + string(getComputeAsmCommonTypes()) +
454 "%buf = OpTypeStruct %f32arr\n"
455 "%bufptr = OpTypePointer Uniform %buf\n"
456 "%indata1 = OpVariable %bufptr Uniform\n"
457 "%indata2 = OpVariable %bufptr Uniform\n"
459 "%buf2 = OpTypeStruct %i32arr\n"
460 "%buf2ptr = OpTypePointer Uniform %buf2\n"
461 "%outdata = OpVariable %buf2ptr Uniform\n"
463 "%id = OpVariable %uvec3ptr Input\n"
464 "%zero = OpConstant %i32 0\n"
465 "%consti1 = OpConstant %i32 1\n"
466 "%constf1 = OpConstant %f32 1.0\n"
468 "%main = OpFunction %void None %voidf\n"
470 "%idval = OpLoad %uvec3 %id\n"
471 "%x = OpCompositeExtract %u32 %idval 0\n"
473 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
474 "%inval1 = OpLoad %f32 %inloc1\n"
475 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
476 "%inval2 = OpLoad %f32 %inloc2\n"
477 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
479 "%result = ${OPCODE} %bool %inval1 %inval2\n"
480 "%int_res = OpSelect %i32 %result %consti1 %zero\n"
481 " OpStore %outloc %int_res\n"
486 ADD_OPFUNORD_CASE(equal, "OpFUnordEqual", ==);
487 ADD_OPFUNORD_CASE(less, "OpFUnordLessThan", <);
488 ADD_OPFUNORD_CASE(lessequal, "OpFUnordLessThanEqual", <=);
489 ADD_OPFUNORD_CASE(greater, "OpFUnordGreaterThan", >);
490 ADD_OPFUNORD_CASE(greaterequal, "OpFUnordGreaterThanEqual", >=);
491 ADD_OPFUNORD_CASE(notequal, "OpFUnordNotEqual", !=);
493 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
495 map<string, string> specializations;
496 ComputeShaderSpec spec;
497 const float NaN = std::numeric_limits<float>::quiet_NaN();
498 vector<float> inputFloats1 (numElements, 0);
499 vector<float> inputFloats2 (numElements, 0);
500 vector<deInt32> expectedInts (numElements, 0);
502 specializations["OPCODE"] = cases[caseNdx].opCode;
503 spec.assembly = shaderTemplate.specialize(specializations);
505 fillRandomScalars(rnd, 1.f, 100.f, &inputFloats1[0], numElements);
506 for (size_t ndx = 0; ndx < numElements; ++ndx)
510 case 0: inputFloats2[ndx] = inputFloats1[ndx] + 1.0f; break;
511 case 1: inputFloats2[ndx] = inputFloats1[ndx] - 1.0f; break;
512 case 2: inputFloats2[ndx] = inputFloats1[ndx]; break;
513 case 3: inputFloats2[ndx] = NaN; break;
514 case 4: inputFloats2[ndx] = inputFloats1[ndx]; inputFloats1[ndx] = NaN; break;
515 case 5: inputFloats2[ndx] = NaN; inputFloats1[ndx] = NaN; break;
517 expectedInts[ndx] = tcu::Float32(inputFloats1[ndx]).isNaN() || tcu::Float32(inputFloats2[ndx]).isNaN() || cases[caseNdx].compareFunc(inputFloats1[ndx], inputFloats2[ndx]);
520 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
521 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
522 spec.outputs.push_back(BufferSp(new Int32Buffer(expectedInts)));
523 spec.numWorkGroups = IVec3(numElements, 1, 1);
524 spec.verifyIO = &compareFUnord;
525 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
528 return group.release();
534 const char* assembly;
535 const char* retValAssembly;
536 OpAtomicType opAtomic;
537 deInt32 numOutputElements;
539 OpAtomicCase(const char* _name, const char* _assembly, const char* _retValAssembly, OpAtomicType _opAtomic, deInt32 _numOutputElements)
541 , assembly (_assembly)
542 , retValAssembly (_retValAssembly)
543 , opAtomic (_opAtomic)
544 , numOutputElements (_numOutputElements) {}
547 tcu::TestCaseGroup* createOpAtomicGroup (tcu::TestContext& testCtx, bool useStorageBuffer, int numElements = 65535, bool verifyReturnValues = false)
549 std::string groupName ("opatomic");
550 if (useStorageBuffer)
551 groupName += "_storage_buffer";
552 if (verifyReturnValues)
553 groupName += "_return_values";
554 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, groupName.c_str(), "Test the OpAtomic* opcodes"));
555 vector<OpAtomicCase> cases;
557 const StringTemplate shaderTemplate (
559 string("OpCapability Shader\n") +
560 (useStorageBuffer ? "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" : "") +
561 "OpMemoryModel Logical GLSL450\n"
562 "OpEntryPoint GLCompute %main \"main\" %id\n"
563 "OpExecutionMode %main LocalSize 1 1 1\n" +
565 "OpSource GLSL 430\n"
566 "OpName %main \"main\"\n"
567 "OpName %id \"gl_GlobalInvocationID\"\n"
569 "OpDecorate %id BuiltIn GlobalInvocationId\n"
571 "OpDecorate %buf ${BLOCK_DECORATION}\n"
572 "OpDecorate %indata DescriptorSet 0\n"
573 "OpDecorate %indata Binding 0\n"
574 "OpDecorate %i32arr ArrayStride 4\n"
575 "OpMemberDecorate %buf 0 Offset 0\n"
577 "OpDecorate %sumbuf ${BLOCK_DECORATION}\n"
578 "OpDecorate %sum DescriptorSet 0\n"
579 "OpDecorate %sum Binding 1\n"
580 "OpMemberDecorate %sumbuf 0 Coherent\n"
581 "OpMemberDecorate %sumbuf 0 Offset 0\n"
583 "${RETVAL_BUF_DECORATE}"
585 + getComputeAsmCommonTypes("${BLOCK_POINTER_TYPE}") +
587 "%buf = OpTypeStruct %i32arr\n"
588 "%bufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %buf\n"
589 "%indata = OpVariable %bufptr ${BLOCK_POINTER_TYPE}\n"
591 "%sumbuf = OpTypeStruct %i32arr\n"
592 "%sumbufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %sumbuf\n"
593 "%sum = OpVariable %sumbufptr ${BLOCK_POINTER_TYPE}\n"
597 "%id = OpVariable %uvec3ptr Input\n"
598 "%minusone = OpConstant %i32 -1\n"
599 "%zero = OpConstant %i32 0\n"
600 "%one = OpConstant %u32 1\n"
601 "%two = OpConstant %i32 2\n"
603 "%main = OpFunction %void None %voidf\n"
605 "%idval = OpLoad %uvec3 %id\n"
606 "%x = OpCompositeExtract %u32 %idval 0\n"
608 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
609 "%inval = OpLoad %i32 %inloc\n"
611 "%outloc = OpAccessChain %i32ptr %sum %zero ${INDEX}\n"
618 #define ADD_OPATOMIC_CASE(NAME, ASSEMBLY, RETVAL_ASSEMBLY, OPATOMIC, NUM_OUTPUT_ELEMENTS) \
620 DE_ASSERT((NUM_OUTPUT_ELEMENTS) == 1 || (NUM_OUTPUT_ELEMENTS) == numElements); \
621 cases.push_back(OpAtomicCase(#NAME, ASSEMBLY, RETVAL_ASSEMBLY, OPATOMIC, NUM_OUTPUT_ELEMENTS)); \
622 } while (deGetFalse())
623 #define ADD_OPATOMIC_CASE_1(NAME, ASSEMBLY, RETVAL_ASSEMBLY, OPATOMIC) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, RETVAL_ASSEMBLY, OPATOMIC, 1)
624 #define ADD_OPATOMIC_CASE_N(NAME, ASSEMBLY, RETVAL_ASSEMBLY, OPATOMIC) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, RETVAL_ASSEMBLY, OPATOMIC, numElements)
626 ADD_OPATOMIC_CASE_1(iadd, "%retv = OpAtomicIAdd %i32 %outloc %one %zero %inval\n",
627 " OpStore %retloc %retv\n", OPATOMIC_IADD );
628 ADD_OPATOMIC_CASE_1(isub, "%retv = OpAtomicISub %i32 %outloc %one %zero %inval\n",
629 " OpStore %retloc %retv\n", OPATOMIC_ISUB );
630 ADD_OPATOMIC_CASE_1(iinc, "%retv = OpAtomicIIncrement %i32 %outloc %one %zero\n",
631 " OpStore %retloc %retv\n", OPATOMIC_IINC );
632 ADD_OPATOMIC_CASE_1(idec, "%retv = OpAtomicIDecrement %i32 %outloc %one %zero\n",
633 " OpStore %retloc %retv\n", OPATOMIC_IDEC );
634 if (!verifyReturnValues)
636 ADD_OPATOMIC_CASE_N(load, "%inval2 = OpAtomicLoad %i32 %inloc %one %zero\n"
637 " OpStore %outloc %inval2\n", "", OPATOMIC_LOAD );
638 ADD_OPATOMIC_CASE_N(store, " OpAtomicStore %outloc %one %zero %inval\n", "", OPATOMIC_STORE );
641 ADD_OPATOMIC_CASE_N(compex, "%even = OpSMod %i32 %inval %two\n"
642 " OpStore %outloc %even\n"
643 "%retv = OpAtomicCompareExchange %i32 %outloc %one %zero %zero %minusone %zero\n",
644 " OpStore %retloc %retv\n", OPATOMIC_COMPEX );
647 #undef ADD_OPATOMIC_CASE
648 #undef ADD_OPATOMIC_CASE_1
649 #undef ADD_OPATOMIC_CASE_N
651 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
653 map<string, string> specializations;
654 ComputeShaderSpec spec;
655 vector<deInt32> inputInts (numElements, 0);
656 vector<deInt32> expected (cases[caseNdx].numOutputElements, -1);
658 specializations["INDEX"] = (cases[caseNdx].numOutputElements == 1) ? "%zero" : "%x";
659 specializations["INSTRUCTION"] = cases[caseNdx].assembly;
660 specializations["BLOCK_DECORATION"] = useStorageBuffer ? "Block" : "BufferBlock";
661 specializations["BLOCK_POINTER_TYPE"] = useStorageBuffer ? "StorageBuffer" : "Uniform";
663 if (verifyReturnValues)
665 const StringTemplate blockDecoration (
667 "OpDecorate %retbuf ${BLOCK_DECORATION}\n"
668 "OpDecorate %ret DescriptorSet 0\n"
669 "OpDecorate %ret Binding 2\n"
670 "OpMemberDecorate %retbuf 0 Offset 0\n\n");
672 const StringTemplate blockDeclaration (
674 "%retbuf = OpTypeStruct %i32arr\n"
675 "%retbufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %retbuf\n"
676 "%ret = OpVariable %retbufptr ${BLOCK_POINTER_TYPE}\n\n");
678 specializations["RETVAL_ASSEMBLY"] =
679 "%retloc = OpAccessChain %i32ptr %ret %zero %x\n"
680 + std::string(cases[caseNdx].retValAssembly);
682 specializations["RETVAL_BUF_DECORATE"] = blockDecoration.specialize(specializations);
683 specializations["RETVAL_BUF_DECL"] = blockDeclaration.specialize(specializations);
687 specializations["RETVAL_ASSEMBLY"] = "";
688 specializations["RETVAL_BUF_DECORATE"] = "";
689 specializations["RETVAL_BUF_DECL"] = "";
692 spec.assembly = shaderTemplate.specialize(specializations);
694 if (useStorageBuffer)
695 spec.extensions.push_back("VK_KHR_storage_buffer_storage_class");
697 spec.inputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_INPUT)));
698 spec.outputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_EXPECTED)));
699 if (verifyReturnValues)
700 spec.outputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_ATOMIC_RET)));
701 spec.numWorkGroups = IVec3(numElements, 1, 1);
703 if (verifyReturnValues)
705 switch (cases[caseNdx].opAtomic)
708 spec.verifyIO = OpAtomicBuffer::compareWithRetvals<OPATOMIC_IADD>;
711 spec.verifyIO = OpAtomicBuffer::compareWithRetvals<OPATOMIC_ISUB>;
714 spec.verifyIO = OpAtomicBuffer::compareWithRetvals<OPATOMIC_IINC>;
717 spec.verifyIO = OpAtomicBuffer::compareWithRetvals<OPATOMIC_IDEC>;
719 case OPATOMIC_COMPEX:
720 spec.verifyIO = OpAtomicBuffer::compareWithRetvals<OPATOMIC_COMPEX>;
723 DE_FATAL("Unsupported OpAtomic type for return value verification");
726 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
729 return group.release();
732 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
734 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
735 ComputeShaderSpec spec;
736 de::Random rnd (deStringHash(group->getName()));
737 const int numElements = 100;
738 vector<float> positiveFloats (numElements, 0);
739 vector<float> negativeFloats (numElements, 0);
741 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
743 for (size_t ndx = 0; ndx < numElements; ++ndx)
744 negativeFloats[ndx] = -positiveFloats[ndx];
747 string(getComputeAsmShaderPreamble()) +
749 "%fname1 = OpString \"negateInputs.comp\"\n"
750 "%fname2 = OpString \"negateInputs\"\n"
752 "OpSource GLSL 430\n"
753 "OpName %main \"main\"\n"
754 "OpName %id \"gl_GlobalInvocationID\"\n"
756 "OpDecorate %id BuiltIn GlobalInvocationId\n"
758 + string(getComputeAsmInputOutputBufferTraits()) +
760 "OpLine %fname1 0 0\n" // At the earliest possible position
762 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
764 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
765 "OpLine %fname2 1 0\n" // Different filenames
766 "OpLine %fname1 1000 100000\n"
768 "%id = OpVariable %uvec3ptr Input\n"
769 "%zero = OpConstant %i32 0\n"
771 "OpLine %fname1 1 1\n" // Before a function
773 "%main = OpFunction %void None %voidf\n"
776 "OpLine %fname1 1 1\n" // In a function
778 "%idval = OpLoad %uvec3 %id\n"
779 "%x = OpCompositeExtract %u32 %idval 0\n"
780 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
781 "%inval = OpLoad %f32 %inloc\n"
782 "%neg = OpFNegate %f32 %inval\n"
783 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
784 " OpStore %outloc %neg\n"
787 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
788 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
789 spec.numWorkGroups = IVec3(numElements, 1, 1);
791 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
793 return group.release();
796 bool veryfiBinaryShader (const ProgramBinary& binary)
798 const size_t paternCount = 3u;
799 bool paternsCheck[paternCount] =
803 const string patersns[paternCount] =
809 size_t paternNdx = 0u;
811 for (size_t ndx = 0u; ndx < binary.getSize(); ++ndx)
813 if (false == paternsCheck[paternNdx] &&
814 patersns[paternNdx][0] == static_cast<char>(binary.getBinary()[ndx]) &&
815 deMemoryEqual((const char*)&binary.getBinary()[ndx], &patersns[paternNdx][0], patersns[paternNdx].length()))
817 paternsCheck[paternNdx]= true;
819 if (paternNdx == paternCount)
824 for (size_t ndx = 0u; ndx < paternCount; ++ndx)
826 if (!paternsCheck[ndx])
833 tcu::TestCaseGroup* createOpModuleProcessedGroup (tcu::TestContext& testCtx)
835 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opmoduleprocessed", "Test the OpModuleProcessed instruction"));
836 ComputeShaderSpec spec;
837 de::Random rnd (deStringHash(group->getName()));
838 const int numElements = 10;
839 vector<float> positiveFloats (numElements, 0);
840 vector<float> negativeFloats (numElements, 0);
842 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
844 for (size_t ndx = 0; ndx < numElements; ++ndx)
845 negativeFloats[ndx] = -positiveFloats[ndx];
848 string(getComputeAsmShaderPreamble()) +
849 "%fname = OpString \"negateInputs.comp\"\n"
851 "OpSource GLSL 430\n"
852 "OpName %main \"main\"\n"
853 "OpName %id \"gl_GlobalInvocationID\"\n"
854 "OpModuleProcessed \"VULKAN CTS\"\n" //OpModuleProcessed;
855 "OpModuleProcessed \"Negative values\"\n"
856 "OpModuleProcessed \"Date: 2017/09/21\"\n"
857 "OpDecorate %id BuiltIn GlobalInvocationId\n"
859 + string(getComputeAsmInputOutputBufferTraits())
861 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
863 "OpLine %fname 0 1\n"
865 "OpLine %fname 1000 1\n"
867 "%id = OpVariable %uvec3ptr Input\n"
868 "%zero = OpConstant %i32 0\n"
869 "%main = OpFunction %void None %voidf\n"
872 "%idval = OpLoad %uvec3 %id\n"
873 "%x = OpCompositeExtract %u32 %idval 0\n"
875 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
876 "%inval = OpLoad %f32 %inloc\n"
877 "%neg = OpFNegate %f32 %inval\n"
878 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
879 " OpStore %outloc %neg\n"
882 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
883 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
884 spec.numWorkGroups = IVec3(numElements, 1, 1);
885 spec.verifyBinary = veryfiBinaryShader;
886 spec.spirvVersion = SPIRV_VERSION_1_3;
888 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpModuleProcessed Tests", spec));
890 return group.release();
893 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
895 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
896 ComputeShaderSpec spec;
897 de::Random rnd (deStringHash(group->getName()));
898 const int numElements = 100;
899 vector<float> positiveFloats (numElements, 0);
900 vector<float> negativeFloats (numElements, 0);
902 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
904 for (size_t ndx = 0; ndx < numElements; ++ndx)
905 negativeFloats[ndx] = -positiveFloats[ndx];
908 string(getComputeAsmShaderPreamble()) +
910 "%fname = OpString \"negateInputs.comp\"\n"
912 "OpSource GLSL 430\n"
913 "OpName %main \"main\"\n"
914 "OpName %id \"gl_GlobalInvocationID\"\n"
916 "OpDecorate %id BuiltIn GlobalInvocationId\n"
918 + string(getComputeAsmInputOutputBufferTraits()) +
920 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
922 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
924 "OpLine %fname 0 1\n"
925 "OpNoLine\n" // Immediately following a preceding OpLine
927 "OpLine %fname 1000 1\n"
929 "%id = OpVariable %uvec3ptr Input\n"
930 "%zero = OpConstant %i32 0\n"
932 "OpNoLine\n" // Contents after the previous OpLine
934 "%main = OpFunction %void None %voidf\n"
936 "%idval = OpLoad %uvec3 %id\n"
937 "%x = OpCompositeExtract %u32 %idval 0\n"
939 "OpNoLine\n" // Multiple OpNoLine
943 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
944 "%inval = OpLoad %f32 %inloc\n"
945 "%neg = OpFNegate %f32 %inval\n"
946 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
947 " OpStore %outloc %neg\n"
950 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
951 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
952 spec.numWorkGroups = IVec3(numElements, 1, 1);
954 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
956 return group.release();
959 // Compare instruction for the contraction compute case.
960 // Returns true if the output is what is expected from the test case.
961 bool compareNoContractCase(const std::vector<Resource>&, const vector<AllocationSp>& outputAllocs, const std::vector<Resource>& expectedOutputs, TestLog&)
963 if (outputAllocs.size() != 1)
966 // Only size is needed because we are not comparing the exact values.
967 size_t byteSize = expectedOutputs[0].getByteSize();
969 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
971 for(size_t i = 0; i < byteSize / sizeof(float); ++i) {
972 if (outputAsFloat[i] != 0.f &&
973 outputAsFloat[i] != -ldexp(1, -24)) {
981 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
983 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
984 vector<CaseParameter> cases;
985 const int numElements = 100;
986 vector<float> inputFloats1 (numElements, 0);
987 vector<float> inputFloats2 (numElements, 0);
988 vector<float> outputFloats (numElements, 0);
989 const StringTemplate shaderTemplate (
990 string(getComputeAsmShaderPreamble()) +
992 "OpName %main \"main\"\n"
993 "OpName %id \"gl_GlobalInvocationID\"\n"
995 "OpDecorate %id BuiltIn GlobalInvocationId\n"
999 "OpDecorate %buf BufferBlock\n"
1000 "OpDecorate %indata1 DescriptorSet 0\n"
1001 "OpDecorate %indata1 Binding 0\n"
1002 "OpDecorate %indata2 DescriptorSet 0\n"
1003 "OpDecorate %indata2 Binding 1\n"
1004 "OpDecorate %outdata DescriptorSet 0\n"
1005 "OpDecorate %outdata Binding 2\n"
1006 "OpDecorate %f32arr ArrayStride 4\n"
1007 "OpMemberDecorate %buf 0 Offset 0\n"
1009 + string(getComputeAsmCommonTypes()) +
1011 "%buf = OpTypeStruct %f32arr\n"
1012 "%bufptr = OpTypePointer Uniform %buf\n"
1013 "%indata1 = OpVariable %bufptr Uniform\n"
1014 "%indata2 = OpVariable %bufptr Uniform\n"
1015 "%outdata = OpVariable %bufptr Uniform\n"
1017 "%id = OpVariable %uvec3ptr Input\n"
1018 "%zero = OpConstant %i32 0\n"
1019 "%c_f_m1 = OpConstant %f32 -1.\n"
1021 "%main = OpFunction %void None %voidf\n"
1022 "%label = OpLabel\n"
1023 "%idval = OpLoad %uvec3 %id\n"
1024 "%x = OpCompositeExtract %u32 %idval 0\n"
1025 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1026 "%inval1 = OpLoad %f32 %inloc1\n"
1027 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1028 "%inval2 = OpLoad %f32 %inloc2\n"
1029 "%mul = OpFMul %f32 %inval1 %inval2\n"
1030 "%add = OpFAdd %f32 %mul %c_f_m1\n"
1031 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1032 " OpStore %outloc %add\n"
1034 " OpFunctionEnd\n");
1036 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
1037 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
1038 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
1040 for (size_t ndx = 0; ndx < numElements; ++ndx)
1042 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
1043 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
1044 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
1045 // conducted separately and the result is rounded to 1, or 0x1.fffffcp-1
1046 // So the final result will be 0.f or 0x1p-24.
1047 // If the operation is combined into a precise fused multiply-add, then the result would be
1048 // 2^-46 (0xa8800000).
1049 outputFloats[ndx] = 0.f;
1052 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
1054 map<string, string> specializations;
1055 ComputeShaderSpec spec;
1057 specializations["DECORATION"] = cases[caseNdx].param;
1058 spec.assembly = shaderTemplate.specialize(specializations);
1059 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1060 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1061 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1062 spec.numWorkGroups = IVec3(numElements, 1, 1);
1063 // Check against the two possible answers based on rounding mode.
1064 spec.verifyIO = &compareNoContractCase;
1066 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
1068 return group.release();
1071 bool compareFRem(const std::vector<Resource>&, const vector<AllocationSp>& outputAllocs, const std::vector<Resource>& expectedOutputs, TestLog&)
1073 if (outputAllocs.size() != 1)
1076 vector<deUint8> expectedBytes;
1077 expectedOutputs[0].getBytes(expectedBytes);
1079 const float* expectedOutputAsFloat = reinterpret_cast<const float*>(&expectedBytes.front());
1080 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1082 for (size_t idx = 0; idx < expectedBytes.size() / sizeof(float); ++idx)
1084 const float f0 = expectedOutputAsFloat[idx];
1085 const float f1 = outputAsFloat[idx];
1086 // \todo relative error needs to be fairly high because FRem may be implemented as
1087 // (roughly) frac(a/b)*b, so LSB errors can be magnified. But this should be fine for now.
1088 if (deFloatAbs((f1 - f0) / f0) > 0.02)
1095 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
1097 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
1098 ComputeShaderSpec spec;
1099 de::Random rnd (deStringHash(group->getName()));
1100 const int numElements = 200;
1101 vector<float> inputFloats1 (numElements, 0);
1102 vector<float> inputFloats2 (numElements, 0);
1103 vector<float> outputFloats (numElements, 0);
1105 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1106 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
1108 for (size_t ndx = 0; ndx < numElements; ++ndx)
1110 // Guard against divisors near zero.
1111 if (std::fabs(inputFloats2[ndx]) < 1e-3)
1112 inputFloats2[ndx] = 8.f;
1114 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1115 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
1119 string(getComputeAsmShaderPreamble()) +
1121 "OpName %main \"main\"\n"
1122 "OpName %id \"gl_GlobalInvocationID\"\n"
1124 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1126 "OpDecorate %buf BufferBlock\n"
1127 "OpDecorate %indata1 DescriptorSet 0\n"
1128 "OpDecorate %indata1 Binding 0\n"
1129 "OpDecorate %indata2 DescriptorSet 0\n"
1130 "OpDecorate %indata2 Binding 1\n"
1131 "OpDecorate %outdata DescriptorSet 0\n"
1132 "OpDecorate %outdata Binding 2\n"
1133 "OpDecorate %f32arr ArrayStride 4\n"
1134 "OpMemberDecorate %buf 0 Offset 0\n"
1136 + string(getComputeAsmCommonTypes()) +
1138 "%buf = OpTypeStruct %f32arr\n"
1139 "%bufptr = OpTypePointer Uniform %buf\n"
1140 "%indata1 = OpVariable %bufptr Uniform\n"
1141 "%indata2 = OpVariable %bufptr Uniform\n"
1142 "%outdata = OpVariable %bufptr Uniform\n"
1144 "%id = OpVariable %uvec3ptr Input\n"
1145 "%zero = OpConstant %i32 0\n"
1147 "%main = OpFunction %void None %voidf\n"
1148 "%label = OpLabel\n"
1149 "%idval = OpLoad %uvec3 %id\n"
1150 "%x = OpCompositeExtract %u32 %idval 0\n"
1151 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1152 "%inval1 = OpLoad %f32 %inloc1\n"
1153 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1154 "%inval2 = OpLoad %f32 %inloc2\n"
1155 "%rem = OpFRem %f32 %inval1 %inval2\n"
1156 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1157 " OpStore %outloc %rem\n"
1161 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1162 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1163 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1164 spec.numWorkGroups = IVec3(numElements, 1, 1);
1165 spec.verifyIO = &compareFRem;
1167 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1169 return group.release();
1172 bool compareNMin (const std::vector<Resource>&, const vector<AllocationSp>& outputAllocs, const std::vector<Resource>& expectedOutputs, TestLog&)
1174 if (outputAllocs.size() != 1)
1177 const BufferSp& expectedOutput (expectedOutputs[0].getBuffer());
1178 std::vector<deUint8> data;
1179 expectedOutput->getBytes(data);
1181 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1182 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1184 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float); ++idx)
1186 const float f0 = expectedOutputAsFloat[idx];
1187 const float f1 = outputAsFloat[idx];
1189 // For NMin, we accept NaN as output if both inputs were NaN.
1190 // Otherwise the NaN is the wrong choise, as on architectures that
1191 // do not handle NaN, those are huge values.
1192 if (!(tcu::Float32(f1).isNaN() && tcu::Float32(f0).isNaN()) && deFloatAbs(f1 - f0) > 0.00001f)
1199 tcu::TestCaseGroup* createOpNMinGroup (tcu::TestContext& testCtx)
1201 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnmin", "Test the OpNMin instruction"));
1202 ComputeShaderSpec spec;
1203 de::Random rnd (deStringHash(group->getName()));
1204 const int numElements = 200;
1205 vector<float> inputFloats1 (numElements, 0);
1206 vector<float> inputFloats2 (numElements, 0);
1207 vector<float> outputFloats (numElements, 0);
1209 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1210 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1212 // Make the first case a full-NAN case.
1213 inputFloats1[0] = TCU_NAN;
1214 inputFloats2[0] = TCU_NAN;
1216 for (size_t ndx = 0; ndx < numElements; ++ndx)
1218 // By default, pick the smallest
1219 outputFloats[ndx] = std::min(inputFloats1[ndx], inputFloats2[ndx]);
1221 // Make half of the cases NaN cases
1224 // Alternate between the NaN operand
1227 outputFloats[ndx] = inputFloats2[ndx];
1228 inputFloats1[ndx] = TCU_NAN;
1232 outputFloats[ndx] = inputFloats1[ndx];
1233 inputFloats2[ndx] = TCU_NAN;
1239 "OpCapability Shader\n"
1240 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1241 "OpMemoryModel Logical GLSL450\n"
1242 "OpEntryPoint GLCompute %main \"main\" %id\n"
1243 "OpExecutionMode %main LocalSize 1 1 1\n"
1245 "OpName %main \"main\"\n"
1246 "OpName %id \"gl_GlobalInvocationID\"\n"
1248 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1250 "OpDecorate %buf BufferBlock\n"
1251 "OpDecorate %indata1 DescriptorSet 0\n"
1252 "OpDecorate %indata1 Binding 0\n"
1253 "OpDecorate %indata2 DescriptorSet 0\n"
1254 "OpDecorate %indata2 Binding 1\n"
1255 "OpDecorate %outdata DescriptorSet 0\n"
1256 "OpDecorate %outdata Binding 2\n"
1257 "OpDecorate %f32arr ArrayStride 4\n"
1258 "OpMemberDecorate %buf 0 Offset 0\n"
1260 + string(getComputeAsmCommonTypes()) +
1262 "%buf = OpTypeStruct %f32arr\n"
1263 "%bufptr = OpTypePointer Uniform %buf\n"
1264 "%indata1 = OpVariable %bufptr Uniform\n"
1265 "%indata2 = OpVariable %bufptr Uniform\n"
1266 "%outdata = OpVariable %bufptr Uniform\n"
1268 "%id = OpVariable %uvec3ptr Input\n"
1269 "%zero = OpConstant %i32 0\n"
1271 "%main = OpFunction %void None %voidf\n"
1272 "%label = OpLabel\n"
1273 "%idval = OpLoad %uvec3 %id\n"
1274 "%x = OpCompositeExtract %u32 %idval 0\n"
1275 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1276 "%inval1 = OpLoad %f32 %inloc1\n"
1277 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1278 "%inval2 = OpLoad %f32 %inloc2\n"
1279 "%rem = OpExtInst %f32 %std450 NMin %inval1 %inval2\n"
1280 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1281 " OpStore %outloc %rem\n"
1285 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1286 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1287 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1288 spec.numWorkGroups = IVec3(numElements, 1, 1);
1289 spec.verifyIO = &compareNMin;
1291 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1293 return group.release();
1296 bool compareNMax (const std::vector<Resource>&, const vector<AllocationSp>& outputAllocs, const std::vector<Resource>& expectedOutputs, TestLog&)
1298 if (outputAllocs.size() != 1)
1301 const BufferSp& expectedOutput = expectedOutputs[0].getBuffer();
1302 std::vector<deUint8> data;
1303 expectedOutput->getBytes(data);
1305 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1306 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1308 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float); ++idx)
1310 const float f0 = expectedOutputAsFloat[idx];
1311 const float f1 = outputAsFloat[idx];
1313 // For NMax, NaN is considered acceptable result, since in
1314 // architectures that do not handle NaNs, those are huge values.
1315 if (!tcu::Float32(f1).isNaN() && deFloatAbs(f1 - f0) > 0.00001f)
1322 tcu::TestCaseGroup* createOpNMaxGroup (tcu::TestContext& testCtx)
1324 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, "opnmax", "Test the OpNMax instruction"));
1325 ComputeShaderSpec spec;
1326 de::Random rnd (deStringHash(group->getName()));
1327 const int numElements = 200;
1328 vector<float> inputFloats1 (numElements, 0);
1329 vector<float> inputFloats2 (numElements, 0);
1330 vector<float> outputFloats (numElements, 0);
1332 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1333 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1335 // Make the first case a full-NAN case.
1336 inputFloats1[0] = TCU_NAN;
1337 inputFloats2[0] = TCU_NAN;
1339 for (size_t ndx = 0; ndx < numElements; ++ndx)
1341 // By default, pick the biggest
1342 outputFloats[ndx] = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1344 // Make half of the cases NaN cases
1347 // Alternate between the NaN operand
1350 outputFloats[ndx] = inputFloats2[ndx];
1351 inputFloats1[ndx] = TCU_NAN;
1355 outputFloats[ndx] = inputFloats1[ndx];
1356 inputFloats2[ndx] = TCU_NAN;
1362 "OpCapability Shader\n"
1363 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1364 "OpMemoryModel Logical GLSL450\n"
1365 "OpEntryPoint GLCompute %main \"main\" %id\n"
1366 "OpExecutionMode %main LocalSize 1 1 1\n"
1368 "OpName %main \"main\"\n"
1369 "OpName %id \"gl_GlobalInvocationID\"\n"
1371 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1373 "OpDecorate %buf BufferBlock\n"
1374 "OpDecorate %indata1 DescriptorSet 0\n"
1375 "OpDecorate %indata1 Binding 0\n"
1376 "OpDecorate %indata2 DescriptorSet 0\n"
1377 "OpDecorate %indata2 Binding 1\n"
1378 "OpDecorate %outdata DescriptorSet 0\n"
1379 "OpDecorate %outdata Binding 2\n"
1380 "OpDecorate %f32arr ArrayStride 4\n"
1381 "OpMemberDecorate %buf 0 Offset 0\n"
1383 + string(getComputeAsmCommonTypes()) +
1385 "%buf = OpTypeStruct %f32arr\n"
1386 "%bufptr = OpTypePointer Uniform %buf\n"
1387 "%indata1 = OpVariable %bufptr Uniform\n"
1388 "%indata2 = OpVariable %bufptr Uniform\n"
1389 "%outdata = OpVariable %bufptr Uniform\n"
1391 "%id = OpVariable %uvec3ptr Input\n"
1392 "%zero = OpConstant %i32 0\n"
1394 "%main = OpFunction %void None %voidf\n"
1395 "%label = OpLabel\n"
1396 "%idval = OpLoad %uvec3 %id\n"
1397 "%x = OpCompositeExtract %u32 %idval 0\n"
1398 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1399 "%inval1 = OpLoad %f32 %inloc1\n"
1400 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1401 "%inval2 = OpLoad %f32 %inloc2\n"
1402 "%rem = OpExtInst %f32 %std450 NMax %inval1 %inval2\n"
1403 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1404 " OpStore %outloc %rem\n"
1408 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1409 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1410 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1411 spec.numWorkGroups = IVec3(numElements, 1, 1);
1412 spec.verifyIO = &compareNMax;
1414 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1416 return group.release();
1419 bool compareNClamp (const std::vector<Resource>&, const vector<AllocationSp>& outputAllocs, const std::vector<Resource>& expectedOutputs, TestLog&)
1421 if (outputAllocs.size() != 1)
1424 const BufferSp& expectedOutput = expectedOutputs[0].getBuffer();
1425 std::vector<deUint8> data;
1426 expectedOutput->getBytes(data);
1428 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1429 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1431 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float) / 2; ++idx)
1433 const float e0 = expectedOutputAsFloat[idx * 2];
1434 const float e1 = expectedOutputAsFloat[idx * 2 + 1];
1435 const float res = outputAsFloat[idx];
1437 // For NClamp, we have two possible outcomes based on
1438 // whether NaNs are handled or not.
1439 // If either min or max value is NaN, the result is undefined,
1440 // so this test doesn't stress those. If the clamped value is
1441 // NaN, and NaNs are handled, the result is min; if NaNs are not
1442 // handled, they are big values that result in max.
1443 // If all three parameters are NaN, the result should be NaN.
1444 if (!((tcu::Float32(e0).isNaN() && tcu::Float32(res).isNaN()) ||
1445 (deFloatAbs(e0 - res) < 0.00001f) ||
1446 (deFloatAbs(e1 - res) < 0.00001f)))
1453 tcu::TestCaseGroup* createOpNClampGroup (tcu::TestContext& testCtx)
1455 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnclamp", "Test the OpNClamp instruction"));
1456 ComputeShaderSpec spec;
1457 de::Random rnd (deStringHash(group->getName()));
1458 const int numElements = 200;
1459 vector<float> inputFloats1 (numElements, 0);
1460 vector<float> inputFloats2 (numElements, 0);
1461 vector<float> inputFloats3 (numElements, 0);
1462 vector<float> outputFloats (numElements * 2, 0);
1464 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1465 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1466 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats3[0], numElements);
1468 for (size_t ndx = 0; ndx < numElements; ++ndx)
1470 // Results are only defined if max value is bigger than min value.
1471 if (inputFloats2[ndx] > inputFloats3[ndx])
1473 float t = inputFloats2[ndx];
1474 inputFloats2[ndx] = inputFloats3[ndx];
1475 inputFloats3[ndx] = t;
1478 // By default, do the clamp, setting both possible answers
1479 float defaultRes = std::min(std::max(inputFloats1[ndx], inputFloats2[ndx]), inputFloats3[ndx]);
1481 float maxResA = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1482 float maxResB = maxResA;
1484 // Alternate between the NaN cases
1487 inputFloats1[ndx] = TCU_NAN;
1488 // If NaN is handled, the result should be same as the clamp minimum.
1489 // If NaN is not handled, the result should clamp to the clamp maximum.
1490 maxResA = inputFloats2[ndx];
1491 maxResB = inputFloats3[ndx];
1495 // Not a NaN case - only one legal result.
1496 maxResA = defaultRes;
1497 maxResB = defaultRes;
1500 outputFloats[ndx * 2] = maxResA;
1501 outputFloats[ndx * 2 + 1] = maxResB;
1504 // Make the first case a full-NAN case.
1505 inputFloats1[0] = TCU_NAN;
1506 inputFloats2[0] = TCU_NAN;
1507 inputFloats3[0] = TCU_NAN;
1508 outputFloats[0] = TCU_NAN;
1509 outputFloats[1] = TCU_NAN;
1512 "OpCapability Shader\n"
1513 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1514 "OpMemoryModel Logical GLSL450\n"
1515 "OpEntryPoint GLCompute %main \"main\" %id\n"
1516 "OpExecutionMode %main LocalSize 1 1 1\n"
1518 "OpName %main \"main\"\n"
1519 "OpName %id \"gl_GlobalInvocationID\"\n"
1521 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1523 "OpDecorate %buf BufferBlock\n"
1524 "OpDecorate %indata1 DescriptorSet 0\n"
1525 "OpDecorate %indata1 Binding 0\n"
1526 "OpDecorate %indata2 DescriptorSet 0\n"
1527 "OpDecorate %indata2 Binding 1\n"
1528 "OpDecorate %indata3 DescriptorSet 0\n"
1529 "OpDecorate %indata3 Binding 2\n"
1530 "OpDecorate %outdata DescriptorSet 0\n"
1531 "OpDecorate %outdata Binding 3\n"
1532 "OpDecorate %f32arr ArrayStride 4\n"
1533 "OpMemberDecorate %buf 0 Offset 0\n"
1535 + string(getComputeAsmCommonTypes()) +
1537 "%buf = OpTypeStruct %f32arr\n"
1538 "%bufptr = OpTypePointer Uniform %buf\n"
1539 "%indata1 = OpVariable %bufptr Uniform\n"
1540 "%indata2 = OpVariable %bufptr Uniform\n"
1541 "%indata3 = OpVariable %bufptr Uniform\n"
1542 "%outdata = OpVariable %bufptr Uniform\n"
1544 "%id = OpVariable %uvec3ptr Input\n"
1545 "%zero = OpConstant %i32 0\n"
1547 "%main = OpFunction %void None %voidf\n"
1548 "%label = OpLabel\n"
1549 "%idval = OpLoad %uvec3 %id\n"
1550 "%x = OpCompositeExtract %u32 %idval 0\n"
1551 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1552 "%inval1 = OpLoad %f32 %inloc1\n"
1553 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1554 "%inval2 = OpLoad %f32 %inloc2\n"
1555 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1556 "%inval3 = OpLoad %f32 %inloc3\n"
1557 "%rem = OpExtInst %f32 %std450 NClamp %inval1 %inval2 %inval3\n"
1558 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1559 " OpStore %outloc %rem\n"
1563 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1564 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1565 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1566 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1567 spec.numWorkGroups = IVec3(numElements, 1, 1);
1568 spec.verifyIO = &compareNClamp;
1570 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1572 return group.release();
1575 tcu::TestCaseGroup* createOpSRemComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1577 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem", "Test the OpSRem instruction"));
1578 de::Random rnd (deStringHash(group->getName()));
1579 const int numElements = 200;
1581 const struct CaseParams
1584 const char* failMessage; // customized status message
1585 qpTestResult failResult; // override status on failure
1586 int op1Min, op1Max; // operand ranges
1590 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1591 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1593 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1595 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1597 const CaseParams& params = cases[caseNdx];
1598 ComputeShaderSpec spec;
1599 vector<deInt32> inputInts1 (numElements, 0);
1600 vector<deInt32> inputInts2 (numElements, 0);
1601 vector<deInt32> outputInts (numElements, 0);
1603 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1604 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1606 for (int ndx = 0; ndx < numElements; ++ndx)
1608 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1609 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1613 string(getComputeAsmShaderPreamble()) +
1615 "OpName %main \"main\"\n"
1616 "OpName %id \"gl_GlobalInvocationID\"\n"
1618 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1620 "OpDecorate %buf BufferBlock\n"
1621 "OpDecorate %indata1 DescriptorSet 0\n"
1622 "OpDecorate %indata1 Binding 0\n"
1623 "OpDecorate %indata2 DescriptorSet 0\n"
1624 "OpDecorate %indata2 Binding 1\n"
1625 "OpDecorate %outdata DescriptorSet 0\n"
1626 "OpDecorate %outdata Binding 2\n"
1627 "OpDecorate %i32arr ArrayStride 4\n"
1628 "OpMemberDecorate %buf 0 Offset 0\n"
1630 + string(getComputeAsmCommonTypes()) +
1632 "%buf = OpTypeStruct %i32arr\n"
1633 "%bufptr = OpTypePointer Uniform %buf\n"
1634 "%indata1 = OpVariable %bufptr Uniform\n"
1635 "%indata2 = OpVariable %bufptr Uniform\n"
1636 "%outdata = OpVariable %bufptr Uniform\n"
1638 "%id = OpVariable %uvec3ptr Input\n"
1639 "%zero = OpConstant %i32 0\n"
1641 "%main = OpFunction %void None %voidf\n"
1642 "%label = OpLabel\n"
1643 "%idval = OpLoad %uvec3 %id\n"
1644 "%x = OpCompositeExtract %u32 %idval 0\n"
1645 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1646 "%inval1 = OpLoad %i32 %inloc1\n"
1647 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1648 "%inval2 = OpLoad %i32 %inloc2\n"
1649 "%rem = OpSRem %i32 %inval1 %inval2\n"
1650 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1651 " OpStore %outloc %rem\n"
1655 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1656 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1657 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1658 spec.numWorkGroups = IVec3(numElements, 1, 1);
1659 spec.failResult = params.failResult;
1660 spec.failMessage = params.failMessage;
1662 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1665 return group.release();
1668 tcu::TestCaseGroup* createOpSRemComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1670 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem64", "Test the 64-bit OpSRem instruction"));
1671 de::Random rnd (deStringHash(group->getName()));
1672 const int numElements = 200;
1674 const struct CaseParams
1677 const char* failMessage; // customized status message
1678 qpTestResult failResult; // override status on failure
1682 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1683 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1685 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1687 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1689 const CaseParams& params = cases[caseNdx];
1690 ComputeShaderSpec spec;
1691 vector<deInt64> inputInts1 (numElements, 0);
1692 vector<deInt64> inputInts2 (numElements, 0);
1693 vector<deInt64> outputInts (numElements, 0);
1695 if (params.positive)
1697 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1698 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1702 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1703 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1706 for (int ndx = 0; ndx < numElements; ++ndx)
1708 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1709 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1713 "OpCapability Int64\n"
1715 + string(getComputeAsmShaderPreamble()) +
1717 "OpName %main \"main\"\n"
1718 "OpName %id \"gl_GlobalInvocationID\"\n"
1720 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1722 "OpDecorate %buf BufferBlock\n"
1723 "OpDecorate %indata1 DescriptorSet 0\n"
1724 "OpDecorate %indata1 Binding 0\n"
1725 "OpDecorate %indata2 DescriptorSet 0\n"
1726 "OpDecorate %indata2 Binding 1\n"
1727 "OpDecorate %outdata DescriptorSet 0\n"
1728 "OpDecorate %outdata Binding 2\n"
1729 "OpDecorate %i64arr ArrayStride 8\n"
1730 "OpMemberDecorate %buf 0 Offset 0\n"
1732 + string(getComputeAsmCommonTypes())
1733 + string(getComputeAsmCommonInt64Types()) +
1735 "%buf = OpTypeStruct %i64arr\n"
1736 "%bufptr = OpTypePointer Uniform %buf\n"
1737 "%indata1 = OpVariable %bufptr Uniform\n"
1738 "%indata2 = OpVariable %bufptr Uniform\n"
1739 "%outdata = OpVariable %bufptr Uniform\n"
1741 "%id = OpVariable %uvec3ptr Input\n"
1742 "%zero = OpConstant %i64 0\n"
1744 "%main = OpFunction %void None %voidf\n"
1745 "%label = OpLabel\n"
1746 "%idval = OpLoad %uvec3 %id\n"
1747 "%x = OpCompositeExtract %u32 %idval 0\n"
1748 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1749 "%inval1 = OpLoad %i64 %inloc1\n"
1750 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1751 "%inval2 = OpLoad %i64 %inloc2\n"
1752 "%rem = OpSRem %i64 %inval1 %inval2\n"
1753 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1754 " OpStore %outloc %rem\n"
1758 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1759 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1760 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1761 spec.numWorkGroups = IVec3(numElements, 1, 1);
1762 spec.failResult = params.failResult;
1763 spec.failMessage = params.failMessage;
1765 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1768 return group.release();
1771 tcu::TestCaseGroup* createOpSModComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1773 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod", "Test the OpSMod instruction"));
1774 de::Random rnd (deStringHash(group->getName()));
1775 const int numElements = 200;
1777 const struct CaseParams
1780 const char* failMessage; // customized status message
1781 qpTestResult failResult; // override status on failure
1782 int op1Min, op1Max; // operand ranges
1786 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1787 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1789 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1791 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1793 const CaseParams& params = cases[caseNdx];
1795 ComputeShaderSpec spec;
1796 vector<deInt32> inputInts1 (numElements, 0);
1797 vector<deInt32> inputInts2 (numElements, 0);
1798 vector<deInt32> outputInts (numElements, 0);
1800 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1801 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1803 for (int ndx = 0; ndx < numElements; ++ndx)
1805 deInt32 rem = inputInts1[ndx] % inputInts2[ndx];
1808 outputInts[ndx] = 0;
1810 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1812 // They have the same sign
1813 outputInts[ndx] = rem;
1817 // They have opposite sign. The remainder operation takes the
1818 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1819 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1820 // the result has the correct sign and that it is still
1821 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1823 // See also http://mathforum.org/library/drmath/view/52343.html
1824 outputInts[ndx] = rem + inputInts2[ndx];
1829 string(getComputeAsmShaderPreamble()) +
1831 "OpName %main \"main\"\n"
1832 "OpName %id \"gl_GlobalInvocationID\"\n"
1834 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1836 "OpDecorate %buf BufferBlock\n"
1837 "OpDecorate %indata1 DescriptorSet 0\n"
1838 "OpDecorate %indata1 Binding 0\n"
1839 "OpDecorate %indata2 DescriptorSet 0\n"
1840 "OpDecorate %indata2 Binding 1\n"
1841 "OpDecorate %outdata DescriptorSet 0\n"
1842 "OpDecorate %outdata Binding 2\n"
1843 "OpDecorate %i32arr ArrayStride 4\n"
1844 "OpMemberDecorate %buf 0 Offset 0\n"
1846 + string(getComputeAsmCommonTypes()) +
1848 "%buf = OpTypeStruct %i32arr\n"
1849 "%bufptr = OpTypePointer Uniform %buf\n"
1850 "%indata1 = OpVariable %bufptr Uniform\n"
1851 "%indata2 = OpVariable %bufptr Uniform\n"
1852 "%outdata = OpVariable %bufptr Uniform\n"
1854 "%id = OpVariable %uvec3ptr Input\n"
1855 "%zero = OpConstant %i32 0\n"
1857 "%main = OpFunction %void None %voidf\n"
1858 "%label = OpLabel\n"
1859 "%idval = OpLoad %uvec3 %id\n"
1860 "%x = OpCompositeExtract %u32 %idval 0\n"
1861 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1862 "%inval1 = OpLoad %i32 %inloc1\n"
1863 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1864 "%inval2 = OpLoad %i32 %inloc2\n"
1865 "%rem = OpSMod %i32 %inval1 %inval2\n"
1866 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1867 " OpStore %outloc %rem\n"
1871 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1872 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1873 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1874 spec.numWorkGroups = IVec3(numElements, 1, 1);
1875 spec.failResult = params.failResult;
1876 spec.failMessage = params.failMessage;
1878 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1881 return group.release();
1884 tcu::TestCaseGroup* createOpSModComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1886 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod64", "Test the OpSMod instruction"));
1887 de::Random rnd (deStringHash(group->getName()));
1888 const int numElements = 200;
1890 const struct CaseParams
1893 const char* failMessage; // customized status message
1894 qpTestResult failResult; // override status on failure
1898 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1899 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1901 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1903 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1905 const CaseParams& params = cases[caseNdx];
1907 ComputeShaderSpec spec;
1908 vector<deInt64> inputInts1 (numElements, 0);
1909 vector<deInt64> inputInts2 (numElements, 0);
1910 vector<deInt64> outputInts (numElements, 0);
1913 if (params.positive)
1915 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1916 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1920 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1921 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1924 for (int ndx = 0; ndx < numElements; ++ndx)
1926 deInt64 rem = inputInts1[ndx] % inputInts2[ndx];
1929 outputInts[ndx] = 0;
1931 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1933 // They have the same sign
1934 outputInts[ndx] = rem;
1938 // They have opposite sign. The remainder operation takes the
1939 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1940 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1941 // the result has the correct sign and that it is still
1942 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1944 // See also http://mathforum.org/library/drmath/view/52343.html
1945 outputInts[ndx] = rem + inputInts2[ndx];
1950 "OpCapability Int64\n"
1952 + string(getComputeAsmShaderPreamble()) +
1954 "OpName %main \"main\"\n"
1955 "OpName %id \"gl_GlobalInvocationID\"\n"
1957 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1959 "OpDecorate %buf BufferBlock\n"
1960 "OpDecorate %indata1 DescriptorSet 0\n"
1961 "OpDecorate %indata1 Binding 0\n"
1962 "OpDecorate %indata2 DescriptorSet 0\n"
1963 "OpDecorate %indata2 Binding 1\n"
1964 "OpDecorate %outdata DescriptorSet 0\n"
1965 "OpDecorate %outdata Binding 2\n"
1966 "OpDecorate %i64arr ArrayStride 8\n"
1967 "OpMemberDecorate %buf 0 Offset 0\n"
1969 + string(getComputeAsmCommonTypes())
1970 + string(getComputeAsmCommonInt64Types()) +
1972 "%buf = OpTypeStruct %i64arr\n"
1973 "%bufptr = OpTypePointer Uniform %buf\n"
1974 "%indata1 = OpVariable %bufptr Uniform\n"
1975 "%indata2 = OpVariable %bufptr Uniform\n"
1976 "%outdata = OpVariable %bufptr Uniform\n"
1978 "%id = OpVariable %uvec3ptr Input\n"
1979 "%zero = OpConstant %i64 0\n"
1981 "%main = OpFunction %void None %voidf\n"
1982 "%label = OpLabel\n"
1983 "%idval = OpLoad %uvec3 %id\n"
1984 "%x = OpCompositeExtract %u32 %idval 0\n"
1985 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1986 "%inval1 = OpLoad %i64 %inloc1\n"
1987 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1988 "%inval2 = OpLoad %i64 %inloc2\n"
1989 "%rem = OpSMod %i64 %inval1 %inval2\n"
1990 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1991 " OpStore %outloc %rem\n"
1995 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1996 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1997 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1998 spec.numWorkGroups = IVec3(numElements, 1, 1);
1999 spec.failResult = params.failResult;
2000 spec.failMessage = params.failMessage;
2002 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
2005 return group.release();
2008 // Copy contents in the input buffer to the output buffer.
2009 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
2011 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
2012 de::Random rnd (deStringHash(group->getName()));
2013 const int numElements = 100;
2015 // 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.
2016 ComputeShaderSpec spec1;
2017 vector<Vec4> inputFloats1 (numElements);
2018 vector<Vec4> outputFloats1 (numElements);
2020 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
2022 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2023 floorAll(inputFloats1);
2025 for (size_t ndx = 0; ndx < numElements; ++ndx)
2026 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
2029 string(getComputeAsmShaderPreamble()) +
2031 "OpName %main \"main\"\n"
2032 "OpName %id \"gl_GlobalInvocationID\"\n"
2034 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2035 "OpDecorate %vec4arr ArrayStride 16\n"
2037 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2039 "%vec4 = OpTypeVector %f32 4\n"
2040 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
2041 "%vec4ptr_f = OpTypePointer Function %vec4\n"
2042 "%vec4arr = OpTypeRuntimeArray %vec4\n"
2043 "%buf = OpTypeStruct %vec4arr\n"
2044 "%bufptr = OpTypePointer Uniform %buf\n"
2045 "%indata = OpVariable %bufptr Uniform\n"
2046 "%outdata = OpVariable %bufptr Uniform\n"
2048 "%id = OpVariable %uvec3ptr Input\n"
2049 "%zero = OpConstant %i32 0\n"
2050 "%c_f_0 = OpConstant %f32 0.\n"
2051 "%c_f_0_5 = OpConstant %f32 0.5\n"
2052 "%c_f_1_5 = OpConstant %f32 1.5\n"
2053 "%c_f_2_5 = OpConstant %f32 2.5\n"
2054 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
2056 "%main = OpFunction %void None %voidf\n"
2057 "%label = OpLabel\n"
2058 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
2059 "%idval = OpLoad %uvec3 %id\n"
2060 "%x = OpCompositeExtract %u32 %idval 0\n"
2061 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
2062 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
2063 " OpCopyMemory %v_vec4 %inloc\n"
2064 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
2065 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
2066 " OpStore %outloc %add\n"
2070 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
2071 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
2072 spec1.numWorkGroups = IVec3(numElements, 1, 1);
2074 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
2076 // The following case copies a float[100] variable from the input buffer to the output buffer.
2077 ComputeShaderSpec spec2;
2078 vector<float> inputFloats2 (numElements);
2079 vector<float> outputFloats2 (numElements);
2081 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
2083 for (size_t ndx = 0; ndx < numElements; ++ndx)
2084 outputFloats2[ndx] = inputFloats2[ndx];
2087 string(getComputeAsmShaderPreamble()) +
2089 "OpName %main \"main\"\n"
2090 "OpName %id \"gl_GlobalInvocationID\"\n"
2092 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2093 "OpDecorate %f32arr100 ArrayStride 4\n"
2095 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2097 "%hundred = OpConstant %u32 100\n"
2098 "%f32arr100 = OpTypeArray %f32 %hundred\n"
2099 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
2100 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
2101 "%buf = OpTypeStruct %f32arr100\n"
2102 "%bufptr = OpTypePointer Uniform %buf\n"
2103 "%indata = OpVariable %bufptr Uniform\n"
2104 "%outdata = OpVariable %bufptr Uniform\n"
2106 "%id = OpVariable %uvec3ptr Input\n"
2107 "%zero = OpConstant %i32 0\n"
2109 "%main = OpFunction %void None %voidf\n"
2110 "%label = OpLabel\n"
2111 "%var = OpVariable %f32arr100ptr_f Function\n"
2112 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
2113 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
2114 " OpCopyMemory %var %inarr\n"
2115 " OpCopyMemory %outarr %var\n"
2119 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2120 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2121 spec2.numWorkGroups = IVec3(1, 1, 1);
2123 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
2125 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
2126 ComputeShaderSpec spec3;
2127 vector<float> inputFloats3 (16);
2128 vector<float> outputFloats3 (16);
2130 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
2132 for (size_t ndx = 0; ndx < 16; ++ndx)
2133 outputFloats3[ndx] = inputFloats3[ndx];
2136 string(getComputeAsmShaderPreamble()) +
2138 "OpName %main \"main\"\n"
2139 "OpName %id \"gl_GlobalInvocationID\"\n"
2141 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2142 "OpMemberDecorate %buf 0 Offset 0\n"
2143 "OpMemberDecorate %buf 1 Offset 16\n"
2144 "OpMemberDecorate %buf 2 Offset 32\n"
2145 "OpMemberDecorate %buf 3 Offset 48\n"
2147 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2149 "%vec4 = OpTypeVector %f32 4\n"
2150 "%buf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
2151 "%bufptr = OpTypePointer Uniform %buf\n"
2152 "%indata = OpVariable %bufptr Uniform\n"
2153 "%outdata = OpVariable %bufptr Uniform\n"
2154 "%vec4stptr = OpTypePointer Function %buf\n"
2156 "%id = OpVariable %uvec3ptr Input\n"
2157 "%zero = OpConstant %i32 0\n"
2159 "%main = OpFunction %void None %voidf\n"
2160 "%label = OpLabel\n"
2161 "%var = OpVariable %vec4stptr Function\n"
2162 " OpCopyMemory %var %indata\n"
2163 " OpCopyMemory %outdata %var\n"
2167 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2168 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
2169 spec3.numWorkGroups = IVec3(1, 1, 1);
2171 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
2173 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
2174 ComputeShaderSpec spec4;
2175 vector<float> inputFloats4 (numElements);
2176 vector<float> outputFloats4 (numElements);
2178 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
2180 for (size_t ndx = 0; ndx < numElements; ++ndx)
2181 outputFloats4[ndx] = -inputFloats4[ndx];
2184 string(getComputeAsmShaderPreamble()) +
2186 "OpName %main \"main\"\n"
2187 "OpName %id \"gl_GlobalInvocationID\"\n"
2189 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2191 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2193 "%f32ptr_f = OpTypePointer Function %f32\n"
2194 "%id = OpVariable %uvec3ptr Input\n"
2195 "%zero = OpConstant %i32 0\n"
2197 "%main = OpFunction %void None %voidf\n"
2198 "%label = OpLabel\n"
2199 "%var = OpVariable %f32ptr_f Function\n"
2200 "%idval = OpLoad %uvec3 %id\n"
2201 "%x = OpCompositeExtract %u32 %idval 0\n"
2202 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2203 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2204 " OpCopyMemory %var %inloc\n"
2205 "%val = OpLoad %f32 %var\n"
2206 "%neg = OpFNegate %f32 %val\n"
2207 " OpStore %outloc %neg\n"
2211 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2212 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
2213 spec4.numWorkGroups = IVec3(numElements, 1, 1);
2215 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
2217 return group.release();
2220 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
2222 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
2223 ComputeShaderSpec spec;
2224 de::Random rnd (deStringHash(group->getName()));
2225 const int numElements = 100;
2226 vector<float> inputFloats (numElements, 0);
2227 vector<float> outputFloats (numElements, 0);
2229 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
2231 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2232 floorAll(inputFloats);
2234 for (size_t ndx = 0; ndx < numElements; ++ndx)
2235 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
2238 string(getComputeAsmShaderPreamble()) +
2240 "OpName %main \"main\"\n"
2241 "OpName %id \"gl_GlobalInvocationID\"\n"
2243 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2245 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2247 "%fmat = OpTypeMatrix %fvec3 3\n"
2248 "%three = OpConstant %u32 3\n"
2249 "%farr = OpTypeArray %f32 %three\n"
2250 "%fst = OpTypeStruct %f32 %f32\n"
2252 + string(getComputeAsmInputOutputBuffer()) +
2254 "%id = OpVariable %uvec3ptr Input\n"
2255 "%zero = OpConstant %i32 0\n"
2256 "%c_f = OpConstant %f32 1.5\n"
2257 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
2258 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
2259 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
2260 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
2262 "%main = OpFunction %void None %voidf\n"
2263 "%label = OpLabel\n"
2264 "%c_f_copy = OpCopyObject %f32 %c_f\n"
2265 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
2266 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
2267 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
2268 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
2269 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
2270 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
2271 "%farr_elem = OpCompositeExtract %f32 %c_farr_copy 2\n"
2272 "%fst_elem = OpCompositeExtract %f32 %c_fst_copy 1\n"
2273 // Add up. 1.5 * 5 = 7.5.
2274 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
2275 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
2276 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
2277 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
2279 "%idval = OpLoad %uvec3 %id\n"
2280 "%x = OpCompositeExtract %u32 %idval 0\n"
2281 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2282 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2283 "%inval = OpLoad %f32 %inloc\n"
2284 "%add = OpFAdd %f32 %add4 %inval\n"
2285 " OpStore %outloc %add\n"
2288 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2289 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2290 spec.numWorkGroups = IVec3(numElements, 1, 1);
2292 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
2294 return group.release();
2296 // Assembly code used for testing OpUnreachable is based on GLSL source code:
2300 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2301 // float elements[];
2303 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2304 // float elements[];
2307 // void not_called_func() {
2308 // // place OpUnreachable here
2311 // uint modulo4(uint val) {
2312 // switch (val % uint(4)) {
2313 // case 0: return 3;
2314 // case 1: return 2;
2315 // case 2: return 1;
2316 // case 3: return 0;
2317 // default: return 100; // place OpUnreachable here
2323 // // place OpUnreachable here
2327 // uint x = gl_GlobalInvocationID.x;
2328 // if (const5() > modulo4(1000)) {
2329 // output_data.elements[x] = -input_data.elements[x];
2331 // // place OpUnreachable here
2332 // output_data.elements[x] = input_data.elements[x];
2336 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
2338 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
2339 ComputeShaderSpec spec;
2340 de::Random rnd (deStringHash(group->getName()));
2341 const int numElements = 100;
2342 vector<float> positiveFloats (numElements, 0);
2343 vector<float> negativeFloats (numElements, 0);
2345 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2347 for (size_t ndx = 0; ndx < numElements; ++ndx)
2348 negativeFloats[ndx] = -positiveFloats[ndx];
2351 string(getComputeAsmShaderPreamble()) +
2353 "OpSource GLSL 430\n"
2354 "OpName %main \"main\"\n"
2355 "OpName %func_not_called_func \"not_called_func(\"\n"
2356 "OpName %func_modulo4 \"modulo4(u1;\"\n"
2357 "OpName %func_const5 \"const5(\"\n"
2358 "OpName %id \"gl_GlobalInvocationID\"\n"
2360 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2362 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2364 "%u32ptr = OpTypePointer Function %u32\n"
2365 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
2366 "%unitf = OpTypeFunction %u32\n"
2368 "%id = OpVariable %uvec3ptr Input\n"
2369 "%zero = OpConstant %u32 0\n"
2370 "%one = OpConstant %u32 1\n"
2371 "%two = OpConstant %u32 2\n"
2372 "%three = OpConstant %u32 3\n"
2373 "%four = OpConstant %u32 4\n"
2374 "%five = OpConstant %u32 5\n"
2375 "%hundred = OpConstant %u32 100\n"
2376 "%thousand = OpConstant %u32 1000\n"
2378 + string(getComputeAsmInputOutputBuffer()) +
2381 "%main = OpFunction %void None %voidf\n"
2382 "%main_entry = OpLabel\n"
2383 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
2384 "%idval = OpLoad %uvec3 %id\n"
2385 "%x = OpCompositeExtract %u32 %idval 0\n"
2386 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2387 "%inval = OpLoad %f32 %inloc\n"
2388 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2389 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
2390 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
2391 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
2392 " OpSelectionMerge %if_end None\n"
2393 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2394 "%if_true = OpLabel\n"
2395 "%negate = OpFNegate %f32 %inval\n"
2396 " OpStore %outloc %negate\n"
2397 " OpBranch %if_end\n"
2398 "%if_false = OpLabel\n"
2399 " OpUnreachable\n" // Unreachable else branch for if statement
2400 "%if_end = OpLabel\n"
2404 // not_called_function()
2405 "%func_not_called_func = OpFunction %void None %voidf\n"
2406 "%not_called_func_entry = OpLabel\n"
2407 " OpUnreachable\n" // Unreachable entry block in not called static function
2411 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
2412 "%valptr = OpFunctionParameter %u32ptr\n"
2413 "%modulo4_entry = OpLabel\n"
2414 "%val = OpLoad %u32 %valptr\n"
2415 "%modulo = OpUMod %u32 %val %four\n"
2416 " OpSelectionMerge %switch_merge None\n"
2417 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
2418 "%case0 = OpLabel\n"
2419 " OpReturnValue %three\n"
2420 "%case1 = OpLabel\n"
2421 " OpReturnValue %two\n"
2422 "%case2 = OpLabel\n"
2423 " OpReturnValue %one\n"
2424 "%case3 = OpLabel\n"
2425 " OpReturnValue %zero\n"
2426 "%default = OpLabel\n"
2427 " OpUnreachable\n" // Unreachable default case for switch statement
2428 "%switch_merge = OpLabel\n"
2429 " OpUnreachable\n" // Unreachable merge block for switch statement
2433 "%func_const5 = OpFunction %u32 None %unitf\n"
2434 "%const5_entry = OpLabel\n"
2435 " OpReturnValue %five\n"
2436 "%unreachable = OpLabel\n"
2437 " OpUnreachable\n" // Unreachable block in function
2439 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2440 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2441 spec.numWorkGroups = IVec3(numElements, 1, 1);
2443 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
2445 return group.release();
2448 // Assembly code used for testing decoration group is based on GLSL source code:
2452 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
2453 // float elements[];
2455 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
2456 // float elements[];
2458 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
2459 // float elements[];
2461 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
2462 // float elements[];
2464 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
2465 // float elements[];
2467 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
2468 // float elements[];
2472 // uint x = gl_GlobalInvocationID.x;
2473 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
2475 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
2477 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
2478 ComputeShaderSpec spec;
2479 de::Random rnd (deStringHash(group->getName()));
2480 const int numElements = 100;
2481 vector<float> inputFloats0 (numElements, 0);
2482 vector<float> inputFloats1 (numElements, 0);
2483 vector<float> inputFloats2 (numElements, 0);
2484 vector<float> inputFloats3 (numElements, 0);
2485 vector<float> inputFloats4 (numElements, 0);
2486 vector<float> outputFloats (numElements, 0);
2488 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
2489 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
2490 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
2491 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
2492 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
2494 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2495 floorAll(inputFloats0);
2496 floorAll(inputFloats1);
2497 floorAll(inputFloats2);
2498 floorAll(inputFloats3);
2499 floorAll(inputFloats4);
2501 for (size_t ndx = 0; ndx < numElements; ++ndx)
2502 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
2505 string(getComputeAsmShaderPreamble()) +
2507 "OpSource GLSL 430\n"
2508 "OpName %main \"main\"\n"
2509 "OpName %id \"gl_GlobalInvocationID\"\n"
2511 // Not using group decoration on variable.
2512 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2513 // Not using group decoration on type.
2514 "OpDecorate %f32arr ArrayStride 4\n"
2516 "OpDecorate %groups BufferBlock\n"
2517 "OpDecorate %groupm Offset 0\n"
2518 "%groups = OpDecorationGroup\n"
2519 "%groupm = OpDecorationGroup\n"
2521 // Group decoration on multiple structs.
2522 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
2523 // Group decoration on multiple struct members.
2524 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
2526 "OpDecorate %group1 DescriptorSet 0\n"
2527 "OpDecorate %group3 DescriptorSet 0\n"
2528 "OpDecorate %group3 NonWritable\n"
2529 "OpDecorate %group3 Restrict\n"
2530 "%group0 = OpDecorationGroup\n"
2531 "%group1 = OpDecorationGroup\n"
2532 "%group3 = OpDecorationGroup\n"
2534 // Applying the same decoration group multiple times.
2535 "OpGroupDecorate %group1 %outdata\n"
2536 "OpGroupDecorate %group1 %outdata\n"
2537 "OpGroupDecorate %group1 %outdata\n"
2538 "OpDecorate %outdata DescriptorSet 0\n"
2539 "OpDecorate %outdata Binding 5\n"
2540 // Applying decoration group containing nothing.
2541 "OpGroupDecorate %group0 %indata0\n"
2542 "OpDecorate %indata0 DescriptorSet 0\n"
2543 "OpDecorate %indata0 Binding 0\n"
2544 // Applying decoration group containing one decoration.
2545 "OpGroupDecorate %group1 %indata1\n"
2546 "OpDecorate %indata1 Binding 1\n"
2547 // Applying decoration group containing multiple decorations.
2548 "OpGroupDecorate %group3 %indata2 %indata3\n"
2549 "OpDecorate %indata2 Binding 2\n"
2550 "OpDecorate %indata3 Binding 3\n"
2551 // Applying multiple decoration groups (with overlapping).
2552 "OpGroupDecorate %group0 %indata4\n"
2553 "OpGroupDecorate %group1 %indata4\n"
2554 "OpGroupDecorate %group3 %indata4\n"
2555 "OpDecorate %indata4 Binding 4\n"
2557 + string(getComputeAsmCommonTypes()) +
2559 "%id = OpVariable %uvec3ptr Input\n"
2560 "%zero = OpConstant %i32 0\n"
2562 "%outbuf = OpTypeStruct %f32arr\n"
2563 "%outbufptr = OpTypePointer Uniform %outbuf\n"
2564 "%outdata = OpVariable %outbufptr Uniform\n"
2565 "%inbuf0 = OpTypeStruct %f32arr\n"
2566 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
2567 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
2568 "%inbuf1 = OpTypeStruct %f32arr\n"
2569 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
2570 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
2571 "%inbuf2 = OpTypeStruct %f32arr\n"
2572 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
2573 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
2574 "%inbuf3 = OpTypeStruct %f32arr\n"
2575 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
2576 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
2577 "%inbuf4 = OpTypeStruct %f32arr\n"
2578 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
2579 "%indata4 = OpVariable %inbufptr Uniform\n"
2581 "%main = OpFunction %void None %voidf\n"
2582 "%label = OpLabel\n"
2583 "%idval = OpLoad %uvec3 %id\n"
2584 "%x = OpCompositeExtract %u32 %idval 0\n"
2585 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
2586 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
2587 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
2588 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
2589 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
2590 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2591 "%inval0 = OpLoad %f32 %inloc0\n"
2592 "%inval1 = OpLoad %f32 %inloc1\n"
2593 "%inval2 = OpLoad %f32 %inloc2\n"
2594 "%inval3 = OpLoad %f32 %inloc3\n"
2595 "%inval4 = OpLoad %f32 %inloc4\n"
2596 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
2597 "%add1 = OpFAdd %f32 %add0 %inval2\n"
2598 "%add2 = OpFAdd %f32 %add1 %inval3\n"
2599 "%add = OpFAdd %f32 %add2 %inval4\n"
2600 " OpStore %outloc %add\n"
2603 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
2604 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
2605 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2606 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2607 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2608 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2609 spec.numWorkGroups = IVec3(numElements, 1, 1);
2611 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
2613 return group.release();
2616 struct SpecConstantTwoIntCase
2618 const char* caseName;
2619 const char* scDefinition0;
2620 const char* scDefinition1;
2621 const char* scResultType;
2622 const char* scOperation;
2623 deInt32 scActualValue0;
2624 deInt32 scActualValue1;
2625 const char* resultOperation;
2626 vector<deInt32> expectedOutput;
2628 SpecConstantTwoIntCase (const char* name,
2629 const char* definition0,
2630 const char* definition1,
2631 const char* resultType,
2632 const char* operation,
2635 const char* resultOp,
2636 const vector<deInt32>& output)
2638 , scDefinition0 (definition0)
2639 , scDefinition1 (definition1)
2640 , scResultType (resultType)
2641 , scOperation (operation)
2642 , scActualValue0 (value0)
2643 , scActualValue1 (value1)
2644 , resultOperation (resultOp)
2645 , expectedOutput (output) {}
2648 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
2650 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
2651 vector<SpecConstantTwoIntCase> cases;
2652 de::Random rnd (deStringHash(group->getName()));
2653 const int numElements = 100;
2654 vector<deInt32> inputInts (numElements, 0);
2655 vector<deInt32> outputInts1 (numElements, 0);
2656 vector<deInt32> outputInts2 (numElements, 0);
2657 vector<deInt32> outputInts3 (numElements, 0);
2658 vector<deInt32> outputInts4 (numElements, 0);
2659 const StringTemplate shaderTemplate (
2660 "${CAPABILITIES:opt}"
2661 + string(getComputeAsmShaderPreamble()) +
2663 "OpName %main \"main\"\n"
2664 "OpName %id \"gl_GlobalInvocationID\"\n"
2666 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2667 "OpDecorate %sc_0 SpecId 0\n"
2668 "OpDecorate %sc_1 SpecId 1\n"
2669 "OpDecorate %i32arr ArrayStride 4\n"
2671 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2673 "${OPTYPE_DEFINITIONS:opt}"
2674 "%buf = OpTypeStruct %i32arr\n"
2675 "%bufptr = OpTypePointer Uniform %buf\n"
2676 "%indata = OpVariable %bufptr Uniform\n"
2677 "%outdata = OpVariable %bufptr Uniform\n"
2679 "%id = OpVariable %uvec3ptr Input\n"
2680 "%zero = OpConstant %i32 0\n"
2682 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
2683 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
2684 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
2686 "%main = OpFunction %void None %voidf\n"
2687 "%label = OpLabel\n"
2688 "${TYPE_CONVERT:opt}"
2689 "%idval = OpLoad %uvec3 %id\n"
2690 "%x = OpCompositeExtract %u32 %idval 0\n"
2691 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2692 "%inval = OpLoad %i32 %inloc\n"
2693 "%final = ${GEN_RESULT}\n"
2694 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2695 " OpStore %outloc %final\n"
2697 " OpFunctionEnd\n");
2699 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
2701 for (size_t ndx = 0; ndx < numElements; ++ndx)
2703 outputInts1[ndx] = inputInts[ndx] + 42;
2704 outputInts2[ndx] = inputInts[ndx];
2705 outputInts3[ndx] = inputInts[ndx] - 11200;
2706 outputInts4[ndx] = inputInts[ndx] + 1;
2709 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
2710 const char addSc32ToInput[] = "OpIAdd %i32 %inval %sc_final32";
2711 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
2712 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
2714 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
2715 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
2716 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
2717 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
2718 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
2719 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2720 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2721 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
2722 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
2723 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
2724 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
2725 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2726 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2727 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
2728 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
2729 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
2730 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2731 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
2732 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
2733 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
2734 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2735 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
2736 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
2737 cases.push_back(SpecConstantTwoIntCase("inotequal", " %i32 0", " %i32 0", "%bool", "INotEqual %sc_0 %sc_1", 42, 24, selectTrueUsingSc, outputInts2));
2738 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2739 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2740 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2741 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2742 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
2743 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
2744 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
2745 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
2746 cases.push_back(SpecConstantTwoIntCase("sconvert", " %i32 0", " %i32 0", "%i16", "SConvert %sc_0", -11200, 0, addSc32ToInput, outputInts3));
2747 // -969998336 stored as 32-bit two's complement is the binary representation of -11200 as IEEE-754 Float
2748 cases.push_back(SpecConstantTwoIntCase("fconvert", " %f32 0", " %f32 0", "%f64", "FConvert %sc_0", -969998336, 0, addSc32ToInput, outputInts3));
2750 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2752 map<string, string> specializations;
2753 ComputeShaderSpec spec;
2754 ComputeTestFeatures features = COMPUTE_TEST_USES_NONE;
2756 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
2757 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
2758 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
2759 specializations["SC_OP"] = cases[caseNdx].scOperation;
2760 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
2762 // Special SPIR-V code for SConvert-case
2763 if (strcmp(cases[caseNdx].caseName, "sconvert") == 0)
2765 features = COMPUTE_TEST_USES_INT16;
2766 specializations["CAPABILITIES"] = "OpCapability Int16\n"; // Adds 16-bit integer capability
2767 specializations["OPTYPE_DEFINITIONS"] = "%i16 = OpTypeInt 16 1\n"; // Adds 16-bit integer type
2768 specializations["TYPE_CONVERT"] = "%sc_final32 = OpSConvert %i32 %sc_final\n"; // Converts 16-bit integer to 32-bit integer
2771 // Special SPIR-V code for FConvert-case
2772 if (strcmp(cases[caseNdx].caseName, "fconvert") == 0)
2774 features = COMPUTE_TEST_USES_FLOAT64;
2775 specializations["CAPABILITIES"] = "OpCapability Float64\n"; // Adds 64-bit float capability
2776 specializations["OPTYPE_DEFINITIONS"] = "%f64 = OpTypeFloat 64\n"; // Adds 64-bit float type
2777 specializations["TYPE_CONVERT"] = "%sc_final32 = OpConvertFToS %i32 %sc_final\n"; // Converts 64-bit float to 32-bit integer
2780 spec.assembly = shaderTemplate.specialize(specializations);
2781 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2782 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
2783 spec.numWorkGroups = IVec3(numElements, 1, 1);
2784 spec.specConstants.append(cases[caseNdx].scActualValue0);
2785 spec.specConstants.append(cases[caseNdx].scActualValue1);
2787 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec, features));
2790 ComputeShaderSpec spec;
2793 string(getComputeAsmShaderPreamble()) +
2795 "OpName %main \"main\"\n"
2796 "OpName %id \"gl_GlobalInvocationID\"\n"
2798 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2799 "OpDecorate %sc_0 SpecId 0\n"
2800 "OpDecorate %sc_1 SpecId 1\n"
2801 "OpDecorate %sc_2 SpecId 2\n"
2802 "OpDecorate %i32arr ArrayStride 4\n"
2804 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2806 "%ivec3 = OpTypeVector %i32 3\n"
2807 "%buf = OpTypeStruct %i32arr\n"
2808 "%bufptr = OpTypePointer Uniform %buf\n"
2809 "%indata = OpVariable %bufptr Uniform\n"
2810 "%outdata = OpVariable %bufptr Uniform\n"
2812 "%id = OpVariable %uvec3ptr Input\n"
2813 "%zero = OpConstant %i32 0\n"
2814 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
2815 "%vec3_undef = OpUndef %ivec3\n"
2817 "%sc_0 = OpSpecConstant %i32 0\n"
2818 "%sc_1 = OpSpecConstant %i32 0\n"
2819 "%sc_2 = OpSpecConstant %i32 0\n"
2820 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
2821 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
2822 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
2823 "%sc_vec3_0_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
2824 "%sc_vec3_1_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
2825 "%sc_vec3_2_s = OpSpecConstantOp %ivec3 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
2826 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
2827 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
2828 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
2829 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
2830 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
2831 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
2832 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
2834 "%main = OpFunction %void None %voidf\n"
2835 "%label = OpLabel\n"
2836 "%idval = OpLoad %uvec3 %id\n"
2837 "%x = OpCompositeExtract %u32 %idval 0\n"
2838 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2839 "%inval = OpLoad %i32 %inloc\n"
2840 "%final = OpIAdd %i32 %inval %sc_final\n"
2841 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2842 " OpStore %outloc %final\n"
2845 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2846 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
2847 spec.numWorkGroups = IVec3(numElements, 1, 1);
2848 spec.specConstants.append<deInt32>(123);
2849 spec.specConstants.append<deInt32>(56);
2850 spec.specConstants.append<deInt32>(-77);
2852 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
2854 return group.release();
2857 void createOpPhiVartypeTests (de::MovePtr<tcu::TestCaseGroup>& group, tcu::TestContext& testCtx)
2859 ComputeShaderSpec specInt;
2860 ComputeShaderSpec specFloat;
2861 ComputeShaderSpec specVec3;
2862 ComputeShaderSpec specMat4;
2863 ComputeShaderSpec specArray;
2864 ComputeShaderSpec specStruct;
2865 de::Random rnd (deStringHash(group->getName()));
2866 const int numElements = 100;
2867 vector<float> inputFloats (numElements, 0);
2868 vector<float> outputFloats (numElements, 0);
2870 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
2872 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2873 floorAll(inputFloats);
2875 for (size_t ndx = 0; ndx < numElements; ++ndx)
2877 // Just check if the value is positive or not
2878 outputFloats[ndx] = (inputFloats[ndx] > 0) ? 1.0f : -1.0f;
2881 // All of the tests are of the form:
2885 // if (inputdata > 0)
2892 specFloat.assembly =
2893 string(getComputeAsmShaderPreamble()) +
2895 "OpSource GLSL 430\n"
2896 "OpName %main \"main\"\n"
2897 "OpName %id \"gl_GlobalInvocationID\"\n"
2899 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2901 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2903 "%id = OpVariable %uvec3ptr Input\n"
2904 "%zero = OpConstant %i32 0\n"
2905 "%float_0 = OpConstant %f32 0.0\n"
2906 "%float_1 = OpConstant %f32 1.0\n"
2907 "%float_n1 = OpConstant %f32 -1.0\n"
2909 "%main = OpFunction %void None %voidf\n"
2910 "%entry = OpLabel\n"
2911 "%idval = OpLoad %uvec3 %id\n"
2912 "%x = OpCompositeExtract %u32 %idval 0\n"
2913 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2914 "%inval = OpLoad %f32 %inloc\n"
2916 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2917 " OpSelectionMerge %cm None\n"
2918 " OpBranchConditional %comp %tb %fb\n"
2924 "%res = OpPhi %f32 %float_1 %tb %float_n1 %fb\n"
2926 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2927 " OpStore %outloc %res\n"
2931 specFloat.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2932 specFloat.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2933 specFloat.numWorkGroups = IVec3(numElements, 1, 1);
2936 string(getComputeAsmShaderPreamble()) +
2938 "OpSource GLSL 430\n"
2939 "OpName %main \"main\"\n"
2940 "OpName %id \"gl_GlobalInvocationID\"\n"
2942 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2944 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2946 "%id = OpVariable %uvec3ptr Input\n"
2947 "%v4f32 = OpTypeVector %f32 4\n"
2948 "%mat4v4f32 = OpTypeMatrix %v4f32 4\n"
2949 "%zero = OpConstant %i32 0\n"
2950 "%float_0 = OpConstant %f32 0.0\n"
2951 "%float_1 = OpConstant %f32 1.0\n"
2952 "%float_n1 = OpConstant %f32 -1.0\n"
2953 "%m11 = OpConstantComposite %v4f32 %float_1 %float_0 %float_0 %float_0\n"
2954 "%m12 = OpConstantComposite %v4f32 %float_0 %float_1 %float_0 %float_0\n"
2955 "%m13 = OpConstantComposite %v4f32 %float_0 %float_0 %float_1 %float_0\n"
2956 "%m14 = OpConstantComposite %v4f32 %float_0 %float_0 %float_0 %float_1\n"
2957 "%m1 = OpConstantComposite %mat4v4f32 %m11 %m12 %m13 %m14\n"
2958 "%m21 = OpConstantComposite %v4f32 %float_n1 %float_0 %float_0 %float_0\n"
2959 "%m22 = OpConstantComposite %v4f32 %float_0 %float_n1 %float_0 %float_0\n"
2960 "%m23 = OpConstantComposite %v4f32 %float_0 %float_0 %float_n1 %float_0\n"
2961 "%m24 = OpConstantComposite %v4f32 %float_0 %float_0 %float_0 %float_n1\n"
2962 "%m2 = OpConstantComposite %mat4v4f32 %m21 %m22 %m23 %m24\n"
2964 "%main = OpFunction %void None %voidf\n"
2965 "%entry = OpLabel\n"
2966 "%idval = OpLoad %uvec3 %id\n"
2967 "%x = OpCompositeExtract %u32 %idval 0\n"
2968 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2969 "%inval = OpLoad %f32 %inloc\n"
2971 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2972 " OpSelectionMerge %cm None\n"
2973 " OpBranchConditional %comp %tb %fb\n"
2979 "%mres = OpPhi %mat4v4f32 %m1 %tb %m2 %fb\n"
2980 "%res = OpCompositeExtract %f32 %mres 2 2\n"
2982 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2983 " OpStore %outloc %res\n"
2987 specMat4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2988 specMat4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2989 specMat4.numWorkGroups = IVec3(numElements, 1, 1);
2992 string(getComputeAsmShaderPreamble()) +
2994 "OpSource GLSL 430\n"
2995 "OpName %main \"main\"\n"
2996 "OpName %id \"gl_GlobalInvocationID\"\n"
2998 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3000 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3002 "%id = OpVariable %uvec3ptr Input\n"
3003 "%zero = OpConstant %i32 0\n"
3004 "%float_0 = OpConstant %f32 0.0\n"
3005 "%float_1 = OpConstant %f32 1.0\n"
3006 "%float_n1 = OpConstant %f32 -1.0\n"
3007 "%v1 = OpConstantComposite %fvec3 %float_1 %float_1 %float_1\n"
3008 "%v2 = OpConstantComposite %fvec3 %float_n1 %float_n1 %float_n1\n"
3010 "%main = OpFunction %void None %voidf\n"
3011 "%entry = OpLabel\n"
3012 "%idval = OpLoad %uvec3 %id\n"
3013 "%x = OpCompositeExtract %u32 %idval 0\n"
3014 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3015 "%inval = OpLoad %f32 %inloc\n"
3017 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3018 " OpSelectionMerge %cm None\n"
3019 " OpBranchConditional %comp %tb %fb\n"
3025 "%vres = OpPhi %fvec3 %v1 %tb %v2 %fb\n"
3026 "%res = OpCompositeExtract %f32 %vres 2\n"
3028 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3029 " OpStore %outloc %res\n"
3033 specVec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3034 specVec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3035 specVec3.numWorkGroups = IVec3(numElements, 1, 1);
3038 string(getComputeAsmShaderPreamble()) +
3040 "OpSource GLSL 430\n"
3041 "OpName %main \"main\"\n"
3042 "OpName %id \"gl_GlobalInvocationID\"\n"
3044 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3046 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3048 "%id = OpVariable %uvec3ptr Input\n"
3049 "%zero = OpConstant %i32 0\n"
3050 "%float_0 = OpConstant %f32 0.0\n"
3051 "%i1 = OpConstant %i32 1\n"
3052 "%i2 = OpConstant %i32 -1\n"
3054 "%main = OpFunction %void None %voidf\n"
3055 "%entry = OpLabel\n"
3056 "%idval = OpLoad %uvec3 %id\n"
3057 "%x = OpCompositeExtract %u32 %idval 0\n"
3058 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3059 "%inval = OpLoad %f32 %inloc\n"
3061 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3062 " OpSelectionMerge %cm None\n"
3063 " OpBranchConditional %comp %tb %fb\n"
3069 "%ires = OpPhi %i32 %i1 %tb %i2 %fb\n"
3070 "%res = OpConvertSToF %f32 %ires\n"
3072 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3073 " OpStore %outloc %res\n"
3077 specInt.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3078 specInt.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3079 specInt.numWorkGroups = IVec3(numElements, 1, 1);
3081 specArray.assembly =
3082 string(getComputeAsmShaderPreamble()) +
3084 "OpSource GLSL 430\n"
3085 "OpName %main \"main\"\n"
3086 "OpName %id \"gl_GlobalInvocationID\"\n"
3088 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3090 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3092 "%id = OpVariable %uvec3ptr Input\n"
3093 "%zero = OpConstant %i32 0\n"
3094 "%u7 = OpConstant %u32 7\n"
3095 "%float_0 = OpConstant %f32 0.0\n"
3096 "%float_1 = OpConstant %f32 1.0\n"
3097 "%float_n1 = OpConstant %f32 -1.0\n"
3098 "%f32a7 = OpTypeArray %f32 %u7\n"
3099 "%a1 = OpConstantComposite %f32a7 %float_1 %float_1 %float_1 %float_1 %float_1 %float_1 %float_1\n"
3100 "%a2 = OpConstantComposite %f32a7 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1\n"
3101 "%main = OpFunction %void None %voidf\n"
3102 "%entry = OpLabel\n"
3103 "%idval = OpLoad %uvec3 %id\n"
3104 "%x = OpCompositeExtract %u32 %idval 0\n"
3105 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3106 "%inval = OpLoad %f32 %inloc\n"
3108 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3109 " OpSelectionMerge %cm None\n"
3110 " OpBranchConditional %comp %tb %fb\n"
3116 "%ares = OpPhi %f32a7 %a1 %tb %a2 %fb\n"
3117 "%res = OpCompositeExtract %f32 %ares 5\n"
3119 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3120 " OpStore %outloc %res\n"
3124 specArray.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3125 specArray.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3126 specArray.numWorkGroups = IVec3(numElements, 1, 1);
3128 specStruct.assembly =
3129 string(getComputeAsmShaderPreamble()) +
3131 "OpSource GLSL 430\n"
3132 "OpName %main \"main\"\n"
3133 "OpName %id \"gl_GlobalInvocationID\"\n"
3135 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3137 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3139 "%id = OpVariable %uvec3ptr Input\n"
3140 "%zero = OpConstant %i32 0\n"
3141 "%float_0 = OpConstant %f32 0.0\n"
3142 "%float_1 = OpConstant %f32 1.0\n"
3143 "%float_n1 = OpConstant %f32 -1.0\n"
3145 "%v2f32 = OpTypeVector %f32 2\n"
3146 "%Data2 = OpTypeStruct %f32 %v2f32\n"
3147 "%Data = OpTypeStruct %Data2 %f32\n"
3149 "%in1a = OpConstantComposite %v2f32 %float_1 %float_1\n"
3150 "%in1b = OpConstantComposite %Data2 %float_1 %in1a\n"
3151 "%s1 = OpConstantComposite %Data %in1b %float_1\n"
3152 "%in2a = OpConstantComposite %v2f32 %float_n1 %float_n1\n"
3153 "%in2b = OpConstantComposite %Data2 %float_n1 %in2a\n"
3154 "%s2 = OpConstantComposite %Data %in2b %float_n1\n"
3156 "%main = OpFunction %void None %voidf\n"
3157 "%entry = OpLabel\n"
3158 "%idval = OpLoad %uvec3 %id\n"
3159 "%x = OpCompositeExtract %u32 %idval 0\n"
3160 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3161 "%inval = OpLoad %f32 %inloc\n"
3163 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3164 " OpSelectionMerge %cm None\n"
3165 " OpBranchConditional %comp %tb %fb\n"
3171 "%sres = OpPhi %Data %s1 %tb %s2 %fb\n"
3172 "%res = OpCompositeExtract %f32 %sres 0 0\n"
3174 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3175 " OpStore %outloc %res\n"
3179 specStruct.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3180 specStruct.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3181 specStruct.numWorkGroups = IVec3(numElements, 1, 1);
3183 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_int", "OpPhi with int variables", specInt));
3184 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_float", "OpPhi with float variables", specFloat));
3185 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_vec3", "OpPhi with vec3 variables", specVec3));
3186 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_mat4", "OpPhi with mat4 variables", specMat4));
3187 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_array", "OpPhi with array variables", specArray));
3188 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_struct", "OpPhi with struct variables", specStruct));
3191 string generateConstantDefinitions (int count)
3193 std::ostringstream r;
3194 for (int i = 0; i < count; i++)
3195 r << "%cf" << (i * 10 + 5) << " = OpConstant %f32 " <<(i * 10 + 5) << ".0\n";
3200 string generateSwitchCases (int count)
3202 std::ostringstream r;
3203 for (int i = 0; i < count; i++)
3204 r << " " << i << " %case" << i;
3209 string generateSwitchTargets (int count)
3211 std::ostringstream r;
3212 for (int i = 0; i < count; i++)
3213 r << "%case" << i << " = OpLabel\n OpBranch %phi\n";
3218 string generateOpPhiParams (int count)
3220 std::ostringstream r;
3221 for (int i = 0; i < count; i++)
3222 r << " %cf" << (i * 10 + 5) << " %case" << i;
3227 string generateIntWidth (int value)
3229 std::ostringstream r;
3234 // Expand input string by injecting "ABC" between the input
3235 // string characters. The acc/add/treshold parameters are used
3236 // to skip some of the injections to make the result less
3237 // uniform (and a lot shorter).
3238 string expandOpPhiCase5 (const string& s, int &acc, int add, int treshold)
3240 std::ostringstream res;
3241 const char* p = s.c_str();
3257 // Calculate expected result based on the code string
3258 float calcOpPhiCase5 (float val, const string& s)
3260 const char* p = s.c_str();
3263 const float tv[8] = { 0.5f, 1.5f, 3.5f, 7.5f, 15.5f, 31.5f, 63.5f, 127.5f };
3264 const float v = deFloatAbs(val);
3269 for (int i = 7; i >= 0; --i)
3270 x[i] = std::fmod((float)v, (float)(2 << i));
3271 for (int i = 7; i >= 0; --i)
3272 b[i] = x[i] > tv[i];
3279 if (skip == 0 && b[depth])
3290 if (b[depth] || skip)
3304 // In the code string, the letters represent the following:
3307 // if (certain bit is set)
3318 // AABCBC leads to if(){r++;if(){r++;}else{}}else{}
3319 // ABABCC leads to if(){r++;}else{if(){r++;}else{}}
3320 // ABCABC leads to if(){r++;}else{}if(){r++;}else{}
3322 // Code generation gets a bit complicated due to the else-branches,
3323 // which do not generate new values. Thus, the generator needs to
3324 // keep track of the previous variable change seen by the else
3326 string generateOpPhiCase5 (const string& s)
3328 std::stack<int> idStack;
3329 std::stack<std::string> value;
3330 std::stack<std::string> valueLabel;
3331 std::stack<std::string> mergeLeft;
3332 std::stack<std::string> mergeRight;
3333 std::ostringstream res;
3334 const char* p = s.c_str();
3340 value.push("%f32_0");
3341 valueLabel.push("%f32_0 %entry");
3349 idStack.push(currId);
3350 res << "\tOpSelectionMerge %m" << currId << " None\n";
3351 res << "\tOpBranchConditional %b" << depth << " %t" << currId << " %f" << currId << "\n";
3352 res << "%t" << currId << " = OpLabel\n";
3353 res << "%rt" << currId << " = OpFAdd %f32 " << value.top() << " %f32_1\n";
3354 std::ostringstream tag;
3355 tag << "%rt" << currId;
3356 value.push(tag.str());
3357 tag << " %t" << currId;
3358 valueLabel.push(tag.str());
3363 mergeLeft.push(valueLabel.top());
3366 res << "\tOpBranch %m" << currId << "\n";
3367 res << "%f" << currId << " = OpLabel\n";
3368 std::ostringstream tag;
3369 tag << value.top() << " %f" << currId;
3371 valueLabel.push(tag.str());
3376 mergeRight.push(valueLabel.top());
3377 res << "\tOpBranch %m" << currId << "\n";
3378 res << "%m" << currId << " = OpLabel\n";
3380 res << "%res"; // last result goes to %res
3382 res << "%rm" << currId;
3383 res << " = OpPhi %f32 " << mergeLeft.top() << " " << mergeRight.top() << "\n";
3384 std::ostringstream tag;
3385 tag << "%rm" << currId;
3387 value.push(tag.str());
3388 tag << " %m" << currId;
3390 valueLabel.push(tag.str());
3395 currId = idStack.top();
3403 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
3405 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
3406 ComputeShaderSpec spec1;
3407 ComputeShaderSpec spec2;
3408 ComputeShaderSpec spec3;
3409 ComputeShaderSpec spec4;
3410 ComputeShaderSpec spec5;
3411 de::Random rnd (deStringHash(group->getName()));
3412 const int numElements = 100;
3413 vector<float> inputFloats (numElements, 0);
3414 vector<float> outputFloats1 (numElements, 0);
3415 vector<float> outputFloats2 (numElements, 0);
3416 vector<float> outputFloats3 (numElements, 0);
3417 vector<float> outputFloats4 (numElements, 0);
3418 vector<float> outputFloats5 (numElements, 0);
3419 std::string codestring = "ABC";
3420 const int test4Width = 1024;
3422 // Build case 5 code string. Each iteration makes the hierarchy more complicated.
3423 // 9 iterations with (7, 24) parameters makes the hierarchy 8 deep with about 1500 lines of
3425 for (int i = 0, acc = 0; i < 9; i++)
3426 codestring = expandOpPhiCase5(codestring, acc, 7, 24);
3428 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
3430 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3431 floorAll(inputFloats);
3433 for (size_t ndx = 0; ndx < numElements; ++ndx)
3437 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
3438 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
3439 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
3442 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
3443 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
3445 int index4 = (int)deFloor(deAbs((float)ndx * inputFloats[ndx]));
3446 outputFloats4[ndx] = (float)(index4 % test4Width) * 10.0f + 5.0f;
3448 outputFloats5[ndx] = calcOpPhiCase5(inputFloats[ndx], codestring);
3452 string(getComputeAsmShaderPreamble()) +
3454 "OpSource GLSL 430\n"
3455 "OpName %main \"main\"\n"
3456 "OpName %id \"gl_GlobalInvocationID\"\n"
3458 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3460 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3462 "%id = OpVariable %uvec3ptr Input\n"
3463 "%zero = OpConstant %i32 0\n"
3464 "%three = OpConstant %u32 3\n"
3465 "%constf5p5 = OpConstant %f32 5.5\n"
3466 "%constf20p5 = OpConstant %f32 20.5\n"
3467 "%constf1p75 = OpConstant %f32 1.75\n"
3468 "%constf8p5 = OpConstant %f32 8.5\n"
3469 "%constf6p5 = OpConstant %f32 6.5\n"
3471 "%main = OpFunction %void None %voidf\n"
3472 "%entry = OpLabel\n"
3473 "%idval = OpLoad %uvec3 %id\n"
3474 "%x = OpCompositeExtract %u32 %idval 0\n"
3475 "%selector = OpUMod %u32 %x %three\n"
3476 " OpSelectionMerge %phi None\n"
3477 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
3479 // Case 1 before OpPhi.
3480 "%case1 = OpLabel\n"
3483 "%default = OpLabel\n"
3487 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
3488 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3489 "%inval = OpLoad %f32 %inloc\n"
3490 "%add = OpFAdd %f32 %inval %operand\n"
3491 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3492 " OpStore %outloc %add\n"
3495 // Case 0 after OpPhi.
3496 "%case0 = OpLabel\n"
3500 // Case 2 after OpPhi.
3501 "%case2 = OpLabel\n"
3505 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3506 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
3507 spec1.numWorkGroups = IVec3(numElements, 1, 1);
3509 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
3512 string(getComputeAsmShaderPreamble()) +
3514 "OpName %main \"main\"\n"
3515 "OpName %id \"gl_GlobalInvocationID\"\n"
3517 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3519 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3521 "%id = OpVariable %uvec3ptr Input\n"
3522 "%zero = OpConstant %i32 0\n"
3523 "%one = OpConstant %i32 1\n"
3524 "%three = OpConstant %i32 3\n"
3525 "%constf6p5 = OpConstant %f32 6.5\n"
3527 "%main = OpFunction %void None %voidf\n"
3528 "%entry = OpLabel\n"
3529 "%idval = OpLoad %uvec3 %id\n"
3530 "%x = OpCompositeExtract %u32 %idval 0\n"
3531 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3532 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3533 "%inval = OpLoad %f32 %inloc\n"
3537 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
3538 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
3539 "%step_next = OpIAdd %i32 %step %one\n"
3540 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
3541 "%still_loop = OpSLessThan %bool %step %three\n"
3542 " OpLoopMerge %exit %phi None\n"
3543 " OpBranchConditional %still_loop %phi %exit\n"
3546 " OpStore %outloc %accum\n"
3549 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3550 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
3551 spec2.numWorkGroups = IVec3(numElements, 1, 1);
3553 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
3556 string(getComputeAsmShaderPreamble()) +
3558 "OpName %main \"main\"\n"
3559 "OpName %id \"gl_GlobalInvocationID\"\n"
3561 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3563 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3565 "%f32ptr_f = OpTypePointer Function %f32\n"
3566 "%id = OpVariable %uvec3ptr Input\n"
3567 "%true = OpConstantTrue %bool\n"
3568 "%false = OpConstantFalse %bool\n"
3569 "%zero = OpConstant %i32 0\n"
3570 "%constf8p5 = OpConstant %f32 8.5\n"
3572 "%main = OpFunction %void None %voidf\n"
3573 "%entry = OpLabel\n"
3574 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
3575 "%idval = OpLoad %uvec3 %id\n"
3576 "%x = OpCompositeExtract %u32 %idval 0\n"
3577 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3578 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3579 "%a_init = OpLoad %f32 %inloc\n"
3580 "%b_init = OpLoad %f32 %b\n"
3584 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
3585 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
3586 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
3587 " OpLoopMerge %exit %phi None\n"
3588 " OpBranchConditional %still_loop %phi %exit\n"
3591 "%sub = OpFSub %f32 %a_next %b_next\n"
3592 " OpStore %outloc %sub\n"
3595 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3596 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
3597 spec3.numWorkGroups = IVec3(numElements, 1, 1);
3599 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
3602 "OpCapability Shader\n"
3603 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
3604 "OpMemoryModel Logical GLSL450\n"
3605 "OpEntryPoint GLCompute %main \"main\" %id\n"
3606 "OpExecutionMode %main LocalSize 1 1 1\n"
3608 "OpSource GLSL 430\n"
3609 "OpName %main \"main\"\n"
3610 "OpName %id \"gl_GlobalInvocationID\"\n"
3612 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3614 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3616 "%id = OpVariable %uvec3ptr Input\n"
3617 "%zero = OpConstant %i32 0\n"
3618 "%cimod = OpConstant %u32 " + generateIntWidth(test4Width) + "\n"
3620 + generateConstantDefinitions(test4Width) +
3622 "%main = OpFunction %void None %voidf\n"
3623 "%entry = OpLabel\n"
3624 "%idval = OpLoad %uvec3 %id\n"
3625 "%x = OpCompositeExtract %u32 %idval 0\n"
3626 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3627 "%inval = OpLoad %f32 %inloc\n"
3628 "%xf = OpConvertUToF %f32 %x\n"
3629 "%xm = OpFMul %f32 %xf %inval\n"
3630 "%xa = OpExtInst %f32 %ext FAbs %xm\n"
3631 "%xi = OpConvertFToU %u32 %xa\n"
3632 "%selector = OpUMod %u32 %xi %cimod\n"
3633 " OpSelectionMerge %phi None\n"
3634 " OpSwitch %selector %default "
3636 + generateSwitchCases(test4Width) +
3638 "%default = OpLabel\n"
3641 + generateSwitchTargets(test4Width) +
3644 "%result = OpPhi %f32"
3646 + generateOpPhiParams(test4Width) +
3648 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3649 " OpStore %outloc %result\n"
3653 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3654 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
3655 spec4.numWorkGroups = IVec3(numElements, 1, 1);
3657 group->addChild(new SpvAsmComputeShaderCase(testCtx, "wide", "OpPhi with a lot of parameters", spec4));
3660 "OpCapability Shader\n"
3661 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
3662 "OpMemoryModel Logical GLSL450\n"
3663 "OpEntryPoint GLCompute %main \"main\" %id\n"
3664 "OpExecutionMode %main LocalSize 1 1 1\n"
3665 "%code = OpString \"" + codestring + "\"\n"
3667 "OpSource GLSL 430\n"
3668 "OpName %main \"main\"\n"
3669 "OpName %id \"gl_GlobalInvocationID\"\n"
3671 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3673 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3675 "%id = OpVariable %uvec3ptr Input\n"
3676 "%zero = OpConstant %i32 0\n"
3677 "%f32_0 = OpConstant %f32 0.0\n"
3678 "%f32_0_5 = OpConstant %f32 0.5\n"
3679 "%f32_1 = OpConstant %f32 1.0\n"
3680 "%f32_1_5 = OpConstant %f32 1.5\n"
3681 "%f32_2 = OpConstant %f32 2.0\n"
3682 "%f32_3_5 = OpConstant %f32 3.5\n"
3683 "%f32_4 = OpConstant %f32 4.0\n"
3684 "%f32_7_5 = OpConstant %f32 7.5\n"
3685 "%f32_8 = OpConstant %f32 8.0\n"
3686 "%f32_15_5 = OpConstant %f32 15.5\n"
3687 "%f32_16 = OpConstant %f32 16.0\n"
3688 "%f32_31_5 = OpConstant %f32 31.5\n"
3689 "%f32_32 = OpConstant %f32 32.0\n"
3690 "%f32_63_5 = OpConstant %f32 63.5\n"
3691 "%f32_64 = OpConstant %f32 64.0\n"
3692 "%f32_127_5 = OpConstant %f32 127.5\n"
3693 "%f32_128 = OpConstant %f32 128.0\n"
3694 "%f32_256 = OpConstant %f32 256.0\n"
3696 "%main = OpFunction %void None %voidf\n"
3697 "%entry = OpLabel\n"
3698 "%idval = OpLoad %uvec3 %id\n"
3699 "%x = OpCompositeExtract %u32 %idval 0\n"
3700 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3701 "%inval = OpLoad %f32 %inloc\n"
3703 "%xabs = OpExtInst %f32 %ext FAbs %inval\n"
3704 "%x8 = OpFMod %f32 %xabs %f32_256\n"
3705 "%x7 = OpFMod %f32 %xabs %f32_128\n"
3706 "%x6 = OpFMod %f32 %xabs %f32_64\n"
3707 "%x5 = OpFMod %f32 %xabs %f32_32\n"
3708 "%x4 = OpFMod %f32 %xabs %f32_16\n"
3709 "%x3 = OpFMod %f32 %xabs %f32_8\n"
3710 "%x2 = OpFMod %f32 %xabs %f32_4\n"
3711 "%x1 = OpFMod %f32 %xabs %f32_2\n"
3713 "%b7 = OpFOrdGreaterThanEqual %bool %x8 %f32_127_5\n"
3714 "%b6 = OpFOrdGreaterThanEqual %bool %x7 %f32_63_5\n"
3715 "%b5 = OpFOrdGreaterThanEqual %bool %x6 %f32_31_5\n"
3716 "%b4 = OpFOrdGreaterThanEqual %bool %x5 %f32_15_5\n"
3717 "%b3 = OpFOrdGreaterThanEqual %bool %x4 %f32_7_5\n"
3718 "%b2 = OpFOrdGreaterThanEqual %bool %x3 %f32_3_5\n"
3719 "%b1 = OpFOrdGreaterThanEqual %bool %x2 %f32_1_5\n"
3720 "%b0 = OpFOrdGreaterThanEqual %bool %x1 %f32_0_5\n"
3722 + generateOpPhiCase5(codestring) +
3724 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3725 " OpStore %outloc %res\n"
3729 spec5.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3730 spec5.outputs.push_back(BufferSp(new Float32Buffer(outputFloats5)));
3731 spec5.numWorkGroups = IVec3(numElements, 1, 1);
3733 group->addChild(new SpvAsmComputeShaderCase(testCtx, "nested", "Stress OpPhi with a lot of nesting", spec5));
3735 createOpPhiVartypeTests(group, testCtx);
3737 return group.release();
3740 // Assembly code used for testing block order is based on GLSL source code:
3744 // layout(std140, set = 0, binding = 0) readonly buffer Input {
3745 // float elements[];
3747 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
3748 // float elements[];
3752 // uint x = gl_GlobalInvocationID.x;
3753 // output_data.elements[x] = input_data.elements[x];
3754 // if (x > uint(50)) {
3755 // switch (x % uint(3)) {
3756 // case 0: output_data.elements[x] += 1.5f; break;
3757 // case 1: output_data.elements[x] += 42.f; break;
3758 // case 2: output_data.elements[x] -= 27.f; break;
3762 // output_data.elements[x] = -input_data.elements[x];
3765 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
3767 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
3768 ComputeShaderSpec spec;
3769 de::Random rnd (deStringHash(group->getName()));
3770 const int numElements = 100;
3771 vector<float> inputFloats (numElements, 0);
3772 vector<float> outputFloats (numElements, 0);
3774 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3776 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3777 floorAll(inputFloats);
3779 for (size_t ndx = 0; ndx <= 50; ++ndx)
3780 outputFloats[ndx] = -inputFloats[ndx];
3782 for (size_t ndx = 51; ndx < numElements; ++ndx)
3786 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
3787 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
3788 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
3794 string(getComputeAsmShaderPreamble()) +
3796 "OpSource GLSL 430\n"
3797 "OpName %main \"main\"\n"
3798 "OpName %id \"gl_GlobalInvocationID\"\n"
3800 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3802 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3804 "%u32ptr = OpTypePointer Function %u32\n"
3805 "%u32ptr_input = OpTypePointer Input %u32\n"
3807 + string(getComputeAsmInputOutputBuffer()) +
3809 "%id = OpVariable %uvec3ptr Input\n"
3810 "%zero = OpConstant %i32 0\n"
3811 "%const3 = OpConstant %u32 3\n"
3812 "%const50 = OpConstant %u32 50\n"
3813 "%constf1p5 = OpConstant %f32 1.5\n"
3814 "%constf27 = OpConstant %f32 27.0\n"
3815 "%constf42 = OpConstant %f32 42.0\n"
3817 "%main = OpFunction %void None %voidf\n"
3820 "%entry = OpLabel\n"
3822 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
3823 "%xvar = OpVariable %u32ptr Function\n"
3824 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
3825 "%x = OpLoad %u32 %xptr\n"
3826 " OpStore %xvar %x\n"
3828 "%cmp = OpUGreaterThan %bool %x %const50\n"
3829 " OpSelectionMerge %if_merge None\n"
3830 " OpBranchConditional %cmp %if_true %if_false\n"
3832 // False branch for if-statement: placed in the middle of switch cases and before true branch.
3833 "%if_false = OpLabel\n"
3834 "%x_f = OpLoad %u32 %xvar\n"
3835 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
3836 "%inval_f = OpLoad %f32 %inloc_f\n"
3837 "%negate = OpFNegate %f32 %inval_f\n"
3838 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
3839 " OpStore %outloc_f %negate\n"
3840 " OpBranch %if_merge\n"
3842 // Merge block for if-statement: placed in the middle of true and false branch.
3843 "%if_merge = OpLabel\n"
3846 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
3847 "%if_true = OpLabel\n"
3848 "%xval_t = OpLoad %u32 %xvar\n"
3849 "%mod = OpUMod %u32 %xval_t %const3\n"
3850 " OpSelectionMerge %switch_merge None\n"
3851 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
3853 // Merge block for switch-statement: placed before the case
3854 // bodies. But it must follow OpSwitch which dominates it.
3855 "%switch_merge = OpLabel\n"
3856 " OpBranch %if_merge\n"
3858 // Case 1 for switch-statement: placed before case 0.
3859 // It must follow the OpSwitch that dominates it.
3860 "%case1 = OpLabel\n"
3861 "%x_1 = OpLoad %u32 %xvar\n"
3862 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
3863 "%inval_1 = OpLoad %f32 %inloc_1\n"
3864 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
3865 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
3866 " OpStore %outloc_1 %addf42\n"
3867 " OpBranch %switch_merge\n"
3869 // Case 2 for switch-statement.
3870 "%case2 = OpLabel\n"
3871 "%x_2 = OpLoad %u32 %xvar\n"
3872 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
3873 "%inval_2 = OpLoad %f32 %inloc_2\n"
3874 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
3875 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
3876 " OpStore %outloc_2 %subf27\n"
3877 " OpBranch %switch_merge\n"
3879 // Default case for switch-statement: placed in the middle of normal cases.
3880 "%default = OpLabel\n"
3881 " OpBranch %switch_merge\n"
3883 // Case 0 for switch-statement: out of order.
3884 "%case0 = OpLabel\n"
3885 "%x_0 = OpLoad %u32 %xvar\n"
3886 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
3887 "%inval_0 = OpLoad %f32 %inloc_0\n"
3888 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
3889 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
3890 " OpStore %outloc_0 %addf1p5\n"
3891 " OpBranch %switch_merge\n"
3894 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3895 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3896 spec.numWorkGroups = IVec3(numElements, 1, 1);
3898 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
3900 return group.release();
3903 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
3905 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
3906 ComputeShaderSpec spec1;
3907 ComputeShaderSpec spec2;
3908 de::Random rnd (deStringHash(group->getName()));
3909 const int numElements = 100;
3910 vector<float> inputFloats (numElements, 0);
3911 vector<float> outputFloats1 (numElements, 0);
3912 vector<float> outputFloats2 (numElements, 0);
3913 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
3915 for (size_t ndx = 0; ndx < numElements; ++ndx)
3917 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
3918 outputFloats2[ndx] = -inputFloats[ndx];
3921 const string assembly(
3922 "OpCapability Shader\n"
3923 "OpCapability ClipDistance\n"
3924 "OpMemoryModel Logical GLSL450\n"
3925 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
3926 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
3927 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
3928 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
3929 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
3930 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
3932 "OpName %comp_main1 \"entrypoint1\"\n"
3933 "OpName %comp_main2 \"entrypoint2\"\n"
3934 "OpName %vert_main \"entrypoint2\"\n"
3935 "OpName %id \"gl_GlobalInvocationID\"\n"
3936 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
3937 "OpName %vertexIndex \"gl_VertexIndex\"\n"
3938 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
3939 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
3940 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
3941 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
3943 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3944 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
3945 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
3946 "OpDecorate %vert_builtin_st Block\n"
3947 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
3948 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
3949 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
3951 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3953 "%zero = OpConstant %i32 0\n"
3954 "%one = OpConstant %u32 1\n"
3955 "%c_f32_1 = OpConstant %f32 1\n"
3957 "%i32inputptr = OpTypePointer Input %i32\n"
3958 "%vec4 = OpTypeVector %f32 4\n"
3959 "%vec4ptr = OpTypePointer Output %vec4\n"
3960 "%f32arr1 = OpTypeArray %f32 %one\n"
3961 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
3962 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
3963 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
3965 "%id = OpVariable %uvec3ptr Input\n"
3966 "%vertexIndex = OpVariable %i32inputptr Input\n"
3967 "%instanceIndex = OpVariable %i32inputptr Input\n"
3968 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
3970 // gl_Position = vec4(1.);
3971 "%vert_main = OpFunction %void None %voidf\n"
3972 "%vert_entry = OpLabel\n"
3973 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
3974 " OpStore %position %c_vec4_1\n"
3979 "%comp_main1 = OpFunction %void None %voidf\n"
3980 "%comp1_entry = OpLabel\n"
3981 "%idval1 = OpLoad %uvec3 %id\n"
3982 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
3983 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
3984 "%inval1 = OpLoad %f32 %inloc1\n"
3985 "%add = OpFAdd %f32 %inval1 %inval1\n"
3986 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
3987 " OpStore %outloc1 %add\n"
3992 "%comp_main2 = OpFunction %void None %voidf\n"
3993 "%comp2_entry = OpLabel\n"
3994 "%idval2 = OpLoad %uvec3 %id\n"
3995 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
3996 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
3997 "%inval2 = OpLoad %f32 %inloc2\n"
3998 "%neg = OpFNegate %f32 %inval2\n"
3999 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
4000 " OpStore %outloc2 %neg\n"
4002 " OpFunctionEnd\n");
4004 spec1.assembly = assembly;
4005 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4006 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
4007 spec1.numWorkGroups = IVec3(numElements, 1, 1);
4008 spec1.entryPoint = "entrypoint1";
4010 spec2.assembly = assembly;
4011 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4012 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
4013 spec2.numWorkGroups = IVec3(numElements, 1, 1);
4014 spec2.entryPoint = "entrypoint2";
4016 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
4017 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
4019 return group.release();
4022 inline std::string makeLongUTF8String (size_t num4ByteChars)
4024 // An example of a longest valid UTF-8 character. Be explicit about the
4025 // character type because Microsoft compilers can otherwise interpret the
4026 // character string as being over wide (16-bit) characters. Ideally, we
4027 // would just use a C++11 UTF-8 string literal, but we want to support older
4028 // Microsoft compilers.
4029 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
4030 std::string longString;
4031 longString.reserve(num4ByteChars * 4);
4032 for (size_t count = 0; count < num4ByteChars; count++)
4034 longString += earthAfrica;
4039 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
4041 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
4042 vector<CaseParameter> cases;
4043 de::Random rnd (deStringHash(group->getName()));
4044 const int numElements = 100;
4045 vector<float> positiveFloats (numElements, 0);
4046 vector<float> negativeFloats (numElements, 0);
4047 const StringTemplate shaderTemplate (
4048 "OpCapability Shader\n"
4049 "OpMemoryModel Logical GLSL450\n"
4051 "OpEntryPoint GLCompute %main \"main\" %id\n"
4052 "OpExecutionMode %main LocalSize 1 1 1\n"
4056 "OpName %main \"main\"\n"
4057 "OpName %id \"gl_GlobalInvocationID\"\n"
4059 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4061 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4063 "%id = OpVariable %uvec3ptr Input\n"
4064 "%zero = OpConstant %i32 0\n"
4066 "%main = OpFunction %void None %voidf\n"
4067 "%label = OpLabel\n"
4068 "%idval = OpLoad %uvec3 %id\n"
4069 "%x = OpCompositeExtract %u32 %idval 0\n"
4070 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4071 "%inval = OpLoad %f32 %inloc\n"
4072 "%neg = OpFNegate %f32 %inval\n"
4073 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4074 " OpStore %outloc %neg\n"
4076 " OpFunctionEnd\n");
4078 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
4079 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
4080 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
4081 "OpSource GLSL 430 %fname"));
4082 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
4083 "OpSource GLSL 430 %fname"));
4084 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
4085 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
4086 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
4087 "OpSource GLSL 430 %fname \"\""));
4088 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
4089 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
4090 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
4091 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
4092 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
4093 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
4094 "OpSourceContinued \"id main() {}\""));
4095 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
4096 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4097 "OpSourceContinued \"\""));
4098 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
4099 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4100 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
4101 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
4102 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4103 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
4104 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
4105 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
4106 "OpSourceContinued \"void\"\n"
4107 "OpSourceContinued \"main()\"\n"
4108 "OpSourceContinued \"{}\""));
4109 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
4110 "OpSource GLSL 430 %fname \"\"\n"
4111 "OpSourceContinued \"#version 430\nvoid main() {}\""));
4113 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4115 for (size_t ndx = 0; ndx < numElements; ++ndx)
4116 negativeFloats[ndx] = -positiveFloats[ndx];
4118 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4120 map<string, string> specializations;
4121 ComputeShaderSpec spec;
4123 specializations["SOURCE"] = cases[caseNdx].param;
4124 spec.assembly = shaderTemplate.specialize(specializations);
4125 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4126 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4127 spec.numWorkGroups = IVec3(numElements, 1, 1);
4129 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4132 return group.release();
4135 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
4137 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
4138 vector<CaseParameter> cases;
4139 de::Random rnd (deStringHash(group->getName()));
4140 const int numElements = 100;
4141 vector<float> inputFloats (numElements, 0);
4142 vector<float> outputFloats (numElements, 0);
4143 const StringTemplate shaderTemplate (
4144 string(getComputeAsmShaderPreamble()) +
4146 "OpSourceExtension \"${EXTENSION}\"\n"
4148 "OpName %main \"main\"\n"
4149 "OpName %id \"gl_GlobalInvocationID\"\n"
4151 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4153 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4155 "%id = OpVariable %uvec3ptr Input\n"
4156 "%zero = OpConstant %i32 0\n"
4158 "%main = OpFunction %void None %voidf\n"
4159 "%label = OpLabel\n"
4160 "%idval = OpLoad %uvec3 %id\n"
4161 "%x = OpCompositeExtract %u32 %idval 0\n"
4162 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4163 "%inval = OpLoad %f32 %inloc\n"
4164 "%neg = OpFNegate %f32 %inval\n"
4165 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4166 " OpStore %outloc %neg\n"
4168 " OpFunctionEnd\n");
4170 cases.push_back(CaseParameter("empty_extension", ""));
4171 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
4172 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
4173 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
4174 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
4176 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
4178 for (size_t ndx = 0; ndx < numElements; ++ndx)
4179 outputFloats[ndx] = -inputFloats[ndx];
4181 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4183 map<string, string> specializations;
4184 ComputeShaderSpec spec;
4186 specializations["EXTENSION"] = cases[caseNdx].param;
4187 spec.assembly = shaderTemplate.specialize(specializations);
4188 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4189 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4190 spec.numWorkGroups = IVec3(numElements, 1, 1);
4192 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4195 return group.release();
4198 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
4199 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
4201 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
4202 vector<CaseParameter> cases;
4203 de::Random rnd (deStringHash(group->getName()));
4204 const int numElements = 100;
4205 vector<float> positiveFloats (numElements, 0);
4206 vector<float> negativeFloats (numElements, 0);
4207 const StringTemplate shaderTemplate (
4208 string(getComputeAsmShaderPreamble()) +
4210 "OpSource GLSL 430\n"
4211 "OpName %main \"main\"\n"
4212 "OpName %id \"gl_GlobalInvocationID\"\n"
4214 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4216 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4217 "%uvec2 = OpTypeVector %u32 2\n"
4218 "%bvec3 = OpTypeVector %bool 3\n"
4219 "%fvec4 = OpTypeVector %f32 4\n"
4220 "%fmat33 = OpTypeMatrix %fvec3 3\n"
4221 "%const100 = OpConstant %u32 100\n"
4222 "%uarr100 = OpTypeArray %i32 %const100\n"
4223 "%struct = OpTypeStruct %f32 %i32 %u32\n"
4224 "%pointer = OpTypePointer Function %i32\n"
4225 + string(getComputeAsmInputOutputBuffer()) +
4227 "%null = OpConstantNull ${TYPE}\n"
4229 "%id = OpVariable %uvec3ptr Input\n"
4230 "%zero = OpConstant %i32 0\n"
4232 "%main = OpFunction %void None %voidf\n"
4233 "%label = OpLabel\n"
4234 "%idval = OpLoad %uvec3 %id\n"
4235 "%x = OpCompositeExtract %u32 %idval 0\n"
4236 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4237 "%inval = OpLoad %f32 %inloc\n"
4238 "%neg = OpFNegate %f32 %inval\n"
4239 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4240 " OpStore %outloc %neg\n"
4242 " OpFunctionEnd\n");
4244 cases.push_back(CaseParameter("bool", "%bool"));
4245 cases.push_back(CaseParameter("sint32", "%i32"));
4246 cases.push_back(CaseParameter("uint32", "%u32"));
4247 cases.push_back(CaseParameter("float32", "%f32"));
4248 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
4249 cases.push_back(CaseParameter("vec3bool", "%bvec3"));
4250 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
4251 cases.push_back(CaseParameter("matrix", "%fmat33"));
4252 cases.push_back(CaseParameter("array", "%uarr100"));
4253 cases.push_back(CaseParameter("struct", "%struct"));
4254 cases.push_back(CaseParameter("pointer", "%pointer"));
4256 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4258 for (size_t ndx = 0; ndx < numElements; ++ndx)
4259 negativeFloats[ndx] = -positiveFloats[ndx];
4261 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4263 map<string, string> specializations;
4264 ComputeShaderSpec spec;
4266 specializations["TYPE"] = cases[caseNdx].param;
4267 spec.assembly = shaderTemplate.specialize(specializations);
4268 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4269 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4270 spec.numWorkGroups = IVec3(numElements, 1, 1);
4272 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4275 return group.release();
4278 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
4279 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
4281 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
4282 vector<CaseParameter> cases;
4283 de::Random rnd (deStringHash(group->getName()));
4284 const int numElements = 100;
4285 vector<float> positiveFloats (numElements, 0);
4286 vector<float> negativeFloats (numElements, 0);
4287 const StringTemplate shaderTemplate (
4288 string(getComputeAsmShaderPreamble()) +
4290 "OpSource GLSL 430\n"
4291 "OpName %main \"main\"\n"
4292 "OpName %id \"gl_GlobalInvocationID\"\n"
4294 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4296 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4298 "%id = OpVariable %uvec3ptr Input\n"
4299 "%zero = OpConstant %i32 0\n"
4303 "%main = OpFunction %void None %voidf\n"
4304 "%label = OpLabel\n"
4305 "%idval = OpLoad %uvec3 %id\n"
4306 "%x = OpCompositeExtract %u32 %idval 0\n"
4307 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4308 "%inval = OpLoad %f32 %inloc\n"
4309 "%neg = OpFNegate %f32 %inval\n"
4310 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4311 " OpStore %outloc %neg\n"
4313 " OpFunctionEnd\n");
4315 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
4316 "%const = OpConstantComposite %uvec3 %five %zero %five"));
4317 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
4318 "%ten = OpConstant %f32 10.\n"
4319 "%fzero = OpConstant %f32 0.\n"
4320 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
4321 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
4322 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
4323 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
4324 "%fzero = OpConstant %f32 0.\n"
4325 "%one = OpConstant %f32 1.\n"
4326 "%point5 = OpConstant %f32 0.5\n"
4327 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
4328 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
4329 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
4330 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
4331 "%st2 = OpTypeStruct %i32 %i32\n"
4332 "%struct = OpTypeStruct %st1 %st2\n"
4333 "%point5 = OpConstant %f32 0.5\n"
4334 "%one = OpConstant %u32 1\n"
4335 "%ten = OpConstant %i32 10\n"
4336 "%st1val = OpConstantComposite %st1 %one %point5\n"
4337 "%st2val = OpConstantComposite %st2 %ten %ten\n"
4338 "%const = OpConstantComposite %struct %st1val %st2val"));
4340 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4342 for (size_t ndx = 0; ndx < numElements; ++ndx)
4343 negativeFloats[ndx] = -positiveFloats[ndx];
4345 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4347 map<string, string> specializations;
4348 ComputeShaderSpec spec;
4350 specializations["CONSTANT"] = cases[caseNdx].param;
4351 spec.assembly = shaderTemplate.specialize(specializations);
4352 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4353 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4354 spec.numWorkGroups = IVec3(numElements, 1, 1);
4356 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4359 return group.release();
4362 // Creates a floating point number with the given exponent, and significand
4363 // bits set. It can only create normalized numbers. Only the least significant
4364 // 24 bits of the significand will be examined. The final bit of the
4365 // significand will also be ignored. This allows alignment to be written
4366 // similarly to C99 hex-floats.
4367 // For example if you wanted to write 0x1.7f34p-12 you would call
4368 // constructNormalizedFloat(-12, 0x7f3400)
4369 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
4373 for (deInt32 idx = 0; idx < 23; ++idx)
4375 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
4379 return std::ldexp(f, exponent);
4382 // Compare instruction for the OpQuantizeF16 compute exact case.
4383 // Returns true if the output is what is expected from the test case.
4384 bool compareOpQuantizeF16ComputeExactCase (const std::vector<Resource>&, const vector<AllocationSp>& outputAllocs, const std::vector<Resource>& expectedOutputs, TestLog&)
4386 if (outputAllocs.size() != 1)
4389 // Only size is needed because we cannot compare Nans.
4390 size_t byteSize = expectedOutputs[0].getByteSize();
4392 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
4394 if (byteSize != 4*sizeof(float)) {
4398 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
4399 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
4404 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
4405 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
4410 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
4411 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
4416 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
4417 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
4424 // Checks that every output from a test-case is a float NaN.
4425 bool compareNan (const std::vector<Resource>&, const vector<AllocationSp>& outputAllocs, const std::vector<Resource>& expectedOutputs, TestLog&)
4427 if (outputAllocs.size() != 1)
4430 // Only size is needed because we cannot compare Nans.
4431 size_t byteSize = expectedOutputs[0].getByteSize();
4433 const float* const output_as_float = static_cast<const float*>(outputAllocs[0]->getHostPtr());
4435 for (size_t idx = 0; idx < byteSize / sizeof(float); ++idx)
4437 if (!deFloatIsNaN(output_as_float[idx]))
4446 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
4447 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
4449 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
4451 const std::string shader (
4452 string(getComputeAsmShaderPreamble()) +
4454 "OpSource GLSL 430\n"
4455 "OpName %main \"main\"\n"
4456 "OpName %id \"gl_GlobalInvocationID\"\n"
4458 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4460 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4462 "%id = OpVariable %uvec3ptr Input\n"
4463 "%zero = OpConstant %i32 0\n"
4465 "%main = OpFunction %void None %voidf\n"
4466 "%label = OpLabel\n"
4467 "%idval = OpLoad %uvec3 %id\n"
4468 "%x = OpCompositeExtract %u32 %idval 0\n"
4469 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4470 "%inval = OpLoad %f32 %inloc\n"
4471 "%quant = OpQuantizeToF16 %f32 %inval\n"
4472 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4473 " OpStore %outloc %quant\n"
4475 " OpFunctionEnd\n");
4478 ComputeShaderSpec spec;
4479 const deUint32 numElements = 100;
4480 vector<float> infinities;
4481 vector<float> results;
4483 infinities.reserve(numElements);
4484 results.reserve(numElements);
4486 for (size_t idx = 0; idx < numElements; ++idx)
4491 infinities.push_back(std::numeric_limits<float>::infinity());
4492 results.push_back(std::numeric_limits<float>::infinity());
4495 infinities.push_back(-std::numeric_limits<float>::infinity());
4496 results.push_back(-std::numeric_limits<float>::infinity());
4499 infinities.push_back(std::ldexp(1.0f, 16));
4500 results.push_back(std::numeric_limits<float>::infinity());
4503 infinities.push_back(std::ldexp(-1.0f, 32));
4504 results.push_back(-std::numeric_limits<float>::infinity());
4509 spec.assembly = shader;
4510 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
4511 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
4512 spec.numWorkGroups = IVec3(numElements, 1, 1);
4514 group->addChild(new SpvAsmComputeShaderCase(
4515 testCtx, "infinities", "Check that infinities propagated and created", spec));
4519 ComputeShaderSpec spec;
4521 const deUint32 numElements = 100;
4523 nans.reserve(numElements);
4525 for (size_t idx = 0; idx < numElements; ++idx)
4529 nans.push_back(std::numeric_limits<float>::quiet_NaN());
4533 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
4537 spec.assembly = shader;
4538 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
4539 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
4540 spec.numWorkGroups = IVec3(numElements, 1, 1);
4541 spec.verifyIO = &compareNan;
4543 group->addChild(new SpvAsmComputeShaderCase(
4544 testCtx, "propagated_nans", "Check that nans are propagated", spec));
4548 ComputeShaderSpec spec;
4549 vector<float> small;
4550 vector<float> zeros;
4551 const deUint32 numElements = 100;
4553 small.reserve(numElements);
4554 zeros.reserve(numElements);
4556 for (size_t idx = 0; idx < numElements; ++idx)
4561 small.push_back(0.f);
4562 zeros.push_back(0.f);
4565 small.push_back(-0.f);
4566 zeros.push_back(-0.f);
4569 small.push_back(std::ldexp(1.0f, -16));
4570 zeros.push_back(0.f);
4573 small.push_back(std::ldexp(-1.0f, -32));
4574 zeros.push_back(-0.f);
4577 small.push_back(std::ldexp(1.0f, -127));
4578 zeros.push_back(0.f);
4581 small.push_back(-std::ldexp(1.0f, -128));
4582 zeros.push_back(-0.f);
4587 spec.assembly = shader;
4588 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
4589 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
4590 spec.numWorkGroups = IVec3(numElements, 1, 1);
4592 group->addChild(new SpvAsmComputeShaderCase(
4593 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
4597 ComputeShaderSpec spec;
4598 vector<float> exact;
4599 const deUint32 numElements = 200;
4601 exact.reserve(numElements);
4603 for (size_t idx = 0; idx < numElements; ++idx)
4604 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
4606 spec.assembly = shader;
4607 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
4608 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
4609 spec.numWorkGroups = IVec3(numElements, 1, 1);
4611 group->addChild(new SpvAsmComputeShaderCase(
4612 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
4616 ComputeShaderSpec spec;
4617 vector<float> inputs;
4618 const deUint32 numElements = 4;
4620 inputs.push_back(constructNormalizedFloat(8, 0x300300));
4621 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
4622 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
4623 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
4625 spec.assembly = shader;
4626 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
4627 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4628 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
4629 spec.numWorkGroups = IVec3(numElements, 1, 1);
4631 group->addChild(new SpvAsmComputeShaderCase(
4632 testCtx, "rounded", "Check that are rounded when needed", spec));
4635 return group.release();
4638 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
4640 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
4642 const std::string shader (
4643 string(getComputeAsmShaderPreamble()) +
4645 "OpName %main \"main\"\n"
4646 "OpName %id \"gl_GlobalInvocationID\"\n"
4648 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4650 "OpDecorate %sc_0 SpecId 0\n"
4651 "OpDecorate %sc_1 SpecId 1\n"
4652 "OpDecorate %sc_2 SpecId 2\n"
4653 "OpDecorate %sc_3 SpecId 3\n"
4654 "OpDecorate %sc_4 SpecId 4\n"
4655 "OpDecorate %sc_5 SpecId 5\n"
4657 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4659 "%id = OpVariable %uvec3ptr Input\n"
4660 "%zero = OpConstant %i32 0\n"
4661 "%c_u32_6 = OpConstant %u32 6\n"
4663 "%sc_0 = OpSpecConstant %f32 0.\n"
4664 "%sc_1 = OpSpecConstant %f32 0.\n"
4665 "%sc_2 = OpSpecConstant %f32 0.\n"
4666 "%sc_3 = OpSpecConstant %f32 0.\n"
4667 "%sc_4 = OpSpecConstant %f32 0.\n"
4668 "%sc_5 = OpSpecConstant %f32 0.\n"
4670 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
4671 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
4672 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
4673 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
4674 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
4675 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
4677 "%main = OpFunction %void None %voidf\n"
4678 "%label = OpLabel\n"
4679 "%idval = OpLoad %uvec3 %id\n"
4680 "%x = OpCompositeExtract %u32 %idval 0\n"
4681 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4682 "%selector = OpUMod %u32 %x %c_u32_6\n"
4683 " OpSelectionMerge %exit None\n"
4684 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
4686 "%case0 = OpLabel\n"
4687 " OpStore %outloc %sc_0_quant\n"
4690 "%case1 = OpLabel\n"
4691 " OpStore %outloc %sc_1_quant\n"
4694 "%case2 = OpLabel\n"
4695 " OpStore %outloc %sc_2_quant\n"
4698 "%case3 = OpLabel\n"
4699 " OpStore %outloc %sc_3_quant\n"
4702 "%case4 = OpLabel\n"
4703 " OpStore %outloc %sc_4_quant\n"
4706 "%case5 = OpLabel\n"
4707 " OpStore %outloc %sc_5_quant\n"
4713 " OpFunctionEnd\n");
4716 ComputeShaderSpec spec;
4717 const deUint8 numCases = 4;
4718 vector<float> inputs (numCases, 0.f);
4719 vector<float> outputs;
4721 spec.assembly = shader;
4722 spec.numWorkGroups = IVec3(numCases, 1, 1);
4724 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
4725 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
4726 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
4727 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
4729 outputs.push_back(std::numeric_limits<float>::infinity());
4730 outputs.push_back(-std::numeric_limits<float>::infinity());
4731 outputs.push_back(std::numeric_limits<float>::infinity());
4732 outputs.push_back(-std::numeric_limits<float>::infinity());
4734 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4735 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4737 group->addChild(new SpvAsmComputeShaderCase(
4738 testCtx, "infinities", "Check that infinities propagated and created", spec));
4742 ComputeShaderSpec spec;
4743 const deUint8 numCases = 2;
4744 vector<float> inputs (numCases, 0.f);
4745 vector<float> outputs;
4747 spec.assembly = shader;
4748 spec.numWorkGroups = IVec3(numCases, 1, 1);
4749 spec.verifyIO = &compareNan;
4751 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
4752 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
4754 for (deUint8 idx = 0; idx < numCases; ++idx)
4755 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(outputs[idx]));
4757 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4758 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4760 group->addChild(new SpvAsmComputeShaderCase(
4761 testCtx, "propagated_nans", "Check that nans are propagated", spec));
4765 ComputeShaderSpec spec;
4766 const deUint8 numCases = 6;
4767 vector<float> inputs (numCases, 0.f);
4768 vector<float> outputs;
4770 spec.assembly = shader;
4771 spec.numWorkGroups = IVec3(numCases, 1, 1);
4773 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(0.f));
4774 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(-0.f));
4775 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
4776 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
4777 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
4778 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
4780 outputs.push_back(0.f);
4781 outputs.push_back(-0.f);
4782 outputs.push_back(0.f);
4783 outputs.push_back(-0.f);
4784 outputs.push_back(0.f);
4785 outputs.push_back(-0.f);
4787 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4788 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4790 group->addChild(new SpvAsmComputeShaderCase(
4791 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
4795 ComputeShaderSpec spec;
4796 const deUint8 numCases = 6;
4797 vector<float> inputs (numCases, 0.f);
4798 vector<float> outputs;
4800 spec.assembly = shader;
4801 spec.numWorkGroups = IVec3(numCases, 1, 1);
4803 for (deUint8 idx = 0; idx < 6; ++idx)
4805 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
4806 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(f));
4807 outputs.push_back(f);
4810 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4811 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4813 group->addChild(new SpvAsmComputeShaderCase(
4814 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
4818 ComputeShaderSpec spec;
4819 const deUint8 numCases = 4;
4820 vector<float> inputs (numCases, 0.f);
4821 vector<float> outputs;
4823 spec.assembly = shader;
4824 spec.numWorkGroups = IVec3(numCases, 1, 1);
4825 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
4827 outputs.push_back(constructNormalizedFloat(8, 0x300300));
4828 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
4829 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
4830 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
4832 for (deUint8 idx = 0; idx < numCases; ++idx)
4833 spec.specConstants.append<deInt32>(bitwiseCast<deUint32>(outputs[idx]));
4835 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4836 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4838 group->addChild(new SpvAsmComputeShaderCase(
4839 testCtx, "rounded", "Check that are rounded when needed", spec));
4842 return group.release();
4845 // Checks that constant null/composite values can be used in computation.
4846 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
4848 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
4849 ComputeShaderSpec spec;
4850 de::Random rnd (deStringHash(group->getName()));
4851 const int numElements = 100;
4852 vector<float> positiveFloats (numElements, 0);
4853 vector<float> negativeFloats (numElements, 0);
4855 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4857 for (size_t ndx = 0; ndx < numElements; ++ndx)
4858 negativeFloats[ndx] = -positiveFloats[ndx];
4861 "OpCapability Shader\n"
4862 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
4863 "OpMemoryModel Logical GLSL450\n"
4864 "OpEntryPoint GLCompute %main \"main\" %id\n"
4865 "OpExecutionMode %main LocalSize 1 1 1\n"
4867 "OpSource GLSL 430\n"
4868 "OpName %main \"main\"\n"
4869 "OpName %id \"gl_GlobalInvocationID\"\n"
4871 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4873 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4875 "%fmat = OpTypeMatrix %fvec3 3\n"
4876 "%ten = OpConstant %u32 10\n"
4877 "%f32arr10 = OpTypeArray %f32 %ten\n"
4878 "%fst = OpTypeStruct %f32 %f32\n"
4880 + string(getComputeAsmInputOutputBuffer()) +
4882 "%id = OpVariable %uvec3ptr Input\n"
4883 "%zero = OpConstant %i32 0\n"
4885 // Create a bunch of null values
4886 "%unull = OpConstantNull %u32\n"
4887 "%fnull = OpConstantNull %f32\n"
4888 "%vnull = OpConstantNull %fvec3\n"
4889 "%mnull = OpConstantNull %fmat\n"
4890 "%anull = OpConstantNull %f32arr10\n"
4891 "%snull = OpConstantComposite %fst %fnull %fnull\n"
4893 "%main = OpFunction %void None %voidf\n"
4894 "%label = OpLabel\n"
4895 "%idval = OpLoad %uvec3 %id\n"
4896 "%x = OpCompositeExtract %u32 %idval 0\n"
4897 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4898 "%inval = OpLoad %f32 %inloc\n"
4899 "%neg = OpFNegate %f32 %inval\n"
4901 // Get the abs() of (a certain element of) those null values
4902 "%unull_cov = OpConvertUToF %f32 %unull\n"
4903 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
4904 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
4905 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
4906 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
4907 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
4908 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
4909 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
4910 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
4911 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
4912 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
4915 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
4916 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
4917 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
4918 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
4919 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
4920 "%final = OpFAdd %f32 %add5 %snull_abs\n"
4922 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4923 " OpStore %outloc %final\n" // write to output
4926 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4927 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4928 spec.numWorkGroups = IVec3(numElements, 1, 1);
4930 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
4932 return group.release();
4935 // Assembly code used for testing loop control is based on GLSL source code:
4938 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4939 // float elements[];
4941 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4942 // float elements[];
4946 // uint x = gl_GlobalInvocationID.x;
4947 // output_data.elements[x] = input_data.elements[x];
4948 // for (uint i = 0; i < 4; ++i)
4949 // output_data.elements[x] += 1.f;
4951 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
4953 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
4954 vector<CaseParameter> cases;
4955 de::Random rnd (deStringHash(group->getName()));
4956 const int numElements = 100;
4957 vector<float> inputFloats (numElements, 0);
4958 vector<float> outputFloats (numElements, 0);
4959 const StringTemplate shaderTemplate (
4960 string(getComputeAsmShaderPreamble()) +
4962 "OpSource GLSL 430\n"
4963 "OpName %main \"main\"\n"
4964 "OpName %id \"gl_GlobalInvocationID\"\n"
4966 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4968 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4970 "%u32ptr = OpTypePointer Function %u32\n"
4972 "%id = OpVariable %uvec3ptr Input\n"
4973 "%zero = OpConstant %i32 0\n"
4974 "%uzero = OpConstant %u32 0\n"
4975 "%one = OpConstant %i32 1\n"
4976 "%constf1 = OpConstant %f32 1.0\n"
4977 "%four = OpConstant %u32 4\n"
4979 "%main = OpFunction %void None %voidf\n"
4980 "%entry = OpLabel\n"
4981 "%i = OpVariable %u32ptr Function\n"
4982 " OpStore %i %uzero\n"
4984 "%idval = OpLoad %uvec3 %id\n"
4985 "%x = OpCompositeExtract %u32 %idval 0\n"
4986 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4987 "%inval = OpLoad %f32 %inloc\n"
4988 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4989 " OpStore %outloc %inval\n"
4990 " OpBranch %loop_entry\n"
4992 "%loop_entry = OpLabel\n"
4993 "%i_val = OpLoad %u32 %i\n"
4994 "%cmp_lt = OpULessThan %bool %i_val %four\n"
4995 " OpLoopMerge %loop_merge %loop_body ${CONTROL}\n"
4996 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
4997 "%loop_body = OpLabel\n"
4998 "%outval = OpLoad %f32 %outloc\n"
4999 "%addf1 = OpFAdd %f32 %outval %constf1\n"
5000 " OpStore %outloc %addf1\n"
5001 "%new_i = OpIAdd %u32 %i_val %one\n"
5002 " OpStore %i %new_i\n"
5003 " OpBranch %loop_entry\n"
5004 "%loop_merge = OpLabel\n"
5006 " OpFunctionEnd\n");
5008 cases.push_back(CaseParameter("none", "None"));
5009 cases.push_back(CaseParameter("unroll", "Unroll"));
5010 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
5011 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
5013 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5015 for (size_t ndx = 0; ndx < numElements; ++ndx)
5016 outputFloats[ndx] = inputFloats[ndx] + 4.f;
5018 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5020 map<string, string> specializations;
5021 ComputeShaderSpec spec;
5023 specializations["CONTROL"] = cases[caseNdx].param;
5024 spec.assembly = shaderTemplate.specialize(specializations);
5025 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5026 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5027 spec.numWorkGroups = IVec3(numElements, 1, 1);
5029 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5032 group->addChild(new SpvAsmLoopControlDependencyLengthCase(testCtx, "dependency_length", "dependency_length"));
5033 group->addChild(new SpvAsmLoopControlDependencyInfiniteCase(testCtx, "dependency_infinite", "dependency_infinite"));
5035 return group.release();
5038 // Assembly code used for testing selection control is based on GLSL source code:
5041 // layout(std140, set = 0, binding = 0) readonly buffer Input {
5042 // float elements[];
5044 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
5045 // float elements[];
5049 // uint x = gl_GlobalInvocationID.x;
5050 // float val = input_data.elements[x];
5052 // output_data.elements[x] = val + 1.f;
5054 // output_data.elements[x] = val - 1.f;
5056 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
5058 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
5059 vector<CaseParameter> cases;
5060 de::Random rnd (deStringHash(group->getName()));
5061 const int numElements = 100;
5062 vector<float> inputFloats (numElements, 0);
5063 vector<float> outputFloats (numElements, 0);
5064 const StringTemplate shaderTemplate (
5065 string(getComputeAsmShaderPreamble()) +
5067 "OpSource GLSL 430\n"
5068 "OpName %main \"main\"\n"
5069 "OpName %id \"gl_GlobalInvocationID\"\n"
5071 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5073 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5075 "%id = OpVariable %uvec3ptr Input\n"
5076 "%zero = OpConstant %i32 0\n"
5077 "%constf1 = OpConstant %f32 1.0\n"
5078 "%constf10 = OpConstant %f32 10.0\n"
5080 "%main = OpFunction %void None %voidf\n"
5081 "%entry = OpLabel\n"
5082 "%idval = OpLoad %uvec3 %id\n"
5083 "%x = OpCompositeExtract %u32 %idval 0\n"
5084 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5085 "%inval = OpLoad %f32 %inloc\n"
5086 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5087 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
5089 " OpSelectionMerge %if_end ${CONTROL}\n"
5090 " OpBranchConditional %cmp_gt %if_true %if_false\n"
5091 "%if_true = OpLabel\n"
5092 "%addf1 = OpFAdd %f32 %inval %constf1\n"
5093 " OpStore %outloc %addf1\n"
5094 " OpBranch %if_end\n"
5095 "%if_false = OpLabel\n"
5096 "%subf1 = OpFSub %f32 %inval %constf1\n"
5097 " OpStore %outloc %subf1\n"
5098 " OpBranch %if_end\n"
5099 "%if_end = OpLabel\n"
5101 " OpFunctionEnd\n");
5103 cases.push_back(CaseParameter("none", "None"));
5104 cases.push_back(CaseParameter("flatten", "Flatten"));
5105 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
5106 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
5108 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5110 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
5111 floorAll(inputFloats);
5113 for (size_t ndx = 0; ndx < numElements; ++ndx)
5114 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
5116 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5118 map<string, string> specializations;
5119 ComputeShaderSpec spec;
5121 specializations["CONTROL"] = cases[caseNdx].param;
5122 spec.assembly = shaderTemplate.specialize(specializations);
5123 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5124 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5125 spec.numWorkGroups = IVec3(numElements, 1, 1);
5127 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5130 return group.release();
5133 tcu::TestCaseGroup* createOpNameGroup(tcu::TestContext& testCtx)
5135 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opname", "Tests OpName cases"));
5136 de::MovePtr<tcu::TestCaseGroup> entryMainGroup (new tcu::TestCaseGroup(testCtx, "entry_main", "OpName tests with entry main"));
5137 de::MovePtr<tcu::TestCaseGroup> entryNotGroup (new tcu::TestCaseGroup(testCtx, "entry_rdc", "OpName tests with entry rdc"));
5138 vector<CaseParameter> cases;
5139 vector<string> testFunc;
5140 de::Random rnd (deStringHash(group->getName()));
5141 const int numElements = 100;
5142 vector<float> inputFloats (numElements, 0);
5143 vector<float> outputFloats (numElements, 0);
5145 fillRandomScalars(rnd, -100.0f, 100.0f, &inputFloats[0], numElements);
5147 for(size_t ndx = 0; ndx < numElements; ++ndx)
5148 outputFloats[ndx] = -inputFloats[ndx];
5150 const StringTemplate shaderTemplate (
5151 "OpCapability Shader\n"
5152 "OpMemoryModel Logical GLSL450\n"
5153 "OpEntryPoint GLCompute %main \"${ENTRY}\" %id\n"
5154 "OpExecutionMode %main LocalSize 1 1 1\n"
5156 "OpName %${FUNC_ID} \"${NAME}\"\n"
5158 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5160 + string(getComputeAsmInputOutputBufferTraits())
5162 + string(getComputeAsmCommonTypes())
5164 + string(getComputeAsmInputOutputBuffer()) +
5166 "%id = OpVariable %uvec3ptr Input\n"
5167 "%zero = OpConstant %i32 0\n"
5169 "%func = OpFunction %void None %voidf\n"
5174 "%main = OpFunction %void None %voidf\n"
5175 "%entry = OpLabel\n"
5176 "%7 = OpFunctionCall %void %func\n"
5178 "%idval = OpLoad %uvec3 %id\n"
5179 "%x = OpCompositeExtract %u32 %idval 0\n"
5181 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5182 "%inval = OpLoad %f32 %inloc\n"
5183 "%neg = OpFNegate %f32 %inval\n"
5184 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5185 " OpStore %outloc %neg\n"
5189 " OpFunctionEnd\n");
5191 cases.push_back(CaseParameter("_is_main", "main"));
5192 cases.push_back(CaseParameter("_is_not_main", "not_main"));
5194 testFunc.push_back("main");
5195 testFunc.push_back("func");
5197 for(size_t fNdx = 0; fNdx < testFunc.size(); ++fNdx)
5199 for(size_t ndx = 0; ndx < cases.size(); ++ndx)
5201 map<string, string> specializations;
5202 ComputeShaderSpec spec;
5204 specializations["ENTRY"] = "main";
5205 specializations["FUNC_ID"] = testFunc[fNdx];
5206 specializations["NAME"] = cases[ndx].param;
5207 spec.assembly = shaderTemplate.specialize(specializations);
5208 spec.numWorkGroups = IVec3(numElements, 1, 1);
5209 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5210 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5212 entryMainGroup->addChild(new SpvAsmComputeShaderCase(testCtx, (testFunc[fNdx] + cases[ndx].name).c_str(), cases[ndx].name, spec));
5216 cases.push_back(CaseParameter("_is_entry", "rdc"));
5218 for(size_t fNdx = 0; fNdx < testFunc.size(); ++fNdx)
5220 for(size_t ndx = 0; ndx < cases.size(); ++ndx)
5222 map<string, string> specializations;
5223 ComputeShaderSpec spec;
5225 specializations["ENTRY"] = "rdc";
5226 specializations["FUNC_ID"] = testFunc[fNdx];
5227 specializations["NAME"] = cases[ndx].param;
5228 spec.assembly = shaderTemplate.specialize(specializations);
5229 spec.numWorkGroups = IVec3(numElements, 1, 1);
5230 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5231 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5232 spec.entryPoint = "rdc";
5234 entryNotGroup->addChild(new SpvAsmComputeShaderCase(testCtx, (testFunc[fNdx] + cases[ndx].name).c_str(), cases[ndx].name, spec));
5238 group->addChild(entryMainGroup.release());
5239 group->addChild(entryNotGroup.release());
5241 return group.release();
5244 // Assembly code used for testing function control is based on GLSL source code:
5248 // layout(std140, set = 0, binding = 0) readonly buffer Input {
5249 // float elements[];
5251 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
5252 // float elements[];
5255 // float const10() { return 10.f; }
5258 // uint x = gl_GlobalInvocationID.x;
5259 // output_data.elements[x] = input_data.elements[x] + const10();
5261 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
5263 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
5264 vector<CaseParameter> cases;
5265 de::Random rnd (deStringHash(group->getName()));
5266 const int numElements = 100;
5267 vector<float> inputFloats (numElements, 0);
5268 vector<float> outputFloats (numElements, 0);
5269 const StringTemplate shaderTemplate (
5270 string(getComputeAsmShaderPreamble()) +
5272 "OpSource GLSL 430\n"
5273 "OpName %main \"main\"\n"
5274 "OpName %func_const10 \"const10(\"\n"
5275 "OpName %id \"gl_GlobalInvocationID\"\n"
5277 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5279 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5281 "%f32f = OpTypeFunction %f32\n"
5282 "%id = OpVariable %uvec3ptr Input\n"
5283 "%zero = OpConstant %i32 0\n"
5284 "%constf10 = OpConstant %f32 10.0\n"
5286 "%main = OpFunction %void None %voidf\n"
5287 "%entry = OpLabel\n"
5288 "%idval = OpLoad %uvec3 %id\n"
5289 "%x = OpCompositeExtract %u32 %idval 0\n"
5290 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5291 "%inval = OpLoad %f32 %inloc\n"
5292 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
5293 "%fadd = OpFAdd %f32 %inval %ret_10\n"
5294 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5295 " OpStore %outloc %fadd\n"
5299 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
5300 "%label = OpLabel\n"
5301 " OpReturnValue %constf10\n"
5302 " OpFunctionEnd\n");
5304 cases.push_back(CaseParameter("none", "None"));
5305 cases.push_back(CaseParameter("inline", "Inline"));
5306 cases.push_back(CaseParameter("dont_inline", "DontInline"));
5307 cases.push_back(CaseParameter("pure", "Pure"));
5308 cases.push_back(CaseParameter("const", "Const"));
5309 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
5310 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
5311 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
5312 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
5314 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5316 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
5317 floorAll(inputFloats);
5319 for (size_t ndx = 0; ndx < numElements; ++ndx)
5320 outputFloats[ndx] = inputFloats[ndx] + 10.f;
5322 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5324 map<string, string> specializations;
5325 ComputeShaderSpec spec;
5327 specializations["CONTROL"] = cases[caseNdx].param;
5328 spec.assembly = shaderTemplate.specialize(specializations);
5329 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5330 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5331 spec.numWorkGroups = IVec3(numElements, 1, 1);
5333 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5336 return group.release();
5339 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
5341 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
5342 vector<CaseParameter> cases;
5343 de::Random rnd (deStringHash(group->getName()));
5344 const int numElements = 100;
5345 vector<float> inputFloats (numElements, 0);
5346 vector<float> outputFloats (numElements, 0);
5347 const StringTemplate shaderTemplate (
5348 string(getComputeAsmShaderPreamble()) +
5350 "OpSource GLSL 430\n"
5351 "OpName %main \"main\"\n"
5352 "OpName %id \"gl_GlobalInvocationID\"\n"
5354 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5356 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5358 "%f32ptr_f = OpTypePointer Function %f32\n"
5360 "%id = OpVariable %uvec3ptr Input\n"
5361 "%zero = OpConstant %i32 0\n"
5362 "%four = OpConstant %i32 4\n"
5364 "%main = OpFunction %void None %voidf\n"
5365 "%label = OpLabel\n"
5366 "%copy = OpVariable %f32ptr_f Function\n"
5367 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
5368 "%x = OpCompositeExtract %u32 %idval 0\n"
5369 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5370 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5371 " OpCopyMemory %copy %inloc ${ACCESS}\n"
5372 "%val1 = OpLoad %f32 %copy\n"
5373 "%val2 = OpLoad %f32 %inloc\n"
5374 "%add = OpFAdd %f32 %val1 %val2\n"
5375 " OpStore %outloc %add ${ACCESS}\n"
5377 " OpFunctionEnd\n");
5379 cases.push_back(CaseParameter("null", ""));
5380 cases.push_back(CaseParameter("none", "None"));
5381 cases.push_back(CaseParameter("volatile", "Volatile"));
5382 cases.push_back(CaseParameter("aligned", "Aligned 4"));
5383 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
5384 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
5385 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
5387 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5389 for (size_t ndx = 0; ndx < numElements; ++ndx)
5390 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
5392 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5394 map<string, string> specializations;
5395 ComputeShaderSpec spec;
5397 specializations["ACCESS"] = cases[caseNdx].param;
5398 spec.assembly = shaderTemplate.specialize(specializations);
5399 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5400 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5401 spec.numWorkGroups = IVec3(numElements, 1, 1);
5403 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5406 return group.release();
5409 // Checks that we can get undefined values for various types, without exercising a computation with it.
5410 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
5412 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
5413 vector<CaseParameter> cases;
5414 de::Random rnd (deStringHash(group->getName()));
5415 const int numElements = 100;
5416 vector<float> positiveFloats (numElements, 0);
5417 vector<float> negativeFloats (numElements, 0);
5418 const StringTemplate shaderTemplate (
5419 string(getComputeAsmShaderPreamble()) +
5421 "OpSource GLSL 430\n"
5422 "OpName %main \"main\"\n"
5423 "OpName %id \"gl_GlobalInvocationID\"\n"
5425 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5427 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
5428 "%uvec2 = OpTypeVector %u32 2\n"
5429 "%fvec4 = OpTypeVector %f32 4\n"
5430 "%fmat33 = OpTypeMatrix %fvec3 3\n"
5431 "%image = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
5432 "%sampler = OpTypeSampler\n"
5433 "%simage = OpTypeSampledImage %image\n"
5434 "%const100 = OpConstant %u32 100\n"
5435 "%uarr100 = OpTypeArray %i32 %const100\n"
5436 "%struct = OpTypeStruct %f32 %i32 %u32\n"
5437 "%pointer = OpTypePointer Function %i32\n"
5438 + string(getComputeAsmInputOutputBuffer()) +
5440 "%id = OpVariable %uvec3ptr Input\n"
5441 "%zero = OpConstant %i32 0\n"
5443 "%main = OpFunction %void None %voidf\n"
5444 "%label = OpLabel\n"
5446 "%undef = OpUndef ${TYPE}\n"
5448 "%idval = OpLoad %uvec3 %id\n"
5449 "%x = OpCompositeExtract %u32 %idval 0\n"
5451 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5452 "%inval = OpLoad %f32 %inloc\n"
5453 "%neg = OpFNegate %f32 %inval\n"
5454 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5455 " OpStore %outloc %neg\n"
5457 " OpFunctionEnd\n");
5459 cases.push_back(CaseParameter("bool", "%bool"));
5460 cases.push_back(CaseParameter("sint32", "%i32"));
5461 cases.push_back(CaseParameter("uint32", "%u32"));
5462 cases.push_back(CaseParameter("float32", "%f32"));
5463 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
5464 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
5465 cases.push_back(CaseParameter("matrix", "%fmat33"));
5466 cases.push_back(CaseParameter("image", "%image"));
5467 cases.push_back(CaseParameter("sampler", "%sampler"));
5468 cases.push_back(CaseParameter("sampledimage", "%simage"));
5469 cases.push_back(CaseParameter("array", "%uarr100"));
5470 cases.push_back(CaseParameter("runtimearray", "%f32arr"));
5471 cases.push_back(CaseParameter("struct", "%struct"));
5472 cases.push_back(CaseParameter("pointer", "%pointer"));
5474 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
5476 for (size_t ndx = 0; ndx < numElements; ++ndx)
5477 negativeFloats[ndx] = -positiveFloats[ndx];
5479 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5481 map<string, string> specializations;
5482 ComputeShaderSpec spec;
5484 specializations["TYPE"] = cases[caseNdx].param;
5485 spec.assembly = shaderTemplate.specialize(specializations);
5486 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
5487 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
5488 spec.numWorkGroups = IVec3(numElements, 1, 1);
5490 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5493 return group.release();
5497 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
5499 struct NameCodePair { string name, code; };
5500 RGBA defaultColors[4];
5501 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
5502 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
5503 map<string, string> fragments = passthruFragments();
5504 const NameCodePair tests[] =
5506 {"unknown", "OpSource Unknown 321"},
5507 {"essl", "OpSource ESSL 310"},
5508 {"glsl", "OpSource GLSL 450"},
5509 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
5510 {"opencl_c", "OpSource OpenCL_C 120"},
5511 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
5512 {"file", opsourceGLSLWithFile},
5513 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
5514 // Longest possible source string: SPIR-V limits instructions to 65535
5515 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
5516 // contain 65530 UTF8 characters (one word each) plus one last word
5517 // containing 3 ASCII characters and \0.
5518 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
5521 getDefaultColors(defaultColors);
5522 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5524 fragments["debug"] = tests[testNdx].code;
5525 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5528 return opSourceTests.release();
5531 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
5533 struct NameCodePair { string name, code; };
5534 RGBA defaultColors[4];
5535 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
5536 map<string, string> fragments = passthruFragments();
5537 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
5538 const NameCodePair tests[] =
5540 {"empty", opsource + "OpSourceContinued \"\""},
5541 {"short", opsource + "OpSourceContinued \"abcde\""},
5542 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
5543 // Longest possible source string: SPIR-V limits instructions to 65535
5544 // words, of which the first one is OpSourceContinued/length; the rest
5545 // will contain 65533 UTF8 characters (one word each) plus one last word
5546 // containing 3 ASCII characters and \0.
5547 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
5550 getDefaultColors(defaultColors);
5551 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5553 fragments["debug"] = tests[testNdx].code;
5554 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5557 return opSourceTests.release();
5559 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
5561 RGBA defaultColors[4];
5562 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
5563 map<string, string> fragments;
5564 getDefaultColors(defaultColors);
5565 fragments["debug"] =
5566 "%name = OpString \"name\"\n";
5568 fragments["pre_main"] =
5571 "OpLine %name 1 1\n"
5573 "OpLine %name 1 1\n"
5574 "OpLine %name 1 1\n"
5575 "%second_function = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5577 "OpLine %name 1 1\n"
5579 "OpLine %name 1 1\n"
5580 "OpLine %name 1 1\n"
5581 "%second_param1 = OpFunctionParameter %v4f32\n"
5584 "%label_secondfunction = OpLabel\n"
5586 "OpReturnValue %second_param1\n"
5591 fragments["testfun"] =
5592 // A %test_code function that returns its argument unchanged.
5595 "OpLine %name 1 1\n"
5596 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5598 "%param1 = OpFunctionParameter %v4f32\n"
5601 "%label_testfun = OpLabel\n"
5603 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5604 "OpReturnValue %val1\n"
5606 "OpLine %name 1 1\n"
5609 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
5611 return opLineTests.release();
5614 tcu::TestCaseGroup* createOpModuleProcessedTests(tcu::TestContext& testCtx)
5616 RGBA defaultColors[4];
5617 de::MovePtr<tcu::TestCaseGroup> opModuleProcessedTests (new tcu::TestCaseGroup(testCtx, "opmoduleprocessed", "OpModuleProcessed instruction"));
5618 map<string, string> fragments;
5619 std::vector<std::string> noExtensions;
5620 GraphicsResources resources;
5622 getDefaultColors(defaultColors);
5623 resources.verifyBinary = veryfiBinaryShader;
5624 resources.spirvVersion = SPIRV_VERSION_1_3;
5626 fragments["moduleprocessed"] =
5627 "OpModuleProcessed \"VULKAN CTS\"\n"
5628 "OpModuleProcessed \"Negative values\"\n"
5629 "OpModuleProcessed \"Date: 2017/09/21\"\n";
5631 fragments["pre_main"] =
5632 "%second_function = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5633 "%second_param1 = OpFunctionParameter %v4f32\n"
5634 "%label_secondfunction = OpLabel\n"
5635 "OpReturnValue %second_param1\n"
5638 fragments["testfun"] =
5639 // A %test_code function that returns its argument unchanged.
5640 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5641 "%param1 = OpFunctionParameter %v4f32\n"
5642 "%label_testfun = OpLabel\n"
5643 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5644 "OpReturnValue %val1\n"
5647 createTestsForAllStages ("opmoduleprocessed", defaultColors, defaultColors, fragments, resources, noExtensions, opModuleProcessedTests.get());
5649 return opModuleProcessedTests.release();
5653 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
5655 RGBA defaultColors[4];
5656 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
5657 map<string, string> fragments;
5658 std::vector<std::pair<std::string, std::string> > problemStrings;
5660 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
5661 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
5662 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
5663 getDefaultColors(defaultColors);
5665 fragments["debug"] =
5666 "%other_name = OpString \"other_name\"\n";
5668 fragments["pre_main"] =
5669 "OpLine %file_name 32 0\n"
5670 "OpLine %file_name 32 32\n"
5671 "OpLine %file_name 32 40\n"
5672 "OpLine %other_name 32 40\n"
5673 "OpLine %other_name 0 100\n"
5674 "OpLine %other_name 0 4294967295\n"
5675 "OpLine %other_name 4294967295 0\n"
5676 "OpLine %other_name 32 40\n"
5677 "OpLine %file_name 0 0\n"
5678 "%second_function = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5679 "OpLine %file_name 1 0\n"
5680 "%second_param1 = OpFunctionParameter %v4f32\n"
5681 "OpLine %file_name 1 3\n"
5682 "OpLine %file_name 1 2\n"
5683 "%label_secondfunction = OpLabel\n"
5684 "OpLine %file_name 0 2\n"
5685 "OpReturnValue %second_param1\n"
5687 "OpLine %file_name 0 2\n"
5688 "OpLine %file_name 0 2\n";
5690 fragments["testfun"] =
5691 // A %test_code function that returns its argument unchanged.
5692 "OpLine %file_name 1 0\n"
5693 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5694 "OpLine %file_name 16 330\n"
5695 "%param1 = OpFunctionParameter %v4f32\n"
5696 "OpLine %file_name 14 442\n"
5697 "%label_testfun = OpLabel\n"
5698 "OpLine %file_name 11 1024\n"
5699 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5700 "OpLine %file_name 2 97\n"
5701 "OpReturnValue %val1\n"
5703 "OpLine %file_name 5 32\n";
5705 for (size_t i = 0; i < problemStrings.size(); ++i)
5707 map<string, string> testFragments = fragments;
5708 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
5709 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
5712 return opLineTests.release();
5715 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
5717 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
5721 const char functionStart[] =
5722 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5723 "%param1 = OpFunctionParameter %v4f32\n"
5726 const char functionEnd[] =
5727 "OpReturnValue %transformed_param\n"
5730 struct NameConstantsCode
5737 NameConstantsCode tests[] =
5741 "%cnull = OpConstantNull %v4f32\n",
5742 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
5746 "%cnull = OpConstantNull %f32\n",
5747 "%vp = OpVariable %fp_v4f32 Function\n"
5748 "%v = OpLoad %v4f32 %vp\n"
5749 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
5750 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
5751 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
5752 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
5753 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
5757 "%cnull = OpConstantNull %bool\n",
5758 "%v = OpVariable %fp_v4f32 Function\n"
5759 " OpStore %v %param1\n"
5760 " OpSelectionMerge %false_label None\n"
5761 " OpBranchConditional %cnull %true_label %false_label\n"
5762 "%true_label = OpLabel\n"
5763 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
5764 " OpBranch %false_label\n"
5765 "%false_label = OpLabel\n"
5766 "%transformed_param = OpLoad %v4f32 %v\n"
5770 "%cnull = OpConstantNull %i32\n",
5771 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
5772 "%b = OpIEqual %bool %cnull %c_i32_0\n"
5773 " OpSelectionMerge %false_label None\n"
5774 " OpBranchConditional %b %true_label %false_label\n"
5775 "%true_label = OpLabel\n"
5776 " OpStore %v %param1\n"
5777 " OpBranch %false_label\n"
5778 "%false_label = OpLabel\n"
5779 "%transformed_param = OpLoad %v4f32 %v\n"
5783 "%stype = OpTypeStruct %f32 %v4f32\n"
5784 "%fp_stype = OpTypePointer Function %stype\n"
5785 "%cnull = OpConstantNull %stype\n",
5786 "%v = OpVariable %fp_stype Function %cnull\n"
5787 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
5788 "%f_val = OpLoad %v4f32 %f\n"
5789 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
5793 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
5794 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
5795 "%cnull = OpConstantNull %a4_v4f32\n",
5796 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
5797 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5798 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
5799 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
5800 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
5801 "%f_val = OpLoad %v4f32 %f\n"
5802 "%f1_val = OpLoad %v4f32 %f1\n"
5803 "%f2_val = OpLoad %v4f32 %f2\n"
5804 "%f3_val = OpLoad %v4f32 %f3\n"
5805 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
5806 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
5807 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
5808 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
5812 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5813 "%cnull = OpConstantNull %mat4x4_f32\n",
5814 // Our null matrix * any vector should result in a zero vector.
5815 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
5816 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
5820 getHalfColorsFullAlpha(colors);
5822 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5824 map<string, string> fragments;
5825 fragments["pre_main"] = tests[testNdx].constants;
5826 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5827 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
5829 return opConstantNullTests.release();
5831 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
5833 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
5834 RGBA inputColors[4];
5835 RGBA outputColors[4];
5838 const char functionStart[] =
5839 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5840 "%param1 = OpFunctionParameter %v4f32\n"
5843 const char functionEnd[] =
5844 "OpReturnValue %transformed_param\n"
5847 struct NameConstantsCode
5854 NameConstantsCode tests[] =
5859 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
5860 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
5865 "%stype = OpTypeStruct %v4f32 %f32\n"
5866 "%fp_stype = OpTypePointer Function %stype\n"
5867 "%f32_n_1 = OpConstant %f32 -1.0\n"
5868 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5869 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
5870 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
5872 "%v = OpVariable %fp_stype Function %cval\n"
5873 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5874 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
5875 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
5876 "%f32_val = OpLoad %f32 %f32_ptr\n"
5877 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
5878 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
5879 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
5882 // [1|0|0|0.5] [x] = x + 0.5
5883 // [0|1|0|0.5] [y] = y + 0.5
5884 // [0|0|1|0.5] [z] = z + 0.5
5885 // [0|0|0|1 ] [1] = 1
5888 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5889 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
5890 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
5891 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
5892 "%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"
5893 "%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",
5895 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
5900 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5901 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5902 "%f32_n_1 = OpConstant %f32 -1.0\n"
5903 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5904 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
5906 "%v = OpVariable %fp_a4f32 Function %carr\n"
5907 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
5908 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
5909 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5910 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
5911 "%f_val = OpLoad %f32 %f\n"
5912 "%f1_val = OpLoad %f32 %f1\n"
5913 "%f2_val = OpLoad %f32 %f2\n"
5914 "%f3_val = OpLoad %f32 %f3\n"
5915 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
5916 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
5917 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
5918 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
5919 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5926 // [ 1.0, 1.0, 1.0, 1.0]
5930 // [ 0.0, 0.5, 0.0, 0.0]
5934 // [ 1.0, 1.0, 1.0, 1.0]
5937 "array_of_struct_of_array",
5939 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5940 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5941 "%stype = OpTypeStruct %f32 %a4f32\n"
5942 "%a3stype = OpTypeArray %stype %c_u32_3\n"
5943 "%fp_a3stype = OpTypePointer Function %a3stype\n"
5944 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
5945 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5946 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
5947 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
5948 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
5950 "%v = OpVariable %fp_a3stype Function %carr\n"
5951 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
5952 "%f_l = OpLoad %f32 %f\n"
5953 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
5954 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5958 getHalfColorsFullAlpha(inputColors);
5959 outputColors[0] = RGBA(255, 255, 255, 255);
5960 outputColors[1] = RGBA(255, 127, 127, 255);
5961 outputColors[2] = RGBA(127, 255, 127, 255);
5962 outputColors[3] = RGBA(127, 127, 255, 255);
5964 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5966 map<string, string> fragments;
5967 fragments["pre_main"] = tests[testNdx].constants;
5968 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5969 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
5971 return opConstantCompositeTests.release();
5974 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
5976 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
5977 RGBA inputColors[4];
5978 RGBA outputColors[4];
5979 map<string, string> fragments;
5981 // vec4 test_code(vec4 param) {
5982 // vec4 result = param;
5983 // for (int i = 0; i < 4; ++i) {
5984 // if (i == 0) result[i] = 0.;
5985 // else result[i] = 1. - result[i];
5989 const char function[] =
5990 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5991 "%param1 = OpFunctionParameter %v4f32\n"
5993 "%iptr = OpVariable %fp_i32 Function\n"
5994 "%result = OpVariable %fp_v4f32 Function\n"
5995 " OpStore %iptr %c_i32_0\n"
5996 " OpStore %result %param1\n"
5999 // Loop entry block.
6001 "%ival = OpLoad %i32 %iptr\n"
6002 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6003 " OpLoopMerge %exit %if_entry None\n"
6004 " OpBranchConditional %lt_4 %if_entry %exit\n"
6006 // Merge block for loop.
6008 "%ret = OpLoad %v4f32 %result\n"
6009 " OpReturnValue %ret\n"
6011 // If-statement entry block.
6012 "%if_entry = OpLabel\n"
6013 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6014 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
6015 " OpSelectionMerge %if_exit None\n"
6016 " OpBranchConditional %eq_0 %if_true %if_false\n"
6018 // False branch for if-statement.
6019 "%if_false = OpLabel\n"
6020 "%val = OpLoad %f32 %loc\n"
6021 "%sub = OpFSub %f32 %c_f32_1 %val\n"
6022 " OpStore %loc %sub\n"
6023 " OpBranch %if_exit\n"
6025 // Merge block for if-statement.
6026 "%if_exit = OpLabel\n"
6027 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6028 " OpStore %iptr %ival_next\n"
6031 // True branch for if-statement.
6032 "%if_true = OpLabel\n"
6033 " OpStore %loc %c_f32_0\n"
6034 " OpBranch %if_exit\n"
6038 fragments["testfun"] = function;
6040 inputColors[0] = RGBA(127, 127, 127, 0);
6041 inputColors[1] = RGBA(127, 0, 0, 0);
6042 inputColors[2] = RGBA(0, 127, 0, 0);
6043 inputColors[3] = RGBA(0, 0, 127, 0);
6045 outputColors[0] = RGBA(0, 128, 128, 255);
6046 outputColors[1] = RGBA(0, 255, 255, 255);
6047 outputColors[2] = RGBA(0, 128, 255, 255);
6048 outputColors[3] = RGBA(0, 255, 128, 255);
6050 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
6052 return group.release();
6055 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
6057 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
6058 RGBA inputColors[4];
6059 RGBA outputColors[4];
6060 map<string, string> fragments;
6062 const char typesAndConstants[] =
6063 "%c_f32_p2 = OpConstant %f32 0.2\n"
6064 "%c_f32_p4 = OpConstant %f32 0.4\n"
6065 "%c_f32_p6 = OpConstant %f32 0.6\n"
6066 "%c_f32_p8 = OpConstant %f32 0.8\n";
6068 // vec4 test_code(vec4 param) {
6069 // vec4 result = param;
6070 // for (int i = 0; i < 4; ++i) {
6072 // case 0: result[i] += .2; break;
6073 // case 1: result[i] += .6; break;
6074 // case 2: result[i] += .4; break;
6075 // case 3: result[i] += .8; break;
6076 // default: break; // unreachable
6081 const char function[] =
6082 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6083 "%param1 = OpFunctionParameter %v4f32\n"
6085 "%iptr = OpVariable %fp_i32 Function\n"
6086 "%result = OpVariable %fp_v4f32 Function\n"
6087 " OpStore %iptr %c_i32_0\n"
6088 " OpStore %result %param1\n"
6091 // Loop entry block.
6093 "%ival = OpLoad %i32 %iptr\n"
6094 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6095 " OpLoopMerge %exit %switch_exit None\n"
6096 " OpBranchConditional %lt_4 %switch_entry %exit\n"
6098 // Merge block for loop.
6100 "%ret = OpLoad %v4f32 %result\n"
6101 " OpReturnValue %ret\n"
6103 // Switch-statement entry block.
6104 "%switch_entry = OpLabel\n"
6105 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6106 "%val = OpLoad %f32 %loc\n"
6107 " OpSelectionMerge %switch_exit None\n"
6108 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6110 "%case2 = OpLabel\n"
6111 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
6112 " OpStore %loc %addp4\n"
6113 " OpBranch %switch_exit\n"
6115 "%switch_default = OpLabel\n"
6118 "%case3 = OpLabel\n"
6119 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
6120 " OpStore %loc %addp8\n"
6121 " OpBranch %switch_exit\n"
6123 "%case0 = OpLabel\n"
6124 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
6125 " OpStore %loc %addp2\n"
6126 " OpBranch %switch_exit\n"
6128 // Merge block for switch-statement.
6129 "%switch_exit = OpLabel\n"
6130 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6131 " OpStore %iptr %ival_next\n"
6134 "%case1 = OpLabel\n"
6135 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
6136 " OpStore %loc %addp6\n"
6137 " OpBranch %switch_exit\n"
6141 fragments["pre_main"] = typesAndConstants;
6142 fragments["testfun"] = function;
6144 inputColors[0] = RGBA(127, 27, 127, 51);
6145 inputColors[1] = RGBA(127, 0, 0, 51);
6146 inputColors[2] = RGBA(0, 27, 0, 51);
6147 inputColors[3] = RGBA(0, 0, 127, 51);
6149 outputColors[0] = RGBA(178, 180, 229, 255);
6150 outputColors[1] = RGBA(178, 153, 102, 255);
6151 outputColors[2] = RGBA(51, 180, 102, 255);
6152 outputColors[3] = RGBA(51, 153, 229, 255);
6154 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
6156 return group.release();
6159 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
6161 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
6162 RGBA inputColors[4];
6163 RGBA outputColors[4];
6164 map<string, string> fragments;
6166 const char decorations[] =
6167 "OpDecorate %array_group ArrayStride 4\n"
6168 "OpDecorate %struct_member_group Offset 0\n"
6169 "%array_group = OpDecorationGroup\n"
6170 "%struct_member_group = OpDecorationGroup\n"
6172 "OpDecorate %group1 RelaxedPrecision\n"
6173 "OpDecorate %group3 RelaxedPrecision\n"
6174 "OpDecorate %group3 Invariant\n"
6175 "OpDecorate %group3 Restrict\n"
6176 "%group0 = OpDecorationGroup\n"
6177 "%group1 = OpDecorationGroup\n"
6178 "%group3 = OpDecorationGroup\n";
6180 const char typesAndConstants[] =
6181 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
6182 "%struct1 = OpTypeStruct %a3f32\n"
6183 "%struct2 = OpTypeStruct %a3f32\n"
6184 "%fp_struct1 = OpTypePointer Function %struct1\n"
6185 "%fp_struct2 = OpTypePointer Function %struct2\n"
6186 "%c_f32_2 = OpConstant %f32 2.\n"
6187 "%c_f32_n2 = OpConstant %f32 -2.\n"
6189 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
6190 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
6191 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
6192 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
6194 const char function[] =
6195 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6196 "%param = OpFunctionParameter %v4f32\n"
6197 "%entry = OpLabel\n"
6198 "%result = OpVariable %fp_v4f32 Function\n"
6199 "%v_struct1 = OpVariable %fp_struct1 Function\n"
6200 "%v_struct2 = OpVariable %fp_struct2 Function\n"
6201 " OpStore %result %param\n"
6202 " OpStore %v_struct1 %c_struct1\n"
6203 " OpStore %v_struct2 %c_struct2\n"
6204 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
6205 "%val1 = OpLoad %f32 %ptr1\n"
6206 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
6207 "%val2 = OpLoad %f32 %ptr2\n"
6208 "%addvalues = OpFAdd %f32 %val1 %val2\n"
6209 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
6210 "%val = OpLoad %f32 %ptr\n"
6211 "%addresult = OpFAdd %f32 %addvalues %val\n"
6212 " OpStore %ptr %addresult\n"
6213 "%ret = OpLoad %v4f32 %result\n"
6214 " OpReturnValue %ret\n"
6217 struct CaseNameDecoration
6223 CaseNameDecoration tests[] =
6226 "same_decoration_group_on_multiple_types",
6227 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
6230 "empty_decoration_group",
6231 "OpGroupDecorate %group0 %a3f32\n"
6232 "OpGroupDecorate %group0 %result\n"
6235 "one_element_decoration_group",
6236 "OpGroupDecorate %array_group %a3f32\n"
6239 "multiple_elements_decoration_group",
6240 "OpGroupDecorate %group3 %v_struct1\n"
6243 "multiple_decoration_groups_on_same_variable",
6244 "OpGroupDecorate %group0 %v_struct2\n"
6245 "OpGroupDecorate %group1 %v_struct2\n"
6246 "OpGroupDecorate %group3 %v_struct2\n"
6249 "same_decoration_group_multiple_times",
6250 "OpGroupDecorate %group1 %addvalues\n"
6251 "OpGroupDecorate %group1 %addvalues\n"
6252 "OpGroupDecorate %group1 %addvalues\n"
6257 getHalfColorsFullAlpha(inputColors);
6258 getHalfColorsFullAlpha(outputColors);
6260 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
6262 fragments["decoration"] = decorations + tests[idx].decoration;
6263 fragments["pre_main"] = typesAndConstants;
6264 fragments["testfun"] = function;
6266 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
6269 return group.release();
6272 struct SpecConstantTwoIntGraphicsCase
6274 const char* caseName;
6275 const char* scDefinition0;
6276 const char* scDefinition1;
6277 const char* scResultType;
6278 const char* scOperation;
6279 deInt32 scActualValue0;
6280 deInt32 scActualValue1;
6281 const char* resultOperation;
6282 RGBA expectedColors[4];
6284 SpecConstantTwoIntGraphicsCase (const char* name,
6285 const char* definition0,
6286 const char* definition1,
6287 const char* resultType,
6288 const char* operation,
6291 const char* resultOp,
6292 const RGBA (&output)[4])
6294 , scDefinition0 (definition0)
6295 , scDefinition1 (definition1)
6296 , scResultType (resultType)
6297 , scOperation (operation)
6298 , scActualValue0 (value0)
6299 , scActualValue1 (value1)
6300 , resultOperation (resultOp)
6302 expectedColors[0] = output[0];
6303 expectedColors[1] = output[1];
6304 expectedColors[2] = output[2];
6305 expectedColors[3] = output[3];
6309 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
6311 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
6312 vector<SpecConstantTwoIntGraphicsCase> cases;
6313 RGBA inputColors[4];
6314 RGBA outputColors0[4];
6315 RGBA outputColors1[4];
6316 RGBA outputColors2[4];
6318 const char decorations1[] =
6319 "OpDecorate %sc_0 SpecId 0\n"
6320 "OpDecorate %sc_1 SpecId 1\n";
6322 const char typesAndConstants1[] =
6323 "${OPTYPE_DEFINITIONS:opt}"
6324 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
6325 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
6326 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
6328 const char function1[] =
6329 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6330 "%param = OpFunctionParameter %v4f32\n"
6331 "%label = OpLabel\n"
6332 "%result = OpVariable %fp_v4f32 Function\n"
6333 "${TYPE_CONVERT:opt}"
6334 " OpStore %result %param\n"
6335 "%gen = ${GEN_RESULT}\n"
6336 "%index = OpIAdd %i32 %gen %c_i32_1\n"
6337 "%loc = OpAccessChain %fp_f32 %result %index\n"
6338 "%val = OpLoad %f32 %loc\n"
6339 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6340 " OpStore %loc %add\n"
6341 "%ret = OpLoad %v4f32 %result\n"
6342 " OpReturnValue %ret\n"
6345 inputColors[0] = RGBA(127, 127, 127, 255);
6346 inputColors[1] = RGBA(127, 0, 0, 255);
6347 inputColors[2] = RGBA(0, 127, 0, 255);
6348 inputColors[3] = RGBA(0, 0, 127, 255);
6350 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
6351 outputColors0[0] = RGBA(255, 127, 127, 255);
6352 outputColors0[1] = RGBA(255, 0, 0, 255);
6353 outputColors0[2] = RGBA(128, 127, 0, 255);
6354 outputColors0[3] = RGBA(128, 0, 127, 255);
6356 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
6357 outputColors1[0] = RGBA(127, 255, 127, 255);
6358 outputColors1[1] = RGBA(127, 128, 0, 255);
6359 outputColors1[2] = RGBA(0, 255, 0, 255);
6360 outputColors1[3] = RGBA(0, 128, 127, 255);
6362 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
6363 outputColors2[0] = RGBA(127, 127, 255, 255);
6364 outputColors2[1] = RGBA(127, 0, 128, 255);
6365 outputColors2[2] = RGBA(0, 127, 128, 255);
6366 outputColors2[3] = RGBA(0, 0, 255, 255);
6368 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
6369 const char addZeroToSc32[] = "OpIAdd %i32 %c_i32_0 %sc_op32";
6370 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
6371 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
6373 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
6374 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
6375 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
6376 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
6377 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
6378 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6379 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6380 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
6381 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
6382 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
6383 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
6384 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
6385 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
6386 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
6387 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
6388 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
6389 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6390 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
6391 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
6392 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
6393 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6394 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
6395 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
6396 cases.push_back(SpecConstantTwoIntGraphicsCase("inotequal", " %i32 0", " %i32 0", "%bool", "INotEqual %sc_0 %sc_1", 42, 24, selectTrueUsingSc, outputColors2));
6397 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6398 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6399 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6400 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6401 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
6402 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
6403 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
6404 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
6405 cases.push_back(SpecConstantTwoIntGraphicsCase("sconvert", " %i32 0", " %i32 0", "%i16", "SConvert %sc_0", -1, 0, addZeroToSc32, outputColors0));
6406 // -1082130432 stored as 32-bit two's complement is the binary representation of -1 as IEEE-754 Float
6407 cases.push_back(SpecConstantTwoIntGraphicsCase("fconvert", " %f32 0", " %f32 0", "%f64", "FConvert %sc_0", -1082130432, 0, addZeroToSc32, outputColors0));
6408 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
6410 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
6412 map<string, string> specializations;
6413 map<string, string> fragments;
6414 SpecConstants specConstants;
6415 vector<string> features;
6416 PushConstants noPushConstants;
6417 GraphicsResources noResources;
6418 GraphicsInterfaces noInterfaces;
6419 std::vector<std::string> noExtensions;
6421 // Special SPIR-V code for SConvert-case
6422 if (strcmp(cases[caseNdx].caseName, "sconvert") == 0)
6424 features.push_back("shaderInt16");
6425 fragments["capability"] = "OpCapability Int16\n"; // Adds 16-bit integer capability
6426 specializations["OPTYPE_DEFINITIONS"] = "%i16 = OpTypeInt 16 1\n"; // Adds 16-bit integer type
6427 specializations["TYPE_CONVERT"] = "%sc_op32 = OpSConvert %i32 %sc_op\n"; // Converts 16-bit integer to 32-bit integer
6430 // Special SPIR-V code for FConvert-case
6431 if (strcmp(cases[caseNdx].caseName, "fconvert") == 0)
6433 features.push_back("shaderFloat64");
6434 fragments["capability"] = "OpCapability Float64\n"; // Adds 64-bit float capability
6435 specializations["OPTYPE_DEFINITIONS"] = "%f64 = OpTypeFloat 64\n"; // Adds 64-bit float type
6436 specializations["TYPE_CONVERT"] = "%sc_op32 = OpConvertFToS %i32 %sc_op\n"; // Converts 64-bit float to 32-bit integer
6439 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
6440 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
6441 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
6442 specializations["SC_OP"] = cases[caseNdx].scOperation;
6443 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
6445 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
6446 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
6447 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
6449 specConstants.append(cases[caseNdx].scActualValue0);
6450 specConstants.append(cases[caseNdx].scActualValue1);
6452 createTestsForAllStages(
6453 cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants,
6454 noPushConstants, noResources, noInterfaces, noExtensions, features, VulkanFeatures(), group.get());
6457 const char decorations2[] =
6458 "OpDecorate %sc_0 SpecId 0\n"
6459 "OpDecorate %sc_1 SpecId 1\n"
6460 "OpDecorate %sc_2 SpecId 2\n";
6462 const char typesAndConstants2[] =
6463 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
6464 "%vec3_undef = OpUndef %v3i32\n"
6466 "%sc_0 = OpSpecConstant %i32 0\n"
6467 "%sc_1 = OpSpecConstant %i32 0\n"
6468 "%sc_2 = OpSpecConstant %i32 0\n"
6469 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
6470 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
6471 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
6472 "%sc_vec3_0_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
6473 "%sc_vec3_1_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
6474 "%sc_vec3_2_s = OpSpecConstantOp %v3i32 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
6475 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
6476 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
6477 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
6478 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
6479 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
6480 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
6481 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
6483 const char function2[] =
6484 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6485 "%param = OpFunctionParameter %v4f32\n"
6486 "%label = OpLabel\n"
6487 "%result = OpVariable %fp_v4f32 Function\n"
6488 " OpStore %result %param\n"
6489 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
6490 "%val = OpLoad %f32 %loc\n"
6491 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6492 " OpStore %loc %add\n"
6493 "%ret = OpLoad %v4f32 %result\n"
6494 " OpReturnValue %ret\n"
6497 map<string, string> fragments;
6498 SpecConstants specConstants;
6500 fragments["decoration"] = decorations2;
6501 fragments["pre_main"] = typesAndConstants2;
6502 fragments["testfun"] = function2;
6504 specConstants.append<deInt32>(56789);
6505 specConstants.append<deInt32>(-2);
6506 specConstants.append<deInt32>(56788);
6508 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
6510 return group.release();
6513 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
6515 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
6516 RGBA inputColors[4];
6517 RGBA outputColors1[4];
6518 RGBA outputColors2[4];
6519 RGBA outputColors3[4];
6520 map<string, string> fragments1;
6521 map<string, string> fragments2;
6522 map<string, string> fragments3;
6524 const char typesAndConstants1[] =
6525 "%c_f32_p2 = OpConstant %f32 0.2\n"
6526 "%c_f32_p4 = OpConstant %f32 0.4\n"
6527 "%c_f32_p5 = OpConstant %f32 0.5\n"
6528 "%c_f32_p8 = OpConstant %f32 0.8\n";
6530 // vec4 test_code(vec4 param) {
6531 // vec4 result = param;
6532 // for (int i = 0; i < 4; ++i) {
6535 // case 0: operand = .2; break;
6536 // case 1: operand = .5; break;
6537 // case 2: operand = .4; break;
6538 // case 3: operand = .0; break;
6539 // default: break; // unreachable
6541 // result[i] += operand;
6545 const char function1[] =
6546 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6547 "%param1 = OpFunctionParameter %v4f32\n"
6549 "%iptr = OpVariable %fp_i32 Function\n"
6550 "%result = OpVariable %fp_v4f32 Function\n"
6551 " OpStore %iptr %c_i32_0\n"
6552 " OpStore %result %param1\n"
6556 "%ival = OpLoad %i32 %iptr\n"
6557 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6558 " OpLoopMerge %exit %phi None\n"
6559 " OpBranchConditional %lt_4 %entry %exit\n"
6561 "%entry = OpLabel\n"
6562 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6563 "%val = OpLoad %f32 %loc\n"
6564 " OpSelectionMerge %phi None\n"
6565 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6567 "%case0 = OpLabel\n"
6569 "%case1 = OpLabel\n"
6571 "%case2 = OpLabel\n"
6573 "%case3 = OpLabel\n"
6576 "%default = OpLabel\n"
6580 "%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
6581 "%add = OpFAdd %f32 %val %operand\n"
6582 " OpStore %loc %add\n"
6583 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6584 " OpStore %iptr %ival_next\n"
6588 "%ret = OpLoad %v4f32 %result\n"
6589 " OpReturnValue %ret\n"
6593 fragments1["pre_main"] = typesAndConstants1;
6594 fragments1["testfun"] = function1;
6596 getHalfColorsFullAlpha(inputColors);
6598 outputColors1[0] = RGBA(178, 255, 229, 255);
6599 outputColors1[1] = RGBA(178, 127, 102, 255);
6600 outputColors1[2] = RGBA(51, 255, 102, 255);
6601 outputColors1[3] = RGBA(51, 127, 229, 255);
6603 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
6605 const char typesAndConstants2[] =
6606 "%c_f32_p2 = OpConstant %f32 0.2\n";
6608 // Add .4 to the second element of the given parameter.
6609 const char function2[] =
6610 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6611 "%param = OpFunctionParameter %v4f32\n"
6612 "%entry = OpLabel\n"
6613 "%result = OpVariable %fp_v4f32 Function\n"
6614 " OpStore %result %param\n"
6615 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6616 "%val = OpLoad %f32 %loc\n"
6620 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
6621 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
6622 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
6623 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
6624 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
6625 " OpLoopMerge %exit %phi None\n"
6626 " OpBranchConditional %still_loop %phi %exit\n"
6629 " OpStore %loc %accum\n"
6630 "%ret = OpLoad %v4f32 %result\n"
6631 " OpReturnValue %ret\n"
6635 fragments2["pre_main"] = typesAndConstants2;
6636 fragments2["testfun"] = function2;
6638 outputColors2[0] = RGBA(127, 229, 127, 255);
6639 outputColors2[1] = RGBA(127, 102, 0, 255);
6640 outputColors2[2] = RGBA(0, 229, 0, 255);
6641 outputColors2[3] = RGBA(0, 102, 127, 255);
6643 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
6645 const char typesAndConstants3[] =
6646 "%true = OpConstantTrue %bool\n"
6647 "%false = OpConstantFalse %bool\n"
6648 "%c_f32_p2 = OpConstant %f32 0.2\n";
6650 // Swap the second and the third element of the given parameter.
6651 const char function3[] =
6652 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6653 "%param = OpFunctionParameter %v4f32\n"
6654 "%entry = OpLabel\n"
6655 "%result = OpVariable %fp_v4f32 Function\n"
6656 " OpStore %result %param\n"
6657 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6658 "%a_init = OpLoad %f32 %a_loc\n"
6659 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
6660 "%b_init = OpLoad %f32 %b_loc\n"
6664 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
6665 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
6666 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
6667 " OpLoopMerge %exit %phi None\n"
6668 " OpBranchConditional %still_loop %phi %exit\n"
6671 " OpStore %a_loc %a_next\n"
6672 " OpStore %b_loc %b_next\n"
6673 "%ret = OpLoad %v4f32 %result\n"
6674 " OpReturnValue %ret\n"
6678 fragments3["pre_main"] = typesAndConstants3;
6679 fragments3["testfun"] = function3;
6681 outputColors3[0] = RGBA(127, 127, 127, 255);
6682 outputColors3[1] = RGBA(127, 0, 0, 255);
6683 outputColors3[2] = RGBA(0, 0, 127, 255);
6684 outputColors3[3] = RGBA(0, 127, 0, 255);
6686 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
6688 return group.release();
6691 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
6693 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
6694 RGBA inputColors[4];
6695 RGBA outputColors[4];
6697 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
6698 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
6699 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
6700 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
6701 const char constantsAndTypes[] =
6702 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
6703 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
6704 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
6705 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
6706 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n";
6708 const char function[] =
6709 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6710 "%param = OpFunctionParameter %v4f32\n"
6711 "%label = OpLabel\n"
6712 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
6713 "%var2 = OpVariable %fp_f32 Function\n"
6714 "%red = OpCompositeExtract %f32 %param 0\n"
6715 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
6716 " OpStore %var2 %plus_red\n"
6717 "%val1 = OpLoad %f32 %var1\n"
6718 "%val2 = OpLoad %f32 %var2\n"
6719 "%mul = OpFMul %f32 %val1 %val2\n"
6720 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
6721 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
6722 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
6723 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
6724 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
6725 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
6726 " OpReturnValue %ret\n"
6729 struct CaseNameDecoration
6736 CaseNameDecoration tests[] = {
6737 {"multiplication", "OpDecorate %mul NoContraction"},
6738 {"addition", "OpDecorate %add NoContraction"},
6739 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
6742 getHalfColorsFullAlpha(inputColors);
6744 for (deUint8 idx = 0; idx < 4; ++idx)
6746 inputColors[idx].setRed(0);
6747 outputColors[idx] = RGBA(0, 0, 0, 255);
6750 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
6752 map<string, string> fragments;
6754 fragments["decoration"] = tests[testNdx].decoration;
6755 fragments["pre_main"] = constantsAndTypes;
6756 fragments["testfun"] = function;
6758 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
6761 return group.release();
6764 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
6766 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
6769 const char constantsAndTypes[] =
6770 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
6771 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
6772 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
6773 "%fp_stype = OpTypePointer Function %stype\n";
6775 const char function[] =
6776 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6777 "%param1 = OpFunctionParameter %v4f32\n"
6779 "%v1 = OpVariable %fp_v4f32 Function\n"
6780 "%v2 = OpVariable %fp_a2f32 Function\n"
6781 "%v3 = OpVariable %fp_f32 Function\n"
6782 "%v = OpVariable %fp_stype Function\n"
6783 "%vv = OpVariable %fp_stype Function\n"
6784 "%vvv = OpVariable %fp_f32 Function\n"
6786 " OpStore %v1 %c_v4f32_1_1_1_1\n"
6787 " OpStore %v2 %c_a2f32_1\n"
6788 " OpStore %v3 %c_f32_1\n"
6790 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
6791 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
6792 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
6793 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
6794 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
6795 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
6797 " OpStore %p_v4f32 %v1_v ${access_type}\n"
6798 " OpStore %p_a2f32 %v2_v ${access_type}\n"
6799 " OpStore %p_f32 %v3_v ${access_type}\n"
6801 " OpCopyMemory %vv %v ${access_type}\n"
6802 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
6804 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
6805 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
6806 "%v_f32_3 = OpLoad %f32 %vvv\n"
6808 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
6809 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
6810 " OpReturnValue %ret2\n"
6813 struct NameMemoryAccess
6820 NameMemoryAccess tests[] =
6823 { "volatile", "Volatile" },
6824 { "aligned", "Aligned 1" },
6825 { "volatile_aligned", "Volatile|Aligned 1" },
6826 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
6827 { "volatile_nontemporal", "Volatile|Nontemporal" },
6828 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
6831 getHalfColorsFullAlpha(colors);
6833 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
6835 map<string, string> fragments;
6836 map<string, string> memoryAccess;
6837 memoryAccess["access_type"] = tests[testNdx].accessType;
6839 fragments["pre_main"] = constantsAndTypes;
6840 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
6841 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
6843 return memoryAccessTests.release();
6845 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
6847 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
6848 RGBA defaultColors[4];
6849 map<string, string> fragments;
6850 getDefaultColors(defaultColors);
6852 // First, simple cases that don't do anything with the OpUndef result.
6853 struct NameCodePair { string name, decl, type; };
6854 const NameCodePair tests[] =
6856 {"bool", "", "%bool"},
6857 {"vec2uint32", "", "%v2u32"},
6858 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown", "%type"},
6859 {"sampler", "%type = OpTypeSampler", "%type"},
6860 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img", "%type"},
6861 {"pointer", "", "%fp_i32"},
6862 {"runtimearray", "%type = OpTypeRuntimeArray %f32", "%type"},
6863 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100", "%type"},
6864 {"struct", "%type = OpTypeStruct %f32 %i32 %u32", "%type"}};
6865 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
6867 fragments["undef_type"] = tests[testNdx].type;
6868 fragments["testfun"] = StringTemplate(
6869 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6870 "%param1 = OpFunctionParameter %v4f32\n"
6871 "%label_testfun = OpLabel\n"
6872 "%undef = OpUndef ${undef_type}\n"
6873 "OpReturnValue %param1\n"
6874 "OpFunctionEnd\n").specialize(fragments);
6875 fragments["pre_main"] = tests[testNdx].decl;
6876 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
6880 fragments["testfun"] =
6881 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6882 "%param1 = OpFunctionParameter %v4f32\n"
6883 "%label_testfun = OpLabel\n"
6884 "%undef = OpUndef %f32\n"
6885 "%zero = OpFMul %f32 %undef %c_f32_0\n"
6886 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
6887 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
6888 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6889 "%b = OpFAdd %f32 %a %actually_zero\n"
6890 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
6891 "OpReturnValue %ret\n"
6894 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6896 fragments["testfun"] =
6897 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6898 "%param1 = OpFunctionParameter %v4f32\n"
6899 "%label_testfun = OpLabel\n"
6900 "%undef = OpUndef %i32\n"
6901 "%zero = OpIMul %i32 %undef %c_i32_0\n"
6902 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6903 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6904 "OpReturnValue %ret\n"
6907 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6909 fragments["testfun"] =
6910 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6911 "%param1 = OpFunctionParameter %v4f32\n"
6912 "%label_testfun = OpLabel\n"
6913 "%undef = OpUndef %u32\n"
6914 "%zero = OpIMul %u32 %undef %c_i32_0\n"
6915 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6916 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6917 "OpReturnValue %ret\n"
6920 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6922 fragments["testfun"] =
6923 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6924 "%param1 = OpFunctionParameter %v4f32\n"
6925 "%label_testfun = OpLabel\n"
6926 "%undef = OpUndef %v4f32\n"
6927 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
6928 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
6929 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
6930 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
6931 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
6932 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6933 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6934 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6935 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6936 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6937 "%actually_zero_1 = OpSelect %f32 %is_nan_1 %c_f32_0 %zero_1\n"
6938 "%actually_zero_2 = OpSelect %f32 %is_nan_2 %c_f32_0 %zero_2\n"
6939 "%actually_zero_3 = OpSelect %f32 %is_nan_3 %c_f32_0 %zero_3\n"
6940 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6941 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6942 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6943 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6944 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6945 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6946 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6947 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6948 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6949 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6950 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6951 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6952 "OpReturnValue %ret\n"
6955 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6957 fragments["pre_main"] =
6958 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
6959 fragments["testfun"] =
6960 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6961 "%param1 = OpFunctionParameter %v4f32\n"
6962 "%label_testfun = OpLabel\n"
6963 "%undef = OpUndef %m2x2f32\n"
6964 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
6965 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
6966 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
6967 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
6968 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
6969 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6970 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6971 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6972 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6973 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6974 "%actually_zero_1 = OpSelect %f32 %is_nan_1 %c_f32_0 %zero_1\n"
6975 "%actually_zero_2 = OpSelect %f32 %is_nan_2 %c_f32_0 %zero_2\n"
6976 "%actually_zero_3 = OpSelect %f32 %is_nan_3 %c_f32_0 %zero_3\n"
6977 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6978 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6979 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6980 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6981 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6982 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6983 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6984 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6985 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6986 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6987 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6988 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6989 "OpReturnValue %ret\n"
6992 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
6994 return opUndefTests.release();
6997 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
6999 const RGBA inputColors[4] =
7002 RGBA(0, 0, 255, 255),
7003 RGBA(0, 255, 0, 255),
7004 RGBA(0, 255, 255, 255)
7007 const RGBA expectedColors[4] =
7009 RGBA(255, 0, 0, 255),
7010 RGBA(255, 0, 0, 255),
7011 RGBA(255, 0, 0, 255),
7012 RGBA(255, 0, 0, 255)
7015 const struct SingleFP16Possibility
7018 const char* constant; // Value to assign to %test_constant.
7020 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
7026 -constructNormalizedFloat(1, 0x300000),
7027 "%cond = OpFOrdEqual %bool %c %test_constant\n"
7032 constructNormalizedFloat(7, 0x000000),
7033 "%cond = OpFOrdEqual %bool %c %test_constant\n"
7035 // SPIR-V requires that OpQuantizeToF16 flushes
7036 // any numbers that would end up denormalized in F16 to zero.
7040 std::ldexp(1.5f, -140),
7041 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
7046 -std::ldexp(1.5f, -140),
7047 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
7052 std::ldexp(1.0f, -16),
7053 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
7054 }, // too small positive
7056 "negative_too_small",
7058 -std::ldexp(1.0f, -32),
7059 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
7060 }, // too small negative
7064 -std::ldexp(1.0f, 128),
7066 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
7067 "%inf = OpIsInf %bool %c\n"
7068 "%cond = OpLogicalAnd %bool %gz %inf\n"
7073 std::ldexp(1.0f, 128),
7075 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
7076 "%inf = OpIsInf %bool %c\n"
7077 "%cond = OpLogicalAnd %bool %gz %inf\n"
7080 "round_to_negative_inf",
7082 -std::ldexp(1.0f, 32),
7084 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
7085 "%inf = OpIsInf %bool %c\n"
7086 "%cond = OpLogicalAnd %bool %gz %inf\n"
7091 std::ldexp(1.0f, 16),
7093 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
7094 "%inf = OpIsInf %bool %c\n"
7095 "%cond = OpLogicalAnd %bool %gz %inf\n"
7100 std::numeric_limits<float>::quiet_NaN(),
7102 // Test for any NaN value, as NaNs are not preserved
7103 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
7104 "%cond = OpIsNan %bool %direct_quant\n"
7109 std::numeric_limits<float>::quiet_NaN(),
7111 // Test for any NaN value, as NaNs are not preserved
7112 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
7113 "%cond = OpIsNan %bool %direct_quant\n"
7116 const char* constants =
7117 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
7119 StringTemplate function (
7120 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7121 "%param1 = OpFunctionParameter %v4f32\n"
7122 "%label_testfun = OpLabel\n"
7123 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7124 "%b = OpFAdd %f32 %test_constant %a\n"
7125 "%c = OpQuantizeToF16 %f32 %b\n"
7127 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7128 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
7129 " OpReturnValue %retval\n"
7133 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
7134 const char* specConstants =
7135 "%test_constant = OpSpecConstant %f32 0.\n"
7136 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
7138 StringTemplate specConstantFunction(
7139 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7140 "%param1 = OpFunctionParameter %v4f32\n"
7141 "%label_testfun = OpLabel\n"
7143 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7144 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
7145 " OpReturnValue %retval\n"
7149 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
7151 map<string, string> codeSpecialization;
7152 map<string, string> fragments;
7153 codeSpecialization["condition"] = tests[idx].condition;
7154 fragments["testfun"] = function.specialize(codeSpecialization);
7155 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
7156 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7159 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
7161 map<string, string> codeSpecialization;
7162 map<string, string> fragments;
7163 SpecConstants passConstants;
7164 deInt32 specConstant;
7166 codeSpecialization["condition"] = tests[idx].condition;
7167 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
7168 fragments["decoration"] = specDecorations;
7169 fragments["pre_main"] = specConstants;
7171 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
7172 passConstants.append(specConstant);
7174 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7178 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
7180 RGBA inputColors[4] = {
7182 RGBA(0, 0, 255, 255),
7183 RGBA(0, 255, 0, 255),
7184 RGBA(0, 255, 255, 255)
7187 RGBA expectedColors[4] =
7189 RGBA(255, 0, 0, 255),
7190 RGBA(255, 0, 0, 255),
7191 RGBA(255, 0, 0, 255),
7192 RGBA(255, 0, 0, 255)
7195 struct DualFP16Possibility
7200 const char* possibleOutput1;
7201 const char* possibleOutput2;
7204 "positive_round_up_or_round_down",
7206 constructNormalizedFloat(8, 0x300300),
7211 "negative_round_up_or_round_down",
7213 -constructNormalizedFloat(-7, 0x600800),
7220 constructNormalizedFloat(2, 0x01e000),
7225 "carry_to_exponent",
7227 constructNormalizedFloat(1, 0xffe000),
7232 StringTemplate constants (
7233 "%input_const = OpConstant %f32 ${input}\n"
7234 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7235 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7238 StringTemplate specConstants (
7239 "%input_const = OpSpecConstant %f32 0.\n"
7240 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7241 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7244 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
7246 const char* function =
7247 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7248 "%param1 = OpFunctionParameter %v4f32\n"
7249 "%label_testfun = OpLabel\n"
7250 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7251 // For the purposes of this test we assume that 0.f will always get
7252 // faithfully passed through the pipeline stages.
7253 "%b = OpFAdd %f32 %input_const %a\n"
7254 "%c = OpQuantizeToF16 %f32 %b\n"
7255 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
7256 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
7257 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
7258 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7259 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
7260 " OpReturnValue %retval\n"
7263 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7264 map<string, string> fragments;
7265 map<string, string> constantSpecialization;
7267 constantSpecialization["input"] = tests[idx].input;
7268 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7269 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7270 fragments["testfun"] = function;
7271 fragments["pre_main"] = constants.specialize(constantSpecialization);
7272 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7275 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7276 map<string, string> fragments;
7277 map<string, string> constantSpecialization;
7278 SpecConstants passConstants;
7279 deInt32 specConstant;
7281 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7282 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7283 fragments["testfun"] = function;
7284 fragments["decoration"] = specDecorations;
7285 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
7287 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
7288 passConstants.append(specConstant);
7290 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7294 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
7296 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
7297 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
7298 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
7299 return opQuantizeTests.release();
7302 struct ShaderPermutation
7304 deUint8 vertexPermutation;
7305 deUint8 geometryPermutation;
7306 deUint8 tesscPermutation;
7307 deUint8 tessePermutation;
7308 deUint8 fragmentPermutation;
7311 ShaderPermutation getShaderPermutation(deUint8 inputValue)
7313 ShaderPermutation permutation =
7315 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
7316 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
7317 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
7318 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
7319 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
7324 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
7326 RGBA defaultColors[4];
7327 RGBA invertedColors[4];
7328 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
7330 const ShaderElement combinedPipeline[] =
7332 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
7333 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
7334 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7335 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7336 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
7339 getDefaultColors(defaultColors);
7340 getInvertedDefaultColors(invertedColors);
7341 addFunctionCaseWithPrograms<InstanceContext>(
7342 moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline,
7343 createInstanceContext(combinedPipeline, map<string, string>()));
7345 const char* numbers[] =
7350 for (deInt8 idx = 0; idx < 32; ++idx)
7352 ShaderPermutation permutation = getShaderPermutation(idx);
7353 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
7354 const ShaderElement pipeline[] =
7356 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
7357 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
7358 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7359 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7360 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
7363 // If there are an even number of swaps, then it should be no-op.
7364 // If there are an odd number, the color should be flipped.
7365 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
7367 addFunctionCaseWithPrograms<InstanceContext>(
7368 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
7369 createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
7373 addFunctionCaseWithPrograms<InstanceContext>(
7374 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
7375 createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
7378 return moduleTests.release();
7381 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
7383 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
7384 RGBA defaultColors[4];
7385 getDefaultColors(defaultColors);
7386 map<string, string> fragments;
7387 fragments["pre_main"] =
7388 "%c_f32_5 = OpConstant %f32 5.\n";
7390 // A loop with a single block. The Continue Target is the loop block
7391 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
7392 // -- the "continue construct" forms the entire loop.
7393 fragments["testfun"] =
7394 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7395 "%param1 = OpFunctionParameter %v4f32\n"
7397 "%entry = OpLabel\n"
7398 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7401 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7403 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7404 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
7405 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7406 "%val = OpFAdd %f32 %val1 %delta\n"
7407 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
7408 "%count__ = OpISub %i32 %count %c_i32_1\n"
7409 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7410 "OpLoopMerge %exit %loop None\n"
7411 "OpBranchConditional %again %loop %exit\n"
7414 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7415 "OpReturnValue %result\n"
7419 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
7421 // Body comprised of multiple basic blocks.
7422 const StringTemplate multiBlock(
7423 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7424 "%param1 = OpFunctionParameter %v4f32\n"
7426 "%entry = OpLabel\n"
7427 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7430 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7432 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
7433 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
7434 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
7435 // There are several possibilities for the Continue Target below. Each
7436 // will be specialized into a separate test case.
7437 "OpLoopMerge %exit ${continue_target} None\n"
7441 ";delta_next = (delta > 0) ? -1 : 1;\n"
7442 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
7443 "OpSelectionMerge %gather DontFlatten\n"
7444 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
7447 "OpBranch %gather\n"
7450 "OpBranch %gather\n"
7452 "%gather = OpLabel\n"
7453 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
7454 "%val = OpFAdd %f32 %val1 %delta\n"
7455 "%count__ = OpISub %i32 %count %c_i32_1\n"
7456 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7457 "OpBranchConditional %again %loop %exit\n"
7460 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7461 "OpReturnValue %result\n"
7465 map<string, string> continue_target;
7467 // The Continue Target is the loop block itself.
7468 continue_target["continue_target"] = "%loop";
7469 fragments["testfun"] = multiBlock.specialize(continue_target);
7470 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
7472 // The Continue Target is at the end of the loop.
7473 continue_target["continue_target"] = "%gather";
7474 fragments["testfun"] = multiBlock.specialize(continue_target);
7475 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
7477 // A loop with continue statement.
7478 fragments["testfun"] =
7479 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7480 "%param1 = OpFunctionParameter %v4f32\n"
7482 "%entry = OpLabel\n"
7483 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7486 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
7488 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7489 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
7490 "OpLoopMerge %exit %continue None\n"
7494 ";skip if %count==2\n"
7495 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
7496 "OpSelectionMerge %continue DontFlatten\n"
7497 "OpBranchConditional %eq2 %continue %body\n"
7500 "%fcount = OpConvertSToF %f32 %count\n"
7501 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7502 "OpBranch %continue\n"
7504 "%continue = OpLabel\n"
7505 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
7506 "%count__ = OpISub %i32 %count %c_i32_1\n"
7507 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7508 "OpBranchConditional %again %loop %exit\n"
7511 "%same = OpFSub %f32 %val %c_f32_8\n"
7512 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7513 "OpReturnValue %result\n"
7515 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
7517 // A loop with break.
7518 fragments["testfun"] =
7519 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7520 "%param1 = OpFunctionParameter %v4f32\n"
7522 "%entry = OpLabel\n"
7523 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7524 "%dot = OpDot %f32 %param1 %param1\n"
7525 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7526 "%zero = OpConvertFToU %u32 %div\n"
7527 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7528 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7531 ";adds 4 and 3 to %val0 (exits early)\n"
7533 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7534 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7535 "OpLoopMerge %exit %continue None\n"
7539 ";end loop if %count==%two\n"
7540 "%above2 = OpSGreaterThan %bool %count %two\n"
7541 "OpSelectionMerge %continue DontFlatten\n"
7542 "OpBranchConditional %above2 %body %exit\n"
7545 "%fcount = OpConvertSToF %f32 %count\n"
7546 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7547 "OpBranch %continue\n"
7549 "%continue = OpLabel\n"
7550 "%count__ = OpISub %i32 %count %c_i32_1\n"
7551 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7552 "OpBranchConditional %again %loop %exit\n"
7555 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
7556 "%same = OpFSub %f32 %val_post %c_f32_7\n"
7557 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7558 "OpReturnValue %result\n"
7560 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
7562 // A loop with return.
7563 fragments["testfun"] =
7564 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7565 "%param1 = OpFunctionParameter %v4f32\n"
7567 "%entry = OpLabel\n"
7568 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7569 "%dot = OpDot %f32 %param1 %param1\n"
7570 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7571 "%zero = OpConvertFToU %u32 %div\n"
7572 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7573 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7576 ";returns early without modifying %param1\n"
7578 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7579 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7580 "OpLoopMerge %exit %continue None\n"
7584 ";return if %count==%two\n"
7585 "%above2 = OpSGreaterThan %bool %count %two\n"
7586 "OpSelectionMerge %continue DontFlatten\n"
7587 "OpBranchConditional %above2 %body %early_exit\n"
7589 "%early_exit = OpLabel\n"
7590 "OpReturnValue %param1\n"
7593 "%fcount = OpConvertSToF %f32 %count\n"
7594 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7595 "OpBranch %continue\n"
7597 "%continue = OpLabel\n"
7598 "%count__ = OpISub %i32 %count %c_i32_1\n"
7599 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7600 "OpBranchConditional %again %loop %exit\n"
7603 ";should never get here, so return an incorrect result\n"
7604 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
7605 "OpReturnValue %result\n"
7607 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
7609 // Continue inside a switch block to break to enclosing loop's merge block.
7610 // Matches roughly the following GLSL code:
7611 // for (; keep_going; keep_going = false)
7613 // switch (int(param1.x))
7615 // case 0: continue;
7616 // case 1: continue;
7617 // default: continue;
7619 // dead code: modify return value to invalid result.
7621 fragments["pre_main"] =
7622 "%fp_bool = OpTypePointer Function %bool\n"
7623 "%true = OpConstantTrue %bool\n"
7624 "%false = OpConstantFalse %bool\n";
7626 fragments["testfun"] =
7627 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7628 "%param1 = OpFunctionParameter %v4f32\n"
7630 "%entry = OpLabel\n"
7631 "%keep_going = OpVariable %fp_bool Function\n"
7632 "%val_ptr = OpVariable %fp_f32 Function\n"
7633 "%param1_x = OpCompositeExtract %f32 %param1 0\n"
7634 "OpStore %keep_going %true\n"
7635 "OpBranch %forloop_begin\n"
7637 "%forloop_begin = OpLabel\n"
7638 "OpLoopMerge %forloop_merge %forloop_continue None\n"
7639 "OpBranch %forloop\n"
7641 "%forloop = OpLabel\n"
7642 "%for_condition = OpLoad %bool %keep_going\n"
7643 "OpBranchConditional %for_condition %forloop_body %forloop_merge\n"
7645 "%forloop_body = OpLabel\n"
7646 "OpStore %val_ptr %param1_x\n"
7647 "%param1_x_int = OpConvertFToS %i32 %param1_x\n"
7649 "OpSelectionMerge %switch_merge None\n"
7650 "OpSwitch %param1_x_int %default 0 %case_0 1 %case_1\n"
7651 "%case_0 = OpLabel\n"
7652 "OpBranch %forloop_continue\n"
7653 "%case_1 = OpLabel\n"
7654 "OpBranch %forloop_continue\n"
7655 "%default = OpLabel\n"
7656 "OpBranch %forloop_continue\n"
7657 "%switch_merge = OpLabel\n"
7658 ";should never get here, so change the return value to invalid result\n"
7659 "OpStore %val_ptr %c_f32_1\n"
7660 "OpBranch %forloop_continue\n"
7662 "%forloop_continue = OpLabel\n"
7663 "OpStore %keep_going %false\n"
7664 "OpBranch %forloop_begin\n"
7665 "%forloop_merge = OpLabel\n"
7667 "%val = OpLoad %f32 %val_ptr\n"
7668 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7669 "OpReturnValue %result\n"
7671 createTestsForAllStages("switch_continue", defaultColors, defaultColors, fragments, testGroup.get());
7673 return testGroup.release();
7676 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
7677 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
7679 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
7680 map<string, string> fragments;
7682 // A barrier inside a function body.
7683 fragments["pre_main"] =
7684 "%Workgroup = OpConstant %i32 2\n"
7685 "%WorkgroupAcquireRelease = OpConstant %i32 0x108\n";
7686 fragments["testfun"] =
7687 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7688 "%param1 = OpFunctionParameter %v4f32\n"
7689 "%label_testfun = OpLabel\n"
7690 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7691 "OpReturnValue %param1\n"
7693 addTessCtrlTest(testGroup.get(), "in_function", fragments);
7695 // Common setup code for the following tests.
7696 fragments["pre_main"] =
7697 "%Workgroup = OpConstant %i32 2\n"
7698 "%WorkgroupAcquireRelease = OpConstant %i32 0x108\n"
7699 "%c_f32_5 = OpConstant %f32 5.\n";
7700 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
7701 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7702 "%param1 = OpFunctionParameter %v4f32\n"
7703 "%entry = OpLabel\n"
7704 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7705 "%dot = OpDot %f32 %param1 %param1\n"
7706 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7707 "%zero = OpConvertFToU %u32 %div\n";
7709 // Barriers inside OpSwitch branches.
7710 fragments["testfun"] =
7712 "OpSelectionMerge %switch_exit None\n"
7713 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
7715 "%case1 = OpLabel\n"
7716 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7717 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7718 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7719 "OpBranch %switch_exit\n"
7721 "%switch_default = OpLabel\n"
7722 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7723 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7724 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7725 "OpBranch %switch_exit\n"
7727 "%case0 = OpLabel\n"
7728 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7729 "OpBranch %switch_exit\n"
7731 "%switch_exit = OpLabel\n"
7732 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
7733 "OpReturnValue %ret\n"
7735 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
7737 // Barriers inside if-then-else.
7738 fragments["testfun"] =
7740 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
7741 "OpSelectionMerge %exit DontFlatten\n"
7742 "OpBranchConditional %eq0 %then %else\n"
7745 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7746 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7747 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7751 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7755 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
7756 "OpReturnValue %ret\n"
7758 addTessCtrlTest(testGroup.get(), "in_if", fragments);
7760 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
7761 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
7762 fragments["testfun"] =
7764 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
7765 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
7766 "OpSelectionMerge %exit DontFlatten\n"
7767 "OpBranchConditional %thread0 %then %else\n"
7770 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7774 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
7778 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
7779 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7780 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
7781 "OpReturnValue %ret\n"
7783 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
7785 // A barrier inside a loop.
7786 fragments["pre_main"] =
7787 "%Workgroup = OpConstant %i32 2\n"
7788 "%WorkgroupAcquireRelease = OpConstant %i32 0x108\n"
7789 "%c_f32_10 = OpConstant %f32 10.\n";
7790 fragments["testfun"] =
7791 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7792 "%param1 = OpFunctionParameter %v4f32\n"
7793 "%entry = OpLabel\n"
7794 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7797 ";adds 4, 3, 2, and 1 to %val0\n"
7799 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7800 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7801 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7802 "%fcount = OpConvertSToF %f32 %count\n"
7803 "%val = OpFAdd %f32 %val1 %fcount\n"
7804 "%count__ = OpISub %i32 %count %c_i32_1\n"
7805 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7806 "OpLoopMerge %exit %loop None\n"
7807 "OpBranchConditional %again %loop %exit\n"
7810 "%same = OpFSub %f32 %val %c_f32_10\n"
7811 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7812 "OpReturnValue %ret\n"
7814 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
7816 return testGroup.release();
7819 // Test for the OpFRem instruction.
7820 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
7822 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
7823 map<string, string> fragments;
7824 RGBA inputColors[4];
7825 RGBA outputColors[4];
7827 fragments["pre_main"] =
7828 "%c_f32_3 = OpConstant %f32 3.0\n"
7829 "%c_f32_n3 = OpConstant %f32 -3.0\n"
7830 "%c_f32_4 = OpConstant %f32 4.0\n"
7831 "%c_f32_p75 = OpConstant %f32 0.75\n"
7832 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
7833 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
7834 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
7836 // The test does the following.
7837 // vec4 result = (param1 * 8.0) - 4.0;
7838 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
7839 fragments["testfun"] =
7840 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7841 "%param1 = OpFunctionParameter %v4f32\n"
7842 "%label_testfun = OpLabel\n"
7843 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
7844 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
7845 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
7846 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
7847 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
7848 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
7849 "OpReturnValue %xy_0_1\n"
7853 inputColors[0] = RGBA(16, 16, 0, 255);
7854 inputColors[1] = RGBA(232, 232, 0, 255);
7855 inputColors[2] = RGBA(232, 16, 0, 255);
7856 inputColors[3] = RGBA(16, 232, 0, 255);
7858 outputColors[0] = RGBA(64, 64, 0, 255);
7859 outputColors[1] = RGBA(255, 255, 0, 255);
7860 outputColors[2] = RGBA(255, 64, 0, 255);
7861 outputColors[3] = RGBA(64, 255, 0, 255);
7863 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
7864 return testGroup.release();
7867 // Test for the OpSRem instruction.
7868 tcu::TestCaseGroup* createOpSRemGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
7870 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "srem", "OpSRem"));
7871 map<string, string> fragments;
7873 fragments["pre_main"] =
7874 "%c_f32_255 = OpConstant %f32 255.0\n"
7875 "%c_i32_128 = OpConstant %i32 128\n"
7876 "%c_i32_255 = OpConstant %i32 255\n"
7877 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
7878 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
7879 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
7881 // The test does the following.
7882 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
7883 // ivec4 result = ivec4(srem(ints.x, ints.y), srem(ints.y, ints.z), srem(ints.z, ints.x), 255);
7884 // return float(result + 128) / 255.0;
7885 fragments["testfun"] =
7886 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7887 "%param1 = OpFunctionParameter %v4f32\n"
7888 "%label_testfun = OpLabel\n"
7889 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
7890 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
7891 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
7892 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
7893 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
7894 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
7895 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
7896 "%x_out = OpSRem %i32 %x_in %y_in\n"
7897 "%y_out = OpSRem %i32 %y_in %z_in\n"
7898 "%z_out = OpSRem %i32 %z_in %x_in\n"
7899 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
7900 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
7901 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
7902 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
7903 "OpReturnValue %float_out\n"
7906 const struct CaseParams
7909 const char* failMessageTemplate; // customized status message
7910 qpTestResult failResult; // override status on failure
7911 int operands[4][3]; // four (x, y, z) vectors of operands
7912 int results[4][3]; // four (x, y, z) vectors of results
7918 QP_TEST_RESULT_FAIL,
7919 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
7920 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
7924 "Inconsistent results, but within specification: ${reason}",
7925 negFailResult, // negative operands, not required by the spec
7926 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
7927 { { 5, 12, -2 }, { 0, -5, 2 }, { 3, 8, -6 }, { 25, -60, 0 } }, // results
7930 // If either operand is negative the result is undefined. Some implementations may still return correct values.
7932 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
7934 const CaseParams& params = cases[caseNdx];
7935 RGBA inputColors[4];
7936 RGBA outputColors[4];
7938 for (int i = 0; i < 4; ++i)
7940 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
7941 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
7944 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
7947 return testGroup.release();
7950 // Test for the OpSMod instruction.
7951 tcu::TestCaseGroup* createOpSModGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
7953 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "smod", "OpSMod"));
7954 map<string, string> fragments;
7956 fragments["pre_main"] =
7957 "%c_f32_255 = OpConstant %f32 255.0\n"
7958 "%c_i32_128 = OpConstant %i32 128\n"
7959 "%c_i32_255 = OpConstant %i32 255\n"
7960 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
7961 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
7962 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
7964 // The test does the following.
7965 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
7966 // ivec4 result = ivec4(smod(ints.x, ints.y), smod(ints.y, ints.z), smod(ints.z, ints.x), 255);
7967 // return float(result + 128) / 255.0;
7968 fragments["testfun"] =
7969 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7970 "%param1 = OpFunctionParameter %v4f32\n"
7971 "%label_testfun = OpLabel\n"
7972 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
7973 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
7974 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
7975 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
7976 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
7977 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
7978 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
7979 "%x_out = OpSMod %i32 %x_in %y_in\n"
7980 "%y_out = OpSMod %i32 %y_in %z_in\n"
7981 "%z_out = OpSMod %i32 %z_in %x_in\n"
7982 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
7983 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
7984 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
7985 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
7986 "OpReturnValue %float_out\n"
7989 const struct CaseParams
7992 const char* failMessageTemplate; // customized status message
7993 qpTestResult failResult; // override status on failure
7994 int operands[4][3]; // four (x, y, z) vectors of operands
7995 int results[4][3]; // four (x, y, z) vectors of results
8001 QP_TEST_RESULT_FAIL,
8002 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
8003 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
8007 "Inconsistent results, but within specification: ${reason}",
8008 negFailResult, // negative operands, not required by the spec
8009 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
8010 { { 5, -5, 3 }, { 0, 2, -3 }, { 3, -73, 69 }, { -35, 40, 0 } }, // results
8013 // If either operand is negative the result is undefined. Some implementations may still return correct values.
8015 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
8017 const CaseParams& params = cases[caseNdx];
8018 RGBA inputColors[4];
8019 RGBA outputColors[4];
8021 for (int i = 0; i < 4; ++i)
8023 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
8024 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
8027 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
8029 return testGroup.release();
8032 enum ConversionDataType
8034 DATA_TYPE_SIGNED_16,
8035 DATA_TYPE_SIGNED_32,
8036 DATA_TYPE_SIGNED_64,
8037 DATA_TYPE_UNSIGNED_16,
8038 DATA_TYPE_UNSIGNED_32,
8039 DATA_TYPE_UNSIGNED_64,
8042 DATA_TYPE_VEC2_SIGNED_16,
8043 DATA_TYPE_VEC2_SIGNED_32
8046 const string getBitWidthStr (ConversionDataType type)
8050 case DATA_TYPE_SIGNED_16:
8051 case DATA_TYPE_UNSIGNED_16:
8054 case DATA_TYPE_SIGNED_32:
8055 case DATA_TYPE_UNSIGNED_32:
8056 case DATA_TYPE_FLOAT_32:
8057 case DATA_TYPE_VEC2_SIGNED_16:
8060 case DATA_TYPE_SIGNED_64:
8061 case DATA_TYPE_UNSIGNED_64:
8062 case DATA_TYPE_FLOAT_64:
8063 case DATA_TYPE_VEC2_SIGNED_32:
8072 const string getByteWidthStr (ConversionDataType type)
8076 case DATA_TYPE_SIGNED_16:
8077 case DATA_TYPE_UNSIGNED_16:
8080 case DATA_TYPE_SIGNED_32:
8081 case DATA_TYPE_UNSIGNED_32:
8082 case DATA_TYPE_FLOAT_32:
8083 case DATA_TYPE_VEC2_SIGNED_16:
8086 case DATA_TYPE_SIGNED_64:
8087 case DATA_TYPE_UNSIGNED_64:
8088 case DATA_TYPE_FLOAT_64:
8089 case DATA_TYPE_VEC2_SIGNED_32:
8098 bool isSigned (ConversionDataType type)
8102 case DATA_TYPE_SIGNED_16:
8103 case DATA_TYPE_SIGNED_32:
8104 case DATA_TYPE_SIGNED_64:
8105 case DATA_TYPE_FLOAT_32:
8106 case DATA_TYPE_FLOAT_64:
8107 case DATA_TYPE_VEC2_SIGNED_16:
8108 case DATA_TYPE_VEC2_SIGNED_32:
8111 case DATA_TYPE_UNSIGNED_16:
8112 case DATA_TYPE_UNSIGNED_32:
8113 case DATA_TYPE_UNSIGNED_64:
8122 bool isInt (ConversionDataType type)
8126 case DATA_TYPE_SIGNED_16:
8127 case DATA_TYPE_SIGNED_32:
8128 case DATA_TYPE_SIGNED_64:
8129 case DATA_TYPE_UNSIGNED_16:
8130 case DATA_TYPE_UNSIGNED_32:
8131 case DATA_TYPE_UNSIGNED_64:
8134 case DATA_TYPE_FLOAT_32:
8135 case DATA_TYPE_FLOAT_64:
8136 case DATA_TYPE_VEC2_SIGNED_16:
8137 case DATA_TYPE_VEC2_SIGNED_32:
8146 bool isFloat (ConversionDataType type)
8150 case DATA_TYPE_SIGNED_16:
8151 case DATA_TYPE_SIGNED_32:
8152 case DATA_TYPE_SIGNED_64:
8153 case DATA_TYPE_UNSIGNED_16:
8154 case DATA_TYPE_UNSIGNED_32:
8155 case DATA_TYPE_UNSIGNED_64:
8156 case DATA_TYPE_VEC2_SIGNED_16:
8157 case DATA_TYPE_VEC2_SIGNED_32:
8160 case DATA_TYPE_FLOAT_32:
8161 case DATA_TYPE_FLOAT_64:
8170 const string getTypeName (ConversionDataType type)
8172 string prefix = isSigned(type) ? "" : "u";
8174 if (isInt(type)) return prefix + "int" + getBitWidthStr(type);
8175 else if (isFloat(type)) return prefix + "float" + getBitWidthStr(type);
8176 else if (type == DATA_TYPE_VEC2_SIGNED_16) return "i16vec2";
8177 else if (type == DATA_TYPE_VEC2_SIGNED_32) return "i32vec2";
8178 else DE_ASSERT(false);
8183 const string getTestName (ConversionDataType from, ConversionDataType to)
8185 return getTypeName(from) + "_to_" + getTypeName(to);
8188 const string getAsmTypeName (ConversionDataType type)
8192 if (isInt(type)) prefix = isSigned(type) ? "i" : "u";
8193 else if (isFloat(type)) prefix = "f";
8194 else if (type == DATA_TYPE_VEC2_SIGNED_16) return "i16vec2";
8195 else if (type == DATA_TYPE_VEC2_SIGNED_32) return "v2i32";
8196 else DE_ASSERT(false);
8198 return prefix + getBitWidthStr(type);
8201 template<typename T>
8202 BufferSp getSpecializedBuffer (deInt64 number)
8204 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
8207 BufferSp getBuffer (ConversionDataType type, deInt64 number)
8211 case DATA_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
8212 case DATA_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
8213 case DATA_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
8214 case DATA_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
8215 case DATA_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
8216 case DATA_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
8217 case DATA_TYPE_FLOAT_32: return getSpecializedBuffer<deUint32>(number);
8218 case DATA_TYPE_FLOAT_64: return getSpecializedBuffer<deUint64>(number);
8219 case DATA_TYPE_VEC2_SIGNED_16: return getSpecializedBuffer<deUint32>(number);
8220 case DATA_TYPE_VEC2_SIGNED_32: return getSpecializedBuffer<deUint64>(number);
8222 default: DE_ASSERT(false);
8223 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
8227 bool usesInt16 (ConversionDataType from, ConversionDataType to)
8229 return (from == DATA_TYPE_SIGNED_16 || from == DATA_TYPE_UNSIGNED_16
8230 || to == DATA_TYPE_SIGNED_16 || to == DATA_TYPE_UNSIGNED_16
8231 || from == DATA_TYPE_VEC2_SIGNED_16 || to == DATA_TYPE_VEC2_SIGNED_16);
8234 bool usesInt32 (ConversionDataType from, ConversionDataType to)
8236 return (from == DATA_TYPE_SIGNED_32 || from == DATA_TYPE_UNSIGNED_32
8237 || to == DATA_TYPE_SIGNED_32 || to == DATA_TYPE_UNSIGNED_32
8238 || from == DATA_TYPE_VEC2_SIGNED_32 || to == DATA_TYPE_VEC2_SIGNED_32);
8241 bool usesInt64 (ConversionDataType from, ConversionDataType to)
8243 return (from == DATA_TYPE_SIGNED_64 || from == DATA_TYPE_UNSIGNED_64
8244 || to == DATA_TYPE_SIGNED_64 || to == DATA_TYPE_UNSIGNED_64);
8247 bool usesFloat64 (ConversionDataType from, ConversionDataType to)
8249 return (from == DATA_TYPE_FLOAT_64 || to == DATA_TYPE_FLOAT_64);
8253 ComputeTestFeatures getConversionUsedFeatures (ConversionDataType from, ConversionDataType to)
8255 if (usesInt16(from, to) && usesInt64(from, to)) return COMPUTE_TEST_USES_INT16_INT64;
8256 else if (usesInt16(from, to) && usesInt32(from, to)) return COMPUTE_TEST_USES_NONE;
8257 else if (usesInt16(from, to)) return COMPUTE_TEST_USES_INT16; // This is not set for int16<-->int32 only conversions
8258 else if (usesInt64(from, to)) return COMPUTE_TEST_USES_INT64;
8259 else if (usesFloat64(from, to)) return COMPUTE_TEST_USES_FLOAT64;
8260 else return COMPUTE_TEST_USES_NONE;
8263 vector<string> getFeatureStringVector (ComputeTestFeatures computeTestFeatures)
8265 vector<string> features;
8266 if (computeTestFeatures == COMPUTE_TEST_USES_INT16_INT64)
8268 features.push_back("shaderInt16");
8269 features.push_back("shaderInt64");
8271 else if (computeTestFeatures == COMPUTE_TEST_USES_INT16) features.push_back("shaderInt16");
8272 else if (computeTestFeatures == COMPUTE_TEST_USES_INT64) features.push_back("shaderInt64");
8273 else if (computeTestFeatures == COMPUTE_TEST_USES_FLOAT64) features.push_back("shaderFloat64");
8274 else if (computeTestFeatures == COMPUTE_TEST_USES_NONE) {}
8275 else DE_ASSERT(false);
8282 ConvertCase (ConversionDataType from, ConversionDataType to, deInt64 number, bool separateOutput = false, deInt64 outputNumber = 0)
8285 , m_features (getConversionUsedFeatures(from, to))
8286 , m_name (getTestName(from, to))
8287 , m_inputBuffer (getBuffer(from, number))
8289 m_asmTypes["inputType"] = getAsmTypeName(from);
8290 m_asmTypes["outputType"] = getAsmTypeName(to);
8293 m_outputBuffer = getBuffer(to, outputNumber);
8295 m_outputBuffer = getBuffer(to, number);
8297 if (m_features == COMPUTE_TEST_USES_INT16)
8299 m_asmTypes["datatype_capabilities"] = "OpCapability Int16\n"
8300 "OpCapability StorageUniformBufferBlock16\n"
8301 "OpCapability StorageUniform16\n";
8302 m_asmTypes["datatype_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
8303 "%u16 = OpTypeInt 16 0\n"
8304 "%i16vec2 = OpTypeVector %i16 2\n";
8305 m_asmTypes["datatype_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
8307 else if (m_features == COMPUTE_TEST_USES_INT64)
8309 m_asmTypes["datatype_capabilities"] = "OpCapability Int64\n";
8310 m_asmTypes["datatype_additional_decl"] = "%i64 = OpTypeInt 64 1\n"
8311 "%u64 = OpTypeInt 64 0\n";
8312 m_asmTypes["datatype_extensions"] = "";
8314 else if (m_features == COMPUTE_TEST_USES_INT16_INT64)
8316 m_asmTypes["datatype_capabilities"] = "OpCapability Int16\n"
8317 "OpCapability StorageUniformBufferBlock16\n"
8318 "OpCapability StorageUniform16\n"
8319 "OpCapability Int64\n";
8320 m_asmTypes["datatype_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
8321 "%u16 = OpTypeInt 16 0\n"
8322 "%i64 = OpTypeInt 64 1\n"
8323 "%u64 = OpTypeInt 64 0\n";
8324 m_asmTypes["datatype_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
8326 else if (m_features == COMPUTE_TEST_USES_FLOAT64)
8328 m_asmTypes["datatype_capabilities"] = "OpCapability Float64\n";
8329 m_asmTypes["datatype_additional_decl"] = "%f64 = OpTypeFloat 64\n";
8331 else if (usesInt16(from, to) && usesInt32(from, to))
8333 m_asmTypes["datatype_capabilities"] = "OpCapability StorageUniformBufferBlock16\n"
8334 "OpCapability StorageUniform16\n";
8335 m_asmTypes["datatype_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
8336 "%u16 = OpTypeInt 16 0\n"
8337 "%i16vec2 = OpTypeVector %i16 2\n";
8338 m_asmTypes["datatype_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
8346 ConversionDataType m_fromType;
8347 ConversionDataType m_toType;
8348 ComputeTestFeatures m_features;
8350 map<string, string> m_asmTypes;
8351 BufferSp m_inputBuffer;
8352 BufferSp m_outputBuffer;
8355 const string getConvertCaseShaderStr (const string& instruction, const ConvertCase& convertCase)
8357 map<string, string> params = convertCase.m_asmTypes;
8359 params["instruction"] = instruction;
8360 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
8361 params["outDecorator"] = getByteWidthStr(convertCase.m_toType);
8363 const StringTemplate shader (
8364 "OpCapability Shader\n"
8365 "${datatype_capabilities}"
8366 "${datatype_extensions:opt}"
8367 "OpMemoryModel Logical GLSL450\n"
8368 "OpEntryPoint GLCompute %main \"main\"\n"
8369 "OpExecutionMode %main LocalSize 1 1 1\n"
8370 "OpSource GLSL 430\n"
8371 "OpName %main \"main\"\n"
8373 "OpDecorate %indata DescriptorSet 0\n"
8374 "OpDecorate %indata Binding 0\n"
8375 "OpDecorate %outdata DescriptorSet 0\n"
8376 "OpDecorate %outdata Binding 1\n"
8377 "OpDecorate %in_buf BufferBlock\n"
8378 "OpDecorate %out_buf BufferBlock\n"
8379 "OpMemberDecorate %in_buf 0 Offset 0\n"
8380 "OpMemberDecorate %out_buf 0 Offset 0\n"
8382 "%void = OpTypeVoid\n"
8383 "%voidf = OpTypeFunction %void\n"
8384 "%u32 = OpTypeInt 32 0\n"
8385 "%i32 = OpTypeInt 32 1\n"
8386 "%f32 = OpTypeFloat 32\n"
8387 "%v2i32 = OpTypeVector %i32 2\n"
8388 "${datatype_additional_decl}"
8389 "%uvec3 = OpTypeVector %u32 3\n"
8391 "%in_ptr = OpTypePointer Uniform %${inputType}\n"
8392 "%out_ptr = OpTypePointer Uniform %${outputType}\n"
8393 "%in_buf = OpTypeStruct %${inputType}\n"
8394 "%out_buf = OpTypeStruct %${outputType}\n"
8395 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
8396 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
8397 "%indata = OpVariable %in_bufptr Uniform\n"
8398 "%outdata = OpVariable %out_bufptr Uniform\n"
8400 "%zero = OpConstant %i32 0\n"
8402 "%main = OpFunction %void None %voidf\n"
8403 "%label = OpLabel\n"
8404 "%inloc = OpAccessChain %in_ptr %indata %zero\n"
8405 "%outloc = OpAccessChain %out_ptr %outdata %zero\n"
8406 "%inval = OpLoad %${inputType} %inloc\n"
8407 "%conv = ${instruction} %${outputType} %inval\n"
8408 " OpStore %outloc %conv\n"
8413 return shader.specialize(params);
8416 void createConvertCases (vector<ConvertCase>& testCases, const string& instruction)
8418 if (instruction == "OpUConvert")
8420 // Convert unsigned int to unsigned int
8421 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_16, DATA_TYPE_UNSIGNED_32, 60653));
8422 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_16, DATA_TYPE_UNSIGNED_64, 17991));
8423 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_32, DATA_TYPE_UNSIGNED_64, 904256275));
8424 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_32, DATA_TYPE_UNSIGNED_16, 6275));
8425 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_64, DATA_TYPE_UNSIGNED_32, 701256243));
8426 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_64, DATA_TYPE_UNSIGNED_16, 4741));
8428 else if (instruction == "OpSConvert")
8430 // Sign extension int->int
8431 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_16, DATA_TYPE_SIGNED_32, 14669));
8432 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_16, DATA_TYPE_SIGNED_64, -3341));
8433 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_32, DATA_TYPE_SIGNED_64, 973610259));
8435 // Truncate for int->int
8436 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_32, DATA_TYPE_SIGNED_16, 12382));
8437 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_64, DATA_TYPE_SIGNED_32, -972812359));
8438 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_64, DATA_TYPE_SIGNED_16, -1067742499291926803ll, true, -4371));
8440 // Sign extension for int->uint
8441 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_16, DATA_TYPE_UNSIGNED_32, 14669));
8442 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_16, DATA_TYPE_UNSIGNED_64, -3341, true, 18446744073709548275ull));
8443 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_32, DATA_TYPE_UNSIGNED_64, 973610259));
8445 // Truncate for int->uint
8446 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_32, DATA_TYPE_UNSIGNED_16, 12382));
8447 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_64, DATA_TYPE_UNSIGNED_32, -972812359, true, 3322154937u));
8448 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_64, DATA_TYPE_UNSIGNED_16, -1067742499291926803ll, true, 61165));
8450 // Sign extension for uint->int
8451 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_16, DATA_TYPE_SIGNED_32, 14669));
8452 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_16, DATA_TYPE_SIGNED_64, 62195, true, -3341));
8453 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_32, DATA_TYPE_SIGNED_64, 973610259));
8455 // Truncate for uint->int
8456 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_32, DATA_TYPE_SIGNED_16, 12382));
8457 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_64, DATA_TYPE_SIGNED_32, 18446744072736739257ull, true, -972812359));
8458 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_64, DATA_TYPE_SIGNED_16, 17379001574417624813ull, true, -4371));
8460 // Convert i16vec2 to i32vec2 and vice versa
8461 // Unsigned values are used here to represent negative signed values and to allow defined shifting behaviour.
8462 // The actual signed value -32123 is used here as uint16 value 33413 and uint32 value 4294935173
8463 testCases.push_back(ConvertCase(DATA_TYPE_VEC2_SIGNED_16, DATA_TYPE_VEC2_SIGNED_32, (33413u << 16) | 27593, true, (4294935173ull << 32) | 27593));
8464 testCases.push_back(ConvertCase(DATA_TYPE_VEC2_SIGNED_32, DATA_TYPE_VEC2_SIGNED_16, (4294935173ull << 32) | 27593, true, (33413u << 16) | 27593));
8466 else if (instruction == "OpFConvert")
8468 // All hexadecimal values below represent 1024.0 as 32/64-bit IEEE 754 float
8469 testCases.push_back(ConvertCase(DATA_TYPE_FLOAT_32, DATA_TYPE_FLOAT_64, 0x449a4000, true, 0x4093480000000000));
8470 testCases.push_back(ConvertCase(DATA_TYPE_FLOAT_64, DATA_TYPE_FLOAT_32, 0x4093480000000000, true, 0x449a4000));
8473 DE_FATAL("Unknown instruction");
8476 const map<string, string> getConvertCaseFragments (string instruction, const ConvertCase& convertCase)
8478 map<string, string> params = convertCase.m_asmTypes;
8479 map<string, string> fragments;
8481 params["instruction"] = instruction;
8482 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
8484 const StringTemplate decoration (
8485 " OpDecorate %SSBOi DescriptorSet 0\n"
8486 " OpDecorate %SSBOo DescriptorSet 0\n"
8487 " OpDecorate %SSBOi Binding 0\n"
8488 " OpDecorate %SSBOo Binding 1\n"
8489 " OpDecorate %s_SSBOi Block\n"
8490 " OpDecorate %s_SSBOo Block\n"
8491 "OpMemberDecorate %s_SSBOi 0 Offset 0\n"
8492 "OpMemberDecorate %s_SSBOo 0 Offset 0\n");
8494 const StringTemplate pre_main (
8495 "${datatype_additional_decl:opt}"
8496 " %ptr_in = OpTypePointer StorageBuffer %${inputType}\n"
8497 " %ptr_out = OpTypePointer StorageBuffer %${outputType}\n"
8498 " %s_SSBOi = OpTypeStruct %${inputType}\n"
8499 " %s_SSBOo = OpTypeStruct %${outputType}\n"
8500 " %ptr_SSBOi = OpTypePointer StorageBuffer %s_SSBOi\n"
8501 " %ptr_SSBOo = OpTypePointer StorageBuffer %s_SSBOo\n"
8502 " %SSBOi = OpVariable %ptr_SSBOi StorageBuffer\n"
8503 " %SSBOo = OpVariable %ptr_SSBOo StorageBuffer\n");
8505 const StringTemplate testfun (
8506 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
8507 "%param = OpFunctionParameter %v4f32\n"
8508 "%label = OpLabel\n"
8509 "%iLoc = OpAccessChain %ptr_in %SSBOi %c_u32_0\n"
8510 "%oLoc = OpAccessChain %ptr_out %SSBOo %c_u32_0\n"
8511 "%valIn = OpLoad %${inputType} %iLoc\n"
8512 "%valOut = ${instruction} %${outputType} %valIn\n"
8513 " OpStore %oLoc %valOut\n"
8514 " OpReturnValue %param\n"
8515 " OpFunctionEnd\n");
8517 params["datatype_extensions"] =
8518 params["datatype_extensions"] +
8519 "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n";
8521 fragments["capability"] = params["datatype_capabilities"];
8522 fragments["extension"] = params["datatype_extensions"];
8523 fragments["decoration"] = decoration.specialize(params);
8524 fragments["pre_main"] = pre_main.specialize(params);
8525 fragments["testfun"] = testfun.specialize(params);
8530 // Test for OpSConvert, OpUConvert and OpFConvert in compute shaders
8531 tcu::TestCaseGroup* createConvertComputeTests (tcu::TestContext& testCtx, const string& instruction, const string& name)
8533 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, name.c_str(), instruction.c_str()));
8534 vector<ConvertCase> testCases;
8535 createConvertCases(testCases, instruction);
8537 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8539 ComputeShaderSpec spec;
8540 spec.assembly = getConvertCaseShaderStr(instruction, *test);
8541 spec.numWorkGroups = IVec3(1, 1, 1);
8542 spec.inputs.push_back (test->m_inputBuffer);
8543 spec.outputs.push_back (test->m_outputBuffer);
8545 if (test->m_features == COMPUTE_TEST_USES_INT16 || test->m_features == COMPUTE_TEST_USES_INT16_INT64 || usesInt16(test->m_fromType, test->m_toType)) {
8546 spec.extensions.push_back("VK_KHR_16bit_storage");
8547 spec.requestedVulkanFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_UNIFORM_BUFFER_BLOCK;
8550 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "", spec, test->m_features));
8552 return group.release();
8555 // Test for OpSConvert, OpUConvert and OpFConvert in graphics shaders
8556 tcu::TestCaseGroup* createConvertGraphicsTests (tcu::TestContext& testCtx, const string& instruction, const string& name)
8558 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, name.c_str(), instruction.c_str()));
8559 vector<ConvertCase> testCases;
8560 createConvertCases(testCases, instruction);
8562 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8564 map<string, string> fragments = getConvertCaseFragments(instruction, *test);
8565 vector<string> features = getFeatureStringVector(test->m_features);
8566 GraphicsResources resources;
8567 vector<string> extensions;
8568 SpecConstants noSpecConstants;
8569 PushConstants noPushConstants;
8570 VulkanFeatures vulkanFeatures;
8571 GraphicsInterfaces noInterfaces;
8572 tcu::RGBA defaultColors[4];
8574 getDefaultColors (defaultColors);
8575 resources.inputs.push_back (Resource(test->m_inputBuffer, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER));
8576 resources.outputs.push_back (Resource(test->m_outputBuffer, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER));
8577 extensions.push_back ("VK_KHR_storage_buffer_storage_class");
8579 if (test->m_features == COMPUTE_TEST_USES_INT16 || test->m_features == COMPUTE_TEST_USES_INT16_INT64 || usesInt16(test->m_fromType, test->m_toType))
8581 extensions.push_back("VK_KHR_16bit_storage");
8582 vulkanFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_UNIFORM_BUFFER_BLOCK;
8585 createTestsForAllStages(
8586 test->m_name, defaultColors, defaultColors, fragments, noSpecConstants,
8587 noPushConstants, resources, noInterfaces, extensions, features, vulkanFeatures, group.get());
8589 return group.release();
8592 const string getNumberTypeName (const NumberType type)
8594 if (type == NUMBERTYPE_INT32)
8598 else if (type == NUMBERTYPE_UINT32)
8602 else if (type == NUMBERTYPE_FLOAT32)
8613 deInt32 getInt(de::Random& rnd)
8615 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
8618 const string repeatString (const string& str, int times)
8621 for (int i = 0; i < times; ++i)
8628 const string getRandomConstantString (const NumberType type, de::Random& rnd)
8630 if (type == NUMBERTYPE_INT32)
8632 return numberToString<deInt32>(getInt(rnd));
8634 else if (type == NUMBERTYPE_UINT32)
8636 return numberToString<deUint32>(rnd.getUint32());
8638 else if (type == NUMBERTYPE_FLOAT32)
8640 return numberToString<float>(rnd.getFloat());
8649 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8651 map<string, string> params;
8654 for (int width = 2; width <= 4; ++width)
8656 const string randomConst = numberToString(getInt(rnd));
8657 const string widthStr = numberToString(width);
8658 const string composite_type = "${customType}vec" + widthStr;
8659 const int index = rnd.getInt(0, width-1);
8661 params["type"] = "vec";
8662 params["name"] = params["type"] + "_" + widthStr;
8663 params["compositeDecl"] = composite_type + " = OpTypeVector ${customType} " + widthStr +"\n";
8664 params["compositeType"] = composite_type;
8665 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8666 params["compositeConstruct"] = "%instance = OpCompositeConstruct " + composite_type + repeatString(" %filler", width) + "\n";
8667 params["indexes"] = numberToString(index);
8668 testCases.push_back(params);
8672 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8674 const int limit = 10;
8675 map<string, string> params;
8677 for (int width = 2; width <= limit; ++width)
8679 string randomConst = numberToString(getInt(rnd));
8680 string widthStr = numberToString(width);
8681 int index = rnd.getInt(0, width-1);
8683 params["type"] = "array";
8684 params["name"] = params["type"] + "_" + widthStr;
8685 params["compositeDecl"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
8686 + "%composite = OpTypeArray ${customType} %arraywidth\n";
8687 params["compositeType"] = "%composite";
8688 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8689 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8690 params["indexes"] = numberToString(index);
8691 testCases.push_back(params);
8695 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8697 const int limit = 10;
8698 map<string, string> params;
8700 for (int width = 2; width <= limit; ++width)
8702 string randomConst = numberToString(getInt(rnd));
8703 int index = rnd.getInt(0, width-1);
8705 params["type"] = "struct";
8706 params["name"] = params["type"] + "_" + numberToString(width);
8707 params["compositeDecl"] = "%composite = OpTypeStruct" + repeatString(" ${customType}", width) + "\n";
8708 params["compositeType"] = "%composite";
8709 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8710 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8711 params["indexes"] = numberToString(index);
8712 testCases.push_back(params);
8716 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8718 map<string, string> params;
8721 for (int width = 2; width <= 4; ++width)
8723 string widthStr = numberToString(width);
8725 for (int column = 2 ; column <= 4; ++column)
8727 int index_0 = rnd.getInt(0, column-1);
8728 int index_1 = rnd.getInt(0, width-1);
8729 string columnStr = numberToString(column);
8731 params["type"] = "matrix";
8732 params["name"] = params["type"] + "_" + widthStr + "x" + columnStr;
8733 params["compositeDecl"] = string("%vectype = OpTypeVector ${customType} " + widthStr + "\n")
8734 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
8735 params["compositeType"] = "%composite";
8737 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n"
8738 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
8740 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
8741 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
8742 testCases.push_back(params);
8747 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8749 createVectorCompositeCases(testCases, rnd, type);
8750 createArrayCompositeCases(testCases, rnd, type);
8751 createStructCompositeCases(testCases, rnd, type);
8752 // Matrix only supports float types
8753 if (type == NUMBERTYPE_FLOAT32)
8755 createMatrixCompositeCases(testCases, rnd, type);
8759 const string getAssemblyTypeDeclaration (const NumberType type)
8763 case NUMBERTYPE_INT32: return "OpTypeInt 32 1";
8764 case NUMBERTYPE_UINT32: return "OpTypeInt 32 0";
8765 case NUMBERTYPE_FLOAT32: return "OpTypeFloat 32";
8766 default: DE_ASSERT(false); return "";
8770 const string getAssemblyTypeName (const NumberType type)
8774 case NUMBERTYPE_INT32: return "%i32";
8775 case NUMBERTYPE_UINT32: return "%u32";
8776 case NUMBERTYPE_FLOAT32: return "%f32";
8777 default: DE_ASSERT(false); return "";
8781 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
8783 map<string, string> parameters(params);
8785 const string customType = getAssemblyTypeName(type);
8786 map<string, string> substCustomType;
8787 substCustomType["customType"] = customType;
8788 parameters["compositeDecl"] = StringTemplate(parameters.at("compositeDecl")).specialize(substCustomType);
8789 parameters["compositeType"] = StringTemplate(parameters.at("compositeType")).specialize(substCustomType);
8790 parameters["compositeConstruct"] = StringTemplate(parameters.at("compositeConstruct")).specialize(substCustomType);
8791 parameters["filler"] = StringTemplate(parameters.at("filler")).specialize(substCustomType);
8792 parameters["customType"] = customType;
8793 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
8795 if (parameters.at("compositeType") != "%u32vec3")
8797 parameters["u32vec3Decl"] = "%u32vec3 = OpTypeVector %u32 3\n";
8800 return StringTemplate(
8801 "OpCapability Shader\n"
8802 "OpCapability Matrix\n"
8803 "OpMemoryModel Logical GLSL450\n"
8804 "OpEntryPoint GLCompute %main \"main\" %id\n"
8805 "OpExecutionMode %main LocalSize 1 1 1\n"
8807 "OpSource GLSL 430\n"
8808 "OpName %main \"main\"\n"
8809 "OpName %id \"gl_GlobalInvocationID\"\n"
8812 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8813 "OpDecorate %buf BufferBlock\n"
8814 "OpDecorate %indata DescriptorSet 0\n"
8815 "OpDecorate %indata Binding 0\n"
8816 "OpDecorate %outdata DescriptorSet 0\n"
8817 "OpDecorate %outdata Binding 1\n"
8818 "OpDecorate %customarr ArrayStride 4\n"
8819 "${compositeDecorator}"
8820 "OpMemberDecorate %buf 0 Offset 0\n"
8823 "%void = OpTypeVoid\n"
8824 "%voidf = OpTypeFunction %void\n"
8825 "%u32 = OpTypeInt 32 0\n"
8826 "%i32 = OpTypeInt 32 1\n"
8827 "%f32 = OpTypeFloat 32\n"
8829 // Composite declaration
8835 "${u32vec3Decl:opt}"
8836 "%uvec3ptr = OpTypePointer Input %u32vec3\n"
8838 // Inherited from custom
8839 "%customptr = OpTypePointer Uniform ${customType}\n"
8840 "%customarr = OpTypeRuntimeArray ${customType}\n"
8841 "%buf = OpTypeStruct %customarr\n"
8842 "%bufptr = OpTypePointer Uniform %buf\n"
8844 "%indata = OpVariable %bufptr Uniform\n"
8845 "%outdata = OpVariable %bufptr Uniform\n"
8847 "%id = OpVariable %uvec3ptr Input\n"
8848 "%zero = OpConstant %i32 0\n"
8850 "%main = OpFunction %void None %voidf\n"
8851 "%label = OpLabel\n"
8852 "%idval = OpLoad %u32vec3 %id\n"
8853 "%x = OpCompositeExtract %u32 %idval 0\n"
8855 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
8856 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
8857 // Read the input value
8858 "%inval = OpLoad ${customType} %inloc\n"
8859 // Create the composite and fill it
8860 "${compositeConstruct}"
8861 // Insert the input value to a place
8862 "%instance2 = OpCompositeInsert ${compositeType} %inval %instance ${indexes}\n"
8863 // Read back the value from the position
8864 "%out_val = OpCompositeExtract ${customType} %instance2 ${indexes}\n"
8865 // Store it in the output position
8866 " OpStore %outloc %out_val\n"
8869 ).specialize(parameters);
8872 template<typename T>
8873 BufferSp createCompositeBuffer(T number)
8875 return BufferSp(new Buffer<T>(vector<T>(1, number)));
8878 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
8880 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
8881 de::Random rnd (deStringHash(group->getName()));
8883 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8885 NumberType numberType = NumberType(type);
8886 const string typeName = getNumberTypeName(numberType);
8887 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
8888 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8889 vector<map<string, string> > testCases;
8891 createCompositeCases(testCases, rnd, numberType);
8893 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8895 ComputeShaderSpec spec;
8897 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
8901 case NUMBERTYPE_INT32:
8903 deInt32 number = getInt(rnd);
8904 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8905 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8908 case NUMBERTYPE_UINT32:
8910 deUint32 number = rnd.getUint32();
8911 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8912 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8915 case NUMBERTYPE_FLOAT32:
8917 float number = rnd.getFloat();
8918 spec.inputs.push_back(createCompositeBuffer<float>(number));
8919 spec.outputs.push_back(createCompositeBuffer<float>(number));
8926 spec.numWorkGroups = IVec3(1, 1, 1);
8927 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
8929 group->addChild(subGroup.release());
8931 return group.release();
8934 struct AssemblyStructInfo
8936 AssemblyStructInfo (const deUint32 comp, const deUint32 idx)
8941 deUint32 components;
8945 const string specializeInBoundsShaderTemplate (const NumberType type, const AssemblyStructInfo& structInfo, const map<string, string>& params)
8947 // Create the full index string
8948 string fullIndex = numberToString(structInfo.index) + " " + params.at("indexes");
8949 // Convert it to list of indexes
8950 vector<string> indexes = de::splitString(fullIndex, ' ');
8952 map<string, string> parameters (params);
8953 parameters["structType"] = repeatString(" ${compositeType}", structInfo.components);
8954 parameters["structConstruct"] = repeatString(" %instance", structInfo.components);
8955 parameters["insertIndexes"] = fullIndex;
8957 // In matrix cases the last two index is the CompositeExtract indexes
8958 const deUint32 extractIndexes = (parameters["type"] == "matrix") ? 2 : 1;
8960 // Construct the extractIndex
8961 for (vector<string>::const_iterator index = indexes.end() - extractIndexes; index != indexes.end(); ++index)
8963 parameters["extractIndexes"] += " " + *index;
8966 // Remove the last 1 or 2 element depends on matrix case or not
8967 indexes.erase(indexes.end() - extractIndexes, indexes.end());
8970 // Generate AccessChain index expressions (except for the last one, because we use ptr to the composite)
8971 for (vector<string>::const_iterator index = indexes.begin(); index != indexes.end(); ++index)
8973 string indexId = "%index_" + numberToString(id++);
8974 parameters["accessChainConstDeclaration"] += indexId + " = OpConstant %u32 " + *index + "\n";
8975 parameters["accessChainIndexes"] += " " + indexId;
8978 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
8980 const string customType = getAssemblyTypeName(type);
8981 map<string, string> substCustomType;
8982 substCustomType["customType"] = customType;
8983 parameters["compositeDecl"] = StringTemplate(parameters.at("compositeDecl")).specialize(substCustomType);
8984 parameters["compositeType"] = StringTemplate(parameters.at("compositeType")).specialize(substCustomType);
8985 parameters["compositeConstruct"] = StringTemplate(parameters.at("compositeConstruct")).specialize(substCustomType);
8986 parameters["filler"] = StringTemplate(parameters.at("filler")).specialize(substCustomType);
8987 parameters["customType"] = customType;
8989 const string compositeType = parameters.at("compositeType");
8990 map<string, string> substCompositeType;
8991 substCompositeType["compositeType"] = compositeType;
8992 parameters["structType"] = StringTemplate(parameters.at("structType")).specialize(substCompositeType);
8993 if (compositeType != "%u32vec3")
8995 parameters["u32vec3Decl"] = "%u32vec3 = OpTypeVector %u32 3\n";
8998 return StringTemplate(
8999 "OpCapability Shader\n"
9000 "OpCapability Matrix\n"
9001 "OpMemoryModel Logical GLSL450\n"
9002 "OpEntryPoint GLCompute %main \"main\" %id\n"
9003 "OpExecutionMode %main LocalSize 1 1 1\n"
9005 "OpSource GLSL 430\n"
9006 "OpName %main \"main\"\n"
9007 "OpName %id \"gl_GlobalInvocationID\"\n"
9009 "OpDecorate %id BuiltIn GlobalInvocationId\n"
9010 "OpDecorate %buf BufferBlock\n"
9011 "OpDecorate %indata DescriptorSet 0\n"
9012 "OpDecorate %indata Binding 0\n"
9013 "OpDecorate %outdata DescriptorSet 0\n"
9014 "OpDecorate %outdata Binding 1\n"
9015 "OpDecorate %customarr ArrayStride 4\n"
9016 "${compositeDecorator}"
9017 "OpMemberDecorate %buf 0 Offset 0\n"
9019 "%void = OpTypeVoid\n"
9020 "%voidf = OpTypeFunction %void\n"
9021 "%i32 = OpTypeInt 32 1\n"
9022 "%u32 = OpTypeInt 32 0\n"
9023 "%f32 = OpTypeFloat 32\n"
9026 // %u32vec3 if not already declared in ${compositeDecl}
9027 "${u32vec3Decl:opt}"
9028 "%uvec3ptr = OpTypePointer Input %u32vec3\n"
9029 // Inherited from composite
9030 "%composite_p = OpTypePointer Function ${compositeType}\n"
9031 "%struct_t = OpTypeStruct${structType}\n"
9032 "%struct_p = OpTypePointer Function %struct_t\n"
9035 "${accessChainConstDeclaration}"
9036 // Inherited from custom
9037 "%customptr = OpTypePointer Uniform ${customType}\n"
9038 "%customarr = OpTypeRuntimeArray ${customType}\n"
9039 "%buf = OpTypeStruct %customarr\n"
9040 "%bufptr = OpTypePointer Uniform %buf\n"
9041 "%indata = OpVariable %bufptr Uniform\n"
9042 "%outdata = OpVariable %bufptr Uniform\n"
9044 "%id = OpVariable %uvec3ptr Input\n"
9045 "%zero = OpConstant %u32 0\n"
9046 "%main = OpFunction %void None %voidf\n"
9047 "%label = OpLabel\n"
9048 "%struct_v = OpVariable %struct_p Function\n"
9049 "%idval = OpLoad %u32vec3 %id\n"
9050 "%x = OpCompositeExtract %u32 %idval 0\n"
9051 // Create the input/output type
9052 "%inloc = OpInBoundsAccessChain %customptr %indata %zero %x\n"
9053 "%outloc = OpInBoundsAccessChain %customptr %outdata %zero %x\n"
9054 // Read the input value
9055 "%inval = OpLoad ${customType} %inloc\n"
9056 // Create the composite and fill it
9057 "${compositeConstruct}"
9058 // Create the struct and fill it with the composite
9059 "%struct = OpCompositeConstruct %struct_t${structConstruct}\n"
9061 "%comp_obj = OpCompositeInsert %struct_t %inval %struct ${insertIndexes}\n"
9063 " OpStore %struct_v %comp_obj\n"
9064 // Get deepest possible composite pointer
9065 "%inner_ptr = OpInBoundsAccessChain %composite_p %struct_v${accessChainIndexes}\n"
9066 "%read_obj = OpLoad ${compositeType} %inner_ptr\n"
9067 // Read back the stored value
9068 "%read_val = OpCompositeExtract ${customType} %read_obj${extractIndexes}\n"
9069 " OpStore %outloc %read_val\n"
9072 ).specialize(parameters);
9075 tcu::TestCaseGroup* createOpInBoundsAccessChainGroup (tcu::TestContext& testCtx)
9077 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opinboundsaccesschain", "Test the OpInBoundsAccessChain instruction"));
9078 de::Random rnd (deStringHash(group->getName()));
9080 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
9082 NumberType numberType = NumberType(type);
9083 const string typeName = getNumberTypeName(numberType);
9084 const string description = "Test the OpInBoundsAccessChain instruction with " + typeName + "s";
9085 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
9087 vector<map<string, string> > testCases;
9088 createCompositeCases(testCases, rnd, numberType);
9090 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
9092 ComputeShaderSpec spec;
9094 // Number of components inside of a struct
9095 deUint32 structComponents = rnd.getInt(2, 8);
9096 // Component index value
9097 deUint32 structIndex = rnd.getInt(0, structComponents - 1);
9098 AssemblyStructInfo structInfo(structComponents, structIndex);
9100 spec.assembly = specializeInBoundsShaderTemplate(numberType, structInfo, *test);
9104 case NUMBERTYPE_INT32:
9106 deInt32 number = getInt(rnd);
9107 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
9108 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
9111 case NUMBERTYPE_UINT32:
9113 deUint32 number = rnd.getUint32();
9114 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
9115 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
9118 case NUMBERTYPE_FLOAT32:
9120 float number = rnd.getFloat();
9121 spec.inputs.push_back(createCompositeBuffer<float>(number));
9122 spec.outputs.push_back(createCompositeBuffer<float>(number));
9128 spec.numWorkGroups = IVec3(1, 1, 1);
9129 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpInBoundsAccessChain test", spec));
9131 group->addChild(subGroup.release());
9133 return group.release();
9136 // If the params missing, uninitialized case
9137 const string specializeDefaultOutputShaderTemplate (const NumberType type, const map<string, string>& params = map<string, string>())
9139 map<string, string> parameters(params);
9141 parameters["customType"] = getAssemblyTypeName(type);
9143 // Declare the const value, and use it in the initializer
9144 if (params.find("constValue") != params.end())
9146 parameters["variableInitializer"] = " %const";
9148 // Uninitialized case
9151 parameters["commentDecl"] = ";";
9154 return StringTemplate(
9155 "OpCapability Shader\n"
9156 "OpMemoryModel Logical GLSL450\n"
9157 "OpEntryPoint GLCompute %main \"main\" %id\n"
9158 "OpExecutionMode %main LocalSize 1 1 1\n"
9159 "OpSource GLSL 430\n"
9160 "OpName %main \"main\"\n"
9161 "OpName %id \"gl_GlobalInvocationID\"\n"
9163 "OpDecorate %id BuiltIn GlobalInvocationId\n"
9164 "OpDecorate %indata DescriptorSet 0\n"
9165 "OpDecorate %indata Binding 0\n"
9166 "OpDecorate %outdata DescriptorSet 0\n"
9167 "OpDecorate %outdata Binding 1\n"
9168 "OpDecorate %in_arr ArrayStride 4\n"
9169 "OpDecorate %in_buf BufferBlock\n"
9170 "OpMemberDecorate %in_buf 0 Offset 0\n"
9172 "%void = OpTypeVoid\n"
9173 "%voidf = OpTypeFunction %void\n"
9174 "%u32 = OpTypeInt 32 0\n"
9175 "%i32 = OpTypeInt 32 1\n"
9176 "%f32 = OpTypeFloat 32\n"
9177 "%uvec3 = OpTypeVector %u32 3\n"
9178 "%uvec3ptr = OpTypePointer Input %uvec3\n"
9179 "${commentDecl:opt}%const = OpConstant ${customType} ${constValue:opt}\n"
9181 "%in_ptr = OpTypePointer Uniform ${customType}\n"
9182 "%in_arr = OpTypeRuntimeArray ${customType}\n"
9183 "%in_buf = OpTypeStruct %in_arr\n"
9184 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
9185 "%indata = OpVariable %in_bufptr Uniform\n"
9186 "%outdata = OpVariable %in_bufptr Uniform\n"
9187 "%id = OpVariable %uvec3ptr Input\n"
9188 "%var_ptr = OpTypePointer Function ${customType}\n"
9190 "%zero = OpConstant %i32 0\n"
9192 "%main = OpFunction %void None %voidf\n"
9193 "%label = OpLabel\n"
9194 "%out_var = OpVariable %var_ptr Function${variableInitializer:opt}\n"
9195 "%idval = OpLoad %uvec3 %id\n"
9196 "%x = OpCompositeExtract %u32 %idval 0\n"
9197 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
9198 "%outloc = OpAccessChain %in_ptr %outdata %zero %x\n"
9200 "%outval = OpLoad ${customType} %out_var\n"
9201 " OpStore %outloc %outval\n"
9204 ).specialize(parameters);
9207 bool compareFloats (const std::vector<Resource>&, const vector<AllocationSp>& outputAllocs, const std::vector<Resource>& expectedOutputs, TestLog& log)
9209 DE_ASSERT(outputAllocs.size() != 0);
9210 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
9212 // Use custom epsilon because of the float->string conversion
9213 const float epsilon = 0.00001f;
9215 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
9217 vector<deUint8> expectedBytes;
9221 expectedOutputs[outputNdx].getBytes(expectedBytes);
9222 memcpy(&expected, &expectedBytes.front(), expectedBytes.size());
9223 memcpy(&actual, outputAllocs[outputNdx]->getHostPtr(), expectedBytes.size());
9225 // Test with epsilon
9226 if (fabs(expected - actual) > epsilon)
9228 log << TestLog::Message << "Error: The actual and expected values not matching."
9229 << " Expected: " << expected << " Actual: " << actual << " Epsilon: " << epsilon << TestLog::EndMessage;
9236 // Checks if the driver crash with uninitialized cases
9237 bool passthruVerify (const std::vector<Resource>&, const vector<AllocationSp>& outputAllocs, const std::vector<Resource>& expectedOutputs, TestLog&)
9239 DE_ASSERT(outputAllocs.size() != 0);
9240 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
9242 // Copy and discard the result.
9243 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
9245 vector<deUint8> expectedBytes;
9246 expectedOutputs[outputNdx].getBytes(expectedBytes);
9248 const size_t width = expectedBytes.size();
9249 vector<char> data (width);
9251 memcpy(&data[0], outputAllocs[outputNdx]->getHostPtr(), width);
9256 tcu::TestCaseGroup* createShaderDefaultOutputGroup (tcu::TestContext& testCtx)
9258 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "shader_default_output", "Test shader default output."));
9259 de::Random rnd (deStringHash(group->getName()));
9261 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
9263 NumberType numberType = NumberType(type);
9264 const string typeName = getNumberTypeName(numberType);
9265 const string description = "Test the OpVariable initializer with " + typeName + ".";
9266 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
9268 // 2 similar subcases (initialized and uninitialized)
9269 for (int subCase = 0; subCase < 2; ++subCase)
9271 ComputeShaderSpec spec;
9272 spec.numWorkGroups = IVec3(1, 1, 1);
9274 map<string, string> params;
9278 case NUMBERTYPE_INT32:
9280 deInt32 number = getInt(rnd);
9281 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
9282 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
9283 params["constValue"] = numberToString(number);
9286 case NUMBERTYPE_UINT32:
9288 deUint32 number = rnd.getUint32();
9289 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
9290 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
9291 params["constValue"] = numberToString(number);
9294 case NUMBERTYPE_FLOAT32:
9296 float number = rnd.getFloat();
9297 spec.inputs.push_back(createCompositeBuffer<float>(number));
9298 spec.outputs.push_back(createCompositeBuffer<float>(number));
9299 spec.verifyIO = &compareFloats;
9300 params["constValue"] = numberToString(number);
9307 // Initialized subcase
9310 spec.assembly = specializeDefaultOutputShaderTemplate(numberType, params);
9311 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "initialized", "OpVariable initializer tests.", spec));
9313 // Uninitialized subcase
9316 spec.assembly = specializeDefaultOutputShaderTemplate(numberType);
9317 spec.verifyIO = &passthruVerify;
9318 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "uninitialized", "OpVariable initializer tests.", spec));
9321 group->addChild(subGroup.release());
9323 return group.release();
9326 tcu::TestCaseGroup* createOpNopTests (tcu::TestContext& testCtx)
9328 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
9329 RGBA defaultColors[4];
9330 map<string, string> opNopFragments;
9332 getDefaultColors(defaultColors);
9334 opNopFragments["testfun"] =
9335 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
9336 "%param1 = OpFunctionParameter %v4f32\n"
9337 "%label_testfun = OpLabel\n"
9346 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
9347 "%b = OpFAdd %f32 %a %a\n"
9349 "%c = OpFSub %f32 %b %a\n"
9350 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
9353 "OpReturnValue %ret\n"
9356 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, testGroup.get());
9358 return testGroup.release();
9361 tcu::TestCaseGroup* createOpNameTests (tcu::TestContext& testCtx)
9363 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opname","Test OpName"));
9364 RGBA defaultColors[4];
9365 map<string, string> opNameFragments;
9367 getDefaultColors(defaultColors);
9369 opNameFragments["debug"] =
9370 "OpName %BP_main \"not_main\"";
9372 opNameFragments["testfun"] =
9373 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
9374 "%param1 = OpFunctionParameter %v4f32\n"
9375 "%label_func = OpLabel\n"
9376 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
9377 "%b = OpFAdd %f32 %a %a\n"
9378 "%c = OpFSub %f32 %b %a\n"
9379 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
9380 "OpReturnValue %ret\n"
9383 createTestsForAllStages("opname", defaultColors, defaultColors, opNameFragments, testGroup.get());
9385 return testGroup.release();
9388 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
9390 const bool testComputePipeline = true;
9392 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
9393 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
9394 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
9396 computeTests->addChild(createSpivVersionCheckTests(testCtx, testComputePipeline));
9397 computeTests->addChild(createLocalSizeGroup(testCtx));
9398 computeTests->addChild(createOpNopGroup(testCtx));
9399 computeTests->addChild(createOpFUnordGroup(testCtx));
9400 computeTests->addChild(createOpAtomicGroup(testCtx, false));
9401 computeTests->addChild(createOpAtomicGroup(testCtx, true)); // Using new StorageBuffer decoration
9402 computeTests->addChild(createOpAtomicGroup(testCtx, false, 1024, true)); // Return value validation
9403 computeTests->addChild(createOpLineGroup(testCtx));
9404 computeTests->addChild(createOpModuleProcessedGroup(testCtx));
9405 computeTests->addChild(createOpNoLineGroup(testCtx));
9406 computeTests->addChild(createOpConstantNullGroup(testCtx));
9407 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
9408 computeTests->addChild(createOpConstantUsageGroup(testCtx));
9409 computeTests->addChild(createSpecConstantGroup(testCtx));
9410 computeTests->addChild(createOpSourceGroup(testCtx));
9411 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
9412 computeTests->addChild(createDecorationGroupGroup(testCtx));
9413 computeTests->addChild(createOpPhiGroup(testCtx));
9414 computeTests->addChild(createLoopControlGroup(testCtx));
9415 computeTests->addChild(createFunctionControlGroup(testCtx));
9416 computeTests->addChild(createSelectionControlGroup(testCtx));
9417 computeTests->addChild(createBlockOrderGroup(testCtx));
9418 computeTests->addChild(createMultipleShaderGroup(testCtx));
9419 computeTests->addChild(createMemoryAccessGroup(testCtx));
9420 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
9421 computeTests->addChild(createOpCopyObjectGroup(testCtx));
9422 computeTests->addChild(createNoContractionGroup(testCtx));
9423 computeTests->addChild(createOpUndefGroup(testCtx));
9424 computeTests->addChild(createOpUnreachableGroup(testCtx));
9425 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
9426 computeTests ->addChild(createOpFRemGroup(testCtx));
9427 computeTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_PASS));
9428 computeTests->addChild(createOpSRemComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
9429 computeTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_PASS));
9430 computeTests->addChild(createOpSModComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
9431 computeTests->addChild(createConvertComputeTests(testCtx, "OpSConvert", "sconvert"));
9432 computeTests->addChild(createConvertComputeTests(testCtx, "OpUConvert", "uconvert"));
9433 computeTests->addChild(createConvertComputeTests(testCtx, "OpFConvert", "fconvert"));
9434 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
9435 computeTests->addChild(createOpInBoundsAccessChainGroup(testCtx));
9436 computeTests->addChild(createShaderDefaultOutputGroup(testCtx));
9437 computeTests->addChild(createOpNMinGroup(testCtx));
9438 computeTests->addChild(createOpNMaxGroup(testCtx));
9439 computeTests->addChild(createOpNClampGroup(testCtx));
9441 de::MovePtr<tcu::TestCaseGroup> computeAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
9443 computeAndroidTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9444 computeAndroidTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9446 computeTests->addChild(computeAndroidTests.release());
9449 computeTests->addChild(create8BitStorageComputeGroup(testCtx));
9450 computeTests->addChild(create16BitStorageComputeGroup(testCtx));
9451 computeTests->addChild(createUboMatrixPaddingComputeGroup(testCtx));
9452 computeTests->addChild(createVariableInitComputeGroup(testCtx));
9453 computeTests->addChild(createConditionalBranchComputeGroup(testCtx));
9454 computeTests->addChild(createIndexingComputeGroup(testCtx));
9455 computeTests->addChild(createVariablePointersComputeGroup(testCtx));
9456 computeTests->addChild(createImageSamplerComputeGroup(testCtx));
9457 computeTests->addChild(createOpNameGroup(testCtx));
9458 computeTests->addChild(createPointerParameterComputeGroup(testCtx));
9459 graphicsTests->addChild(createCrossStageInterfaceTests(testCtx));
9460 graphicsTests->addChild(createSpivVersionCheckTests(testCtx, !testComputePipeline));
9461 graphicsTests->addChild(createOpNopTests(testCtx));
9462 graphicsTests->addChild(createOpSourceTests(testCtx));
9463 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
9464 graphicsTests->addChild(createOpModuleProcessedTests(testCtx));
9465 graphicsTests->addChild(createOpLineTests(testCtx));
9466 graphicsTests->addChild(createOpNoLineTests(testCtx));
9467 graphicsTests->addChild(createOpConstantNullTests(testCtx));
9468 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
9469 graphicsTests->addChild(createMemoryAccessTests(testCtx));
9470 graphicsTests->addChild(createOpUndefTests(testCtx));
9471 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
9472 graphicsTests->addChild(createModuleTests(testCtx));
9473 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
9474 graphicsTests->addChild(createOpPhiTests(testCtx));
9475 graphicsTests->addChild(createNoContractionTests(testCtx));
9476 graphicsTests->addChild(createOpQuantizeTests(testCtx));
9477 graphicsTests->addChild(createLoopTests(testCtx));
9478 graphicsTests->addChild(createSpecConstantTests(testCtx));
9479 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
9480 graphicsTests->addChild(createBarrierTests(testCtx));
9481 graphicsTests->addChild(createDecorationGroupTests(testCtx));
9482 graphicsTests->addChild(createFRemTests(testCtx));
9483 graphicsTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
9484 graphicsTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
9487 de::MovePtr<tcu::TestCaseGroup> graphicsAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
9489 graphicsAndroidTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9490 graphicsAndroidTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9492 graphicsTests->addChild(graphicsAndroidTests.release());
9494 graphicsTests->addChild(createOpNameTests(testCtx));
9496 graphicsTests->addChild(create8BitStorageGraphicsGroup(testCtx));
9497 graphicsTests->addChild(create16BitStorageGraphicsGroup(testCtx));
9498 graphicsTests->addChild(createUboMatrixPaddingGraphicsGroup(testCtx));
9499 graphicsTests->addChild(createVariableInitGraphicsGroup(testCtx));
9500 graphicsTests->addChild(createConditionalBranchGraphicsGroup(testCtx));
9501 graphicsTests->addChild(createIndexingGraphicsGroup(testCtx));
9502 graphicsTests->addChild(createVariablePointersGraphicsGroup(testCtx));
9503 graphicsTests->addChild(createImageSamplerGraphicsGroup(testCtx));
9504 graphicsTests->addChild(createConvertGraphicsTests(testCtx, "OpSConvert", "sconvert"));
9505 graphicsTests->addChild(createConvertGraphicsTests(testCtx, "OpUConvert", "uconvert"));
9506 graphicsTests->addChild(createConvertGraphicsTests(testCtx, "OpFConvert", "fconvert"));
9507 graphicsTests->addChild(createPointerParameterGraphicsGroup(testCtx));
9509 instructionTests->addChild(computeTests.release());
9510 instructionTests->addChild(graphicsTests.release());
9512 return instructionTests.release();