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<BufferSp>& inputs, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& 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<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& 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<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& 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<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1174 if (outputAllocs.size() != 1)
1177 const BufferSp& expectedOutput (expectedOutputs[0]);
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<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1298 if (outputAllocs.size() != 1)
1301 const BufferSp& expectedOutput = expectedOutputs[0];
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<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1421 if (outputAllocs.size() != 1)
1424 const BufferSp& expectedOutput = expectedOutputs[0];
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("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2738 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2739 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2740 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2741 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
2742 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
2743 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
2744 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
2745 cases.push_back(SpecConstantTwoIntCase("sconvert", " %i32 0", " %i32 0", "%i16", "SConvert %sc_0", -11200, 0, addSc32ToInput, outputInts3));
2746 // -969998336 stored as 32-bit two's complement is the binary representation of -11200 as IEEE-754 Float
2747 cases.push_back(SpecConstantTwoIntCase("fconvert", " %f32 0", " %f32 0", "%f64", "FConvert %sc_0", -969998336, 0, addSc32ToInput, outputInts3));
2749 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2751 map<string, string> specializations;
2752 ComputeShaderSpec spec;
2753 ComputeTestFeatures features = COMPUTE_TEST_USES_NONE;
2755 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
2756 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
2757 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
2758 specializations["SC_OP"] = cases[caseNdx].scOperation;
2759 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
2761 // Special SPIR-V code for SConvert-case
2762 if (strcmp(cases[caseNdx].caseName, "sconvert") == 0)
2764 features = COMPUTE_TEST_USES_INT16;
2765 specializations["CAPABILITIES"] = "OpCapability Int16\n"; // Adds 16-bit integer capability
2766 specializations["OPTYPE_DEFINITIONS"] = "%i16 = OpTypeInt 16 1\n"; // Adds 16-bit integer type
2767 specializations["TYPE_CONVERT"] = "%sc_final32 = OpSConvert %i32 %sc_final\n"; // Converts 16-bit integer to 32-bit integer
2770 // Special SPIR-V code for FConvert-case
2771 if (strcmp(cases[caseNdx].caseName, "fconvert") == 0)
2773 features = COMPUTE_TEST_USES_FLOAT64;
2774 specializations["CAPABILITIES"] = "OpCapability Float64\n"; // Adds 64-bit float capability
2775 specializations["OPTYPE_DEFINITIONS"] = "%f64 = OpTypeFloat 64\n"; // Adds 64-bit float type
2776 specializations["TYPE_CONVERT"] = "%sc_final32 = OpConvertFToS %i32 %sc_final\n"; // Converts 64-bit float to 32-bit integer
2779 spec.assembly = shaderTemplate.specialize(specializations);
2780 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2781 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
2782 spec.numWorkGroups = IVec3(numElements, 1, 1);
2783 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
2784 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
2786 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec, features));
2789 ComputeShaderSpec spec;
2792 string(getComputeAsmShaderPreamble()) +
2794 "OpName %main \"main\"\n"
2795 "OpName %id \"gl_GlobalInvocationID\"\n"
2797 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2798 "OpDecorate %sc_0 SpecId 0\n"
2799 "OpDecorate %sc_1 SpecId 1\n"
2800 "OpDecorate %sc_2 SpecId 2\n"
2801 "OpDecorate %i32arr ArrayStride 4\n"
2803 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2805 "%ivec3 = OpTypeVector %i32 3\n"
2806 "%buf = OpTypeStruct %i32arr\n"
2807 "%bufptr = OpTypePointer Uniform %buf\n"
2808 "%indata = OpVariable %bufptr Uniform\n"
2809 "%outdata = OpVariable %bufptr Uniform\n"
2811 "%id = OpVariable %uvec3ptr Input\n"
2812 "%zero = OpConstant %i32 0\n"
2813 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
2814 "%vec3_undef = OpUndef %ivec3\n"
2816 "%sc_0 = OpSpecConstant %i32 0\n"
2817 "%sc_1 = OpSpecConstant %i32 0\n"
2818 "%sc_2 = OpSpecConstant %i32 0\n"
2819 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
2820 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
2821 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
2822 "%sc_vec3_0_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
2823 "%sc_vec3_1_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
2824 "%sc_vec3_2_s = OpSpecConstantOp %ivec3 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
2825 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
2826 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
2827 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
2828 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
2829 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
2830 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
2831 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
2833 "%main = OpFunction %void None %voidf\n"
2834 "%label = OpLabel\n"
2835 "%idval = OpLoad %uvec3 %id\n"
2836 "%x = OpCompositeExtract %u32 %idval 0\n"
2837 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2838 "%inval = OpLoad %i32 %inloc\n"
2839 "%final = OpIAdd %i32 %inval %sc_final\n"
2840 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2841 " OpStore %outloc %final\n"
2844 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2845 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
2846 spec.numWorkGroups = IVec3(numElements, 1, 1);
2847 spec.specConstants.push_back(123);
2848 spec.specConstants.push_back(56);
2849 spec.specConstants.push_back(-77);
2851 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
2853 return group.release();
2856 void createOpPhiVartypeTests (de::MovePtr<tcu::TestCaseGroup>& group, tcu::TestContext& testCtx)
2858 ComputeShaderSpec specInt;
2859 ComputeShaderSpec specFloat;
2860 ComputeShaderSpec specVec3;
2861 ComputeShaderSpec specMat4;
2862 ComputeShaderSpec specArray;
2863 ComputeShaderSpec specStruct;
2864 de::Random rnd (deStringHash(group->getName()));
2865 const int numElements = 100;
2866 vector<float> inputFloats (numElements, 0);
2867 vector<float> outputFloats (numElements, 0);
2869 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
2871 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2872 floorAll(inputFloats);
2874 for (size_t ndx = 0; ndx < numElements; ++ndx)
2876 // Just check if the value is positive or not
2877 outputFloats[ndx] = (inputFloats[ndx] > 0) ? 1.0f : -1.0f;
2880 // All of the tests are of the form:
2884 // if (inputdata > 0)
2891 specFloat.assembly =
2892 string(getComputeAsmShaderPreamble()) +
2894 "OpSource GLSL 430\n"
2895 "OpName %main \"main\"\n"
2896 "OpName %id \"gl_GlobalInvocationID\"\n"
2898 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2900 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2902 "%id = OpVariable %uvec3ptr Input\n"
2903 "%zero = OpConstant %i32 0\n"
2904 "%float_0 = OpConstant %f32 0.0\n"
2905 "%float_1 = OpConstant %f32 1.0\n"
2906 "%float_n1 = OpConstant %f32 -1.0\n"
2908 "%main = OpFunction %void None %voidf\n"
2909 "%entry = OpLabel\n"
2910 "%idval = OpLoad %uvec3 %id\n"
2911 "%x = OpCompositeExtract %u32 %idval 0\n"
2912 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2913 "%inval = OpLoad %f32 %inloc\n"
2915 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2916 " OpSelectionMerge %cm None\n"
2917 " OpBranchConditional %comp %tb %fb\n"
2923 "%res = OpPhi %f32 %float_1 %tb %float_n1 %fb\n"
2925 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2926 " OpStore %outloc %res\n"
2930 specFloat.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2931 specFloat.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2932 specFloat.numWorkGroups = IVec3(numElements, 1, 1);
2935 string(getComputeAsmShaderPreamble()) +
2937 "OpSource GLSL 430\n"
2938 "OpName %main \"main\"\n"
2939 "OpName %id \"gl_GlobalInvocationID\"\n"
2941 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2943 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2945 "%id = OpVariable %uvec3ptr Input\n"
2946 "%v4f32 = OpTypeVector %f32 4\n"
2947 "%mat4v4f32 = OpTypeMatrix %v4f32 4\n"
2948 "%zero = OpConstant %i32 0\n"
2949 "%float_0 = OpConstant %f32 0.0\n"
2950 "%float_1 = OpConstant %f32 1.0\n"
2951 "%float_n1 = OpConstant %f32 -1.0\n"
2952 "%m11 = OpConstantComposite %v4f32 %float_1 %float_0 %float_0 %float_0\n"
2953 "%m12 = OpConstantComposite %v4f32 %float_0 %float_1 %float_0 %float_0\n"
2954 "%m13 = OpConstantComposite %v4f32 %float_0 %float_0 %float_1 %float_0\n"
2955 "%m14 = OpConstantComposite %v4f32 %float_0 %float_0 %float_0 %float_1\n"
2956 "%m1 = OpConstantComposite %mat4v4f32 %m11 %m12 %m13 %m14\n"
2957 "%m21 = OpConstantComposite %v4f32 %float_n1 %float_0 %float_0 %float_0\n"
2958 "%m22 = OpConstantComposite %v4f32 %float_0 %float_n1 %float_0 %float_0\n"
2959 "%m23 = OpConstantComposite %v4f32 %float_0 %float_0 %float_n1 %float_0\n"
2960 "%m24 = OpConstantComposite %v4f32 %float_0 %float_0 %float_0 %float_n1\n"
2961 "%m2 = OpConstantComposite %mat4v4f32 %m21 %m22 %m23 %m24\n"
2963 "%main = OpFunction %void None %voidf\n"
2964 "%entry = OpLabel\n"
2965 "%idval = OpLoad %uvec3 %id\n"
2966 "%x = OpCompositeExtract %u32 %idval 0\n"
2967 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2968 "%inval = OpLoad %f32 %inloc\n"
2970 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2971 " OpSelectionMerge %cm None\n"
2972 " OpBranchConditional %comp %tb %fb\n"
2978 "%mres = OpPhi %mat4v4f32 %m1 %tb %m2 %fb\n"
2979 "%res = OpCompositeExtract %f32 %mres 2 2\n"
2981 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2982 " OpStore %outloc %res\n"
2986 specMat4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2987 specMat4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2988 specMat4.numWorkGroups = IVec3(numElements, 1, 1);
2991 string(getComputeAsmShaderPreamble()) +
2993 "OpSource GLSL 430\n"
2994 "OpName %main \"main\"\n"
2995 "OpName %id \"gl_GlobalInvocationID\"\n"
2997 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2999 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3001 "%id = OpVariable %uvec3ptr Input\n"
3002 "%zero = OpConstant %i32 0\n"
3003 "%float_0 = OpConstant %f32 0.0\n"
3004 "%float_1 = OpConstant %f32 1.0\n"
3005 "%float_n1 = OpConstant %f32 -1.0\n"
3006 "%v1 = OpConstantComposite %fvec3 %float_1 %float_1 %float_1\n"
3007 "%v2 = OpConstantComposite %fvec3 %float_n1 %float_n1 %float_n1\n"
3009 "%main = OpFunction %void None %voidf\n"
3010 "%entry = OpLabel\n"
3011 "%idval = OpLoad %uvec3 %id\n"
3012 "%x = OpCompositeExtract %u32 %idval 0\n"
3013 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3014 "%inval = OpLoad %f32 %inloc\n"
3016 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3017 " OpSelectionMerge %cm None\n"
3018 " OpBranchConditional %comp %tb %fb\n"
3024 "%vres = OpPhi %fvec3 %v1 %tb %v2 %fb\n"
3025 "%res = OpCompositeExtract %f32 %vres 2\n"
3027 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3028 " OpStore %outloc %res\n"
3032 specVec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3033 specVec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3034 specVec3.numWorkGroups = IVec3(numElements, 1, 1);
3037 string(getComputeAsmShaderPreamble()) +
3039 "OpSource GLSL 430\n"
3040 "OpName %main \"main\"\n"
3041 "OpName %id \"gl_GlobalInvocationID\"\n"
3043 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3045 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3047 "%id = OpVariable %uvec3ptr Input\n"
3048 "%zero = OpConstant %i32 0\n"
3049 "%float_0 = OpConstant %f32 0.0\n"
3050 "%i1 = OpConstant %i32 1\n"
3051 "%i2 = OpConstant %i32 -1\n"
3053 "%main = OpFunction %void None %voidf\n"
3054 "%entry = OpLabel\n"
3055 "%idval = OpLoad %uvec3 %id\n"
3056 "%x = OpCompositeExtract %u32 %idval 0\n"
3057 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3058 "%inval = OpLoad %f32 %inloc\n"
3060 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3061 " OpSelectionMerge %cm None\n"
3062 " OpBranchConditional %comp %tb %fb\n"
3068 "%ires = OpPhi %i32 %i1 %tb %i2 %fb\n"
3069 "%res = OpConvertSToF %f32 %ires\n"
3071 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3072 " OpStore %outloc %res\n"
3076 specInt.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3077 specInt.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3078 specInt.numWorkGroups = IVec3(numElements, 1, 1);
3080 specArray.assembly =
3081 string(getComputeAsmShaderPreamble()) +
3083 "OpSource GLSL 430\n"
3084 "OpName %main \"main\"\n"
3085 "OpName %id \"gl_GlobalInvocationID\"\n"
3087 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3089 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3091 "%id = OpVariable %uvec3ptr Input\n"
3092 "%zero = OpConstant %i32 0\n"
3093 "%u7 = OpConstant %u32 7\n"
3094 "%float_0 = OpConstant %f32 0.0\n"
3095 "%float_1 = OpConstant %f32 1.0\n"
3096 "%float_n1 = OpConstant %f32 -1.0\n"
3097 "%f32a7 = OpTypeArray %f32 %u7\n"
3098 "%a1 = OpConstantComposite %f32a7 %float_1 %float_1 %float_1 %float_1 %float_1 %float_1 %float_1\n"
3099 "%a2 = OpConstantComposite %f32a7 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1\n"
3100 "%main = OpFunction %void None %voidf\n"
3101 "%entry = OpLabel\n"
3102 "%idval = OpLoad %uvec3 %id\n"
3103 "%x = OpCompositeExtract %u32 %idval 0\n"
3104 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3105 "%inval = OpLoad %f32 %inloc\n"
3107 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3108 " OpSelectionMerge %cm None\n"
3109 " OpBranchConditional %comp %tb %fb\n"
3115 "%ares = OpPhi %f32a7 %a1 %tb %a2 %fb\n"
3116 "%res = OpCompositeExtract %f32 %ares 5\n"
3118 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3119 " OpStore %outloc %res\n"
3123 specArray.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3124 specArray.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3125 specArray.numWorkGroups = IVec3(numElements, 1, 1);
3127 specStruct.assembly =
3128 string(getComputeAsmShaderPreamble()) +
3130 "OpSource GLSL 430\n"
3131 "OpName %main \"main\"\n"
3132 "OpName %id \"gl_GlobalInvocationID\"\n"
3134 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3136 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3138 "%id = OpVariable %uvec3ptr Input\n"
3139 "%zero = OpConstant %i32 0\n"
3140 "%float_0 = OpConstant %f32 0.0\n"
3141 "%float_1 = OpConstant %f32 1.0\n"
3142 "%float_n1 = OpConstant %f32 -1.0\n"
3144 "%v2f32 = OpTypeVector %f32 2\n"
3145 "%Data2 = OpTypeStruct %f32 %v2f32\n"
3146 "%Data = OpTypeStruct %Data2 %f32\n"
3148 "%in1a = OpConstantComposite %v2f32 %float_1 %float_1\n"
3149 "%in1b = OpConstantComposite %Data2 %float_1 %in1a\n"
3150 "%s1 = OpConstantComposite %Data %in1b %float_1\n"
3151 "%in2a = OpConstantComposite %v2f32 %float_n1 %float_n1\n"
3152 "%in2b = OpConstantComposite %Data2 %float_n1 %in2a\n"
3153 "%s2 = OpConstantComposite %Data %in2b %float_n1\n"
3155 "%main = OpFunction %void None %voidf\n"
3156 "%entry = OpLabel\n"
3157 "%idval = OpLoad %uvec3 %id\n"
3158 "%x = OpCompositeExtract %u32 %idval 0\n"
3159 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3160 "%inval = OpLoad %f32 %inloc\n"
3162 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3163 " OpSelectionMerge %cm None\n"
3164 " OpBranchConditional %comp %tb %fb\n"
3170 "%sres = OpPhi %Data %s1 %tb %s2 %fb\n"
3171 "%res = OpCompositeExtract %f32 %sres 0 0\n"
3173 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3174 " OpStore %outloc %res\n"
3178 specStruct.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3179 specStruct.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3180 specStruct.numWorkGroups = IVec3(numElements, 1, 1);
3182 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_int", "OpPhi with int variables", specInt));
3183 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_float", "OpPhi with float variables", specFloat));
3184 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_vec3", "OpPhi with vec3 variables", specVec3));
3185 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_mat4", "OpPhi with mat4 variables", specMat4));
3186 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_array", "OpPhi with array variables", specArray));
3187 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_struct", "OpPhi with struct variables", specStruct));
3190 string generateConstantDefinitions (int count)
3192 std::ostringstream r;
3193 for (int i = 0; i < count; i++)
3194 r << "%cf" << (i * 10 + 5) << " = OpConstant %f32 " <<(i * 10 + 5) << ".0\n";
3199 string generateSwitchCases (int count)
3201 std::ostringstream r;
3202 for (int i = 0; i < count; i++)
3203 r << " " << i << " %case" << i;
3208 string generateSwitchTargets (int count)
3210 std::ostringstream r;
3211 for (int i = 0; i < count; i++)
3212 r << "%case" << i << " = OpLabel\n OpBranch %phi\n";
3217 string generateOpPhiParams (int count)
3219 std::ostringstream r;
3220 for (int i = 0; i < count; i++)
3221 r << " %cf" << (i * 10 + 5) << " %case" << i;
3226 string generateIntWidth (int value)
3228 std::ostringstream r;
3233 // Expand input string by injecting "ABC" between the input
3234 // string characters. The acc/add/treshold parameters are used
3235 // to skip some of the injections to make the result less
3236 // uniform (and a lot shorter).
3237 string expandOpPhiCase5 (const string& s, int &acc, int add, int treshold)
3239 std::ostringstream res;
3240 const char* p = s.c_str();
3256 // Calculate expected result based on the code string
3257 float calcOpPhiCase5 (float val, const string& s)
3259 const char* p = s.c_str();
3262 const float tv[8] = { 0.5f, 1.5f, 3.5f, 7.5f, 15.5f, 31.5f, 63.5f, 127.5f };
3263 const float v = deFloatAbs(val);
3268 for (int i = 7; i >= 0; --i)
3269 x[i] = std::fmod((float)v, (float)(2 << i));
3270 for (int i = 7; i >= 0; --i)
3271 b[i] = x[i] > tv[i];
3278 if (skip == 0 && b[depth])
3289 if (b[depth] || skip)
3303 // In the code string, the letters represent the following:
3306 // if (certain bit is set)
3317 // AABCBC leads to if(){r++;if(){r++;}else{}}else{}
3318 // ABABCC leads to if(){r++;}else{if(){r++;}else{}}
3319 // ABCABC leads to if(){r++;}else{}if(){r++;}else{}
3321 // Code generation gets a bit complicated due to the else-branches,
3322 // which do not generate new values. Thus, the generator needs to
3323 // keep track of the previous variable change seen by the else
3325 string generateOpPhiCase5 (const string& s)
3327 std::stack<int> idStack;
3328 std::stack<std::string> value;
3329 std::stack<std::string> valueLabel;
3330 std::stack<std::string> mergeLeft;
3331 std::stack<std::string> mergeRight;
3332 std::ostringstream res;
3333 const char* p = s.c_str();
3339 value.push("%f32_0");
3340 valueLabel.push("%f32_0 %entry");
3348 idStack.push(currId);
3349 res << "\tOpSelectionMerge %m" << currId << " None\n";
3350 res << "\tOpBranchConditional %b" << depth << " %t" << currId << " %f" << currId << "\n";
3351 res << "%t" << currId << " = OpLabel\n";
3352 res << "%rt" << currId << " = OpFAdd %f32 " << value.top() << " %f32_1\n";
3353 std::ostringstream tag;
3354 tag << "%rt" << currId;
3355 value.push(tag.str());
3356 tag << " %t" << currId;
3357 valueLabel.push(tag.str());
3362 mergeLeft.push(valueLabel.top());
3365 res << "\tOpBranch %m" << currId << "\n";
3366 res << "%f" << currId << " = OpLabel\n";
3367 std::ostringstream tag;
3368 tag << value.top() << " %f" << currId;
3370 valueLabel.push(tag.str());
3375 mergeRight.push(valueLabel.top());
3376 res << "\tOpBranch %m" << currId << "\n";
3377 res << "%m" << currId << " = OpLabel\n";
3379 res << "%res"; // last result goes to %res
3381 res << "%rm" << currId;
3382 res << " = OpPhi %f32 " << mergeLeft.top() << " " << mergeRight.top() << "\n";
3383 std::ostringstream tag;
3384 tag << "%rm" << currId;
3386 value.push(tag.str());
3387 tag << " %m" << currId;
3389 valueLabel.push(tag.str());
3394 currId = idStack.top();
3402 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
3404 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
3405 ComputeShaderSpec spec1;
3406 ComputeShaderSpec spec2;
3407 ComputeShaderSpec spec3;
3408 ComputeShaderSpec spec4;
3409 ComputeShaderSpec spec5;
3410 de::Random rnd (deStringHash(group->getName()));
3411 const int numElements = 100;
3412 vector<float> inputFloats (numElements, 0);
3413 vector<float> outputFloats1 (numElements, 0);
3414 vector<float> outputFloats2 (numElements, 0);
3415 vector<float> outputFloats3 (numElements, 0);
3416 vector<float> outputFloats4 (numElements, 0);
3417 vector<float> outputFloats5 (numElements, 0);
3418 std::string codestring = "ABC";
3419 const int test4Width = 1024;
3421 // Build case 5 code string. Each iteration makes the hierarchy more complicated.
3422 // 9 iterations with (7, 24) parameters makes the hierarchy 8 deep with about 1500 lines of
3424 for (int i = 0, acc = 0; i < 9; i++)
3425 codestring = expandOpPhiCase5(codestring, acc, 7, 24);
3427 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
3429 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3430 floorAll(inputFloats);
3432 for (size_t ndx = 0; ndx < numElements; ++ndx)
3436 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
3437 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
3438 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
3441 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
3442 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
3444 int index4 = (int)deFloor(deAbs((float)ndx * inputFloats[ndx]));
3445 outputFloats4[ndx] = (float)(index4 % test4Width) * 10.0f + 5.0f;
3447 outputFloats5[ndx] = calcOpPhiCase5(inputFloats[ndx], codestring);
3451 string(getComputeAsmShaderPreamble()) +
3453 "OpSource GLSL 430\n"
3454 "OpName %main \"main\"\n"
3455 "OpName %id \"gl_GlobalInvocationID\"\n"
3457 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3459 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3461 "%id = OpVariable %uvec3ptr Input\n"
3462 "%zero = OpConstant %i32 0\n"
3463 "%three = OpConstant %u32 3\n"
3464 "%constf5p5 = OpConstant %f32 5.5\n"
3465 "%constf20p5 = OpConstant %f32 20.5\n"
3466 "%constf1p75 = OpConstant %f32 1.75\n"
3467 "%constf8p5 = OpConstant %f32 8.5\n"
3468 "%constf6p5 = OpConstant %f32 6.5\n"
3470 "%main = OpFunction %void None %voidf\n"
3471 "%entry = OpLabel\n"
3472 "%idval = OpLoad %uvec3 %id\n"
3473 "%x = OpCompositeExtract %u32 %idval 0\n"
3474 "%selector = OpUMod %u32 %x %three\n"
3475 " OpSelectionMerge %phi None\n"
3476 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
3478 // Case 1 before OpPhi.
3479 "%case1 = OpLabel\n"
3482 "%default = OpLabel\n"
3486 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
3487 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3488 "%inval = OpLoad %f32 %inloc\n"
3489 "%add = OpFAdd %f32 %inval %operand\n"
3490 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3491 " OpStore %outloc %add\n"
3494 // Case 0 after OpPhi.
3495 "%case0 = OpLabel\n"
3499 // Case 2 after OpPhi.
3500 "%case2 = OpLabel\n"
3504 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3505 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
3506 spec1.numWorkGroups = IVec3(numElements, 1, 1);
3508 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
3511 string(getComputeAsmShaderPreamble()) +
3513 "OpName %main \"main\"\n"
3514 "OpName %id \"gl_GlobalInvocationID\"\n"
3516 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3518 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3520 "%id = OpVariable %uvec3ptr Input\n"
3521 "%zero = OpConstant %i32 0\n"
3522 "%one = OpConstant %i32 1\n"
3523 "%three = OpConstant %i32 3\n"
3524 "%constf6p5 = OpConstant %f32 6.5\n"
3526 "%main = OpFunction %void None %voidf\n"
3527 "%entry = OpLabel\n"
3528 "%idval = OpLoad %uvec3 %id\n"
3529 "%x = OpCompositeExtract %u32 %idval 0\n"
3530 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3531 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3532 "%inval = OpLoad %f32 %inloc\n"
3536 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
3537 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
3538 "%step_next = OpIAdd %i32 %step %one\n"
3539 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
3540 "%still_loop = OpSLessThan %bool %step %three\n"
3541 " OpLoopMerge %exit %phi None\n"
3542 " OpBranchConditional %still_loop %phi %exit\n"
3545 " OpStore %outloc %accum\n"
3548 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3549 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
3550 spec2.numWorkGroups = IVec3(numElements, 1, 1);
3552 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
3555 string(getComputeAsmShaderPreamble()) +
3557 "OpName %main \"main\"\n"
3558 "OpName %id \"gl_GlobalInvocationID\"\n"
3560 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3562 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3564 "%f32ptr_f = OpTypePointer Function %f32\n"
3565 "%id = OpVariable %uvec3ptr Input\n"
3566 "%true = OpConstantTrue %bool\n"
3567 "%false = OpConstantFalse %bool\n"
3568 "%zero = OpConstant %i32 0\n"
3569 "%constf8p5 = OpConstant %f32 8.5\n"
3571 "%main = OpFunction %void None %voidf\n"
3572 "%entry = OpLabel\n"
3573 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
3574 "%idval = OpLoad %uvec3 %id\n"
3575 "%x = OpCompositeExtract %u32 %idval 0\n"
3576 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3577 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3578 "%a_init = OpLoad %f32 %inloc\n"
3579 "%b_init = OpLoad %f32 %b\n"
3583 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
3584 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
3585 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
3586 " OpLoopMerge %exit %phi None\n"
3587 " OpBranchConditional %still_loop %phi %exit\n"
3590 "%sub = OpFSub %f32 %a_next %b_next\n"
3591 " OpStore %outloc %sub\n"
3594 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3595 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
3596 spec3.numWorkGroups = IVec3(numElements, 1, 1);
3598 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
3601 "OpCapability Shader\n"
3602 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
3603 "OpMemoryModel Logical GLSL450\n"
3604 "OpEntryPoint GLCompute %main \"main\" %id\n"
3605 "OpExecutionMode %main LocalSize 1 1 1\n"
3607 "OpSource GLSL 430\n"
3608 "OpName %main \"main\"\n"
3609 "OpName %id \"gl_GlobalInvocationID\"\n"
3611 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3613 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3615 "%id = OpVariable %uvec3ptr Input\n"
3616 "%zero = OpConstant %i32 0\n"
3617 "%cimod = OpConstant %u32 " + generateIntWidth(test4Width) + "\n"
3619 + generateConstantDefinitions(test4Width) +
3621 "%main = OpFunction %void None %voidf\n"
3622 "%entry = OpLabel\n"
3623 "%idval = OpLoad %uvec3 %id\n"
3624 "%x = OpCompositeExtract %u32 %idval 0\n"
3625 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3626 "%inval = OpLoad %f32 %inloc\n"
3627 "%xf = OpConvertUToF %f32 %x\n"
3628 "%xm = OpFMul %f32 %xf %inval\n"
3629 "%xa = OpExtInst %f32 %ext FAbs %xm\n"
3630 "%xi = OpConvertFToU %u32 %xa\n"
3631 "%selector = OpUMod %u32 %xi %cimod\n"
3632 " OpSelectionMerge %phi None\n"
3633 " OpSwitch %selector %default "
3635 + generateSwitchCases(test4Width) +
3637 "%default = OpLabel\n"
3640 + generateSwitchTargets(test4Width) +
3643 "%result = OpPhi %f32"
3645 + generateOpPhiParams(test4Width) +
3647 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3648 " OpStore %outloc %result\n"
3652 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3653 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
3654 spec4.numWorkGroups = IVec3(numElements, 1, 1);
3656 group->addChild(new SpvAsmComputeShaderCase(testCtx, "wide", "OpPhi with a lot of parameters", spec4));
3659 "OpCapability Shader\n"
3660 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
3661 "OpMemoryModel Logical GLSL450\n"
3662 "OpEntryPoint GLCompute %main \"main\" %id\n"
3663 "OpExecutionMode %main LocalSize 1 1 1\n"
3664 "%code = OpString \"" + codestring + "\"\n"
3666 "OpSource GLSL 430\n"
3667 "OpName %main \"main\"\n"
3668 "OpName %id \"gl_GlobalInvocationID\"\n"
3670 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3672 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3674 "%id = OpVariable %uvec3ptr Input\n"
3675 "%zero = OpConstant %i32 0\n"
3676 "%f32_0 = OpConstant %f32 0.0\n"
3677 "%f32_0_5 = OpConstant %f32 0.5\n"
3678 "%f32_1 = OpConstant %f32 1.0\n"
3679 "%f32_1_5 = OpConstant %f32 1.5\n"
3680 "%f32_2 = OpConstant %f32 2.0\n"
3681 "%f32_3_5 = OpConstant %f32 3.5\n"
3682 "%f32_4 = OpConstant %f32 4.0\n"
3683 "%f32_7_5 = OpConstant %f32 7.5\n"
3684 "%f32_8 = OpConstant %f32 8.0\n"
3685 "%f32_15_5 = OpConstant %f32 15.5\n"
3686 "%f32_16 = OpConstant %f32 16.0\n"
3687 "%f32_31_5 = OpConstant %f32 31.5\n"
3688 "%f32_32 = OpConstant %f32 32.0\n"
3689 "%f32_63_5 = OpConstant %f32 63.5\n"
3690 "%f32_64 = OpConstant %f32 64.0\n"
3691 "%f32_127_5 = OpConstant %f32 127.5\n"
3692 "%f32_128 = OpConstant %f32 128.0\n"
3693 "%f32_256 = OpConstant %f32 256.0\n"
3695 "%main = OpFunction %void None %voidf\n"
3696 "%entry = OpLabel\n"
3697 "%idval = OpLoad %uvec3 %id\n"
3698 "%x = OpCompositeExtract %u32 %idval 0\n"
3699 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3700 "%inval = OpLoad %f32 %inloc\n"
3702 "%xabs = OpExtInst %f32 %ext FAbs %inval\n"
3703 "%x8 = OpFMod %f32 %xabs %f32_256\n"
3704 "%x7 = OpFMod %f32 %xabs %f32_128\n"
3705 "%x6 = OpFMod %f32 %xabs %f32_64\n"
3706 "%x5 = OpFMod %f32 %xabs %f32_32\n"
3707 "%x4 = OpFMod %f32 %xabs %f32_16\n"
3708 "%x3 = OpFMod %f32 %xabs %f32_8\n"
3709 "%x2 = OpFMod %f32 %xabs %f32_4\n"
3710 "%x1 = OpFMod %f32 %xabs %f32_2\n"
3712 "%b7 = OpFOrdGreaterThanEqual %bool %x8 %f32_127_5\n"
3713 "%b6 = OpFOrdGreaterThanEqual %bool %x7 %f32_63_5\n"
3714 "%b5 = OpFOrdGreaterThanEqual %bool %x6 %f32_31_5\n"
3715 "%b4 = OpFOrdGreaterThanEqual %bool %x5 %f32_15_5\n"
3716 "%b3 = OpFOrdGreaterThanEqual %bool %x4 %f32_7_5\n"
3717 "%b2 = OpFOrdGreaterThanEqual %bool %x3 %f32_3_5\n"
3718 "%b1 = OpFOrdGreaterThanEqual %bool %x2 %f32_1_5\n"
3719 "%b0 = OpFOrdGreaterThanEqual %bool %x1 %f32_0_5\n"
3721 + generateOpPhiCase5(codestring) +
3723 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3724 " OpStore %outloc %res\n"
3728 spec5.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3729 spec5.outputs.push_back(BufferSp(new Float32Buffer(outputFloats5)));
3730 spec5.numWorkGroups = IVec3(numElements, 1, 1);
3732 group->addChild(new SpvAsmComputeShaderCase(testCtx, "nested", "Stress OpPhi with a lot of nesting", spec5));
3734 createOpPhiVartypeTests(group, testCtx);
3736 return group.release();
3739 // Assembly code used for testing block order is based on GLSL source code:
3743 // layout(std140, set = 0, binding = 0) readonly buffer Input {
3744 // float elements[];
3746 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
3747 // float elements[];
3751 // uint x = gl_GlobalInvocationID.x;
3752 // output_data.elements[x] = input_data.elements[x];
3753 // if (x > uint(50)) {
3754 // switch (x % uint(3)) {
3755 // case 0: output_data.elements[x] += 1.5f; break;
3756 // case 1: output_data.elements[x] += 42.f; break;
3757 // case 2: output_data.elements[x] -= 27.f; break;
3761 // output_data.elements[x] = -input_data.elements[x];
3764 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
3766 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
3767 ComputeShaderSpec spec;
3768 de::Random rnd (deStringHash(group->getName()));
3769 const int numElements = 100;
3770 vector<float> inputFloats (numElements, 0);
3771 vector<float> outputFloats (numElements, 0);
3773 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3775 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3776 floorAll(inputFloats);
3778 for (size_t ndx = 0; ndx <= 50; ++ndx)
3779 outputFloats[ndx] = -inputFloats[ndx];
3781 for (size_t ndx = 51; ndx < numElements; ++ndx)
3785 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
3786 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
3787 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
3793 string(getComputeAsmShaderPreamble()) +
3795 "OpSource GLSL 430\n"
3796 "OpName %main \"main\"\n"
3797 "OpName %id \"gl_GlobalInvocationID\"\n"
3799 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3801 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3803 "%u32ptr = OpTypePointer Function %u32\n"
3804 "%u32ptr_input = OpTypePointer Input %u32\n"
3806 + string(getComputeAsmInputOutputBuffer()) +
3808 "%id = OpVariable %uvec3ptr Input\n"
3809 "%zero = OpConstant %i32 0\n"
3810 "%const3 = OpConstant %u32 3\n"
3811 "%const50 = OpConstant %u32 50\n"
3812 "%constf1p5 = OpConstant %f32 1.5\n"
3813 "%constf27 = OpConstant %f32 27.0\n"
3814 "%constf42 = OpConstant %f32 42.0\n"
3816 "%main = OpFunction %void None %voidf\n"
3819 "%entry = OpLabel\n"
3821 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
3822 "%xvar = OpVariable %u32ptr Function\n"
3823 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
3824 "%x = OpLoad %u32 %xptr\n"
3825 " OpStore %xvar %x\n"
3827 "%cmp = OpUGreaterThan %bool %x %const50\n"
3828 " OpSelectionMerge %if_merge None\n"
3829 " OpBranchConditional %cmp %if_true %if_false\n"
3831 // False branch for if-statement: placed in the middle of switch cases and before true branch.
3832 "%if_false = OpLabel\n"
3833 "%x_f = OpLoad %u32 %xvar\n"
3834 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
3835 "%inval_f = OpLoad %f32 %inloc_f\n"
3836 "%negate = OpFNegate %f32 %inval_f\n"
3837 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
3838 " OpStore %outloc_f %negate\n"
3839 " OpBranch %if_merge\n"
3841 // Merge block for if-statement: placed in the middle of true and false branch.
3842 "%if_merge = OpLabel\n"
3845 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
3846 "%if_true = OpLabel\n"
3847 "%xval_t = OpLoad %u32 %xvar\n"
3848 "%mod = OpUMod %u32 %xval_t %const3\n"
3849 " OpSelectionMerge %switch_merge None\n"
3850 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
3852 // Merge block for switch-statement: placed before the case
3853 // bodies. But it must follow OpSwitch which dominates it.
3854 "%switch_merge = OpLabel\n"
3855 " OpBranch %if_merge\n"
3857 // Case 1 for switch-statement: placed before case 0.
3858 // It must follow the OpSwitch that dominates it.
3859 "%case1 = OpLabel\n"
3860 "%x_1 = OpLoad %u32 %xvar\n"
3861 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
3862 "%inval_1 = OpLoad %f32 %inloc_1\n"
3863 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
3864 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
3865 " OpStore %outloc_1 %addf42\n"
3866 " OpBranch %switch_merge\n"
3868 // Case 2 for switch-statement.
3869 "%case2 = OpLabel\n"
3870 "%x_2 = OpLoad %u32 %xvar\n"
3871 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
3872 "%inval_2 = OpLoad %f32 %inloc_2\n"
3873 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
3874 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
3875 " OpStore %outloc_2 %subf27\n"
3876 " OpBranch %switch_merge\n"
3878 // Default case for switch-statement: placed in the middle of normal cases.
3879 "%default = OpLabel\n"
3880 " OpBranch %switch_merge\n"
3882 // Case 0 for switch-statement: out of order.
3883 "%case0 = OpLabel\n"
3884 "%x_0 = OpLoad %u32 %xvar\n"
3885 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
3886 "%inval_0 = OpLoad %f32 %inloc_0\n"
3887 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
3888 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
3889 " OpStore %outloc_0 %addf1p5\n"
3890 " OpBranch %switch_merge\n"
3893 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3894 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3895 spec.numWorkGroups = IVec3(numElements, 1, 1);
3897 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
3899 return group.release();
3902 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
3904 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
3905 ComputeShaderSpec spec1;
3906 ComputeShaderSpec spec2;
3907 de::Random rnd (deStringHash(group->getName()));
3908 const int numElements = 100;
3909 vector<float> inputFloats (numElements, 0);
3910 vector<float> outputFloats1 (numElements, 0);
3911 vector<float> outputFloats2 (numElements, 0);
3912 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
3914 for (size_t ndx = 0; ndx < numElements; ++ndx)
3916 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
3917 outputFloats2[ndx] = -inputFloats[ndx];
3920 const string assembly(
3921 "OpCapability Shader\n"
3922 "OpCapability ClipDistance\n"
3923 "OpMemoryModel Logical GLSL450\n"
3924 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
3925 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
3926 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
3927 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
3928 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
3929 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
3931 "OpName %comp_main1 \"entrypoint1\"\n"
3932 "OpName %comp_main2 \"entrypoint2\"\n"
3933 "OpName %vert_main \"entrypoint2\"\n"
3934 "OpName %id \"gl_GlobalInvocationID\"\n"
3935 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
3936 "OpName %vertexIndex \"gl_VertexIndex\"\n"
3937 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
3938 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
3939 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
3940 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
3942 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3943 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
3944 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
3945 "OpDecorate %vert_builtin_st Block\n"
3946 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
3947 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
3948 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
3950 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3952 "%zero = OpConstant %i32 0\n"
3953 "%one = OpConstant %u32 1\n"
3954 "%c_f32_1 = OpConstant %f32 1\n"
3956 "%i32inputptr = OpTypePointer Input %i32\n"
3957 "%vec4 = OpTypeVector %f32 4\n"
3958 "%vec4ptr = OpTypePointer Output %vec4\n"
3959 "%f32arr1 = OpTypeArray %f32 %one\n"
3960 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
3961 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
3962 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
3964 "%id = OpVariable %uvec3ptr Input\n"
3965 "%vertexIndex = OpVariable %i32inputptr Input\n"
3966 "%instanceIndex = OpVariable %i32inputptr Input\n"
3967 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
3969 // gl_Position = vec4(1.);
3970 "%vert_main = OpFunction %void None %voidf\n"
3971 "%vert_entry = OpLabel\n"
3972 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
3973 " OpStore %position %c_vec4_1\n"
3978 "%comp_main1 = OpFunction %void None %voidf\n"
3979 "%comp1_entry = OpLabel\n"
3980 "%idval1 = OpLoad %uvec3 %id\n"
3981 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
3982 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
3983 "%inval1 = OpLoad %f32 %inloc1\n"
3984 "%add = OpFAdd %f32 %inval1 %inval1\n"
3985 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
3986 " OpStore %outloc1 %add\n"
3991 "%comp_main2 = OpFunction %void None %voidf\n"
3992 "%comp2_entry = OpLabel\n"
3993 "%idval2 = OpLoad %uvec3 %id\n"
3994 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
3995 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
3996 "%inval2 = OpLoad %f32 %inloc2\n"
3997 "%neg = OpFNegate %f32 %inval2\n"
3998 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
3999 " OpStore %outloc2 %neg\n"
4001 " OpFunctionEnd\n");
4003 spec1.assembly = assembly;
4004 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4005 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
4006 spec1.numWorkGroups = IVec3(numElements, 1, 1);
4007 spec1.entryPoint = "entrypoint1";
4009 spec2.assembly = assembly;
4010 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4011 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
4012 spec2.numWorkGroups = IVec3(numElements, 1, 1);
4013 spec2.entryPoint = "entrypoint2";
4015 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
4016 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
4018 return group.release();
4021 inline std::string makeLongUTF8String (size_t num4ByteChars)
4023 // An example of a longest valid UTF-8 character. Be explicit about the
4024 // character type because Microsoft compilers can otherwise interpret the
4025 // character string as being over wide (16-bit) characters. Ideally, we
4026 // would just use a C++11 UTF-8 string literal, but we want to support older
4027 // Microsoft compilers.
4028 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
4029 std::string longString;
4030 longString.reserve(num4ByteChars * 4);
4031 for (size_t count = 0; count < num4ByteChars; count++)
4033 longString += earthAfrica;
4038 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
4040 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
4041 vector<CaseParameter> cases;
4042 de::Random rnd (deStringHash(group->getName()));
4043 const int numElements = 100;
4044 vector<float> positiveFloats (numElements, 0);
4045 vector<float> negativeFloats (numElements, 0);
4046 const StringTemplate shaderTemplate (
4047 "OpCapability Shader\n"
4048 "OpMemoryModel Logical GLSL450\n"
4050 "OpEntryPoint GLCompute %main \"main\" %id\n"
4051 "OpExecutionMode %main LocalSize 1 1 1\n"
4055 "OpName %main \"main\"\n"
4056 "OpName %id \"gl_GlobalInvocationID\"\n"
4058 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4060 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4062 "%id = OpVariable %uvec3ptr Input\n"
4063 "%zero = OpConstant %i32 0\n"
4065 "%main = OpFunction %void None %voidf\n"
4066 "%label = OpLabel\n"
4067 "%idval = OpLoad %uvec3 %id\n"
4068 "%x = OpCompositeExtract %u32 %idval 0\n"
4069 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4070 "%inval = OpLoad %f32 %inloc\n"
4071 "%neg = OpFNegate %f32 %inval\n"
4072 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4073 " OpStore %outloc %neg\n"
4075 " OpFunctionEnd\n");
4077 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
4078 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
4079 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
4080 "OpSource GLSL 430 %fname"));
4081 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
4082 "OpSource GLSL 430 %fname"));
4083 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
4084 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
4085 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
4086 "OpSource GLSL 430 %fname \"\""));
4087 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
4088 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
4089 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
4090 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
4091 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
4092 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
4093 "OpSourceContinued \"id main() {}\""));
4094 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
4095 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4096 "OpSourceContinued \"\""));
4097 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
4098 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4099 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
4100 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
4101 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4102 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
4103 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
4104 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
4105 "OpSourceContinued \"void\"\n"
4106 "OpSourceContinued \"main()\"\n"
4107 "OpSourceContinued \"{}\""));
4108 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
4109 "OpSource GLSL 430 %fname \"\"\n"
4110 "OpSourceContinued \"#version 430\nvoid main() {}\""));
4112 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4114 for (size_t ndx = 0; ndx < numElements; ++ndx)
4115 negativeFloats[ndx] = -positiveFloats[ndx];
4117 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4119 map<string, string> specializations;
4120 ComputeShaderSpec spec;
4122 specializations["SOURCE"] = cases[caseNdx].param;
4123 spec.assembly = shaderTemplate.specialize(specializations);
4124 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4125 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4126 spec.numWorkGroups = IVec3(numElements, 1, 1);
4128 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4131 return group.release();
4134 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
4136 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
4137 vector<CaseParameter> cases;
4138 de::Random rnd (deStringHash(group->getName()));
4139 const int numElements = 100;
4140 vector<float> inputFloats (numElements, 0);
4141 vector<float> outputFloats (numElements, 0);
4142 const StringTemplate shaderTemplate (
4143 string(getComputeAsmShaderPreamble()) +
4145 "OpSourceExtension \"${EXTENSION}\"\n"
4147 "OpName %main \"main\"\n"
4148 "OpName %id \"gl_GlobalInvocationID\"\n"
4150 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4152 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4154 "%id = OpVariable %uvec3ptr Input\n"
4155 "%zero = OpConstant %i32 0\n"
4157 "%main = OpFunction %void None %voidf\n"
4158 "%label = OpLabel\n"
4159 "%idval = OpLoad %uvec3 %id\n"
4160 "%x = OpCompositeExtract %u32 %idval 0\n"
4161 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4162 "%inval = OpLoad %f32 %inloc\n"
4163 "%neg = OpFNegate %f32 %inval\n"
4164 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4165 " OpStore %outloc %neg\n"
4167 " OpFunctionEnd\n");
4169 cases.push_back(CaseParameter("empty_extension", ""));
4170 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
4171 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
4172 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
4173 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
4175 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
4177 for (size_t ndx = 0; ndx < numElements; ++ndx)
4178 outputFloats[ndx] = -inputFloats[ndx];
4180 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4182 map<string, string> specializations;
4183 ComputeShaderSpec spec;
4185 specializations["EXTENSION"] = cases[caseNdx].param;
4186 spec.assembly = shaderTemplate.specialize(specializations);
4187 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4188 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4189 spec.numWorkGroups = IVec3(numElements, 1, 1);
4191 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4194 return group.release();
4197 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
4198 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
4200 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
4201 vector<CaseParameter> cases;
4202 de::Random rnd (deStringHash(group->getName()));
4203 const int numElements = 100;
4204 vector<float> positiveFloats (numElements, 0);
4205 vector<float> negativeFloats (numElements, 0);
4206 const StringTemplate shaderTemplate (
4207 string(getComputeAsmShaderPreamble()) +
4209 "OpSource GLSL 430\n"
4210 "OpName %main \"main\"\n"
4211 "OpName %id \"gl_GlobalInvocationID\"\n"
4213 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4215 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4216 "%uvec2 = OpTypeVector %u32 2\n"
4217 "%bvec3 = OpTypeVector %bool 3\n"
4218 "%fvec4 = OpTypeVector %f32 4\n"
4219 "%fmat33 = OpTypeMatrix %fvec3 3\n"
4220 "%const100 = OpConstant %u32 100\n"
4221 "%uarr100 = OpTypeArray %i32 %const100\n"
4222 "%struct = OpTypeStruct %f32 %i32 %u32\n"
4223 "%pointer = OpTypePointer Function %i32\n"
4224 + string(getComputeAsmInputOutputBuffer()) +
4226 "%null = OpConstantNull ${TYPE}\n"
4228 "%id = OpVariable %uvec3ptr Input\n"
4229 "%zero = OpConstant %i32 0\n"
4231 "%main = OpFunction %void None %voidf\n"
4232 "%label = OpLabel\n"
4233 "%idval = OpLoad %uvec3 %id\n"
4234 "%x = OpCompositeExtract %u32 %idval 0\n"
4235 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4236 "%inval = OpLoad %f32 %inloc\n"
4237 "%neg = OpFNegate %f32 %inval\n"
4238 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4239 " OpStore %outloc %neg\n"
4241 " OpFunctionEnd\n");
4243 cases.push_back(CaseParameter("bool", "%bool"));
4244 cases.push_back(CaseParameter("sint32", "%i32"));
4245 cases.push_back(CaseParameter("uint32", "%u32"));
4246 cases.push_back(CaseParameter("float32", "%f32"));
4247 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
4248 cases.push_back(CaseParameter("vec3bool", "%bvec3"));
4249 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
4250 cases.push_back(CaseParameter("matrix", "%fmat33"));
4251 cases.push_back(CaseParameter("array", "%uarr100"));
4252 cases.push_back(CaseParameter("struct", "%struct"));
4253 cases.push_back(CaseParameter("pointer", "%pointer"));
4255 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4257 for (size_t ndx = 0; ndx < numElements; ++ndx)
4258 negativeFloats[ndx] = -positiveFloats[ndx];
4260 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4262 map<string, string> specializations;
4263 ComputeShaderSpec spec;
4265 specializations["TYPE"] = cases[caseNdx].param;
4266 spec.assembly = shaderTemplate.specialize(specializations);
4267 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4268 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4269 spec.numWorkGroups = IVec3(numElements, 1, 1);
4271 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4274 return group.release();
4277 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
4278 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
4280 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
4281 vector<CaseParameter> cases;
4282 de::Random rnd (deStringHash(group->getName()));
4283 const int numElements = 100;
4284 vector<float> positiveFloats (numElements, 0);
4285 vector<float> negativeFloats (numElements, 0);
4286 const StringTemplate shaderTemplate (
4287 string(getComputeAsmShaderPreamble()) +
4289 "OpSource GLSL 430\n"
4290 "OpName %main \"main\"\n"
4291 "OpName %id \"gl_GlobalInvocationID\"\n"
4293 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4295 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4297 "%id = OpVariable %uvec3ptr Input\n"
4298 "%zero = OpConstant %i32 0\n"
4302 "%main = OpFunction %void None %voidf\n"
4303 "%label = OpLabel\n"
4304 "%idval = OpLoad %uvec3 %id\n"
4305 "%x = OpCompositeExtract %u32 %idval 0\n"
4306 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4307 "%inval = OpLoad %f32 %inloc\n"
4308 "%neg = OpFNegate %f32 %inval\n"
4309 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4310 " OpStore %outloc %neg\n"
4312 " OpFunctionEnd\n");
4314 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
4315 "%const = OpConstantComposite %uvec3 %five %zero %five"));
4316 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
4317 "%ten = OpConstant %f32 10.\n"
4318 "%fzero = OpConstant %f32 0.\n"
4319 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
4320 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
4321 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
4322 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
4323 "%fzero = OpConstant %f32 0.\n"
4324 "%one = OpConstant %f32 1.\n"
4325 "%point5 = OpConstant %f32 0.5\n"
4326 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
4327 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
4328 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
4329 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
4330 "%st2 = OpTypeStruct %i32 %i32\n"
4331 "%struct = OpTypeStruct %st1 %st2\n"
4332 "%point5 = OpConstant %f32 0.5\n"
4333 "%one = OpConstant %u32 1\n"
4334 "%ten = OpConstant %i32 10\n"
4335 "%st1val = OpConstantComposite %st1 %one %point5\n"
4336 "%st2val = OpConstantComposite %st2 %ten %ten\n"
4337 "%const = OpConstantComposite %struct %st1val %st2val"));
4339 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4341 for (size_t ndx = 0; ndx < numElements; ++ndx)
4342 negativeFloats[ndx] = -positiveFloats[ndx];
4344 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4346 map<string, string> specializations;
4347 ComputeShaderSpec spec;
4349 specializations["CONSTANT"] = cases[caseNdx].param;
4350 spec.assembly = shaderTemplate.specialize(specializations);
4351 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4352 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4353 spec.numWorkGroups = IVec3(numElements, 1, 1);
4355 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4358 return group.release();
4361 // Creates a floating point number with the given exponent, and significand
4362 // bits set. It can only create normalized numbers. Only the least significant
4363 // 24 bits of the significand will be examined. The final bit of the
4364 // significand will also be ignored. This allows alignment to be written
4365 // similarly to C99 hex-floats.
4366 // For example if you wanted to write 0x1.7f34p-12 you would call
4367 // constructNormalizedFloat(-12, 0x7f3400)
4368 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
4372 for (deInt32 idx = 0; idx < 23; ++idx)
4374 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
4378 return std::ldexp(f, exponent);
4381 // Compare instruction for the OpQuantizeF16 compute exact case.
4382 // Returns true if the output is what is expected from the test case.
4383 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
4385 if (outputAllocs.size() != 1)
4388 // Only size is needed because we cannot compare Nans.
4389 size_t byteSize = expectedOutputs[0]->getByteSize();
4391 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
4393 if (byteSize != 4*sizeof(float)) {
4397 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
4398 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
4403 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
4404 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
4409 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
4410 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
4415 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
4416 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
4423 // Checks that every output from a test-case is a float NaN.
4424 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
4426 if (outputAllocs.size() != 1)
4429 // Only size is needed because we cannot compare Nans.
4430 size_t byteSize = expectedOutputs[0]->getByteSize();
4432 const float* const output_as_float = static_cast<const float*>(outputAllocs[0]->getHostPtr());
4434 for (size_t idx = 0; idx < byteSize / sizeof(float); ++idx)
4436 if (!deFloatIsNaN(output_as_float[idx]))
4445 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
4446 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
4448 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
4450 const std::string shader (
4451 string(getComputeAsmShaderPreamble()) +
4453 "OpSource GLSL 430\n"
4454 "OpName %main \"main\"\n"
4455 "OpName %id \"gl_GlobalInvocationID\"\n"
4457 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4459 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4461 "%id = OpVariable %uvec3ptr Input\n"
4462 "%zero = OpConstant %i32 0\n"
4464 "%main = OpFunction %void None %voidf\n"
4465 "%label = OpLabel\n"
4466 "%idval = OpLoad %uvec3 %id\n"
4467 "%x = OpCompositeExtract %u32 %idval 0\n"
4468 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4469 "%inval = OpLoad %f32 %inloc\n"
4470 "%quant = OpQuantizeToF16 %f32 %inval\n"
4471 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4472 " OpStore %outloc %quant\n"
4474 " OpFunctionEnd\n");
4477 ComputeShaderSpec spec;
4478 const deUint32 numElements = 100;
4479 vector<float> infinities;
4480 vector<float> results;
4482 infinities.reserve(numElements);
4483 results.reserve(numElements);
4485 for (size_t idx = 0; idx < numElements; ++idx)
4490 infinities.push_back(std::numeric_limits<float>::infinity());
4491 results.push_back(std::numeric_limits<float>::infinity());
4494 infinities.push_back(-std::numeric_limits<float>::infinity());
4495 results.push_back(-std::numeric_limits<float>::infinity());
4498 infinities.push_back(std::ldexp(1.0f, 16));
4499 results.push_back(std::numeric_limits<float>::infinity());
4502 infinities.push_back(std::ldexp(-1.0f, 32));
4503 results.push_back(-std::numeric_limits<float>::infinity());
4508 spec.assembly = shader;
4509 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
4510 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
4511 spec.numWorkGroups = IVec3(numElements, 1, 1);
4513 group->addChild(new SpvAsmComputeShaderCase(
4514 testCtx, "infinities", "Check that infinities propagated and created", spec));
4518 ComputeShaderSpec spec;
4520 const deUint32 numElements = 100;
4522 nans.reserve(numElements);
4524 for (size_t idx = 0; idx < numElements; ++idx)
4528 nans.push_back(std::numeric_limits<float>::quiet_NaN());
4532 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
4536 spec.assembly = shader;
4537 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
4538 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
4539 spec.numWorkGroups = IVec3(numElements, 1, 1);
4540 spec.verifyIO = &compareNan;
4542 group->addChild(new SpvAsmComputeShaderCase(
4543 testCtx, "propagated_nans", "Check that nans are propagated", spec));
4547 ComputeShaderSpec spec;
4548 vector<float> small;
4549 vector<float> zeros;
4550 const deUint32 numElements = 100;
4552 small.reserve(numElements);
4553 zeros.reserve(numElements);
4555 for (size_t idx = 0; idx < numElements; ++idx)
4560 small.push_back(0.f);
4561 zeros.push_back(0.f);
4564 small.push_back(-0.f);
4565 zeros.push_back(-0.f);
4568 small.push_back(std::ldexp(1.0f, -16));
4569 zeros.push_back(0.f);
4572 small.push_back(std::ldexp(-1.0f, -32));
4573 zeros.push_back(-0.f);
4576 small.push_back(std::ldexp(1.0f, -127));
4577 zeros.push_back(0.f);
4580 small.push_back(-std::ldexp(1.0f, -128));
4581 zeros.push_back(-0.f);
4586 spec.assembly = shader;
4587 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
4588 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
4589 spec.numWorkGroups = IVec3(numElements, 1, 1);
4591 group->addChild(new SpvAsmComputeShaderCase(
4592 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
4596 ComputeShaderSpec spec;
4597 vector<float> exact;
4598 const deUint32 numElements = 200;
4600 exact.reserve(numElements);
4602 for (size_t idx = 0; idx < numElements; ++idx)
4603 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
4605 spec.assembly = shader;
4606 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
4607 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
4608 spec.numWorkGroups = IVec3(numElements, 1, 1);
4610 group->addChild(new SpvAsmComputeShaderCase(
4611 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
4615 ComputeShaderSpec spec;
4616 vector<float> inputs;
4617 const deUint32 numElements = 4;
4619 inputs.push_back(constructNormalizedFloat(8, 0x300300));
4620 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
4621 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
4622 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
4624 spec.assembly = shader;
4625 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
4626 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4627 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
4628 spec.numWorkGroups = IVec3(numElements, 1, 1);
4630 group->addChild(new SpvAsmComputeShaderCase(
4631 testCtx, "rounded", "Check that are rounded when needed", spec));
4634 return group.release();
4637 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
4639 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
4641 const std::string shader (
4642 string(getComputeAsmShaderPreamble()) +
4644 "OpName %main \"main\"\n"
4645 "OpName %id \"gl_GlobalInvocationID\"\n"
4647 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4649 "OpDecorate %sc_0 SpecId 0\n"
4650 "OpDecorate %sc_1 SpecId 1\n"
4651 "OpDecorate %sc_2 SpecId 2\n"
4652 "OpDecorate %sc_3 SpecId 3\n"
4653 "OpDecorate %sc_4 SpecId 4\n"
4654 "OpDecorate %sc_5 SpecId 5\n"
4656 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4658 "%id = OpVariable %uvec3ptr Input\n"
4659 "%zero = OpConstant %i32 0\n"
4660 "%c_u32_6 = OpConstant %u32 6\n"
4662 "%sc_0 = OpSpecConstant %f32 0.\n"
4663 "%sc_1 = OpSpecConstant %f32 0.\n"
4664 "%sc_2 = OpSpecConstant %f32 0.\n"
4665 "%sc_3 = OpSpecConstant %f32 0.\n"
4666 "%sc_4 = OpSpecConstant %f32 0.\n"
4667 "%sc_5 = OpSpecConstant %f32 0.\n"
4669 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
4670 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
4671 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
4672 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
4673 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
4674 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
4676 "%main = OpFunction %void None %voidf\n"
4677 "%label = OpLabel\n"
4678 "%idval = OpLoad %uvec3 %id\n"
4679 "%x = OpCompositeExtract %u32 %idval 0\n"
4680 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4681 "%selector = OpUMod %u32 %x %c_u32_6\n"
4682 " OpSelectionMerge %exit None\n"
4683 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
4685 "%case0 = OpLabel\n"
4686 " OpStore %outloc %sc_0_quant\n"
4689 "%case1 = OpLabel\n"
4690 " OpStore %outloc %sc_1_quant\n"
4693 "%case2 = OpLabel\n"
4694 " OpStore %outloc %sc_2_quant\n"
4697 "%case3 = OpLabel\n"
4698 " OpStore %outloc %sc_3_quant\n"
4701 "%case4 = OpLabel\n"
4702 " OpStore %outloc %sc_4_quant\n"
4705 "%case5 = OpLabel\n"
4706 " OpStore %outloc %sc_5_quant\n"
4712 " OpFunctionEnd\n");
4715 ComputeShaderSpec spec;
4716 const deUint8 numCases = 4;
4717 vector<float> inputs (numCases, 0.f);
4718 vector<float> outputs;
4720 spec.assembly = shader;
4721 spec.numWorkGroups = IVec3(numCases, 1, 1);
4723 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
4724 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
4725 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
4726 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
4728 outputs.push_back(std::numeric_limits<float>::infinity());
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());
4733 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4734 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4736 group->addChild(new SpvAsmComputeShaderCase(
4737 testCtx, "infinities", "Check that infinities propagated and created", spec));
4741 ComputeShaderSpec spec;
4742 const deUint8 numCases = 2;
4743 vector<float> inputs (numCases, 0.f);
4744 vector<float> outputs;
4746 spec.assembly = shader;
4747 spec.numWorkGroups = IVec3(numCases, 1, 1);
4748 spec.verifyIO = &compareNan;
4750 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
4751 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
4753 for (deUint8 idx = 0; idx < numCases; ++idx)
4754 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
4756 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4757 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4759 group->addChild(new SpvAsmComputeShaderCase(
4760 testCtx, "propagated_nans", "Check that nans are propagated", spec));
4764 ComputeShaderSpec spec;
4765 const deUint8 numCases = 6;
4766 vector<float> inputs (numCases, 0.f);
4767 vector<float> outputs;
4769 spec.assembly = shader;
4770 spec.numWorkGroups = IVec3(numCases, 1, 1);
4772 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
4773 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
4774 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
4775 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
4776 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
4777 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
4779 outputs.push_back(0.f);
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);
4786 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4787 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4789 group->addChild(new SpvAsmComputeShaderCase(
4790 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
4794 ComputeShaderSpec spec;
4795 const deUint8 numCases = 6;
4796 vector<float> inputs (numCases, 0.f);
4797 vector<float> outputs;
4799 spec.assembly = shader;
4800 spec.numWorkGroups = IVec3(numCases, 1, 1);
4802 for (deUint8 idx = 0; idx < 6; ++idx)
4804 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
4805 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
4806 outputs.push_back(f);
4809 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4810 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4812 group->addChild(new SpvAsmComputeShaderCase(
4813 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
4817 ComputeShaderSpec spec;
4818 const deUint8 numCases = 4;
4819 vector<float> inputs (numCases, 0.f);
4820 vector<float> outputs;
4822 spec.assembly = shader;
4823 spec.numWorkGroups = IVec3(numCases, 1, 1);
4824 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
4826 outputs.push_back(constructNormalizedFloat(8, 0x300300));
4827 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
4828 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
4829 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
4831 for (deUint8 idx = 0; idx < numCases; ++idx)
4832 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
4834 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4835 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4837 group->addChild(new SpvAsmComputeShaderCase(
4838 testCtx, "rounded", "Check that are rounded when needed", spec));
4841 return group.release();
4844 // Checks that constant null/composite values can be used in computation.
4845 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
4847 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
4848 ComputeShaderSpec spec;
4849 de::Random rnd (deStringHash(group->getName()));
4850 const int numElements = 100;
4851 vector<float> positiveFloats (numElements, 0);
4852 vector<float> negativeFloats (numElements, 0);
4854 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4856 for (size_t ndx = 0; ndx < numElements; ++ndx)
4857 negativeFloats[ndx] = -positiveFloats[ndx];
4860 "OpCapability Shader\n"
4861 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
4862 "OpMemoryModel Logical GLSL450\n"
4863 "OpEntryPoint GLCompute %main \"main\" %id\n"
4864 "OpExecutionMode %main LocalSize 1 1 1\n"
4866 "OpSource GLSL 430\n"
4867 "OpName %main \"main\"\n"
4868 "OpName %id \"gl_GlobalInvocationID\"\n"
4870 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4872 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4874 "%fmat = OpTypeMatrix %fvec3 3\n"
4875 "%ten = OpConstant %u32 10\n"
4876 "%f32arr10 = OpTypeArray %f32 %ten\n"
4877 "%fst = OpTypeStruct %f32 %f32\n"
4879 + string(getComputeAsmInputOutputBuffer()) +
4881 "%id = OpVariable %uvec3ptr Input\n"
4882 "%zero = OpConstant %i32 0\n"
4884 // Create a bunch of null values
4885 "%unull = OpConstantNull %u32\n"
4886 "%fnull = OpConstantNull %f32\n"
4887 "%vnull = OpConstantNull %fvec3\n"
4888 "%mnull = OpConstantNull %fmat\n"
4889 "%anull = OpConstantNull %f32arr10\n"
4890 "%snull = OpConstantComposite %fst %fnull %fnull\n"
4892 "%main = OpFunction %void None %voidf\n"
4893 "%label = OpLabel\n"
4894 "%idval = OpLoad %uvec3 %id\n"
4895 "%x = OpCompositeExtract %u32 %idval 0\n"
4896 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4897 "%inval = OpLoad %f32 %inloc\n"
4898 "%neg = OpFNegate %f32 %inval\n"
4900 // Get the abs() of (a certain element of) those null values
4901 "%unull_cov = OpConvertUToF %f32 %unull\n"
4902 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
4903 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
4904 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
4905 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
4906 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
4907 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
4908 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
4909 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
4910 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
4911 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
4914 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
4915 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
4916 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
4917 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
4918 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
4919 "%final = OpFAdd %f32 %add5 %snull_abs\n"
4921 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4922 " OpStore %outloc %final\n" // write to output
4925 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4926 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4927 spec.numWorkGroups = IVec3(numElements, 1, 1);
4929 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
4931 return group.release();
4934 // Assembly code used for testing loop control is based on GLSL source code:
4937 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4938 // float elements[];
4940 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4941 // float elements[];
4945 // uint x = gl_GlobalInvocationID.x;
4946 // output_data.elements[x] = input_data.elements[x];
4947 // for (uint i = 0; i < 4; ++i)
4948 // output_data.elements[x] += 1.f;
4950 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
4952 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
4953 vector<CaseParameter> cases;
4954 de::Random rnd (deStringHash(group->getName()));
4955 const int numElements = 100;
4956 vector<float> inputFloats (numElements, 0);
4957 vector<float> outputFloats (numElements, 0);
4958 const StringTemplate shaderTemplate (
4959 string(getComputeAsmShaderPreamble()) +
4961 "OpSource GLSL 430\n"
4962 "OpName %main \"main\"\n"
4963 "OpName %id \"gl_GlobalInvocationID\"\n"
4965 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4967 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4969 "%u32ptr = OpTypePointer Function %u32\n"
4971 "%id = OpVariable %uvec3ptr Input\n"
4972 "%zero = OpConstant %i32 0\n"
4973 "%uzero = OpConstant %u32 0\n"
4974 "%one = OpConstant %i32 1\n"
4975 "%constf1 = OpConstant %f32 1.0\n"
4976 "%four = OpConstant %u32 4\n"
4978 "%main = OpFunction %void None %voidf\n"
4979 "%entry = OpLabel\n"
4980 "%i = OpVariable %u32ptr Function\n"
4981 " OpStore %i %uzero\n"
4983 "%idval = OpLoad %uvec3 %id\n"
4984 "%x = OpCompositeExtract %u32 %idval 0\n"
4985 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4986 "%inval = OpLoad %f32 %inloc\n"
4987 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4988 " OpStore %outloc %inval\n"
4989 " OpBranch %loop_entry\n"
4991 "%loop_entry = OpLabel\n"
4992 "%i_val = OpLoad %u32 %i\n"
4993 "%cmp_lt = OpULessThan %bool %i_val %four\n"
4994 " OpLoopMerge %loop_merge %loop_body ${CONTROL}\n"
4995 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
4996 "%loop_body = OpLabel\n"
4997 "%outval = OpLoad %f32 %outloc\n"
4998 "%addf1 = OpFAdd %f32 %outval %constf1\n"
4999 " OpStore %outloc %addf1\n"
5000 "%new_i = OpIAdd %u32 %i_val %one\n"
5001 " OpStore %i %new_i\n"
5002 " OpBranch %loop_entry\n"
5003 "%loop_merge = OpLabel\n"
5005 " OpFunctionEnd\n");
5007 cases.push_back(CaseParameter("none", "None"));
5008 cases.push_back(CaseParameter("unroll", "Unroll"));
5009 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
5010 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
5012 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5014 for (size_t ndx = 0; ndx < numElements; ++ndx)
5015 outputFloats[ndx] = inputFloats[ndx] + 4.f;
5017 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5019 map<string, string> specializations;
5020 ComputeShaderSpec spec;
5022 specializations["CONTROL"] = cases[caseNdx].param;
5023 spec.assembly = shaderTemplate.specialize(specializations);
5024 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5025 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5026 spec.numWorkGroups = IVec3(numElements, 1, 1);
5028 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5031 group->addChild(new SpvAsmLoopControlDependencyLengthCase(testCtx, "dependency_length", "dependency_length"));
5032 group->addChild(new SpvAsmLoopControlDependencyInfiniteCase(testCtx, "dependency_infinite", "dependency_infinite"));
5034 return group.release();
5037 // Assembly code used for testing selection control is based on GLSL source code:
5040 // layout(std140, set = 0, binding = 0) readonly buffer Input {
5041 // float elements[];
5043 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
5044 // float elements[];
5048 // uint x = gl_GlobalInvocationID.x;
5049 // float val = input_data.elements[x];
5051 // output_data.elements[x] = val + 1.f;
5053 // output_data.elements[x] = val - 1.f;
5055 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
5057 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
5058 vector<CaseParameter> cases;
5059 de::Random rnd (deStringHash(group->getName()));
5060 const int numElements = 100;
5061 vector<float> inputFloats (numElements, 0);
5062 vector<float> outputFloats (numElements, 0);
5063 const StringTemplate shaderTemplate (
5064 string(getComputeAsmShaderPreamble()) +
5066 "OpSource GLSL 430\n"
5067 "OpName %main \"main\"\n"
5068 "OpName %id \"gl_GlobalInvocationID\"\n"
5070 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5072 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5074 "%id = OpVariable %uvec3ptr Input\n"
5075 "%zero = OpConstant %i32 0\n"
5076 "%constf1 = OpConstant %f32 1.0\n"
5077 "%constf10 = OpConstant %f32 10.0\n"
5079 "%main = OpFunction %void None %voidf\n"
5080 "%entry = OpLabel\n"
5081 "%idval = OpLoad %uvec3 %id\n"
5082 "%x = OpCompositeExtract %u32 %idval 0\n"
5083 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5084 "%inval = OpLoad %f32 %inloc\n"
5085 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5086 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
5088 " OpSelectionMerge %if_end ${CONTROL}\n"
5089 " OpBranchConditional %cmp_gt %if_true %if_false\n"
5090 "%if_true = OpLabel\n"
5091 "%addf1 = OpFAdd %f32 %inval %constf1\n"
5092 " OpStore %outloc %addf1\n"
5093 " OpBranch %if_end\n"
5094 "%if_false = OpLabel\n"
5095 "%subf1 = OpFSub %f32 %inval %constf1\n"
5096 " OpStore %outloc %subf1\n"
5097 " OpBranch %if_end\n"
5098 "%if_end = OpLabel\n"
5100 " OpFunctionEnd\n");
5102 cases.push_back(CaseParameter("none", "None"));
5103 cases.push_back(CaseParameter("flatten", "Flatten"));
5104 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
5105 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
5107 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5109 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
5110 floorAll(inputFloats);
5112 for (size_t ndx = 0; ndx < numElements; ++ndx)
5113 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
5115 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5117 map<string, string> specializations;
5118 ComputeShaderSpec spec;
5120 specializations["CONTROL"] = cases[caseNdx].param;
5121 spec.assembly = shaderTemplate.specialize(specializations);
5122 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5123 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5124 spec.numWorkGroups = IVec3(numElements, 1, 1);
5126 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5129 return group.release();
5132 tcu::TestCaseGroup* createOpNameGroup(tcu::TestContext& testCtx)
5134 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opname", "Tests OpName cases"));
5135 de::MovePtr<tcu::TestCaseGroup> entryMainGroup (new tcu::TestCaseGroup(testCtx, "entry_main", "OpName tests with entry main"));
5136 de::MovePtr<tcu::TestCaseGroup> entryNotGroup (new tcu::TestCaseGroup(testCtx, "entry_rdc", "OpName tests with entry rdc"));
5137 vector<CaseParameter> cases;
5138 vector<string> testFunc;
5139 de::Random rnd (deStringHash(group->getName()));
5140 const int numElements = 100;
5141 vector<float> inputFloats (numElements, 0);
5142 vector<float> outputFloats (numElements, 0);
5144 fillRandomScalars(rnd, -100.0f, 100.0f, &inputFloats[0], numElements);
5146 for(size_t ndx = 0; ndx < numElements; ++ndx)
5147 outputFloats[ndx] = -inputFloats[ndx];
5149 const StringTemplate shaderTemplate (
5150 "OpCapability Shader\n"
5151 "OpMemoryModel Logical GLSL450\n"
5152 "OpEntryPoint GLCompute %main \"${ENTRY}\" %id\n"
5153 "OpExecutionMode %main LocalSize 1 1 1\n"
5155 "OpName %${FUNC_ID} \"${NAME}\"\n"
5157 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5159 + string(getComputeAsmInputOutputBufferTraits())
5161 + string(getComputeAsmCommonTypes())
5163 + string(getComputeAsmInputOutputBuffer()) +
5165 "%id = OpVariable %uvec3ptr Input\n"
5166 "%zero = OpConstant %i32 0\n"
5168 "%func = OpFunction %void None %voidf\n"
5173 "%main = OpFunction %void None %voidf\n"
5174 "%entry = OpLabel\n"
5175 "%7 = OpFunctionCall %void %func\n"
5177 "%idval = OpLoad %uvec3 %id\n"
5178 "%x = OpCompositeExtract %u32 %idval 0\n"
5180 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5181 "%inval = OpLoad %f32 %inloc\n"
5182 "%neg = OpFNegate %f32 %inval\n"
5183 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5184 " OpStore %outloc %neg\n"
5188 " OpFunctionEnd\n");
5190 cases.push_back(CaseParameter("_is_main", "main"));
5191 cases.push_back(CaseParameter("_is_not_main", "not_main"));
5193 testFunc.push_back("main");
5194 testFunc.push_back("func");
5196 for(size_t fNdx = 0; fNdx < testFunc.size(); ++fNdx)
5198 for(size_t ndx = 0; ndx < cases.size(); ++ndx)
5200 map<string, string> specializations;
5201 ComputeShaderSpec spec;
5203 specializations["ENTRY"] = "main";
5204 specializations["FUNC_ID"] = testFunc[fNdx];
5205 specializations["NAME"] = cases[ndx].param;
5206 spec.assembly = shaderTemplate.specialize(specializations);
5207 spec.numWorkGroups = IVec3(numElements, 1, 1);
5208 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5209 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5211 entryMainGroup->addChild(new SpvAsmComputeShaderCase(testCtx, (testFunc[fNdx] + cases[ndx].name).c_str(), cases[ndx].name, spec));
5215 cases.push_back(CaseParameter("_is_entry", "rdc"));
5217 for(size_t fNdx = 0; fNdx < testFunc.size(); ++fNdx)
5219 for(size_t ndx = 0; ndx < cases.size(); ++ndx)
5221 map<string, string> specializations;
5222 ComputeShaderSpec spec;
5224 specializations["ENTRY"] = "rdc";
5225 specializations["FUNC_ID"] = testFunc[fNdx];
5226 specializations["NAME"] = cases[ndx].param;
5227 spec.assembly = shaderTemplate.specialize(specializations);
5228 spec.numWorkGroups = IVec3(numElements, 1, 1);
5229 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5230 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5231 spec.entryPoint = "rdc";
5233 entryNotGroup->addChild(new SpvAsmComputeShaderCase(testCtx, (testFunc[fNdx] + cases[ndx].name).c_str(), cases[ndx].name, spec));
5237 group->addChild(entryMainGroup.release());
5238 group->addChild(entryNotGroup.release());
5240 return group.release();
5243 // Assembly code used for testing function control is based on GLSL source code:
5247 // layout(std140, set = 0, binding = 0) readonly buffer Input {
5248 // float elements[];
5250 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
5251 // float elements[];
5254 // float const10() { return 10.f; }
5257 // uint x = gl_GlobalInvocationID.x;
5258 // output_data.elements[x] = input_data.elements[x] + const10();
5260 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
5262 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
5263 vector<CaseParameter> cases;
5264 de::Random rnd (deStringHash(group->getName()));
5265 const int numElements = 100;
5266 vector<float> inputFloats (numElements, 0);
5267 vector<float> outputFloats (numElements, 0);
5268 const StringTemplate shaderTemplate (
5269 string(getComputeAsmShaderPreamble()) +
5271 "OpSource GLSL 430\n"
5272 "OpName %main \"main\"\n"
5273 "OpName %func_const10 \"const10(\"\n"
5274 "OpName %id \"gl_GlobalInvocationID\"\n"
5276 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5278 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5280 "%f32f = OpTypeFunction %f32\n"
5281 "%id = OpVariable %uvec3ptr Input\n"
5282 "%zero = OpConstant %i32 0\n"
5283 "%constf10 = OpConstant %f32 10.0\n"
5285 "%main = OpFunction %void None %voidf\n"
5286 "%entry = OpLabel\n"
5287 "%idval = OpLoad %uvec3 %id\n"
5288 "%x = OpCompositeExtract %u32 %idval 0\n"
5289 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5290 "%inval = OpLoad %f32 %inloc\n"
5291 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
5292 "%fadd = OpFAdd %f32 %inval %ret_10\n"
5293 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5294 " OpStore %outloc %fadd\n"
5298 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
5299 "%label = OpLabel\n"
5300 " OpReturnValue %constf10\n"
5301 " OpFunctionEnd\n");
5303 cases.push_back(CaseParameter("none", "None"));
5304 cases.push_back(CaseParameter("inline", "Inline"));
5305 cases.push_back(CaseParameter("dont_inline", "DontInline"));
5306 cases.push_back(CaseParameter("pure", "Pure"));
5307 cases.push_back(CaseParameter("const", "Const"));
5308 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
5309 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
5310 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
5311 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
5313 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5315 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
5316 floorAll(inputFloats);
5318 for (size_t ndx = 0; ndx < numElements; ++ndx)
5319 outputFloats[ndx] = inputFloats[ndx] + 10.f;
5321 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5323 map<string, string> specializations;
5324 ComputeShaderSpec spec;
5326 specializations["CONTROL"] = cases[caseNdx].param;
5327 spec.assembly = shaderTemplate.specialize(specializations);
5328 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5329 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5330 spec.numWorkGroups = IVec3(numElements, 1, 1);
5332 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5335 return group.release();
5338 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
5340 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
5341 vector<CaseParameter> cases;
5342 de::Random rnd (deStringHash(group->getName()));
5343 const int numElements = 100;
5344 vector<float> inputFloats (numElements, 0);
5345 vector<float> outputFloats (numElements, 0);
5346 const StringTemplate shaderTemplate (
5347 string(getComputeAsmShaderPreamble()) +
5349 "OpSource GLSL 430\n"
5350 "OpName %main \"main\"\n"
5351 "OpName %id \"gl_GlobalInvocationID\"\n"
5353 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5355 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5357 "%f32ptr_f = OpTypePointer Function %f32\n"
5359 "%id = OpVariable %uvec3ptr Input\n"
5360 "%zero = OpConstant %i32 0\n"
5361 "%four = OpConstant %i32 4\n"
5363 "%main = OpFunction %void None %voidf\n"
5364 "%label = OpLabel\n"
5365 "%copy = OpVariable %f32ptr_f Function\n"
5366 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
5367 "%x = OpCompositeExtract %u32 %idval 0\n"
5368 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5369 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5370 " OpCopyMemory %copy %inloc ${ACCESS}\n"
5371 "%val1 = OpLoad %f32 %copy\n"
5372 "%val2 = OpLoad %f32 %inloc\n"
5373 "%add = OpFAdd %f32 %val1 %val2\n"
5374 " OpStore %outloc %add ${ACCESS}\n"
5376 " OpFunctionEnd\n");
5378 cases.push_back(CaseParameter("null", ""));
5379 cases.push_back(CaseParameter("none", "None"));
5380 cases.push_back(CaseParameter("volatile", "Volatile"));
5381 cases.push_back(CaseParameter("aligned", "Aligned 4"));
5382 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
5383 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
5384 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
5386 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5388 for (size_t ndx = 0; ndx < numElements; ++ndx)
5389 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
5391 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5393 map<string, string> specializations;
5394 ComputeShaderSpec spec;
5396 specializations["ACCESS"] = cases[caseNdx].param;
5397 spec.assembly = shaderTemplate.specialize(specializations);
5398 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5399 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5400 spec.numWorkGroups = IVec3(numElements, 1, 1);
5402 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5405 return group.release();
5408 // Checks that we can get undefined values for various types, without exercising a computation with it.
5409 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
5411 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
5412 vector<CaseParameter> cases;
5413 de::Random rnd (deStringHash(group->getName()));
5414 const int numElements = 100;
5415 vector<float> positiveFloats (numElements, 0);
5416 vector<float> negativeFloats (numElements, 0);
5417 const StringTemplate shaderTemplate (
5418 string(getComputeAsmShaderPreamble()) +
5420 "OpSource GLSL 430\n"
5421 "OpName %main \"main\"\n"
5422 "OpName %id \"gl_GlobalInvocationID\"\n"
5424 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5426 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
5427 "%uvec2 = OpTypeVector %u32 2\n"
5428 "%fvec4 = OpTypeVector %f32 4\n"
5429 "%fmat33 = OpTypeMatrix %fvec3 3\n"
5430 "%image = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
5431 "%sampler = OpTypeSampler\n"
5432 "%simage = OpTypeSampledImage %image\n"
5433 "%const100 = OpConstant %u32 100\n"
5434 "%uarr100 = OpTypeArray %i32 %const100\n"
5435 "%struct = OpTypeStruct %f32 %i32 %u32\n"
5436 "%pointer = OpTypePointer Function %i32\n"
5437 + string(getComputeAsmInputOutputBuffer()) +
5439 "%id = OpVariable %uvec3ptr Input\n"
5440 "%zero = OpConstant %i32 0\n"
5442 "%main = OpFunction %void None %voidf\n"
5443 "%label = OpLabel\n"
5445 "%undef = OpUndef ${TYPE}\n"
5447 "%idval = OpLoad %uvec3 %id\n"
5448 "%x = OpCompositeExtract %u32 %idval 0\n"
5450 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5451 "%inval = OpLoad %f32 %inloc\n"
5452 "%neg = OpFNegate %f32 %inval\n"
5453 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5454 " OpStore %outloc %neg\n"
5456 " OpFunctionEnd\n");
5458 cases.push_back(CaseParameter("bool", "%bool"));
5459 cases.push_back(CaseParameter("sint32", "%i32"));
5460 cases.push_back(CaseParameter("uint32", "%u32"));
5461 cases.push_back(CaseParameter("float32", "%f32"));
5462 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
5463 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
5464 cases.push_back(CaseParameter("matrix", "%fmat33"));
5465 cases.push_back(CaseParameter("image", "%image"));
5466 cases.push_back(CaseParameter("sampler", "%sampler"));
5467 cases.push_back(CaseParameter("sampledimage", "%simage"));
5468 cases.push_back(CaseParameter("array", "%uarr100"));
5469 cases.push_back(CaseParameter("runtimearray", "%f32arr"));
5470 cases.push_back(CaseParameter("struct", "%struct"));
5471 cases.push_back(CaseParameter("pointer", "%pointer"));
5473 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
5475 for (size_t ndx = 0; ndx < numElements; ++ndx)
5476 negativeFloats[ndx] = -positiveFloats[ndx];
5478 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5480 map<string, string> specializations;
5481 ComputeShaderSpec spec;
5483 specializations["TYPE"] = cases[caseNdx].param;
5484 spec.assembly = shaderTemplate.specialize(specializations);
5485 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
5486 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
5487 spec.numWorkGroups = IVec3(numElements, 1, 1);
5489 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5492 return group.release();
5496 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
5498 struct NameCodePair { string name, code; };
5499 RGBA defaultColors[4];
5500 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
5501 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
5502 map<string, string> fragments = passthruFragments();
5503 const NameCodePair tests[] =
5505 {"unknown", "OpSource Unknown 321"},
5506 {"essl", "OpSource ESSL 310"},
5507 {"glsl", "OpSource GLSL 450"},
5508 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
5509 {"opencl_c", "OpSource OpenCL_C 120"},
5510 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
5511 {"file", opsourceGLSLWithFile},
5512 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
5513 // Longest possible source string: SPIR-V limits instructions to 65535
5514 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
5515 // contain 65530 UTF8 characters (one word each) plus one last word
5516 // containing 3 ASCII characters and \0.
5517 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
5520 getDefaultColors(defaultColors);
5521 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5523 fragments["debug"] = tests[testNdx].code;
5524 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5527 return opSourceTests.release();
5530 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
5532 struct NameCodePair { string name, code; };
5533 RGBA defaultColors[4];
5534 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
5535 map<string, string> fragments = passthruFragments();
5536 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
5537 const NameCodePair tests[] =
5539 {"empty", opsource + "OpSourceContinued \"\""},
5540 {"short", opsource + "OpSourceContinued \"abcde\""},
5541 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
5542 // Longest possible source string: SPIR-V limits instructions to 65535
5543 // words, of which the first one is OpSourceContinued/length; the rest
5544 // will contain 65533 UTF8 characters (one word each) plus one last word
5545 // containing 3 ASCII characters and \0.
5546 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
5549 getDefaultColors(defaultColors);
5550 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5552 fragments["debug"] = tests[testNdx].code;
5553 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5556 return opSourceTests.release();
5558 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
5560 RGBA defaultColors[4];
5561 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
5562 map<string, string> fragments;
5563 getDefaultColors(defaultColors);
5564 fragments["debug"] =
5565 "%name = OpString \"name\"\n";
5567 fragments["pre_main"] =
5570 "OpLine %name 1 1\n"
5572 "OpLine %name 1 1\n"
5573 "OpLine %name 1 1\n"
5574 "%second_function = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5576 "OpLine %name 1 1\n"
5578 "OpLine %name 1 1\n"
5579 "OpLine %name 1 1\n"
5580 "%second_param1 = OpFunctionParameter %v4f32\n"
5583 "%label_secondfunction = OpLabel\n"
5585 "OpReturnValue %second_param1\n"
5590 fragments["testfun"] =
5591 // A %test_code function that returns its argument unchanged.
5594 "OpLine %name 1 1\n"
5595 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5597 "%param1 = OpFunctionParameter %v4f32\n"
5600 "%label_testfun = OpLabel\n"
5602 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5603 "OpReturnValue %val1\n"
5605 "OpLine %name 1 1\n"
5608 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
5610 return opLineTests.release();
5613 tcu::TestCaseGroup* createOpModuleProcessedTests(tcu::TestContext& testCtx)
5615 RGBA defaultColors[4];
5616 de::MovePtr<tcu::TestCaseGroup> opModuleProcessedTests (new tcu::TestCaseGroup(testCtx, "opmoduleprocessed", "OpModuleProcessed instruction"));
5617 map<string, string> fragments;
5618 std::vector<std::string> noExtensions;
5619 GraphicsResources resources;
5621 getDefaultColors(defaultColors);
5622 resources.verifyBinary = veryfiBinaryShader;
5623 resources.spirvVersion = SPIRV_VERSION_1_3;
5625 fragments["moduleprocessed"] =
5626 "OpModuleProcessed \"VULKAN CTS\"\n"
5627 "OpModuleProcessed \"Negative values\"\n"
5628 "OpModuleProcessed \"Date: 2017/09/21\"\n";
5630 fragments["pre_main"] =
5631 "%second_function = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5632 "%second_param1 = OpFunctionParameter %v4f32\n"
5633 "%label_secondfunction = OpLabel\n"
5634 "OpReturnValue %second_param1\n"
5637 fragments["testfun"] =
5638 // A %test_code function that returns its argument unchanged.
5639 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5640 "%param1 = OpFunctionParameter %v4f32\n"
5641 "%label_testfun = OpLabel\n"
5642 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5643 "OpReturnValue %val1\n"
5646 createTestsForAllStages ("opmoduleprocessed", defaultColors, defaultColors, fragments, resources, noExtensions, opModuleProcessedTests.get());
5648 return opModuleProcessedTests.release();
5652 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
5654 RGBA defaultColors[4];
5655 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
5656 map<string, string> fragments;
5657 std::vector<std::pair<std::string, std::string> > problemStrings;
5659 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
5660 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
5661 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
5662 getDefaultColors(defaultColors);
5664 fragments["debug"] =
5665 "%other_name = OpString \"other_name\"\n";
5667 fragments["pre_main"] =
5668 "OpLine %file_name 32 0\n"
5669 "OpLine %file_name 32 32\n"
5670 "OpLine %file_name 32 40\n"
5671 "OpLine %other_name 32 40\n"
5672 "OpLine %other_name 0 100\n"
5673 "OpLine %other_name 0 4294967295\n"
5674 "OpLine %other_name 4294967295 0\n"
5675 "OpLine %other_name 32 40\n"
5676 "OpLine %file_name 0 0\n"
5677 "%second_function = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5678 "OpLine %file_name 1 0\n"
5679 "%second_param1 = OpFunctionParameter %v4f32\n"
5680 "OpLine %file_name 1 3\n"
5681 "OpLine %file_name 1 2\n"
5682 "%label_secondfunction = OpLabel\n"
5683 "OpLine %file_name 0 2\n"
5684 "OpReturnValue %second_param1\n"
5686 "OpLine %file_name 0 2\n"
5687 "OpLine %file_name 0 2\n";
5689 fragments["testfun"] =
5690 // A %test_code function that returns its argument unchanged.
5691 "OpLine %file_name 1 0\n"
5692 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5693 "OpLine %file_name 16 330\n"
5694 "%param1 = OpFunctionParameter %v4f32\n"
5695 "OpLine %file_name 14 442\n"
5696 "%label_testfun = OpLabel\n"
5697 "OpLine %file_name 11 1024\n"
5698 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5699 "OpLine %file_name 2 97\n"
5700 "OpReturnValue %val1\n"
5702 "OpLine %file_name 5 32\n";
5704 for (size_t i = 0; i < problemStrings.size(); ++i)
5706 map<string, string> testFragments = fragments;
5707 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
5708 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
5711 return opLineTests.release();
5714 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
5716 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
5720 const char functionStart[] =
5721 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5722 "%param1 = OpFunctionParameter %v4f32\n"
5725 const char functionEnd[] =
5726 "OpReturnValue %transformed_param\n"
5729 struct NameConstantsCode
5736 NameConstantsCode tests[] =
5740 "%cnull = OpConstantNull %v4f32\n",
5741 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
5745 "%cnull = OpConstantNull %f32\n",
5746 "%vp = OpVariable %fp_v4f32 Function\n"
5747 "%v = OpLoad %v4f32 %vp\n"
5748 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
5749 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
5750 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
5751 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
5752 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
5756 "%cnull = OpConstantNull %bool\n",
5757 "%v = OpVariable %fp_v4f32 Function\n"
5758 " OpStore %v %param1\n"
5759 " OpSelectionMerge %false_label None\n"
5760 " OpBranchConditional %cnull %true_label %false_label\n"
5761 "%true_label = OpLabel\n"
5762 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
5763 " OpBranch %false_label\n"
5764 "%false_label = OpLabel\n"
5765 "%transformed_param = OpLoad %v4f32 %v\n"
5769 "%cnull = OpConstantNull %i32\n",
5770 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
5771 "%b = OpIEqual %bool %cnull %c_i32_0\n"
5772 " OpSelectionMerge %false_label None\n"
5773 " OpBranchConditional %b %true_label %false_label\n"
5774 "%true_label = OpLabel\n"
5775 " OpStore %v %param1\n"
5776 " OpBranch %false_label\n"
5777 "%false_label = OpLabel\n"
5778 "%transformed_param = OpLoad %v4f32 %v\n"
5782 "%stype = OpTypeStruct %f32 %v4f32\n"
5783 "%fp_stype = OpTypePointer Function %stype\n"
5784 "%cnull = OpConstantNull %stype\n",
5785 "%v = OpVariable %fp_stype Function %cnull\n"
5786 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
5787 "%f_val = OpLoad %v4f32 %f\n"
5788 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
5792 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
5793 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
5794 "%cnull = OpConstantNull %a4_v4f32\n",
5795 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
5796 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5797 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
5798 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
5799 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
5800 "%f_val = OpLoad %v4f32 %f\n"
5801 "%f1_val = OpLoad %v4f32 %f1\n"
5802 "%f2_val = OpLoad %v4f32 %f2\n"
5803 "%f3_val = OpLoad %v4f32 %f3\n"
5804 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
5805 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
5806 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
5807 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
5811 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5812 "%cnull = OpConstantNull %mat4x4_f32\n",
5813 // Our null matrix * any vector should result in a zero vector.
5814 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
5815 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
5819 getHalfColorsFullAlpha(colors);
5821 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5823 map<string, string> fragments;
5824 fragments["pre_main"] = tests[testNdx].constants;
5825 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5826 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
5828 return opConstantNullTests.release();
5830 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
5832 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
5833 RGBA inputColors[4];
5834 RGBA outputColors[4];
5837 const char functionStart[] =
5838 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5839 "%param1 = OpFunctionParameter %v4f32\n"
5842 const char functionEnd[] =
5843 "OpReturnValue %transformed_param\n"
5846 struct NameConstantsCode
5853 NameConstantsCode tests[] =
5858 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
5859 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
5864 "%stype = OpTypeStruct %v4f32 %f32\n"
5865 "%fp_stype = OpTypePointer Function %stype\n"
5866 "%f32_n_1 = OpConstant %f32 -1.0\n"
5867 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5868 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
5869 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
5871 "%v = OpVariable %fp_stype Function %cval\n"
5872 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5873 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
5874 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
5875 "%f32_val = OpLoad %f32 %f32_ptr\n"
5876 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
5877 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
5878 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
5881 // [1|0|0|0.5] [x] = x + 0.5
5882 // [0|1|0|0.5] [y] = y + 0.5
5883 // [0|0|1|0.5] [z] = z + 0.5
5884 // [0|0|0|1 ] [1] = 1
5887 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5888 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
5889 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
5890 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
5891 "%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"
5892 "%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",
5894 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
5899 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5900 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5901 "%f32_n_1 = OpConstant %f32 -1.0\n"
5902 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5903 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
5905 "%v = OpVariable %fp_a4f32 Function %carr\n"
5906 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
5907 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
5908 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5909 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
5910 "%f_val = OpLoad %f32 %f\n"
5911 "%f1_val = OpLoad %f32 %f1\n"
5912 "%f2_val = OpLoad %f32 %f2\n"
5913 "%f3_val = OpLoad %f32 %f3\n"
5914 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
5915 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
5916 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
5917 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
5918 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5925 // [ 1.0, 1.0, 1.0, 1.0]
5929 // [ 0.0, 0.5, 0.0, 0.0]
5933 // [ 1.0, 1.0, 1.0, 1.0]
5936 "array_of_struct_of_array",
5938 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5939 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5940 "%stype = OpTypeStruct %f32 %a4f32\n"
5941 "%a3stype = OpTypeArray %stype %c_u32_3\n"
5942 "%fp_a3stype = OpTypePointer Function %a3stype\n"
5943 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
5944 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5945 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
5946 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
5947 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
5949 "%v = OpVariable %fp_a3stype Function %carr\n"
5950 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
5951 "%f_l = OpLoad %f32 %f\n"
5952 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
5953 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5957 getHalfColorsFullAlpha(inputColors);
5958 outputColors[0] = RGBA(255, 255, 255, 255);
5959 outputColors[1] = RGBA(255, 127, 127, 255);
5960 outputColors[2] = RGBA(127, 255, 127, 255);
5961 outputColors[3] = RGBA(127, 127, 255, 255);
5963 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5965 map<string, string> fragments;
5966 fragments["pre_main"] = tests[testNdx].constants;
5967 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5968 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
5970 return opConstantCompositeTests.release();
5973 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
5975 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
5976 RGBA inputColors[4];
5977 RGBA outputColors[4];
5978 map<string, string> fragments;
5980 // vec4 test_code(vec4 param) {
5981 // vec4 result = param;
5982 // for (int i = 0; i < 4; ++i) {
5983 // if (i == 0) result[i] = 0.;
5984 // else result[i] = 1. - result[i];
5988 const char function[] =
5989 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
5990 "%param1 = OpFunctionParameter %v4f32\n"
5992 "%iptr = OpVariable %fp_i32 Function\n"
5993 "%result = OpVariable %fp_v4f32 Function\n"
5994 " OpStore %iptr %c_i32_0\n"
5995 " OpStore %result %param1\n"
5998 // Loop entry block.
6000 "%ival = OpLoad %i32 %iptr\n"
6001 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6002 " OpLoopMerge %exit %if_entry None\n"
6003 " OpBranchConditional %lt_4 %if_entry %exit\n"
6005 // Merge block for loop.
6007 "%ret = OpLoad %v4f32 %result\n"
6008 " OpReturnValue %ret\n"
6010 // If-statement entry block.
6011 "%if_entry = OpLabel\n"
6012 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6013 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
6014 " OpSelectionMerge %if_exit None\n"
6015 " OpBranchConditional %eq_0 %if_true %if_false\n"
6017 // False branch for if-statement.
6018 "%if_false = OpLabel\n"
6019 "%val = OpLoad %f32 %loc\n"
6020 "%sub = OpFSub %f32 %c_f32_1 %val\n"
6021 " OpStore %loc %sub\n"
6022 " OpBranch %if_exit\n"
6024 // Merge block for if-statement.
6025 "%if_exit = OpLabel\n"
6026 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6027 " OpStore %iptr %ival_next\n"
6030 // True branch for if-statement.
6031 "%if_true = OpLabel\n"
6032 " OpStore %loc %c_f32_0\n"
6033 " OpBranch %if_exit\n"
6037 fragments["testfun"] = function;
6039 inputColors[0] = RGBA(127, 127, 127, 0);
6040 inputColors[1] = RGBA(127, 0, 0, 0);
6041 inputColors[2] = RGBA(0, 127, 0, 0);
6042 inputColors[3] = RGBA(0, 0, 127, 0);
6044 outputColors[0] = RGBA(0, 128, 128, 255);
6045 outputColors[1] = RGBA(0, 255, 255, 255);
6046 outputColors[2] = RGBA(0, 128, 255, 255);
6047 outputColors[3] = RGBA(0, 255, 128, 255);
6049 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
6051 return group.release();
6054 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
6056 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
6057 RGBA inputColors[4];
6058 RGBA outputColors[4];
6059 map<string, string> fragments;
6061 const char typesAndConstants[] =
6062 "%c_f32_p2 = OpConstant %f32 0.2\n"
6063 "%c_f32_p4 = OpConstant %f32 0.4\n"
6064 "%c_f32_p6 = OpConstant %f32 0.6\n"
6065 "%c_f32_p8 = OpConstant %f32 0.8\n";
6067 // vec4 test_code(vec4 param) {
6068 // vec4 result = param;
6069 // for (int i = 0; i < 4; ++i) {
6071 // case 0: result[i] += .2; break;
6072 // case 1: result[i] += .6; break;
6073 // case 2: result[i] += .4; break;
6074 // case 3: result[i] += .8; break;
6075 // default: break; // unreachable
6080 const char function[] =
6081 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6082 "%param1 = OpFunctionParameter %v4f32\n"
6084 "%iptr = OpVariable %fp_i32 Function\n"
6085 "%result = OpVariable %fp_v4f32 Function\n"
6086 " OpStore %iptr %c_i32_0\n"
6087 " OpStore %result %param1\n"
6090 // Loop entry block.
6092 "%ival = OpLoad %i32 %iptr\n"
6093 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6094 " OpLoopMerge %exit %switch_exit None\n"
6095 " OpBranchConditional %lt_4 %switch_entry %exit\n"
6097 // Merge block for loop.
6099 "%ret = OpLoad %v4f32 %result\n"
6100 " OpReturnValue %ret\n"
6102 // Switch-statement entry block.
6103 "%switch_entry = OpLabel\n"
6104 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6105 "%val = OpLoad %f32 %loc\n"
6106 " OpSelectionMerge %switch_exit None\n"
6107 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6109 "%case2 = OpLabel\n"
6110 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
6111 " OpStore %loc %addp4\n"
6112 " OpBranch %switch_exit\n"
6114 "%switch_default = OpLabel\n"
6117 "%case3 = OpLabel\n"
6118 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
6119 " OpStore %loc %addp8\n"
6120 " OpBranch %switch_exit\n"
6122 "%case0 = OpLabel\n"
6123 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
6124 " OpStore %loc %addp2\n"
6125 " OpBranch %switch_exit\n"
6127 // Merge block for switch-statement.
6128 "%switch_exit = OpLabel\n"
6129 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6130 " OpStore %iptr %ival_next\n"
6133 "%case1 = OpLabel\n"
6134 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
6135 " OpStore %loc %addp6\n"
6136 " OpBranch %switch_exit\n"
6140 fragments["pre_main"] = typesAndConstants;
6141 fragments["testfun"] = function;
6143 inputColors[0] = RGBA(127, 27, 127, 51);
6144 inputColors[1] = RGBA(127, 0, 0, 51);
6145 inputColors[2] = RGBA(0, 27, 0, 51);
6146 inputColors[3] = RGBA(0, 0, 127, 51);
6148 outputColors[0] = RGBA(178, 180, 229, 255);
6149 outputColors[1] = RGBA(178, 153, 102, 255);
6150 outputColors[2] = RGBA(51, 180, 102, 255);
6151 outputColors[3] = RGBA(51, 153, 229, 255);
6153 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
6155 return group.release();
6158 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
6160 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
6161 RGBA inputColors[4];
6162 RGBA outputColors[4];
6163 map<string, string> fragments;
6165 const char decorations[] =
6166 "OpDecorate %array_group ArrayStride 4\n"
6167 "OpDecorate %struct_member_group Offset 0\n"
6168 "%array_group = OpDecorationGroup\n"
6169 "%struct_member_group = OpDecorationGroup\n"
6171 "OpDecorate %group1 RelaxedPrecision\n"
6172 "OpDecorate %group3 RelaxedPrecision\n"
6173 "OpDecorate %group3 Invariant\n"
6174 "OpDecorate %group3 Restrict\n"
6175 "%group0 = OpDecorationGroup\n"
6176 "%group1 = OpDecorationGroup\n"
6177 "%group3 = OpDecorationGroup\n";
6179 const char typesAndConstants[] =
6180 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
6181 "%struct1 = OpTypeStruct %a3f32\n"
6182 "%struct2 = OpTypeStruct %a3f32\n"
6183 "%fp_struct1 = OpTypePointer Function %struct1\n"
6184 "%fp_struct2 = OpTypePointer Function %struct2\n"
6185 "%c_f32_2 = OpConstant %f32 2.\n"
6186 "%c_f32_n2 = OpConstant %f32 -2.\n"
6188 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
6189 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
6190 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
6191 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
6193 const char function[] =
6194 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6195 "%param = OpFunctionParameter %v4f32\n"
6196 "%entry = OpLabel\n"
6197 "%result = OpVariable %fp_v4f32 Function\n"
6198 "%v_struct1 = OpVariable %fp_struct1 Function\n"
6199 "%v_struct2 = OpVariable %fp_struct2 Function\n"
6200 " OpStore %result %param\n"
6201 " OpStore %v_struct1 %c_struct1\n"
6202 " OpStore %v_struct2 %c_struct2\n"
6203 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
6204 "%val1 = OpLoad %f32 %ptr1\n"
6205 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
6206 "%val2 = OpLoad %f32 %ptr2\n"
6207 "%addvalues = OpFAdd %f32 %val1 %val2\n"
6208 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
6209 "%val = OpLoad %f32 %ptr\n"
6210 "%addresult = OpFAdd %f32 %addvalues %val\n"
6211 " OpStore %ptr %addresult\n"
6212 "%ret = OpLoad %v4f32 %result\n"
6213 " OpReturnValue %ret\n"
6216 struct CaseNameDecoration
6222 CaseNameDecoration tests[] =
6225 "same_decoration_group_on_multiple_types",
6226 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
6229 "empty_decoration_group",
6230 "OpGroupDecorate %group0 %a3f32\n"
6231 "OpGroupDecorate %group0 %result\n"
6234 "one_element_decoration_group",
6235 "OpGroupDecorate %array_group %a3f32\n"
6238 "multiple_elements_decoration_group",
6239 "OpGroupDecorate %group3 %v_struct1\n"
6242 "multiple_decoration_groups_on_same_variable",
6243 "OpGroupDecorate %group0 %v_struct2\n"
6244 "OpGroupDecorate %group1 %v_struct2\n"
6245 "OpGroupDecorate %group3 %v_struct2\n"
6248 "same_decoration_group_multiple_times",
6249 "OpGroupDecorate %group1 %addvalues\n"
6250 "OpGroupDecorate %group1 %addvalues\n"
6251 "OpGroupDecorate %group1 %addvalues\n"
6256 getHalfColorsFullAlpha(inputColors);
6257 getHalfColorsFullAlpha(outputColors);
6259 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
6261 fragments["decoration"] = decorations + tests[idx].decoration;
6262 fragments["pre_main"] = typesAndConstants;
6263 fragments["testfun"] = function;
6265 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
6268 return group.release();
6271 struct SpecConstantTwoIntGraphicsCase
6273 const char* caseName;
6274 const char* scDefinition0;
6275 const char* scDefinition1;
6276 const char* scResultType;
6277 const char* scOperation;
6278 deInt32 scActualValue0;
6279 deInt32 scActualValue1;
6280 const char* resultOperation;
6281 RGBA expectedColors[4];
6283 SpecConstantTwoIntGraphicsCase (const char* name,
6284 const char* definition0,
6285 const char* definition1,
6286 const char* resultType,
6287 const char* operation,
6290 const char* resultOp,
6291 const RGBA (&output)[4])
6293 , scDefinition0 (definition0)
6294 , scDefinition1 (definition1)
6295 , scResultType (resultType)
6296 , scOperation (operation)
6297 , scActualValue0 (value0)
6298 , scActualValue1 (value1)
6299 , resultOperation (resultOp)
6301 expectedColors[0] = output[0];
6302 expectedColors[1] = output[1];
6303 expectedColors[2] = output[2];
6304 expectedColors[3] = output[3];
6308 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
6310 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
6311 vector<SpecConstantTwoIntGraphicsCase> cases;
6312 RGBA inputColors[4];
6313 RGBA outputColors0[4];
6314 RGBA outputColors1[4];
6315 RGBA outputColors2[4];
6317 const char decorations1[] =
6318 "OpDecorate %sc_0 SpecId 0\n"
6319 "OpDecorate %sc_1 SpecId 1\n";
6321 const char typesAndConstants1[] =
6322 "${OPTYPE_DEFINITIONS:opt}"
6323 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
6324 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
6325 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
6327 const char function1[] =
6328 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6329 "%param = OpFunctionParameter %v4f32\n"
6330 "%label = OpLabel\n"
6331 "%result = OpVariable %fp_v4f32 Function\n"
6332 "${TYPE_CONVERT:opt}"
6333 " OpStore %result %param\n"
6334 "%gen = ${GEN_RESULT}\n"
6335 "%index = OpIAdd %i32 %gen %c_i32_1\n"
6336 "%loc = OpAccessChain %fp_f32 %result %index\n"
6337 "%val = OpLoad %f32 %loc\n"
6338 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6339 " OpStore %loc %add\n"
6340 "%ret = OpLoad %v4f32 %result\n"
6341 " OpReturnValue %ret\n"
6344 inputColors[0] = RGBA(127, 127, 127, 255);
6345 inputColors[1] = RGBA(127, 0, 0, 255);
6346 inputColors[2] = RGBA(0, 127, 0, 255);
6347 inputColors[3] = RGBA(0, 0, 127, 255);
6349 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
6350 outputColors0[0] = RGBA(255, 127, 127, 255);
6351 outputColors0[1] = RGBA(255, 0, 0, 255);
6352 outputColors0[2] = RGBA(128, 127, 0, 255);
6353 outputColors0[3] = RGBA(128, 0, 127, 255);
6355 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
6356 outputColors1[0] = RGBA(127, 255, 127, 255);
6357 outputColors1[1] = RGBA(127, 128, 0, 255);
6358 outputColors1[2] = RGBA(0, 255, 0, 255);
6359 outputColors1[3] = RGBA(0, 128, 127, 255);
6361 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
6362 outputColors2[0] = RGBA(127, 127, 255, 255);
6363 outputColors2[1] = RGBA(127, 0, 128, 255);
6364 outputColors2[2] = RGBA(0, 127, 128, 255);
6365 outputColors2[3] = RGBA(0, 0, 255, 255);
6367 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
6368 const char addZeroToSc32[] = "OpIAdd %i32 %c_i32_0 %sc_op32";
6369 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
6370 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
6372 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
6373 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
6374 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
6375 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
6376 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
6377 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6378 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6379 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
6380 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
6381 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
6382 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
6383 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
6384 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
6385 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
6386 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
6387 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
6388 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6389 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
6390 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
6391 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
6392 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6393 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
6394 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
6395 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6396 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6397 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6398 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6399 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
6400 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
6401 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
6402 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
6403 cases.push_back(SpecConstantTwoIntGraphicsCase("sconvert", " %i32 0", " %i32 0", "%i16", "SConvert %sc_0", -1, 0, addZeroToSc32, outputColors0));
6404 // -1082130432 stored as 32-bit two's complement is the binary representation of -1 as IEEE-754 Float
6405 cases.push_back(SpecConstantTwoIntGraphicsCase("fconvert", " %f32 0", " %f32 0", "%f64", "FConvert %sc_0", -1082130432, 0, addZeroToSc32, outputColors0));
6406 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
6408 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
6410 map<string, string> specializations;
6411 map<string, string> fragments;
6412 vector<deInt32> specConstants;
6413 vector<string> features;
6414 PushConstants noPushConstants;
6415 GraphicsResources noResources;
6416 GraphicsInterfaces noInterfaces;
6417 std::vector<std::string> noExtensions;
6419 // Special SPIR-V code for SConvert-case
6420 if (strcmp(cases[caseNdx].caseName, "sconvert") == 0)
6422 features.push_back("shaderInt16");
6423 fragments["capability"] = "OpCapability Int16\n"; // Adds 16-bit integer capability
6424 specializations["OPTYPE_DEFINITIONS"] = "%i16 = OpTypeInt 16 1\n"; // Adds 16-bit integer type
6425 specializations["TYPE_CONVERT"] = "%sc_op32 = OpSConvert %i32 %sc_op\n"; // Converts 16-bit integer to 32-bit integer
6428 // Special SPIR-V code for FConvert-case
6429 if (strcmp(cases[caseNdx].caseName, "fconvert") == 0)
6431 features.push_back("shaderFloat64");
6432 fragments["capability"] = "OpCapability Float64\n"; // Adds 64-bit float capability
6433 specializations["OPTYPE_DEFINITIONS"] = "%f64 = OpTypeFloat 64\n"; // Adds 64-bit float type
6434 specializations["TYPE_CONVERT"] = "%sc_op32 = OpConvertFToS %i32 %sc_op\n"; // Converts 64-bit float to 32-bit integer
6437 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
6438 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
6439 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
6440 specializations["SC_OP"] = cases[caseNdx].scOperation;
6441 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
6443 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
6444 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
6445 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
6447 specConstants.push_back(cases[caseNdx].scActualValue0);
6448 specConstants.push_back(cases[caseNdx].scActualValue1);
6450 createTestsForAllStages(
6451 cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants,
6452 noPushConstants, noResources, noInterfaces, noExtensions, features, VulkanFeatures(), group.get());
6455 const char decorations2[] =
6456 "OpDecorate %sc_0 SpecId 0\n"
6457 "OpDecorate %sc_1 SpecId 1\n"
6458 "OpDecorate %sc_2 SpecId 2\n";
6460 const char typesAndConstants2[] =
6461 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
6462 "%vec3_undef = OpUndef %v3i32\n"
6464 "%sc_0 = OpSpecConstant %i32 0\n"
6465 "%sc_1 = OpSpecConstant %i32 0\n"
6466 "%sc_2 = OpSpecConstant %i32 0\n"
6467 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
6468 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
6469 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
6470 "%sc_vec3_0_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
6471 "%sc_vec3_1_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
6472 "%sc_vec3_2_s = OpSpecConstantOp %v3i32 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
6473 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
6474 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
6475 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
6476 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
6477 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
6478 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
6479 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
6481 const char function2[] =
6482 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6483 "%param = OpFunctionParameter %v4f32\n"
6484 "%label = OpLabel\n"
6485 "%result = OpVariable %fp_v4f32 Function\n"
6486 " OpStore %result %param\n"
6487 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
6488 "%val = OpLoad %f32 %loc\n"
6489 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6490 " OpStore %loc %add\n"
6491 "%ret = OpLoad %v4f32 %result\n"
6492 " OpReturnValue %ret\n"
6495 map<string, string> fragments;
6496 vector<deInt32> specConstants;
6498 fragments["decoration"] = decorations2;
6499 fragments["pre_main"] = typesAndConstants2;
6500 fragments["testfun"] = function2;
6502 specConstants.push_back(56789);
6503 specConstants.push_back(-2);
6504 specConstants.push_back(56788);
6506 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
6508 return group.release();
6511 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
6513 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
6514 RGBA inputColors[4];
6515 RGBA outputColors1[4];
6516 RGBA outputColors2[4];
6517 RGBA outputColors3[4];
6518 map<string, string> fragments1;
6519 map<string, string> fragments2;
6520 map<string, string> fragments3;
6522 const char typesAndConstants1[] =
6523 "%c_f32_p2 = OpConstant %f32 0.2\n"
6524 "%c_f32_p4 = OpConstant %f32 0.4\n"
6525 "%c_f32_p5 = OpConstant %f32 0.5\n"
6526 "%c_f32_p8 = OpConstant %f32 0.8\n";
6528 // vec4 test_code(vec4 param) {
6529 // vec4 result = param;
6530 // for (int i = 0; i < 4; ++i) {
6533 // case 0: operand = .2; break;
6534 // case 1: operand = .5; break;
6535 // case 2: operand = .4; break;
6536 // case 3: operand = .0; break;
6537 // default: break; // unreachable
6539 // result[i] += operand;
6543 const char function1[] =
6544 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6545 "%param1 = OpFunctionParameter %v4f32\n"
6547 "%iptr = OpVariable %fp_i32 Function\n"
6548 "%result = OpVariable %fp_v4f32 Function\n"
6549 " OpStore %iptr %c_i32_0\n"
6550 " OpStore %result %param1\n"
6554 "%ival = OpLoad %i32 %iptr\n"
6555 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6556 " OpLoopMerge %exit %phi None\n"
6557 " OpBranchConditional %lt_4 %entry %exit\n"
6559 "%entry = OpLabel\n"
6560 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6561 "%val = OpLoad %f32 %loc\n"
6562 " OpSelectionMerge %phi None\n"
6563 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6565 "%case0 = OpLabel\n"
6567 "%case1 = OpLabel\n"
6569 "%case2 = OpLabel\n"
6571 "%case3 = OpLabel\n"
6574 "%default = OpLabel\n"
6578 "%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
6579 "%add = OpFAdd %f32 %val %operand\n"
6580 " OpStore %loc %add\n"
6581 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6582 " OpStore %iptr %ival_next\n"
6586 "%ret = OpLoad %v4f32 %result\n"
6587 " OpReturnValue %ret\n"
6591 fragments1["pre_main"] = typesAndConstants1;
6592 fragments1["testfun"] = function1;
6594 getHalfColorsFullAlpha(inputColors);
6596 outputColors1[0] = RGBA(178, 255, 229, 255);
6597 outputColors1[1] = RGBA(178, 127, 102, 255);
6598 outputColors1[2] = RGBA(51, 255, 102, 255);
6599 outputColors1[3] = RGBA(51, 127, 229, 255);
6601 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
6603 const char typesAndConstants2[] =
6604 "%c_f32_p2 = OpConstant %f32 0.2\n";
6606 // Add .4 to the second element of the given parameter.
6607 const char function2[] =
6608 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6609 "%param = OpFunctionParameter %v4f32\n"
6610 "%entry = OpLabel\n"
6611 "%result = OpVariable %fp_v4f32 Function\n"
6612 " OpStore %result %param\n"
6613 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6614 "%val = OpLoad %f32 %loc\n"
6618 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
6619 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
6620 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
6621 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
6622 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
6623 " OpLoopMerge %exit %phi None\n"
6624 " OpBranchConditional %still_loop %phi %exit\n"
6627 " OpStore %loc %accum\n"
6628 "%ret = OpLoad %v4f32 %result\n"
6629 " OpReturnValue %ret\n"
6633 fragments2["pre_main"] = typesAndConstants2;
6634 fragments2["testfun"] = function2;
6636 outputColors2[0] = RGBA(127, 229, 127, 255);
6637 outputColors2[1] = RGBA(127, 102, 0, 255);
6638 outputColors2[2] = RGBA(0, 229, 0, 255);
6639 outputColors2[3] = RGBA(0, 102, 127, 255);
6641 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
6643 const char typesAndConstants3[] =
6644 "%true = OpConstantTrue %bool\n"
6645 "%false = OpConstantFalse %bool\n"
6646 "%c_f32_p2 = OpConstant %f32 0.2\n";
6648 // Swap the second and the third element of the given parameter.
6649 const char function3[] =
6650 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6651 "%param = OpFunctionParameter %v4f32\n"
6652 "%entry = OpLabel\n"
6653 "%result = OpVariable %fp_v4f32 Function\n"
6654 " OpStore %result %param\n"
6655 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6656 "%a_init = OpLoad %f32 %a_loc\n"
6657 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
6658 "%b_init = OpLoad %f32 %b_loc\n"
6662 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
6663 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
6664 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
6665 " OpLoopMerge %exit %phi None\n"
6666 " OpBranchConditional %still_loop %phi %exit\n"
6669 " OpStore %a_loc %a_next\n"
6670 " OpStore %b_loc %b_next\n"
6671 "%ret = OpLoad %v4f32 %result\n"
6672 " OpReturnValue %ret\n"
6676 fragments3["pre_main"] = typesAndConstants3;
6677 fragments3["testfun"] = function3;
6679 outputColors3[0] = RGBA(127, 127, 127, 255);
6680 outputColors3[1] = RGBA(127, 0, 0, 255);
6681 outputColors3[2] = RGBA(0, 0, 127, 255);
6682 outputColors3[3] = RGBA(0, 127, 0, 255);
6684 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
6686 return group.release();
6689 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
6691 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
6692 RGBA inputColors[4];
6693 RGBA outputColors[4];
6695 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
6696 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
6697 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
6698 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
6699 const char constantsAndTypes[] =
6700 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
6701 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
6702 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
6703 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
6704 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n";
6706 const char function[] =
6707 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6708 "%param = OpFunctionParameter %v4f32\n"
6709 "%label = OpLabel\n"
6710 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
6711 "%var2 = OpVariable %fp_f32 Function\n"
6712 "%red = OpCompositeExtract %f32 %param 0\n"
6713 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
6714 " OpStore %var2 %plus_red\n"
6715 "%val1 = OpLoad %f32 %var1\n"
6716 "%val2 = OpLoad %f32 %var2\n"
6717 "%mul = OpFMul %f32 %val1 %val2\n"
6718 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
6719 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
6720 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
6721 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
6722 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
6723 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
6724 " OpReturnValue %ret\n"
6727 struct CaseNameDecoration
6734 CaseNameDecoration tests[] = {
6735 {"multiplication", "OpDecorate %mul NoContraction"},
6736 {"addition", "OpDecorate %add NoContraction"},
6737 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
6740 getHalfColorsFullAlpha(inputColors);
6742 for (deUint8 idx = 0; idx < 4; ++idx)
6744 inputColors[idx].setRed(0);
6745 outputColors[idx] = RGBA(0, 0, 0, 255);
6748 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
6750 map<string, string> fragments;
6752 fragments["decoration"] = tests[testNdx].decoration;
6753 fragments["pre_main"] = constantsAndTypes;
6754 fragments["testfun"] = function;
6756 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
6759 return group.release();
6762 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
6764 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
6767 const char constantsAndTypes[] =
6768 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
6769 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
6770 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
6771 "%fp_stype = OpTypePointer Function %stype\n";
6773 const char function[] =
6774 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6775 "%param1 = OpFunctionParameter %v4f32\n"
6777 "%v1 = OpVariable %fp_v4f32 Function\n"
6778 "%v2 = OpVariable %fp_a2f32 Function\n"
6779 "%v3 = OpVariable %fp_f32 Function\n"
6780 "%v = OpVariable %fp_stype Function\n"
6781 "%vv = OpVariable %fp_stype Function\n"
6782 "%vvv = OpVariable %fp_f32 Function\n"
6784 " OpStore %v1 %c_v4f32_1_1_1_1\n"
6785 " OpStore %v2 %c_a2f32_1\n"
6786 " OpStore %v3 %c_f32_1\n"
6788 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
6789 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
6790 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
6791 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
6792 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
6793 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
6795 " OpStore %p_v4f32 %v1_v ${access_type}\n"
6796 " OpStore %p_a2f32 %v2_v ${access_type}\n"
6797 " OpStore %p_f32 %v3_v ${access_type}\n"
6799 " OpCopyMemory %vv %v ${access_type}\n"
6800 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
6802 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
6803 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
6804 "%v_f32_3 = OpLoad %f32 %vvv\n"
6806 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
6807 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
6808 " OpReturnValue %ret2\n"
6811 struct NameMemoryAccess
6818 NameMemoryAccess tests[] =
6821 { "volatile", "Volatile" },
6822 { "aligned", "Aligned 1" },
6823 { "volatile_aligned", "Volatile|Aligned 1" },
6824 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
6825 { "volatile_nontemporal", "Volatile|Nontemporal" },
6826 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
6829 getHalfColorsFullAlpha(colors);
6831 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
6833 map<string, string> fragments;
6834 map<string, string> memoryAccess;
6835 memoryAccess["access_type"] = tests[testNdx].accessType;
6837 fragments["pre_main"] = constantsAndTypes;
6838 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
6839 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
6841 return memoryAccessTests.release();
6843 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
6845 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
6846 RGBA defaultColors[4];
6847 map<string, string> fragments;
6848 getDefaultColors(defaultColors);
6850 // First, simple cases that don't do anything with the OpUndef result.
6851 struct NameCodePair { string name, decl, type; };
6852 const NameCodePair tests[] =
6854 {"bool", "", "%bool"},
6855 {"vec2uint32", "", "%v2u32"},
6856 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown", "%type"},
6857 {"sampler", "%type = OpTypeSampler", "%type"},
6858 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img", "%type"},
6859 {"pointer", "", "%fp_i32"},
6860 {"runtimearray", "%type = OpTypeRuntimeArray %f32", "%type"},
6861 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100", "%type"},
6862 {"struct", "%type = OpTypeStruct %f32 %i32 %u32", "%type"}};
6863 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
6865 fragments["undef_type"] = tests[testNdx].type;
6866 fragments["testfun"] = StringTemplate(
6867 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6868 "%param1 = OpFunctionParameter %v4f32\n"
6869 "%label_testfun = OpLabel\n"
6870 "%undef = OpUndef ${undef_type}\n"
6871 "OpReturnValue %param1\n"
6872 "OpFunctionEnd\n").specialize(fragments);
6873 fragments["pre_main"] = tests[testNdx].decl;
6874 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
6878 fragments["testfun"] =
6879 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6880 "%param1 = OpFunctionParameter %v4f32\n"
6881 "%label_testfun = OpLabel\n"
6882 "%undef = OpUndef %f32\n"
6883 "%zero = OpFMul %f32 %undef %c_f32_0\n"
6884 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
6885 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
6886 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6887 "%b = OpFAdd %f32 %a %actually_zero\n"
6888 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
6889 "OpReturnValue %ret\n"
6892 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6894 fragments["testfun"] =
6895 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6896 "%param1 = OpFunctionParameter %v4f32\n"
6897 "%label_testfun = OpLabel\n"
6898 "%undef = OpUndef %i32\n"
6899 "%zero = OpIMul %i32 %undef %c_i32_0\n"
6900 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6901 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6902 "OpReturnValue %ret\n"
6905 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6907 fragments["testfun"] =
6908 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6909 "%param1 = OpFunctionParameter %v4f32\n"
6910 "%label_testfun = OpLabel\n"
6911 "%undef = OpUndef %u32\n"
6912 "%zero = OpIMul %u32 %undef %c_i32_0\n"
6913 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6914 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6915 "OpReturnValue %ret\n"
6918 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6920 fragments["testfun"] =
6921 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6922 "%param1 = OpFunctionParameter %v4f32\n"
6923 "%label_testfun = OpLabel\n"
6924 "%undef = OpUndef %v4f32\n"
6925 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
6926 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
6927 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
6928 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
6929 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
6930 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6931 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6932 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6933 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6934 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6935 "%actually_zero_1 = OpSelect %f32 %is_nan_1 %c_f32_0 %zero_1\n"
6936 "%actually_zero_2 = OpSelect %f32 %is_nan_2 %c_f32_0 %zero_2\n"
6937 "%actually_zero_3 = OpSelect %f32 %is_nan_3 %c_f32_0 %zero_3\n"
6938 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6939 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6940 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6941 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6942 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6943 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6944 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6945 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6946 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6947 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6948 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6949 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6950 "OpReturnValue %ret\n"
6953 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6955 fragments["pre_main"] =
6956 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
6957 fragments["testfun"] =
6958 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
6959 "%param1 = OpFunctionParameter %v4f32\n"
6960 "%label_testfun = OpLabel\n"
6961 "%undef = OpUndef %m2x2f32\n"
6962 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
6963 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
6964 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
6965 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
6966 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
6967 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6968 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6969 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6970 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6971 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6972 "%actually_zero_1 = OpSelect %f32 %is_nan_1 %c_f32_0 %zero_1\n"
6973 "%actually_zero_2 = OpSelect %f32 %is_nan_2 %c_f32_0 %zero_2\n"
6974 "%actually_zero_3 = OpSelect %f32 %is_nan_3 %c_f32_0 %zero_3\n"
6975 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6976 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6977 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6978 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6979 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6980 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6981 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6982 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6983 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6984 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6985 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6986 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6987 "OpReturnValue %ret\n"
6990 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
6992 return opUndefTests.release();
6995 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
6997 const RGBA inputColors[4] =
7000 RGBA(0, 0, 255, 255),
7001 RGBA(0, 255, 0, 255),
7002 RGBA(0, 255, 255, 255)
7005 const RGBA expectedColors[4] =
7007 RGBA(255, 0, 0, 255),
7008 RGBA(255, 0, 0, 255),
7009 RGBA(255, 0, 0, 255),
7010 RGBA(255, 0, 0, 255)
7013 const struct SingleFP16Possibility
7016 const char* constant; // Value to assign to %test_constant.
7018 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
7024 -constructNormalizedFloat(1, 0x300000),
7025 "%cond = OpFOrdEqual %bool %c %test_constant\n"
7030 constructNormalizedFloat(7, 0x000000),
7031 "%cond = OpFOrdEqual %bool %c %test_constant\n"
7033 // SPIR-V requires that OpQuantizeToF16 flushes
7034 // any numbers that would end up denormalized in F16 to zero.
7038 std::ldexp(1.5f, -140),
7039 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
7044 -std::ldexp(1.5f, -140),
7045 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
7050 std::ldexp(1.0f, -16),
7051 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
7052 }, // too small positive
7054 "negative_too_small",
7056 -std::ldexp(1.0f, -32),
7057 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
7058 }, // too small negative
7062 -std::ldexp(1.0f, 128),
7064 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
7065 "%inf = OpIsInf %bool %c\n"
7066 "%cond = OpLogicalAnd %bool %gz %inf\n"
7071 std::ldexp(1.0f, 128),
7073 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
7074 "%inf = OpIsInf %bool %c\n"
7075 "%cond = OpLogicalAnd %bool %gz %inf\n"
7078 "round_to_negative_inf",
7080 -std::ldexp(1.0f, 32),
7082 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
7083 "%inf = OpIsInf %bool %c\n"
7084 "%cond = OpLogicalAnd %bool %gz %inf\n"
7089 std::ldexp(1.0f, 16),
7091 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
7092 "%inf = OpIsInf %bool %c\n"
7093 "%cond = OpLogicalAnd %bool %gz %inf\n"
7098 std::numeric_limits<float>::quiet_NaN(),
7100 // Test for any NaN value, as NaNs are not preserved
7101 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
7102 "%cond = OpIsNan %bool %direct_quant\n"
7107 std::numeric_limits<float>::quiet_NaN(),
7109 // Test for any NaN value, as NaNs are not preserved
7110 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
7111 "%cond = OpIsNan %bool %direct_quant\n"
7114 const char* constants =
7115 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
7117 StringTemplate function (
7118 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7119 "%param1 = OpFunctionParameter %v4f32\n"
7120 "%label_testfun = OpLabel\n"
7121 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7122 "%b = OpFAdd %f32 %test_constant %a\n"
7123 "%c = OpQuantizeToF16 %f32 %b\n"
7125 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7126 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
7127 " OpReturnValue %retval\n"
7131 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
7132 const char* specConstants =
7133 "%test_constant = OpSpecConstant %f32 0.\n"
7134 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
7136 StringTemplate specConstantFunction(
7137 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7138 "%param1 = OpFunctionParameter %v4f32\n"
7139 "%label_testfun = OpLabel\n"
7141 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7142 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
7143 " OpReturnValue %retval\n"
7147 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
7149 map<string, string> codeSpecialization;
7150 map<string, string> fragments;
7151 codeSpecialization["condition"] = tests[idx].condition;
7152 fragments["testfun"] = function.specialize(codeSpecialization);
7153 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
7154 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7157 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
7159 map<string, string> codeSpecialization;
7160 map<string, string> fragments;
7161 vector<deInt32> passConstants;
7162 deInt32 specConstant;
7164 codeSpecialization["condition"] = tests[idx].condition;
7165 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
7166 fragments["decoration"] = specDecorations;
7167 fragments["pre_main"] = specConstants;
7169 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
7170 passConstants.push_back(specConstant);
7172 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7176 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
7178 RGBA inputColors[4] = {
7180 RGBA(0, 0, 255, 255),
7181 RGBA(0, 255, 0, 255),
7182 RGBA(0, 255, 255, 255)
7185 RGBA expectedColors[4] =
7187 RGBA(255, 0, 0, 255),
7188 RGBA(255, 0, 0, 255),
7189 RGBA(255, 0, 0, 255),
7190 RGBA(255, 0, 0, 255)
7193 struct DualFP16Possibility
7198 const char* possibleOutput1;
7199 const char* possibleOutput2;
7202 "positive_round_up_or_round_down",
7204 constructNormalizedFloat(8, 0x300300),
7209 "negative_round_up_or_round_down",
7211 -constructNormalizedFloat(-7, 0x600800),
7218 constructNormalizedFloat(2, 0x01e000),
7223 "carry_to_exponent",
7225 constructNormalizedFloat(1, 0xffe000),
7230 StringTemplate constants (
7231 "%input_const = OpConstant %f32 ${input}\n"
7232 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7233 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7236 StringTemplate specConstants (
7237 "%input_const = OpSpecConstant %f32 0.\n"
7238 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7239 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7242 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
7244 const char* function =
7245 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7246 "%param1 = OpFunctionParameter %v4f32\n"
7247 "%label_testfun = OpLabel\n"
7248 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7249 // For the purposes of this test we assume that 0.f will always get
7250 // faithfully passed through the pipeline stages.
7251 "%b = OpFAdd %f32 %input_const %a\n"
7252 "%c = OpQuantizeToF16 %f32 %b\n"
7253 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
7254 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
7255 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
7256 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7257 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
7258 " OpReturnValue %retval\n"
7261 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7262 map<string, string> fragments;
7263 map<string, string> constantSpecialization;
7265 constantSpecialization["input"] = tests[idx].input;
7266 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7267 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7268 fragments["testfun"] = function;
7269 fragments["pre_main"] = constants.specialize(constantSpecialization);
7270 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7273 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7274 map<string, string> fragments;
7275 map<string, string> constantSpecialization;
7276 vector<deInt32> passConstants;
7277 deInt32 specConstant;
7279 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7280 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7281 fragments["testfun"] = function;
7282 fragments["decoration"] = specDecorations;
7283 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
7285 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
7286 passConstants.push_back(specConstant);
7288 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7292 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
7294 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
7295 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
7296 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
7297 return opQuantizeTests.release();
7300 struct ShaderPermutation
7302 deUint8 vertexPermutation;
7303 deUint8 geometryPermutation;
7304 deUint8 tesscPermutation;
7305 deUint8 tessePermutation;
7306 deUint8 fragmentPermutation;
7309 ShaderPermutation getShaderPermutation(deUint8 inputValue)
7311 ShaderPermutation permutation =
7313 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
7314 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
7315 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
7316 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
7317 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
7322 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
7324 RGBA defaultColors[4];
7325 RGBA invertedColors[4];
7326 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
7328 const ShaderElement combinedPipeline[] =
7330 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
7331 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
7332 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7333 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7334 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
7337 getDefaultColors(defaultColors);
7338 getInvertedDefaultColors(invertedColors);
7339 addFunctionCaseWithPrograms<InstanceContext>(
7340 moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline,
7341 createInstanceContext(combinedPipeline, map<string, string>()));
7343 const char* numbers[] =
7348 for (deInt8 idx = 0; idx < 32; ++idx)
7350 ShaderPermutation permutation = getShaderPermutation(idx);
7351 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
7352 const ShaderElement pipeline[] =
7354 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
7355 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
7356 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7357 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7358 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
7361 // If there are an even number of swaps, then it should be no-op.
7362 // If there are an odd number, the color should be flipped.
7363 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
7365 addFunctionCaseWithPrograms<InstanceContext>(
7366 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
7367 createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
7371 addFunctionCaseWithPrograms<InstanceContext>(
7372 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
7373 createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
7376 return moduleTests.release();
7379 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
7381 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
7382 RGBA defaultColors[4];
7383 getDefaultColors(defaultColors);
7384 map<string, string> fragments;
7385 fragments["pre_main"] =
7386 "%c_f32_5 = OpConstant %f32 5.\n";
7388 // A loop with a single block. The Continue Target is the loop block
7389 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
7390 // -- the "continue construct" forms the entire loop.
7391 fragments["testfun"] =
7392 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7393 "%param1 = OpFunctionParameter %v4f32\n"
7395 "%entry = OpLabel\n"
7396 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7399 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7401 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7402 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
7403 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7404 "%val = OpFAdd %f32 %val1 %delta\n"
7405 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
7406 "%count__ = OpISub %i32 %count %c_i32_1\n"
7407 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7408 "OpLoopMerge %exit %loop None\n"
7409 "OpBranchConditional %again %loop %exit\n"
7412 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7413 "OpReturnValue %result\n"
7417 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
7419 // Body comprised of multiple basic blocks.
7420 const StringTemplate multiBlock(
7421 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7422 "%param1 = OpFunctionParameter %v4f32\n"
7424 "%entry = OpLabel\n"
7425 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7428 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7430 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
7431 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
7432 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
7433 // There are several possibilities for the Continue Target below. Each
7434 // will be specialized into a separate test case.
7435 "OpLoopMerge %exit ${continue_target} None\n"
7439 ";delta_next = (delta > 0) ? -1 : 1;\n"
7440 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
7441 "OpSelectionMerge %gather DontFlatten\n"
7442 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
7445 "OpBranch %gather\n"
7448 "OpBranch %gather\n"
7450 "%gather = OpLabel\n"
7451 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
7452 "%val = OpFAdd %f32 %val1 %delta\n"
7453 "%count__ = OpISub %i32 %count %c_i32_1\n"
7454 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7455 "OpBranchConditional %again %loop %exit\n"
7458 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7459 "OpReturnValue %result\n"
7463 map<string, string> continue_target;
7465 // The Continue Target is the loop block itself.
7466 continue_target["continue_target"] = "%loop";
7467 fragments["testfun"] = multiBlock.specialize(continue_target);
7468 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
7470 // The Continue Target is at the end of the loop.
7471 continue_target["continue_target"] = "%gather";
7472 fragments["testfun"] = multiBlock.specialize(continue_target);
7473 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
7475 // A loop with continue statement.
7476 fragments["testfun"] =
7477 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7478 "%param1 = OpFunctionParameter %v4f32\n"
7480 "%entry = OpLabel\n"
7481 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7484 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
7486 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7487 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
7488 "OpLoopMerge %exit %continue None\n"
7492 ";skip if %count==2\n"
7493 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
7494 "OpSelectionMerge %continue DontFlatten\n"
7495 "OpBranchConditional %eq2 %continue %body\n"
7498 "%fcount = OpConvertSToF %f32 %count\n"
7499 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7500 "OpBranch %continue\n"
7502 "%continue = OpLabel\n"
7503 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
7504 "%count__ = OpISub %i32 %count %c_i32_1\n"
7505 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7506 "OpBranchConditional %again %loop %exit\n"
7509 "%same = OpFSub %f32 %val %c_f32_8\n"
7510 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7511 "OpReturnValue %result\n"
7513 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
7515 // A loop with break.
7516 fragments["testfun"] =
7517 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7518 "%param1 = OpFunctionParameter %v4f32\n"
7520 "%entry = OpLabel\n"
7521 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7522 "%dot = OpDot %f32 %param1 %param1\n"
7523 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7524 "%zero = OpConvertFToU %u32 %div\n"
7525 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7526 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7529 ";adds 4 and 3 to %val0 (exits early)\n"
7531 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7532 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7533 "OpLoopMerge %exit %continue None\n"
7537 ";end loop if %count==%two\n"
7538 "%above2 = OpSGreaterThan %bool %count %two\n"
7539 "OpSelectionMerge %continue DontFlatten\n"
7540 "OpBranchConditional %above2 %body %exit\n"
7543 "%fcount = OpConvertSToF %f32 %count\n"
7544 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7545 "OpBranch %continue\n"
7547 "%continue = OpLabel\n"
7548 "%count__ = OpISub %i32 %count %c_i32_1\n"
7549 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7550 "OpBranchConditional %again %loop %exit\n"
7553 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
7554 "%same = OpFSub %f32 %val_post %c_f32_7\n"
7555 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7556 "OpReturnValue %result\n"
7558 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
7560 // A loop with return.
7561 fragments["testfun"] =
7562 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7563 "%param1 = OpFunctionParameter %v4f32\n"
7565 "%entry = OpLabel\n"
7566 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7567 "%dot = OpDot %f32 %param1 %param1\n"
7568 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7569 "%zero = OpConvertFToU %u32 %div\n"
7570 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7571 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7574 ";returns early without modifying %param1\n"
7576 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7577 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7578 "OpLoopMerge %exit %continue None\n"
7582 ";return if %count==%two\n"
7583 "%above2 = OpSGreaterThan %bool %count %two\n"
7584 "OpSelectionMerge %continue DontFlatten\n"
7585 "OpBranchConditional %above2 %body %early_exit\n"
7587 "%early_exit = OpLabel\n"
7588 "OpReturnValue %param1\n"
7591 "%fcount = OpConvertSToF %f32 %count\n"
7592 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7593 "OpBranch %continue\n"
7595 "%continue = OpLabel\n"
7596 "%count__ = OpISub %i32 %count %c_i32_1\n"
7597 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7598 "OpBranchConditional %again %loop %exit\n"
7601 ";should never get here, so return an incorrect result\n"
7602 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
7603 "OpReturnValue %result\n"
7605 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
7607 // Continue inside a switch block to break to enclosing loop's merge block.
7608 // Matches roughly the following GLSL code:
7609 // for (; keep_going; keep_going = false)
7611 // switch (int(param1.x))
7613 // case 0: continue;
7614 // case 1: continue;
7615 // default: continue;
7617 // dead code: modify return value to invalid result.
7619 fragments["pre_main"] =
7620 "%fp_bool = OpTypePointer Function %bool\n"
7621 "%true = OpConstantTrue %bool\n"
7622 "%false = OpConstantFalse %bool\n";
7624 fragments["testfun"] =
7625 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7626 "%param1 = OpFunctionParameter %v4f32\n"
7628 "%entry = OpLabel\n"
7629 "%keep_going = OpVariable %fp_bool Function\n"
7630 "%val_ptr = OpVariable %fp_f32 Function\n"
7631 "%param1_x = OpCompositeExtract %f32 %param1 0\n"
7632 "OpStore %keep_going %true\n"
7633 "OpBranch %forloop_begin\n"
7635 "%forloop_begin = OpLabel\n"
7636 "OpLoopMerge %forloop_merge %forloop_continue None\n"
7637 "OpBranch %forloop\n"
7639 "%forloop = OpLabel\n"
7640 "%for_condition = OpLoad %bool %keep_going\n"
7641 "OpBranchConditional %for_condition %forloop_body %forloop_merge\n"
7643 "%forloop_body = OpLabel\n"
7644 "OpStore %val_ptr %param1_x\n"
7645 "%param1_x_int = OpConvertFToS %i32 %param1_x\n"
7647 "OpSelectionMerge %switch_merge None\n"
7648 "OpSwitch %param1_x_int %default 0 %case_0 1 %case_1\n"
7649 "%case_0 = OpLabel\n"
7650 "OpBranch %forloop_continue\n"
7651 "%case_1 = OpLabel\n"
7652 "OpBranch %forloop_continue\n"
7653 "%default = OpLabel\n"
7654 "OpBranch %forloop_continue\n"
7655 "%switch_merge = OpLabel\n"
7656 ";should never get here, so change the return value to invalid result\n"
7657 "OpStore %val_ptr %c_f32_1\n"
7658 "OpBranch %forloop_continue\n"
7660 "%forloop_continue = OpLabel\n"
7661 "OpStore %keep_going %false\n"
7662 "OpBranch %forloop_begin\n"
7663 "%forloop_merge = OpLabel\n"
7665 "%val = OpLoad %f32 %val_ptr\n"
7666 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7667 "OpReturnValue %result\n"
7669 createTestsForAllStages("switch_continue", defaultColors, defaultColors, fragments, testGroup.get());
7671 return testGroup.release();
7674 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
7675 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
7677 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
7678 map<string, string> fragments;
7680 // A barrier inside a function body.
7681 fragments["pre_main"] =
7682 "%Workgroup = OpConstant %i32 2\n"
7683 "%WorkgroupAcquireRelease = OpConstant %i32 0x108\n";
7684 fragments["testfun"] =
7685 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7686 "%param1 = OpFunctionParameter %v4f32\n"
7687 "%label_testfun = OpLabel\n"
7688 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7689 "OpReturnValue %param1\n"
7691 addTessCtrlTest(testGroup.get(), "in_function", fragments);
7693 // Common setup code for the following tests.
7694 fragments["pre_main"] =
7695 "%Workgroup = OpConstant %i32 2\n"
7696 "%WorkgroupAcquireRelease = OpConstant %i32 0x108\n"
7697 "%c_f32_5 = OpConstant %f32 5.\n";
7698 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
7699 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7700 "%param1 = OpFunctionParameter %v4f32\n"
7701 "%entry = OpLabel\n"
7702 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7703 "%dot = OpDot %f32 %param1 %param1\n"
7704 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7705 "%zero = OpConvertFToU %u32 %div\n";
7707 // Barriers inside OpSwitch branches.
7708 fragments["testfun"] =
7710 "OpSelectionMerge %switch_exit None\n"
7711 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
7713 "%case1 = OpLabel\n"
7714 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7715 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7716 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7717 "OpBranch %switch_exit\n"
7719 "%switch_default = OpLabel\n"
7720 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7721 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7722 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7723 "OpBranch %switch_exit\n"
7725 "%case0 = OpLabel\n"
7726 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7727 "OpBranch %switch_exit\n"
7729 "%switch_exit = OpLabel\n"
7730 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
7731 "OpReturnValue %ret\n"
7733 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
7735 // Barriers inside if-then-else.
7736 fragments["testfun"] =
7738 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
7739 "OpSelectionMerge %exit DontFlatten\n"
7740 "OpBranchConditional %eq0 %then %else\n"
7743 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7744 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7745 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7749 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7753 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
7754 "OpReturnValue %ret\n"
7756 addTessCtrlTest(testGroup.get(), "in_if", fragments);
7758 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
7759 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
7760 fragments["testfun"] =
7762 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
7763 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
7764 "OpSelectionMerge %exit DontFlatten\n"
7765 "OpBranchConditional %thread0 %then %else\n"
7768 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7772 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
7776 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
7777 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7778 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
7779 "OpReturnValue %ret\n"
7781 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
7783 // A barrier inside a loop.
7784 fragments["pre_main"] =
7785 "%Workgroup = OpConstant %i32 2\n"
7786 "%WorkgroupAcquireRelease = OpConstant %i32 0x108\n"
7787 "%c_f32_10 = OpConstant %f32 10.\n";
7788 fragments["testfun"] =
7789 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7790 "%param1 = OpFunctionParameter %v4f32\n"
7791 "%entry = OpLabel\n"
7792 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7795 ";adds 4, 3, 2, and 1 to %val0\n"
7797 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7798 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7799 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7800 "%fcount = OpConvertSToF %f32 %count\n"
7801 "%val = OpFAdd %f32 %val1 %fcount\n"
7802 "%count__ = OpISub %i32 %count %c_i32_1\n"
7803 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7804 "OpLoopMerge %exit %loop None\n"
7805 "OpBranchConditional %again %loop %exit\n"
7808 "%same = OpFSub %f32 %val %c_f32_10\n"
7809 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7810 "OpReturnValue %ret\n"
7812 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
7814 return testGroup.release();
7817 // Test for the OpFRem instruction.
7818 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
7820 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
7821 map<string, string> fragments;
7822 RGBA inputColors[4];
7823 RGBA outputColors[4];
7825 fragments["pre_main"] =
7826 "%c_f32_3 = OpConstant %f32 3.0\n"
7827 "%c_f32_n3 = OpConstant %f32 -3.0\n"
7828 "%c_f32_4 = OpConstant %f32 4.0\n"
7829 "%c_f32_p75 = OpConstant %f32 0.75\n"
7830 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
7831 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
7832 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
7834 // The test does the following.
7835 // vec4 result = (param1 * 8.0) - 4.0;
7836 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
7837 fragments["testfun"] =
7838 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7839 "%param1 = OpFunctionParameter %v4f32\n"
7840 "%label_testfun = OpLabel\n"
7841 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
7842 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
7843 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
7844 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
7845 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
7846 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
7847 "OpReturnValue %xy_0_1\n"
7851 inputColors[0] = RGBA(16, 16, 0, 255);
7852 inputColors[1] = RGBA(232, 232, 0, 255);
7853 inputColors[2] = RGBA(232, 16, 0, 255);
7854 inputColors[3] = RGBA(16, 232, 0, 255);
7856 outputColors[0] = RGBA(64, 64, 0, 255);
7857 outputColors[1] = RGBA(255, 255, 0, 255);
7858 outputColors[2] = RGBA(255, 64, 0, 255);
7859 outputColors[3] = RGBA(64, 255, 0, 255);
7861 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
7862 return testGroup.release();
7865 // Test for the OpSRem instruction.
7866 tcu::TestCaseGroup* createOpSRemGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
7868 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "srem", "OpSRem"));
7869 map<string, string> fragments;
7871 fragments["pre_main"] =
7872 "%c_f32_255 = OpConstant %f32 255.0\n"
7873 "%c_i32_128 = OpConstant %i32 128\n"
7874 "%c_i32_255 = OpConstant %i32 255\n"
7875 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
7876 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
7877 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
7879 // The test does the following.
7880 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
7881 // ivec4 result = ivec4(srem(ints.x, ints.y), srem(ints.y, ints.z), srem(ints.z, ints.x), 255);
7882 // return float(result + 128) / 255.0;
7883 fragments["testfun"] =
7884 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7885 "%param1 = OpFunctionParameter %v4f32\n"
7886 "%label_testfun = OpLabel\n"
7887 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
7888 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
7889 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
7890 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
7891 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
7892 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
7893 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
7894 "%x_out = OpSRem %i32 %x_in %y_in\n"
7895 "%y_out = OpSRem %i32 %y_in %z_in\n"
7896 "%z_out = OpSRem %i32 %z_in %x_in\n"
7897 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
7898 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
7899 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
7900 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
7901 "OpReturnValue %float_out\n"
7904 const struct CaseParams
7907 const char* failMessageTemplate; // customized status message
7908 qpTestResult failResult; // override status on failure
7909 int operands[4][3]; // four (x, y, z) vectors of operands
7910 int results[4][3]; // four (x, y, z) vectors of results
7916 QP_TEST_RESULT_FAIL,
7917 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
7918 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
7922 "Inconsistent results, but within specification: ${reason}",
7923 negFailResult, // negative operands, not required by the spec
7924 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
7925 { { 5, 12, -2 }, { 0, -5, 2 }, { 3, 8, -6 }, { 25, -60, 0 } }, // results
7928 // If either operand is negative the result is undefined. Some implementations may still return correct values.
7930 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
7932 const CaseParams& params = cases[caseNdx];
7933 RGBA inputColors[4];
7934 RGBA outputColors[4];
7936 for (int i = 0; i < 4; ++i)
7938 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
7939 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
7942 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
7945 return testGroup.release();
7948 // Test for the OpSMod instruction.
7949 tcu::TestCaseGroup* createOpSModGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
7951 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "smod", "OpSMod"));
7952 map<string, string> fragments;
7954 fragments["pre_main"] =
7955 "%c_f32_255 = OpConstant %f32 255.0\n"
7956 "%c_i32_128 = OpConstant %i32 128\n"
7957 "%c_i32_255 = OpConstant %i32 255\n"
7958 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
7959 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
7960 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
7962 // The test does the following.
7963 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
7964 // ivec4 result = ivec4(smod(ints.x, ints.y), smod(ints.y, ints.z), smod(ints.z, ints.x), 255);
7965 // return float(result + 128) / 255.0;
7966 fragments["testfun"] =
7967 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
7968 "%param1 = OpFunctionParameter %v4f32\n"
7969 "%label_testfun = OpLabel\n"
7970 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
7971 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
7972 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
7973 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
7974 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
7975 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
7976 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
7977 "%x_out = OpSMod %i32 %x_in %y_in\n"
7978 "%y_out = OpSMod %i32 %y_in %z_in\n"
7979 "%z_out = OpSMod %i32 %z_in %x_in\n"
7980 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
7981 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
7982 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
7983 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
7984 "OpReturnValue %float_out\n"
7987 const struct CaseParams
7990 const char* failMessageTemplate; // customized status message
7991 qpTestResult failResult; // override status on failure
7992 int operands[4][3]; // four (x, y, z) vectors of operands
7993 int results[4][3]; // four (x, y, z) vectors of results
7999 QP_TEST_RESULT_FAIL,
8000 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
8001 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
8005 "Inconsistent results, but within specification: ${reason}",
8006 negFailResult, // negative operands, not required by the spec
8007 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
8008 { { 5, -5, 3 }, { 0, 2, -3 }, { 3, -73, 69 }, { -35, 40, 0 } }, // results
8011 // If either operand is negative the result is undefined. Some implementations may still return correct values.
8013 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
8015 const CaseParams& params = cases[caseNdx];
8016 RGBA inputColors[4];
8017 RGBA outputColors[4];
8019 for (int i = 0; i < 4; ++i)
8021 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
8022 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
8025 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
8027 return testGroup.release();
8030 enum ConversionDataType
8032 DATA_TYPE_SIGNED_16,
8033 DATA_TYPE_SIGNED_32,
8034 DATA_TYPE_SIGNED_64,
8035 DATA_TYPE_UNSIGNED_16,
8036 DATA_TYPE_UNSIGNED_32,
8037 DATA_TYPE_UNSIGNED_64,
8040 DATA_TYPE_VEC2_SIGNED_16,
8041 DATA_TYPE_VEC2_SIGNED_32
8044 const string getBitWidthStr (ConversionDataType type)
8048 case DATA_TYPE_SIGNED_16:
8049 case DATA_TYPE_UNSIGNED_16:
8052 case DATA_TYPE_SIGNED_32:
8053 case DATA_TYPE_UNSIGNED_32:
8054 case DATA_TYPE_FLOAT_32:
8055 case DATA_TYPE_VEC2_SIGNED_16:
8058 case DATA_TYPE_SIGNED_64:
8059 case DATA_TYPE_UNSIGNED_64:
8060 case DATA_TYPE_FLOAT_64:
8061 case DATA_TYPE_VEC2_SIGNED_32:
8070 const string getByteWidthStr (ConversionDataType type)
8074 case DATA_TYPE_SIGNED_16:
8075 case DATA_TYPE_UNSIGNED_16:
8078 case DATA_TYPE_SIGNED_32:
8079 case DATA_TYPE_UNSIGNED_32:
8080 case DATA_TYPE_FLOAT_32:
8081 case DATA_TYPE_VEC2_SIGNED_16:
8084 case DATA_TYPE_SIGNED_64:
8085 case DATA_TYPE_UNSIGNED_64:
8086 case DATA_TYPE_FLOAT_64:
8087 case DATA_TYPE_VEC2_SIGNED_32:
8096 bool isSigned (ConversionDataType type)
8100 case DATA_TYPE_SIGNED_16:
8101 case DATA_TYPE_SIGNED_32:
8102 case DATA_TYPE_SIGNED_64:
8103 case DATA_TYPE_FLOAT_32:
8104 case DATA_TYPE_FLOAT_64:
8105 case DATA_TYPE_VEC2_SIGNED_16:
8106 case DATA_TYPE_VEC2_SIGNED_32:
8109 case DATA_TYPE_UNSIGNED_16:
8110 case DATA_TYPE_UNSIGNED_32:
8111 case DATA_TYPE_UNSIGNED_64:
8120 bool isInt (ConversionDataType type)
8124 case DATA_TYPE_SIGNED_16:
8125 case DATA_TYPE_SIGNED_32:
8126 case DATA_TYPE_SIGNED_64:
8127 case DATA_TYPE_UNSIGNED_16:
8128 case DATA_TYPE_UNSIGNED_32:
8129 case DATA_TYPE_UNSIGNED_64:
8132 case DATA_TYPE_FLOAT_32:
8133 case DATA_TYPE_FLOAT_64:
8134 case DATA_TYPE_VEC2_SIGNED_16:
8135 case DATA_TYPE_VEC2_SIGNED_32:
8144 bool isFloat (ConversionDataType type)
8148 case DATA_TYPE_SIGNED_16:
8149 case DATA_TYPE_SIGNED_32:
8150 case DATA_TYPE_SIGNED_64:
8151 case DATA_TYPE_UNSIGNED_16:
8152 case DATA_TYPE_UNSIGNED_32:
8153 case DATA_TYPE_UNSIGNED_64:
8154 case DATA_TYPE_VEC2_SIGNED_16:
8155 case DATA_TYPE_VEC2_SIGNED_32:
8158 case DATA_TYPE_FLOAT_32:
8159 case DATA_TYPE_FLOAT_64:
8168 const string getTypeName (ConversionDataType type)
8170 string prefix = isSigned(type) ? "" : "u";
8172 if (isInt(type)) return prefix + "int" + getBitWidthStr(type);
8173 else if (isFloat(type)) return prefix + "float" + getBitWidthStr(type);
8174 else if (type == DATA_TYPE_VEC2_SIGNED_16) return "i16vec2";
8175 else if (type == DATA_TYPE_VEC2_SIGNED_32) return "i32vec2";
8176 else DE_ASSERT(false);
8181 const string getTestName (ConversionDataType from, ConversionDataType to)
8183 return getTypeName(from) + "_to_" + getTypeName(to);
8186 const string getAsmTypeName (ConversionDataType type)
8190 if (isInt(type)) prefix = isSigned(type) ? "i" : "u";
8191 else if (isFloat(type)) prefix = "f";
8192 else if (type == DATA_TYPE_VEC2_SIGNED_16) return "i16vec2";
8193 else if (type == DATA_TYPE_VEC2_SIGNED_32) return "v2i32";
8194 else DE_ASSERT(false);
8196 return prefix + getBitWidthStr(type);
8199 template<typename T>
8200 BufferSp getSpecializedBuffer (deInt64 number)
8202 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
8205 BufferSp getBuffer (ConversionDataType type, deInt64 number)
8209 case DATA_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
8210 case DATA_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
8211 case DATA_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
8212 case DATA_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
8213 case DATA_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
8214 case DATA_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
8215 case DATA_TYPE_FLOAT_32: return getSpecializedBuffer<deUint32>(number);
8216 case DATA_TYPE_FLOAT_64: return getSpecializedBuffer<deUint64>(number);
8217 case DATA_TYPE_VEC2_SIGNED_16: return getSpecializedBuffer<deUint32>(number);
8218 case DATA_TYPE_VEC2_SIGNED_32: return getSpecializedBuffer<deUint64>(number);
8220 default: DE_ASSERT(false);
8221 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
8225 bool usesInt16 (ConversionDataType from, ConversionDataType to)
8227 return (from == DATA_TYPE_SIGNED_16 || from == DATA_TYPE_UNSIGNED_16
8228 || to == DATA_TYPE_SIGNED_16 || to == DATA_TYPE_UNSIGNED_16
8229 || from == DATA_TYPE_VEC2_SIGNED_16 || to == DATA_TYPE_VEC2_SIGNED_16);
8232 bool usesInt32 (ConversionDataType from, ConversionDataType to)
8234 return (from == DATA_TYPE_SIGNED_32 || from == DATA_TYPE_UNSIGNED_32
8235 || to == DATA_TYPE_SIGNED_32 || to == DATA_TYPE_UNSIGNED_32
8236 || from == DATA_TYPE_VEC2_SIGNED_32 || to == DATA_TYPE_VEC2_SIGNED_32);
8239 bool usesInt64 (ConversionDataType from, ConversionDataType to)
8241 return (from == DATA_TYPE_SIGNED_64 || from == DATA_TYPE_UNSIGNED_64
8242 || to == DATA_TYPE_SIGNED_64 || to == DATA_TYPE_UNSIGNED_64);
8245 bool usesFloat64 (ConversionDataType from, ConversionDataType to)
8247 return (from == DATA_TYPE_FLOAT_64 || to == DATA_TYPE_FLOAT_64);
8251 ComputeTestFeatures getConversionUsedFeatures (ConversionDataType from, ConversionDataType to)
8253 if (usesInt16(from, to) && usesInt64(from, to)) return COMPUTE_TEST_USES_INT16_INT64;
8254 else if (usesInt16(from, to) && usesInt32(from, to)) return COMPUTE_TEST_USES_NONE;
8255 else if (usesInt16(from, to)) return COMPUTE_TEST_USES_INT16; // This is not set for int16<-->int32 only conversions
8256 else if (usesInt64(from, to)) return COMPUTE_TEST_USES_INT64;
8257 else if (usesFloat64(from, to)) return COMPUTE_TEST_USES_FLOAT64;
8258 else return COMPUTE_TEST_USES_NONE;
8261 vector<string> getFeatureStringVector (ComputeTestFeatures computeTestFeatures)
8263 vector<string> features;
8264 if (computeTestFeatures == COMPUTE_TEST_USES_INT16_INT64)
8266 features.push_back("shaderInt16");
8267 features.push_back("shaderInt64");
8269 else if (computeTestFeatures == COMPUTE_TEST_USES_INT16) features.push_back("shaderInt16");
8270 else if (computeTestFeatures == COMPUTE_TEST_USES_INT64) features.push_back("shaderInt64");
8271 else if (computeTestFeatures == COMPUTE_TEST_USES_FLOAT64) features.push_back("shaderFloat64");
8272 else if (computeTestFeatures == COMPUTE_TEST_USES_NONE) {}
8273 else DE_ASSERT(false);
8280 ConvertCase (ConversionDataType from, ConversionDataType to, deInt64 number, bool separateOutput = false, deInt64 outputNumber = 0)
8283 , m_features (getConversionUsedFeatures(from, to))
8284 , m_name (getTestName(from, to))
8285 , m_inputBuffer (getBuffer(from, number))
8287 m_asmTypes["inputType"] = getAsmTypeName(from);
8288 m_asmTypes["outputType"] = getAsmTypeName(to);
8291 m_outputBuffer = getBuffer(to, outputNumber);
8293 m_outputBuffer = getBuffer(to, number);
8295 if (m_features == COMPUTE_TEST_USES_INT16)
8297 m_asmTypes["datatype_capabilities"] = "OpCapability Int16\n"
8298 "OpCapability StorageUniformBufferBlock16\n"
8299 "OpCapability StorageUniform16\n";
8300 m_asmTypes["datatype_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
8301 "%u16 = OpTypeInt 16 0\n"
8302 "%i16vec2 = OpTypeVector %i16 2\n";
8303 m_asmTypes["datatype_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
8305 else if (m_features == COMPUTE_TEST_USES_INT64)
8307 m_asmTypes["datatype_capabilities"] = "OpCapability Int64\n";
8308 m_asmTypes["datatype_additional_decl"] = "%i64 = OpTypeInt 64 1\n"
8309 "%u64 = OpTypeInt 64 0\n";
8310 m_asmTypes["datatype_extensions"] = "";
8312 else if (m_features == COMPUTE_TEST_USES_INT16_INT64)
8314 m_asmTypes["datatype_capabilities"] = "OpCapability Int16\n"
8315 "OpCapability StorageUniformBufferBlock16\n"
8316 "OpCapability StorageUniform16\n"
8317 "OpCapability Int64\n";
8318 m_asmTypes["datatype_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
8319 "%u16 = OpTypeInt 16 0\n"
8320 "%i64 = OpTypeInt 64 1\n"
8321 "%u64 = OpTypeInt 64 0\n";
8322 m_asmTypes["datatype_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
8324 else if (m_features == COMPUTE_TEST_USES_FLOAT64)
8326 m_asmTypes["datatype_capabilities"] = "OpCapability Float64\n";
8327 m_asmTypes["datatype_additional_decl"] = "%f64 = OpTypeFloat 64\n";
8329 else if (usesInt16(from, to) && usesInt32(from, to))
8331 m_asmTypes["datatype_capabilities"] = "OpCapability StorageUniformBufferBlock16\n"
8332 "OpCapability StorageUniform16\n";
8333 m_asmTypes["datatype_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
8334 "%u16 = OpTypeInt 16 0\n"
8335 "%i16vec2 = OpTypeVector %i16 2\n";
8336 m_asmTypes["datatype_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
8344 ConversionDataType m_fromType;
8345 ConversionDataType m_toType;
8346 ComputeTestFeatures m_features;
8348 map<string, string> m_asmTypes;
8349 BufferSp m_inputBuffer;
8350 BufferSp m_outputBuffer;
8353 const string getConvertCaseShaderStr (const string& instruction, const ConvertCase& convertCase)
8355 map<string, string> params = convertCase.m_asmTypes;
8357 params["instruction"] = instruction;
8358 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
8359 params["outDecorator"] = getByteWidthStr(convertCase.m_toType);
8361 const StringTemplate shader (
8362 "OpCapability Shader\n"
8363 "${datatype_capabilities}"
8364 "${datatype_extensions:opt}"
8365 "OpMemoryModel Logical GLSL450\n"
8366 "OpEntryPoint GLCompute %main \"main\"\n"
8367 "OpExecutionMode %main LocalSize 1 1 1\n"
8368 "OpSource GLSL 430\n"
8369 "OpName %main \"main\"\n"
8371 "OpDecorate %indata DescriptorSet 0\n"
8372 "OpDecorate %indata Binding 0\n"
8373 "OpDecorate %outdata DescriptorSet 0\n"
8374 "OpDecorate %outdata Binding 1\n"
8375 "OpDecorate %in_buf BufferBlock\n"
8376 "OpDecorate %out_buf BufferBlock\n"
8377 "OpMemberDecorate %in_buf 0 Offset 0\n"
8378 "OpMemberDecorate %out_buf 0 Offset 0\n"
8380 "%void = OpTypeVoid\n"
8381 "%voidf = OpTypeFunction %void\n"
8382 "%u32 = OpTypeInt 32 0\n"
8383 "%i32 = OpTypeInt 32 1\n"
8384 "%f32 = OpTypeFloat 32\n"
8385 "%v2i32 = OpTypeVector %i32 2\n"
8386 "${datatype_additional_decl}"
8387 "%uvec3 = OpTypeVector %u32 3\n"
8389 "%in_ptr = OpTypePointer Uniform %${inputType}\n"
8390 "%out_ptr = OpTypePointer Uniform %${outputType}\n"
8391 "%in_buf = OpTypeStruct %${inputType}\n"
8392 "%out_buf = OpTypeStruct %${outputType}\n"
8393 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
8394 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
8395 "%indata = OpVariable %in_bufptr Uniform\n"
8396 "%outdata = OpVariable %out_bufptr Uniform\n"
8398 "%zero = OpConstant %i32 0\n"
8400 "%main = OpFunction %void None %voidf\n"
8401 "%label = OpLabel\n"
8402 "%inloc = OpAccessChain %in_ptr %indata %zero\n"
8403 "%outloc = OpAccessChain %out_ptr %outdata %zero\n"
8404 "%inval = OpLoad %${inputType} %inloc\n"
8405 "%conv = ${instruction} %${outputType} %inval\n"
8406 " OpStore %outloc %conv\n"
8411 return shader.specialize(params);
8414 void createConvertCases (vector<ConvertCase>& testCases, const string& instruction)
8416 if (instruction == "OpUConvert")
8418 // Convert unsigned int to unsigned int
8419 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_16, DATA_TYPE_UNSIGNED_32, 60653));
8420 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_16, DATA_TYPE_UNSIGNED_64, 17991));
8421 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_32, DATA_TYPE_UNSIGNED_64, 904256275));
8422 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_32, DATA_TYPE_UNSIGNED_16, 6275));
8423 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_64, DATA_TYPE_UNSIGNED_32, 701256243));
8424 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_64, DATA_TYPE_UNSIGNED_16, 4741));
8426 else if (instruction == "OpSConvert")
8428 // Sign extension int->int
8429 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_16, DATA_TYPE_SIGNED_32, 14669));
8430 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_16, DATA_TYPE_SIGNED_64, -3341));
8431 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_32, DATA_TYPE_SIGNED_64, 973610259));
8433 // Truncate for int->int
8434 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_32, DATA_TYPE_SIGNED_16, 12382));
8435 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_64, DATA_TYPE_SIGNED_32, -972812359));
8436 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_64, DATA_TYPE_SIGNED_16, -1067742499291926803ll, true, -4371));
8438 // Sign extension for int->uint
8439 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_16, DATA_TYPE_UNSIGNED_32, 14669));
8440 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_16, DATA_TYPE_UNSIGNED_64, -3341, true, 18446744073709548275ull));
8441 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_32, DATA_TYPE_UNSIGNED_64, 973610259));
8443 // Truncate for int->uint
8444 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_32, DATA_TYPE_UNSIGNED_16, 12382));
8445 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_64, DATA_TYPE_UNSIGNED_32, -972812359, true, 3322154937u));
8446 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_64, DATA_TYPE_UNSIGNED_16, -1067742499291926803ll, true, 61165));
8448 // Sign extension for uint->int
8449 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_16, DATA_TYPE_SIGNED_32, 14669));
8450 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_16, DATA_TYPE_SIGNED_64, 62195, true, -3341));
8451 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_32, DATA_TYPE_SIGNED_64, 973610259));
8453 // Truncate for uint->int
8454 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_32, DATA_TYPE_SIGNED_16, 12382));
8455 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_64, DATA_TYPE_SIGNED_32, 18446744072736739257ull, true, -972812359));
8456 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_64, DATA_TYPE_SIGNED_16, 17379001574417624813ull, true, -4371));
8458 // Convert i16vec2 to i32vec2 and vice versa
8459 // Unsigned values are used here to represent negative signed values and to allow defined shifting behaviour.
8460 // The actual signed value -32123 is used here as uint16 value 33413 and uint32 value 4294935173
8461 testCases.push_back(ConvertCase(DATA_TYPE_VEC2_SIGNED_16, DATA_TYPE_VEC2_SIGNED_32, (33413u << 16) | 27593, true, (4294935173ull << 32) | 27593));
8462 testCases.push_back(ConvertCase(DATA_TYPE_VEC2_SIGNED_32, DATA_TYPE_VEC2_SIGNED_16, (4294935173ull << 32) | 27593, true, (33413u << 16) | 27593));
8464 else if (instruction == "OpFConvert")
8466 // All hexadecimal values below represent 1024.0 as 32/64-bit IEEE 754 float
8467 testCases.push_back(ConvertCase(DATA_TYPE_FLOAT_32, DATA_TYPE_FLOAT_64, 0x449a4000, true, 0x4093480000000000));
8468 testCases.push_back(ConvertCase(DATA_TYPE_FLOAT_64, DATA_TYPE_FLOAT_32, 0x4093480000000000, true, 0x449a4000));
8471 DE_FATAL("Unknown instruction");
8474 const map<string, string> getConvertCaseFragments (string instruction, const ConvertCase& convertCase)
8476 map<string, string> params = convertCase.m_asmTypes;
8477 map<string, string> fragments;
8479 params["instruction"] = instruction;
8480 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
8482 const StringTemplate decoration (
8483 " OpDecorate %SSBOi DescriptorSet 0\n"
8484 " OpDecorate %SSBOo DescriptorSet 0\n"
8485 " OpDecorate %SSBOi Binding 0\n"
8486 " OpDecorate %SSBOo Binding 1\n"
8487 " OpDecorate %s_SSBOi Block\n"
8488 " OpDecorate %s_SSBOo Block\n"
8489 "OpMemberDecorate %s_SSBOi 0 Offset 0\n"
8490 "OpMemberDecorate %s_SSBOo 0 Offset 0\n");
8492 const StringTemplate pre_main (
8493 "${datatype_additional_decl:opt}"
8494 " %ptr_in = OpTypePointer StorageBuffer %${inputType}\n"
8495 " %ptr_out = OpTypePointer StorageBuffer %${outputType}\n"
8496 " %s_SSBOi = OpTypeStruct %${inputType}\n"
8497 " %s_SSBOo = OpTypeStruct %${outputType}\n"
8498 " %ptr_SSBOi = OpTypePointer StorageBuffer %s_SSBOi\n"
8499 " %ptr_SSBOo = OpTypePointer StorageBuffer %s_SSBOo\n"
8500 " %SSBOi = OpVariable %ptr_SSBOi StorageBuffer\n"
8501 " %SSBOo = OpVariable %ptr_SSBOo StorageBuffer\n");
8503 const StringTemplate testfun (
8504 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
8505 "%param = OpFunctionParameter %v4f32\n"
8506 "%label = OpLabel\n"
8507 "%iLoc = OpAccessChain %ptr_in %SSBOi %c_u32_0\n"
8508 "%oLoc = OpAccessChain %ptr_out %SSBOo %c_u32_0\n"
8509 "%valIn = OpLoad %${inputType} %iLoc\n"
8510 "%valOut = ${instruction} %${outputType} %valIn\n"
8511 " OpStore %oLoc %valOut\n"
8512 " OpReturnValue %param\n"
8513 " OpFunctionEnd\n");
8515 params["datatype_extensions"] =
8516 params["datatype_extensions"] +
8517 "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n";
8519 fragments["capability"] = params["datatype_capabilities"];
8520 fragments["extension"] = params["datatype_extensions"];
8521 fragments["decoration"] = decoration.specialize(params);
8522 fragments["pre_main"] = pre_main.specialize(params);
8523 fragments["testfun"] = testfun.specialize(params);
8528 // Test for OpSConvert, OpUConvert and OpFConvert in compute shaders
8529 tcu::TestCaseGroup* createConvertComputeTests (tcu::TestContext& testCtx, const string& instruction, const string& name)
8531 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, name.c_str(), instruction.c_str()));
8532 vector<ConvertCase> testCases;
8533 createConvertCases(testCases, instruction);
8535 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8537 ComputeShaderSpec spec;
8538 spec.assembly = getConvertCaseShaderStr(instruction, *test);
8539 spec.numWorkGroups = IVec3(1, 1, 1);
8540 spec.inputs.push_back (test->m_inputBuffer);
8541 spec.outputs.push_back (test->m_outputBuffer);
8543 if (test->m_features == COMPUTE_TEST_USES_INT16 || test->m_features == COMPUTE_TEST_USES_INT16_INT64 || usesInt16(test->m_fromType, test->m_toType)) {
8544 spec.extensions.push_back("VK_KHR_16bit_storage");
8545 spec.requestedVulkanFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_UNIFORM_BUFFER_BLOCK;
8548 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "", spec, test->m_features));
8550 return group.release();
8553 // Test for OpSConvert, OpUConvert and OpFConvert in graphics shaders
8554 tcu::TestCaseGroup* createConvertGraphicsTests (tcu::TestContext& testCtx, const string& instruction, const string& name)
8556 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, name.c_str(), instruction.c_str()));
8557 vector<ConvertCase> testCases;
8558 createConvertCases(testCases, instruction);
8560 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8562 map<string, string> fragments = getConvertCaseFragments(instruction, *test);
8563 vector<string> features = getFeatureStringVector(test->m_features);
8564 GraphicsResources resources;
8565 vector<string> extensions;
8566 vector<deInt32> noSpecConstants;
8567 PushConstants noPushConstants;
8568 VulkanFeatures vulkanFeatures;
8569 GraphicsInterfaces noInterfaces;
8570 tcu::RGBA defaultColors[4];
8572 getDefaultColors (defaultColors);
8573 resources.inputs.push_back (std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, test->m_inputBuffer));
8574 resources.outputs.push_back (std::make_pair(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, test->m_outputBuffer));
8575 extensions.push_back ("VK_KHR_storage_buffer_storage_class");
8577 if (test->m_features == COMPUTE_TEST_USES_INT16 || test->m_features == COMPUTE_TEST_USES_INT16_INT64 || usesInt16(test->m_fromType, test->m_toType))
8579 extensions.push_back("VK_KHR_16bit_storage");
8580 vulkanFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_UNIFORM_BUFFER_BLOCK;
8583 createTestsForAllStages(
8584 test->m_name, defaultColors, defaultColors, fragments, noSpecConstants,
8585 noPushConstants, resources, noInterfaces, extensions, features, vulkanFeatures, group.get());
8587 return group.release();
8590 const string getNumberTypeName (const NumberType type)
8592 if (type == NUMBERTYPE_INT32)
8596 else if (type == NUMBERTYPE_UINT32)
8600 else if (type == NUMBERTYPE_FLOAT32)
8611 deInt32 getInt(de::Random& rnd)
8613 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
8616 const string repeatString (const string& str, int times)
8619 for (int i = 0; i < times; ++i)
8626 const string getRandomConstantString (const NumberType type, de::Random& rnd)
8628 if (type == NUMBERTYPE_INT32)
8630 return numberToString<deInt32>(getInt(rnd));
8632 else if (type == NUMBERTYPE_UINT32)
8634 return numberToString<deUint32>(rnd.getUint32());
8636 else if (type == NUMBERTYPE_FLOAT32)
8638 return numberToString<float>(rnd.getFloat());
8647 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8649 map<string, string> params;
8652 for (int width = 2; width <= 4; ++width)
8654 const string randomConst = numberToString(getInt(rnd));
8655 const string widthStr = numberToString(width);
8656 const string composite_type = "${customType}vec" + widthStr;
8657 const int index = rnd.getInt(0, width-1);
8659 params["type"] = "vec";
8660 params["name"] = params["type"] + "_" + widthStr;
8661 params["compositeDecl"] = composite_type + " = OpTypeVector ${customType} " + widthStr +"\n";
8662 params["compositeType"] = composite_type;
8663 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8664 params["compositeConstruct"] = "%instance = OpCompositeConstruct " + composite_type + repeatString(" %filler", width) + "\n";
8665 params["indexes"] = numberToString(index);
8666 testCases.push_back(params);
8670 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8672 const int limit = 10;
8673 map<string, string> params;
8675 for (int width = 2; width <= limit; ++width)
8677 string randomConst = numberToString(getInt(rnd));
8678 string widthStr = numberToString(width);
8679 int index = rnd.getInt(0, width-1);
8681 params["type"] = "array";
8682 params["name"] = params["type"] + "_" + widthStr;
8683 params["compositeDecl"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
8684 + "%composite = OpTypeArray ${customType} %arraywidth\n";
8685 params["compositeType"] = "%composite";
8686 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8687 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8688 params["indexes"] = numberToString(index);
8689 testCases.push_back(params);
8693 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8695 const int limit = 10;
8696 map<string, string> params;
8698 for (int width = 2; width <= limit; ++width)
8700 string randomConst = numberToString(getInt(rnd));
8701 int index = rnd.getInt(0, width-1);
8703 params["type"] = "struct";
8704 params["name"] = params["type"] + "_" + numberToString(width);
8705 params["compositeDecl"] = "%composite = OpTypeStruct" + repeatString(" ${customType}", width) + "\n";
8706 params["compositeType"] = "%composite";
8707 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8708 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8709 params["indexes"] = numberToString(index);
8710 testCases.push_back(params);
8714 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8716 map<string, string> params;
8719 for (int width = 2; width <= 4; ++width)
8721 string widthStr = numberToString(width);
8723 for (int column = 2 ; column <= 4; ++column)
8725 int index_0 = rnd.getInt(0, column-1);
8726 int index_1 = rnd.getInt(0, width-1);
8727 string columnStr = numberToString(column);
8729 params["type"] = "matrix";
8730 params["name"] = params["type"] + "_" + widthStr + "x" + columnStr;
8731 params["compositeDecl"] = string("%vectype = OpTypeVector ${customType} " + widthStr + "\n")
8732 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
8733 params["compositeType"] = "%composite";
8735 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n"
8736 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
8738 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
8739 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
8740 testCases.push_back(params);
8745 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8747 createVectorCompositeCases(testCases, rnd, type);
8748 createArrayCompositeCases(testCases, rnd, type);
8749 createStructCompositeCases(testCases, rnd, type);
8750 // Matrix only supports float types
8751 if (type == NUMBERTYPE_FLOAT32)
8753 createMatrixCompositeCases(testCases, rnd, type);
8757 const string getAssemblyTypeDeclaration (const NumberType type)
8761 case NUMBERTYPE_INT32: return "OpTypeInt 32 1";
8762 case NUMBERTYPE_UINT32: return "OpTypeInt 32 0";
8763 case NUMBERTYPE_FLOAT32: return "OpTypeFloat 32";
8764 default: DE_ASSERT(false); return "";
8768 const string getAssemblyTypeName (const NumberType type)
8772 case NUMBERTYPE_INT32: return "%i32";
8773 case NUMBERTYPE_UINT32: return "%u32";
8774 case NUMBERTYPE_FLOAT32: return "%f32";
8775 default: DE_ASSERT(false); return "";
8779 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
8781 map<string, string> parameters(params);
8783 const string customType = getAssemblyTypeName(type);
8784 map<string, string> substCustomType;
8785 substCustomType["customType"] = customType;
8786 parameters["compositeDecl"] = StringTemplate(parameters.at("compositeDecl")).specialize(substCustomType);
8787 parameters["compositeType"] = StringTemplate(parameters.at("compositeType")).specialize(substCustomType);
8788 parameters["compositeConstruct"] = StringTemplate(parameters.at("compositeConstruct")).specialize(substCustomType);
8789 parameters["filler"] = StringTemplate(parameters.at("filler")).specialize(substCustomType);
8790 parameters["customType"] = customType;
8791 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
8793 if (parameters.at("compositeType") != "%u32vec3")
8795 parameters["u32vec3Decl"] = "%u32vec3 = OpTypeVector %u32 3\n";
8798 return StringTemplate(
8799 "OpCapability Shader\n"
8800 "OpCapability Matrix\n"
8801 "OpMemoryModel Logical GLSL450\n"
8802 "OpEntryPoint GLCompute %main \"main\" %id\n"
8803 "OpExecutionMode %main LocalSize 1 1 1\n"
8805 "OpSource GLSL 430\n"
8806 "OpName %main \"main\"\n"
8807 "OpName %id \"gl_GlobalInvocationID\"\n"
8810 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8811 "OpDecorate %buf BufferBlock\n"
8812 "OpDecorate %indata DescriptorSet 0\n"
8813 "OpDecorate %indata Binding 0\n"
8814 "OpDecorate %outdata DescriptorSet 0\n"
8815 "OpDecorate %outdata Binding 1\n"
8816 "OpDecorate %customarr ArrayStride 4\n"
8817 "${compositeDecorator}"
8818 "OpMemberDecorate %buf 0 Offset 0\n"
8821 "%void = OpTypeVoid\n"
8822 "%voidf = OpTypeFunction %void\n"
8823 "%u32 = OpTypeInt 32 0\n"
8824 "%i32 = OpTypeInt 32 1\n"
8825 "%f32 = OpTypeFloat 32\n"
8827 // Composite declaration
8833 "${u32vec3Decl:opt}"
8834 "%uvec3ptr = OpTypePointer Input %u32vec3\n"
8836 // Inherited from custom
8837 "%customptr = OpTypePointer Uniform ${customType}\n"
8838 "%customarr = OpTypeRuntimeArray ${customType}\n"
8839 "%buf = OpTypeStruct %customarr\n"
8840 "%bufptr = OpTypePointer Uniform %buf\n"
8842 "%indata = OpVariable %bufptr Uniform\n"
8843 "%outdata = OpVariable %bufptr Uniform\n"
8845 "%id = OpVariable %uvec3ptr Input\n"
8846 "%zero = OpConstant %i32 0\n"
8848 "%main = OpFunction %void None %voidf\n"
8849 "%label = OpLabel\n"
8850 "%idval = OpLoad %u32vec3 %id\n"
8851 "%x = OpCompositeExtract %u32 %idval 0\n"
8853 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
8854 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
8855 // Read the input value
8856 "%inval = OpLoad ${customType} %inloc\n"
8857 // Create the composite and fill it
8858 "${compositeConstruct}"
8859 // Insert the input value to a place
8860 "%instance2 = OpCompositeInsert ${compositeType} %inval %instance ${indexes}\n"
8861 // Read back the value from the position
8862 "%out_val = OpCompositeExtract ${customType} %instance2 ${indexes}\n"
8863 // Store it in the output position
8864 " OpStore %outloc %out_val\n"
8867 ).specialize(parameters);
8870 template<typename T>
8871 BufferSp createCompositeBuffer(T number)
8873 return BufferSp(new Buffer<T>(vector<T>(1, number)));
8876 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
8878 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
8879 de::Random rnd (deStringHash(group->getName()));
8881 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8883 NumberType numberType = NumberType(type);
8884 const string typeName = getNumberTypeName(numberType);
8885 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
8886 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8887 vector<map<string, string> > testCases;
8889 createCompositeCases(testCases, rnd, numberType);
8891 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8893 ComputeShaderSpec spec;
8895 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
8899 case NUMBERTYPE_INT32:
8901 deInt32 number = getInt(rnd);
8902 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8903 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8906 case NUMBERTYPE_UINT32:
8908 deUint32 number = rnd.getUint32();
8909 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8910 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8913 case NUMBERTYPE_FLOAT32:
8915 float number = rnd.getFloat();
8916 spec.inputs.push_back(createCompositeBuffer<float>(number));
8917 spec.outputs.push_back(createCompositeBuffer<float>(number));
8924 spec.numWorkGroups = IVec3(1, 1, 1);
8925 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
8927 group->addChild(subGroup.release());
8929 return group.release();
8932 struct AssemblyStructInfo
8934 AssemblyStructInfo (const deUint32 comp, const deUint32 idx)
8939 deUint32 components;
8943 const string specializeInBoundsShaderTemplate (const NumberType type, const AssemblyStructInfo& structInfo, const map<string, string>& params)
8945 // Create the full index string
8946 string fullIndex = numberToString(structInfo.index) + " " + params.at("indexes");
8947 // Convert it to list of indexes
8948 vector<string> indexes = de::splitString(fullIndex, ' ');
8950 map<string, string> parameters (params);
8951 parameters["structType"] = repeatString(" ${compositeType}", structInfo.components);
8952 parameters["structConstruct"] = repeatString(" %instance", structInfo.components);
8953 parameters["insertIndexes"] = fullIndex;
8955 // In matrix cases the last two index is the CompositeExtract indexes
8956 const deUint32 extractIndexes = (parameters["type"] == "matrix") ? 2 : 1;
8958 // Construct the extractIndex
8959 for (vector<string>::const_iterator index = indexes.end() - extractIndexes; index != indexes.end(); ++index)
8961 parameters["extractIndexes"] += " " + *index;
8964 // Remove the last 1 or 2 element depends on matrix case or not
8965 indexes.erase(indexes.end() - extractIndexes, indexes.end());
8968 // Generate AccessChain index expressions (except for the last one, because we use ptr to the composite)
8969 for (vector<string>::const_iterator index = indexes.begin(); index != indexes.end(); ++index)
8971 string indexId = "%index_" + numberToString(id++);
8972 parameters["accessChainConstDeclaration"] += indexId + " = OpConstant %u32 " + *index + "\n";
8973 parameters["accessChainIndexes"] += " " + indexId;
8976 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
8978 const string customType = getAssemblyTypeName(type);
8979 map<string, string> substCustomType;
8980 substCustomType["customType"] = customType;
8981 parameters["compositeDecl"] = StringTemplate(parameters.at("compositeDecl")).specialize(substCustomType);
8982 parameters["compositeType"] = StringTemplate(parameters.at("compositeType")).specialize(substCustomType);
8983 parameters["compositeConstruct"] = StringTemplate(parameters.at("compositeConstruct")).specialize(substCustomType);
8984 parameters["filler"] = StringTemplate(parameters.at("filler")).specialize(substCustomType);
8985 parameters["customType"] = customType;
8987 const string compositeType = parameters.at("compositeType");
8988 map<string, string> substCompositeType;
8989 substCompositeType["compositeType"] = compositeType;
8990 parameters["structType"] = StringTemplate(parameters.at("structType")).specialize(substCompositeType);
8991 if (compositeType != "%u32vec3")
8993 parameters["u32vec3Decl"] = "%u32vec3 = OpTypeVector %u32 3\n";
8996 return StringTemplate(
8997 "OpCapability Shader\n"
8998 "OpCapability Matrix\n"
8999 "OpMemoryModel Logical GLSL450\n"
9000 "OpEntryPoint GLCompute %main \"main\" %id\n"
9001 "OpExecutionMode %main LocalSize 1 1 1\n"
9003 "OpSource GLSL 430\n"
9004 "OpName %main \"main\"\n"
9005 "OpName %id \"gl_GlobalInvocationID\"\n"
9007 "OpDecorate %id BuiltIn GlobalInvocationId\n"
9008 "OpDecorate %buf BufferBlock\n"
9009 "OpDecorate %indata DescriptorSet 0\n"
9010 "OpDecorate %indata Binding 0\n"
9011 "OpDecorate %outdata DescriptorSet 0\n"
9012 "OpDecorate %outdata Binding 1\n"
9013 "OpDecorate %customarr ArrayStride 4\n"
9014 "${compositeDecorator}"
9015 "OpMemberDecorate %buf 0 Offset 0\n"
9017 "%void = OpTypeVoid\n"
9018 "%voidf = OpTypeFunction %void\n"
9019 "%i32 = OpTypeInt 32 1\n"
9020 "%u32 = OpTypeInt 32 0\n"
9021 "%f32 = OpTypeFloat 32\n"
9024 // %u32vec3 if not already declared in ${compositeDecl}
9025 "${u32vec3Decl:opt}"
9026 "%uvec3ptr = OpTypePointer Input %u32vec3\n"
9027 // Inherited from composite
9028 "%composite_p = OpTypePointer Function ${compositeType}\n"
9029 "%struct_t = OpTypeStruct${structType}\n"
9030 "%struct_p = OpTypePointer Function %struct_t\n"
9033 "${accessChainConstDeclaration}"
9034 // Inherited from custom
9035 "%customptr = OpTypePointer Uniform ${customType}\n"
9036 "%customarr = OpTypeRuntimeArray ${customType}\n"
9037 "%buf = OpTypeStruct %customarr\n"
9038 "%bufptr = OpTypePointer Uniform %buf\n"
9039 "%indata = OpVariable %bufptr Uniform\n"
9040 "%outdata = OpVariable %bufptr Uniform\n"
9042 "%id = OpVariable %uvec3ptr Input\n"
9043 "%zero = OpConstant %u32 0\n"
9044 "%main = OpFunction %void None %voidf\n"
9045 "%label = OpLabel\n"
9046 "%struct_v = OpVariable %struct_p Function\n"
9047 "%idval = OpLoad %u32vec3 %id\n"
9048 "%x = OpCompositeExtract %u32 %idval 0\n"
9049 // Create the input/output type
9050 "%inloc = OpInBoundsAccessChain %customptr %indata %zero %x\n"
9051 "%outloc = OpInBoundsAccessChain %customptr %outdata %zero %x\n"
9052 // Read the input value
9053 "%inval = OpLoad ${customType} %inloc\n"
9054 // Create the composite and fill it
9055 "${compositeConstruct}"
9056 // Create the struct and fill it with the composite
9057 "%struct = OpCompositeConstruct %struct_t${structConstruct}\n"
9059 "%comp_obj = OpCompositeInsert %struct_t %inval %struct ${insertIndexes}\n"
9061 " OpStore %struct_v %comp_obj\n"
9062 // Get deepest possible composite pointer
9063 "%inner_ptr = OpInBoundsAccessChain %composite_p %struct_v${accessChainIndexes}\n"
9064 "%read_obj = OpLoad ${compositeType} %inner_ptr\n"
9065 // Read back the stored value
9066 "%read_val = OpCompositeExtract ${customType} %read_obj${extractIndexes}\n"
9067 " OpStore %outloc %read_val\n"
9070 ).specialize(parameters);
9073 tcu::TestCaseGroup* createOpInBoundsAccessChainGroup (tcu::TestContext& testCtx)
9075 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opinboundsaccesschain", "Test the OpInBoundsAccessChain instruction"));
9076 de::Random rnd (deStringHash(group->getName()));
9078 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
9080 NumberType numberType = NumberType(type);
9081 const string typeName = getNumberTypeName(numberType);
9082 const string description = "Test the OpInBoundsAccessChain instruction with " + typeName + "s";
9083 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
9085 vector<map<string, string> > testCases;
9086 createCompositeCases(testCases, rnd, numberType);
9088 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
9090 ComputeShaderSpec spec;
9092 // Number of components inside of a struct
9093 deUint32 structComponents = rnd.getInt(2, 8);
9094 // Component index value
9095 deUint32 structIndex = rnd.getInt(0, structComponents - 1);
9096 AssemblyStructInfo structInfo(structComponents, structIndex);
9098 spec.assembly = specializeInBoundsShaderTemplate(numberType, structInfo, *test);
9102 case NUMBERTYPE_INT32:
9104 deInt32 number = getInt(rnd);
9105 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
9106 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
9109 case NUMBERTYPE_UINT32:
9111 deUint32 number = rnd.getUint32();
9112 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
9113 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
9116 case NUMBERTYPE_FLOAT32:
9118 float number = rnd.getFloat();
9119 spec.inputs.push_back(createCompositeBuffer<float>(number));
9120 spec.outputs.push_back(createCompositeBuffer<float>(number));
9126 spec.numWorkGroups = IVec3(1, 1, 1);
9127 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpInBoundsAccessChain test", spec));
9129 group->addChild(subGroup.release());
9131 return group.release();
9134 // If the params missing, uninitialized case
9135 const string specializeDefaultOutputShaderTemplate (const NumberType type, const map<string, string>& params = map<string, string>())
9137 map<string, string> parameters(params);
9139 parameters["customType"] = getAssemblyTypeName(type);
9141 // Declare the const value, and use it in the initializer
9142 if (params.find("constValue") != params.end())
9144 parameters["variableInitializer"] = " %const";
9146 // Uninitialized case
9149 parameters["commentDecl"] = ";";
9152 return StringTemplate(
9153 "OpCapability Shader\n"
9154 "OpMemoryModel Logical GLSL450\n"
9155 "OpEntryPoint GLCompute %main \"main\" %id\n"
9156 "OpExecutionMode %main LocalSize 1 1 1\n"
9157 "OpSource GLSL 430\n"
9158 "OpName %main \"main\"\n"
9159 "OpName %id \"gl_GlobalInvocationID\"\n"
9161 "OpDecorate %id BuiltIn GlobalInvocationId\n"
9162 "OpDecorate %indata DescriptorSet 0\n"
9163 "OpDecorate %indata Binding 0\n"
9164 "OpDecorate %outdata DescriptorSet 0\n"
9165 "OpDecorate %outdata Binding 1\n"
9166 "OpDecorate %in_arr ArrayStride 4\n"
9167 "OpDecorate %in_buf BufferBlock\n"
9168 "OpMemberDecorate %in_buf 0 Offset 0\n"
9170 "%void = OpTypeVoid\n"
9171 "%voidf = OpTypeFunction %void\n"
9172 "%u32 = OpTypeInt 32 0\n"
9173 "%i32 = OpTypeInt 32 1\n"
9174 "%f32 = OpTypeFloat 32\n"
9175 "%uvec3 = OpTypeVector %u32 3\n"
9176 "%uvec3ptr = OpTypePointer Input %uvec3\n"
9177 "${commentDecl:opt}%const = OpConstant ${customType} ${constValue:opt}\n"
9179 "%in_ptr = OpTypePointer Uniform ${customType}\n"
9180 "%in_arr = OpTypeRuntimeArray ${customType}\n"
9181 "%in_buf = OpTypeStruct %in_arr\n"
9182 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
9183 "%indata = OpVariable %in_bufptr Uniform\n"
9184 "%outdata = OpVariable %in_bufptr Uniform\n"
9185 "%id = OpVariable %uvec3ptr Input\n"
9186 "%var_ptr = OpTypePointer Function ${customType}\n"
9188 "%zero = OpConstant %i32 0\n"
9190 "%main = OpFunction %void None %voidf\n"
9191 "%label = OpLabel\n"
9192 "%out_var = OpVariable %var_ptr Function${variableInitializer:opt}\n"
9193 "%idval = OpLoad %uvec3 %id\n"
9194 "%x = OpCompositeExtract %u32 %idval 0\n"
9195 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
9196 "%outloc = OpAccessChain %in_ptr %outdata %zero %x\n"
9198 "%outval = OpLoad ${customType} %out_var\n"
9199 " OpStore %outloc %outval\n"
9202 ).specialize(parameters);
9205 bool compareFloats (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
9207 DE_ASSERT(outputAllocs.size() != 0);
9208 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
9210 // Use custom epsilon because of the float->string conversion
9211 const float epsilon = 0.00001f;
9213 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
9215 vector<deUint8> expectedBytes;
9219 expectedOutputs[outputNdx]->getBytes(expectedBytes);
9220 memcpy(&expected, &expectedBytes.front(), expectedBytes.size());
9221 memcpy(&actual, outputAllocs[outputNdx]->getHostPtr(), expectedBytes.size());
9223 // Test with epsilon
9224 if (fabs(expected - actual) > epsilon)
9226 log << TestLog::Message << "Error: The actual and expected values not matching."
9227 << " Expected: " << expected << " Actual: " << actual << " Epsilon: " << epsilon << TestLog::EndMessage;
9234 // Checks if the driver crash with uninitialized cases
9235 bool passthruVerify (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
9237 DE_ASSERT(outputAllocs.size() != 0);
9238 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
9240 // Copy and discard the result.
9241 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
9243 vector<deUint8> expectedBytes;
9244 expectedOutputs[outputNdx]->getBytes(expectedBytes);
9246 const size_t width = expectedBytes.size();
9247 vector<char> data (width);
9249 memcpy(&data[0], outputAllocs[outputNdx]->getHostPtr(), width);
9254 tcu::TestCaseGroup* createShaderDefaultOutputGroup (tcu::TestContext& testCtx)
9256 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "shader_default_output", "Test shader default output."));
9257 de::Random rnd (deStringHash(group->getName()));
9259 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
9261 NumberType numberType = NumberType(type);
9262 const string typeName = getNumberTypeName(numberType);
9263 const string description = "Test the OpVariable initializer with " + typeName + ".";
9264 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
9266 // 2 similar subcases (initialized and uninitialized)
9267 for (int subCase = 0; subCase < 2; ++subCase)
9269 ComputeShaderSpec spec;
9270 spec.numWorkGroups = IVec3(1, 1, 1);
9272 map<string, string> params;
9276 case NUMBERTYPE_INT32:
9278 deInt32 number = getInt(rnd);
9279 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
9280 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
9281 params["constValue"] = numberToString(number);
9284 case NUMBERTYPE_UINT32:
9286 deUint32 number = rnd.getUint32();
9287 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
9288 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
9289 params["constValue"] = numberToString(number);
9292 case NUMBERTYPE_FLOAT32:
9294 float number = rnd.getFloat();
9295 spec.inputs.push_back(createCompositeBuffer<float>(number));
9296 spec.outputs.push_back(createCompositeBuffer<float>(number));
9297 spec.verifyIO = &compareFloats;
9298 params["constValue"] = numberToString(number);
9305 // Initialized subcase
9308 spec.assembly = specializeDefaultOutputShaderTemplate(numberType, params);
9309 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "initialized", "OpVariable initializer tests.", spec));
9311 // Uninitialized subcase
9314 spec.assembly = specializeDefaultOutputShaderTemplate(numberType);
9315 spec.verifyIO = &passthruVerify;
9316 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "uninitialized", "OpVariable initializer tests.", spec));
9319 group->addChild(subGroup.release());
9321 return group.release();
9324 tcu::TestCaseGroup* createOpNopTests (tcu::TestContext& testCtx)
9326 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
9327 RGBA defaultColors[4];
9328 map<string, string> opNopFragments;
9330 getDefaultColors(defaultColors);
9332 opNopFragments["testfun"] =
9333 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
9334 "%param1 = OpFunctionParameter %v4f32\n"
9335 "%label_testfun = OpLabel\n"
9344 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
9345 "%b = OpFAdd %f32 %a %a\n"
9347 "%c = OpFSub %f32 %b %a\n"
9348 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
9351 "OpReturnValue %ret\n"
9354 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, testGroup.get());
9356 return testGroup.release();
9359 tcu::TestCaseGroup* createOpNameTests (tcu::TestContext& testCtx)
9361 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opname","Test OpName"));
9362 RGBA defaultColors[4];
9363 map<string, string> opNameFragments;
9365 getDefaultColors(defaultColors);
9367 opNameFragments["debug"] =
9368 "OpName %BP_main \"not_main\"";
9370 opNameFragments["testfun"] =
9371 "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n"
9372 "%param1 = OpFunctionParameter %v4f32\n"
9373 "%label_func = OpLabel\n"
9374 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
9375 "%b = OpFAdd %f32 %a %a\n"
9376 "%c = OpFSub %f32 %b %a\n"
9377 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
9378 "OpReturnValue %ret\n"
9381 createTestsForAllStages("opname", defaultColors, defaultColors, opNameFragments, testGroup.get());
9383 return testGroup.release();
9386 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
9388 const bool testComputePipeline = true;
9390 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
9391 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
9392 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
9394 computeTests->addChild(createSpivVersionCheckTests(testCtx, testComputePipeline));
9395 computeTests->addChild(createLocalSizeGroup(testCtx));
9396 computeTests->addChild(createOpNopGroup(testCtx));
9397 computeTests->addChild(createOpFUnordGroup(testCtx));
9398 computeTests->addChild(createOpAtomicGroup(testCtx, false));
9399 computeTests->addChild(createOpAtomicGroup(testCtx, true)); // Using new StorageBuffer decoration
9400 computeTests->addChild(createOpAtomicGroup(testCtx, false, 1024, true)); // Return value validation
9401 computeTests->addChild(createOpLineGroup(testCtx));
9402 computeTests->addChild(createOpModuleProcessedGroup(testCtx));
9403 computeTests->addChild(createOpNoLineGroup(testCtx));
9404 computeTests->addChild(createOpConstantNullGroup(testCtx));
9405 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
9406 computeTests->addChild(createOpConstantUsageGroup(testCtx));
9407 computeTests->addChild(createSpecConstantGroup(testCtx));
9408 computeTests->addChild(createOpSourceGroup(testCtx));
9409 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
9410 computeTests->addChild(createDecorationGroupGroup(testCtx));
9411 computeTests->addChild(createOpPhiGroup(testCtx));
9412 computeTests->addChild(createLoopControlGroup(testCtx));
9413 computeTests->addChild(createFunctionControlGroup(testCtx));
9414 computeTests->addChild(createSelectionControlGroup(testCtx));
9415 computeTests->addChild(createBlockOrderGroup(testCtx));
9416 computeTests->addChild(createMultipleShaderGroup(testCtx));
9417 computeTests->addChild(createMemoryAccessGroup(testCtx));
9418 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
9419 computeTests->addChild(createOpCopyObjectGroup(testCtx));
9420 computeTests->addChild(createNoContractionGroup(testCtx));
9421 computeTests->addChild(createOpUndefGroup(testCtx));
9422 computeTests->addChild(createOpUnreachableGroup(testCtx));
9423 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
9424 computeTests ->addChild(createOpFRemGroup(testCtx));
9425 computeTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_PASS));
9426 computeTests->addChild(createOpSRemComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
9427 computeTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_PASS));
9428 computeTests->addChild(createOpSModComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
9429 computeTests->addChild(createConvertComputeTests(testCtx, "OpSConvert", "sconvert"));
9430 computeTests->addChild(createConvertComputeTests(testCtx, "OpUConvert", "uconvert"));
9431 computeTests->addChild(createConvertComputeTests(testCtx, "OpFConvert", "fconvert"));
9432 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
9433 computeTests->addChild(createOpInBoundsAccessChainGroup(testCtx));
9434 computeTests->addChild(createShaderDefaultOutputGroup(testCtx));
9435 computeTests->addChild(createOpNMinGroup(testCtx));
9436 computeTests->addChild(createOpNMaxGroup(testCtx));
9437 computeTests->addChild(createOpNClampGroup(testCtx));
9439 de::MovePtr<tcu::TestCaseGroup> computeAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
9441 computeAndroidTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9442 computeAndroidTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9444 computeTests->addChild(computeAndroidTests.release());
9447 computeTests->addChild(create8BitStorageComputeGroup(testCtx));
9448 computeTests->addChild(create16BitStorageComputeGroup(testCtx));
9449 computeTests->addChild(createUboMatrixPaddingComputeGroup(testCtx));
9450 computeTests->addChild(createVariableInitComputeGroup(testCtx));
9451 computeTests->addChild(createConditionalBranchComputeGroup(testCtx));
9452 computeTests->addChild(createIndexingComputeGroup(testCtx));
9453 computeTests->addChild(createVariablePointersComputeGroup(testCtx));
9454 computeTests->addChild(createImageSamplerComputeGroup(testCtx));
9455 computeTests->addChild(createOpNameGroup(testCtx));
9456 computeTests->addChild(createPointerParameterComputeGroup(testCtx));
9457 graphicsTests->addChild(createCrossStageInterfaceTests(testCtx));
9458 graphicsTests->addChild(createSpivVersionCheckTests(testCtx, !testComputePipeline));
9459 graphicsTests->addChild(createOpNopTests(testCtx));
9460 graphicsTests->addChild(createOpSourceTests(testCtx));
9461 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
9462 graphicsTests->addChild(createOpModuleProcessedTests(testCtx));
9463 graphicsTests->addChild(createOpLineTests(testCtx));
9464 graphicsTests->addChild(createOpNoLineTests(testCtx));
9465 graphicsTests->addChild(createOpConstantNullTests(testCtx));
9466 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
9467 graphicsTests->addChild(createMemoryAccessTests(testCtx));
9468 graphicsTests->addChild(createOpUndefTests(testCtx));
9469 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
9470 graphicsTests->addChild(createModuleTests(testCtx));
9471 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
9472 graphicsTests->addChild(createOpPhiTests(testCtx));
9473 graphicsTests->addChild(createNoContractionTests(testCtx));
9474 graphicsTests->addChild(createOpQuantizeTests(testCtx));
9475 graphicsTests->addChild(createLoopTests(testCtx));
9476 graphicsTests->addChild(createSpecConstantTests(testCtx));
9477 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
9478 graphicsTests->addChild(createBarrierTests(testCtx));
9479 graphicsTests->addChild(createDecorationGroupTests(testCtx));
9480 graphicsTests->addChild(createFRemTests(testCtx));
9481 graphicsTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
9482 graphicsTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
9485 de::MovePtr<tcu::TestCaseGroup> graphicsAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
9487 graphicsAndroidTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9488 graphicsAndroidTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9490 graphicsTests->addChild(graphicsAndroidTests.release());
9492 graphicsTests->addChild(createOpNameTests(testCtx));
9494 graphicsTests->addChild(create8BitStorageGraphicsGroup(testCtx));
9495 graphicsTests->addChild(create16BitStorageGraphicsGroup(testCtx));
9496 graphicsTests->addChild(createUboMatrixPaddingGraphicsGroup(testCtx));
9497 graphicsTests->addChild(createVariableInitGraphicsGroup(testCtx));
9498 graphicsTests->addChild(createConditionalBranchGraphicsGroup(testCtx));
9499 graphicsTests->addChild(createIndexingGraphicsGroup(testCtx));
9500 graphicsTests->addChild(createVariablePointersGraphicsGroup(testCtx));
9501 graphicsTests->addChild(createImageSamplerGraphicsGroup(testCtx));
9502 graphicsTests->addChild(createConvertGraphicsTests(testCtx, "OpSConvert", "sconvert"));
9503 graphicsTests->addChild(createConvertGraphicsTests(testCtx, "OpUConvert", "uconvert"));
9504 graphicsTests->addChild(createConvertGraphicsTests(testCtx, "OpFConvert", "fconvert"));
9505 graphicsTests->addChild(createPointerParameterGraphicsGroup(testCtx));
9507 instructionTests->addChild(computeTests.release());
9508 instructionTests->addChild(graphicsTests.release());
9510 return instructionTests.release();