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 "vktSpvAsm16bitStorageTests.hpp"
53 #include "vktSpvAsmUboMatrixPaddingTests.hpp"
54 #include "vktSpvAsmConditionalBranchTests.hpp"
55 #include "vktSpvAsmIndexingTests.hpp"
56 #include "vktSpvAsmImageSamplerTests.hpp"
57 #include "vktSpvAsmComputeShaderCase.hpp"
58 #include "vktSpvAsmComputeShaderTestUtil.hpp"
59 #include "vktSpvAsmGraphicsShaderTestUtil.hpp"
60 #include "vktSpvAsmVariablePointersTests.hpp"
61 #include "vktTestCaseUtil.hpp"
73 namespace SpirVAssembly
87 using tcu::TestStatus;
90 using tcu::StringTemplate;
94 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0)
96 T* const typedPtr = (T*)dst;
97 for (int ndx = 0; ndx < numValues; ndx++)
98 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue);
101 // Filter is a function that returns true if a value should pass, false otherwise.
102 template<typename T, typename FilterT>
103 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, FilterT filter, int offset = 0)
105 T* const typedPtr = (T*)dst;
107 for (int ndx = 0; ndx < numValues; ndx++)
110 value = randomScalar<T>(rnd, minValue, maxValue);
111 while (!filter(value));
113 typedPtr[offset + ndx] = value;
117 // Gets a 64-bit integer with a more logarithmic distribution
118 deInt64 randomInt64LogDistributed (de::Random& rnd)
120 deInt64 val = rnd.getUint64();
121 val &= (1ull << rnd.getInt(1, 63)) - 1;
127 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues)
129 for (int ndx = 0; ndx < numValues; ndx++)
130 dst[ndx] = randomInt64LogDistributed(rnd);
133 template<typename FilterT>
134 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues, FilterT filter)
136 for (int ndx = 0; ndx < numValues; ndx++)
140 value = randomInt64LogDistributed(rnd);
141 } while (!filter(value));
146 inline bool filterNonNegative (const deInt64 value)
151 inline bool filterPositive (const deInt64 value)
156 inline bool filterNotZero (const deInt64 value)
161 static void floorAll (vector<float>& values)
163 for (size_t i = 0; i < values.size(); i++)
164 values[i] = deFloatFloor(values[i]);
167 static void floorAll (vector<Vec4>& values)
169 for (size_t i = 0; i < values.size(); i++)
170 values[i] = floor(values[i]);
178 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {}
181 // Assembly code used for testing LocalSize, OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code:
185 // layout(std140, set = 0, binding = 0) readonly buffer Input {
188 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
192 // layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
195 // uint x = gl_GlobalInvocationID.x;
196 // output_data.elements[x] = -input_data.elements[x];
199 static string getAsmForLocalSizeTest(bool useLiteralLocalSize, bool useSpecConstantWorkgroupSize, IVec3 workGroupSize, deUint32 ndx)
201 std::ostringstream out;
202 out << getComputeAsmShaderPreambleWithoutLocalSize();
204 if (useLiteralLocalSize)
206 out << "OpExecutionMode %main LocalSize "
207 << workGroupSize.x() << " " << workGroupSize.y() << " " << workGroupSize.z() << "\n";
210 out << "OpSource GLSL 430\n"
211 "OpName %main \"main\"\n"
212 "OpName %id \"gl_GlobalInvocationID\"\n"
213 "OpDecorate %id BuiltIn GlobalInvocationId\n";
215 if (useSpecConstantWorkgroupSize)
217 out << "OpDecorate %spec_0 SpecId 100\n"
218 << "OpDecorate %spec_1 SpecId 101\n"
219 << "OpDecorate %spec_2 SpecId 102\n"
220 << "OpDecorate %gl_WorkGroupSize BuiltIn WorkgroupSize\n";
223 out << getComputeAsmInputOutputBufferTraits()
224 << getComputeAsmCommonTypes()
225 << getComputeAsmInputOutputBuffer()
226 << "%id = OpVariable %uvec3ptr Input\n"
227 << "%zero = OpConstant %i32 0 \n";
229 if (useSpecConstantWorkgroupSize)
231 out << "%spec_0 = OpSpecConstant %u32 "<< workGroupSize.x() << "\n"
232 << "%spec_1 = OpSpecConstant %u32 "<< workGroupSize.y() << "\n"
233 << "%spec_2 = OpSpecConstant %u32 "<< workGroupSize.z() << "\n"
234 << "%gl_WorkGroupSize = OpSpecConstantComposite %uvec3 %spec_0 %spec_1 %spec_2\n";
237 out << "%main = OpFunction %void None %voidf\n"
238 << "%label = OpLabel\n"
239 << "%idval = OpLoad %uvec3 %id\n"
240 << "%ndx = OpCompositeExtract %u32 %idval " << ndx << "\n"
242 "%inloc = OpAccessChain %f32ptr %indata %zero %ndx\n"
243 "%inval = OpLoad %f32 %inloc\n"
244 "%neg = OpFNegate %f32 %inval\n"
245 "%outloc = OpAccessChain %f32ptr %outdata %zero %ndx\n"
246 " OpStore %outloc %neg\n"
252 tcu::TestCaseGroup* createLocalSizeGroup (tcu::TestContext& testCtx)
254 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "localsize", ""));
255 ComputeShaderSpec spec;
256 de::Random rnd (deStringHash(group->getName()));
257 const deUint32 numElements = 64u;
258 vector<float> positiveFloats (numElements, 0);
259 vector<float> negativeFloats (numElements, 0);
261 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
263 for (size_t ndx = 0; ndx < numElements; ++ndx)
264 negativeFloats[ndx] = -positiveFloats[ndx];
266 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
267 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
269 spec.numWorkGroups = IVec3(numElements, 1, 1);
271 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, 1, 1), 0u);
272 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize", "", spec));
274 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, 1, 1), 0u);
275 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize", "", spec));
277 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, 1, 1), 0u);
278 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize", "", spec));
280 spec.numWorkGroups = IVec3(1, 1, 1);
282 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(numElements, 1, 1), 0u);
283 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_x", "", spec));
285 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(numElements, 1, 1), 0u);
286 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_x", "", spec));
288 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(numElements, 1, 1), 0u);
289 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_x", "", spec));
291 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, numElements, 1), 1u);
292 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_y", "", spec));
294 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, numElements, 1), 1u);
295 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_y", "", spec));
297 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, numElements, 1), 1u);
298 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_y", "", spec));
300 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, 1, numElements), 2u);
301 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_z", "", spec));
303 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, 1, numElements), 2u);
304 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_z", "", spec));
306 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, 1, numElements), 2u);
307 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_z", "", spec));
309 return group.release();
312 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
314 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
315 ComputeShaderSpec spec;
316 de::Random rnd (deStringHash(group->getName()));
317 const int numElements = 100;
318 vector<float> positiveFloats (numElements, 0);
319 vector<float> negativeFloats (numElements, 0);
321 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
323 for (size_t ndx = 0; ndx < numElements; ++ndx)
324 negativeFloats[ndx] = -positiveFloats[ndx];
327 string(getComputeAsmShaderPreamble()) +
329 "OpSource GLSL 430\n"
330 "OpName %main \"main\"\n"
331 "OpName %id \"gl_GlobalInvocationID\"\n"
333 "OpDecorate %id BuiltIn GlobalInvocationId\n"
335 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes())
337 + string(getComputeAsmInputOutputBuffer()) +
339 "%id = OpVariable %uvec3ptr Input\n"
340 "%zero = OpConstant %i32 0\n"
342 "%main = OpFunction %void None %voidf\n"
344 "%idval = OpLoad %uvec3 %id\n"
345 "%x = OpCompositeExtract %u32 %idval 0\n"
347 " OpNop\n" // Inside a function body
349 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
350 "%inval = OpLoad %f32 %inloc\n"
351 "%neg = OpFNegate %f32 %inval\n"
352 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
353 " OpStore %outloc %neg\n"
356 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
357 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
358 spec.numWorkGroups = IVec3(numElements, 1, 1);
360 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
362 return group.release();
365 bool compareFUnord (const std::vector<BufferSp>& inputs, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
367 if (outputAllocs.size() != 1)
370 vector<deUint8> input1Bytes;
371 vector<deUint8> input2Bytes;
372 vector<deUint8> expectedBytes;
374 inputs[0]->getBytes(input1Bytes);
375 inputs[1]->getBytes(input2Bytes);
376 expectedOutputs[0]->getBytes(expectedBytes);
378 const deInt32* const expectedOutputAsInt = reinterpret_cast<const deInt32* const>(&expectedBytes.front());
379 const deInt32* const outputAsInt = static_cast<const deInt32* const>(outputAllocs[0]->getHostPtr());
380 const float* const input1AsFloat = reinterpret_cast<const float* const>(&input1Bytes.front());
381 const float* const input2AsFloat = reinterpret_cast<const float* const>(&input2Bytes.front());
382 bool returnValue = true;
384 for (size_t idx = 0; idx < expectedBytes.size() / sizeof(deInt32); ++idx)
386 if (outputAsInt[idx] != expectedOutputAsInt[idx])
388 log << TestLog::Message << "ERROR: Sub-case failed. inputs: " << input1AsFloat[idx] << "," << input2AsFloat[idx] << " output: " << outputAsInt[idx]<< " expected output: " << expectedOutputAsInt[idx] << TestLog::EndMessage;
395 typedef VkBool32 (*compareFuncType) (float, float);
401 compareFuncType compareFunc;
403 OpFUnordCase (const char* _name, const char* _opCode, compareFuncType _compareFunc)
406 , compareFunc (_compareFunc) {}
409 #define ADD_OPFUNORD_CASE(NAME, OPCODE, OPERATOR) \
411 struct compare_##NAME { static VkBool32 compare(float x, float y) { return (x OPERATOR y) ? VK_TRUE : VK_FALSE; } }; \
412 cases.push_back(OpFUnordCase(#NAME, OPCODE, compare_##NAME::compare)); \
413 } while (deGetFalse())
415 tcu::TestCaseGroup* createOpFUnordGroup (tcu::TestContext& testCtx)
417 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfunord", "Test the OpFUnord* opcodes"));
418 de::Random rnd (deStringHash(group->getName()));
419 const int numElements = 100;
420 vector<OpFUnordCase> cases;
422 const StringTemplate shaderTemplate (
424 string(getComputeAsmShaderPreamble()) +
426 "OpSource GLSL 430\n"
427 "OpName %main \"main\"\n"
428 "OpName %id \"gl_GlobalInvocationID\"\n"
430 "OpDecorate %id BuiltIn GlobalInvocationId\n"
432 "OpDecorate %buf BufferBlock\n"
433 "OpDecorate %buf2 BufferBlock\n"
434 "OpDecorate %indata1 DescriptorSet 0\n"
435 "OpDecorate %indata1 Binding 0\n"
436 "OpDecorate %indata2 DescriptorSet 0\n"
437 "OpDecorate %indata2 Binding 1\n"
438 "OpDecorate %outdata DescriptorSet 0\n"
439 "OpDecorate %outdata Binding 2\n"
440 "OpDecorate %f32arr ArrayStride 4\n"
441 "OpDecorate %i32arr ArrayStride 4\n"
442 "OpMemberDecorate %buf 0 Offset 0\n"
443 "OpMemberDecorate %buf2 0 Offset 0\n"
445 + string(getComputeAsmCommonTypes()) +
447 "%buf = OpTypeStruct %f32arr\n"
448 "%bufptr = OpTypePointer Uniform %buf\n"
449 "%indata1 = OpVariable %bufptr Uniform\n"
450 "%indata2 = OpVariable %bufptr Uniform\n"
452 "%buf2 = OpTypeStruct %i32arr\n"
453 "%buf2ptr = OpTypePointer Uniform %buf2\n"
454 "%outdata = OpVariable %buf2ptr Uniform\n"
456 "%id = OpVariable %uvec3ptr Input\n"
457 "%zero = OpConstant %i32 0\n"
458 "%consti1 = OpConstant %i32 1\n"
459 "%constf1 = OpConstant %f32 1.0\n"
461 "%main = OpFunction %void None %voidf\n"
463 "%idval = OpLoad %uvec3 %id\n"
464 "%x = OpCompositeExtract %u32 %idval 0\n"
466 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
467 "%inval1 = OpLoad %f32 %inloc1\n"
468 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
469 "%inval2 = OpLoad %f32 %inloc2\n"
470 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
472 "%result = ${OPCODE} %bool %inval1 %inval2\n"
473 "%int_res = OpSelect %i32 %result %consti1 %zero\n"
474 " OpStore %outloc %int_res\n"
479 ADD_OPFUNORD_CASE(equal, "OpFUnordEqual", ==);
480 ADD_OPFUNORD_CASE(less, "OpFUnordLessThan", <);
481 ADD_OPFUNORD_CASE(lessequal, "OpFUnordLessThanEqual", <=);
482 ADD_OPFUNORD_CASE(greater, "OpFUnordGreaterThan", >);
483 ADD_OPFUNORD_CASE(greaterequal, "OpFUnordGreaterThanEqual", >=);
484 ADD_OPFUNORD_CASE(notequal, "OpFUnordNotEqual", !=);
486 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
488 map<string, string> specializations;
489 ComputeShaderSpec spec;
490 const float NaN = std::numeric_limits<float>::quiet_NaN();
491 vector<float> inputFloats1 (numElements, 0);
492 vector<float> inputFloats2 (numElements, 0);
493 vector<deInt32> expectedInts (numElements, 0);
495 specializations["OPCODE"] = cases[caseNdx].opCode;
496 spec.assembly = shaderTemplate.specialize(specializations);
498 fillRandomScalars(rnd, 1.f, 100.f, &inputFloats1[0], numElements);
499 for (size_t ndx = 0; ndx < numElements; ++ndx)
503 case 0: inputFloats2[ndx] = inputFloats1[ndx] + 1.0f; break;
504 case 1: inputFloats2[ndx] = inputFloats1[ndx] - 1.0f; break;
505 case 2: inputFloats2[ndx] = inputFloats1[ndx]; break;
506 case 3: inputFloats2[ndx] = NaN; break;
507 case 4: inputFloats2[ndx] = inputFloats1[ndx]; inputFloats1[ndx] = NaN; break;
508 case 5: inputFloats2[ndx] = NaN; inputFloats1[ndx] = NaN; break;
510 expectedInts[ndx] = tcu::Float32(inputFloats1[ndx]).isNaN() || tcu::Float32(inputFloats2[ndx]).isNaN() || cases[caseNdx].compareFunc(inputFloats1[ndx], inputFloats2[ndx]);
513 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
514 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
515 spec.outputs.push_back(BufferSp(new Int32Buffer(expectedInts)));
516 spec.numWorkGroups = IVec3(numElements, 1, 1);
517 spec.verifyIO = &compareFUnord;
518 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
521 return group.release();
527 const char* assembly;
528 OpAtomicType opAtomic;
529 deInt32 numOutputElements;
531 OpAtomicCase (const char* _name, const char* _assembly, OpAtomicType _opAtomic, deInt32 _numOutputElements)
533 , assembly (_assembly)
534 , opAtomic (_opAtomic)
535 , numOutputElements (_numOutputElements) {}
538 tcu::TestCaseGroup* createOpAtomicGroup (tcu::TestContext& testCtx, bool useStorageBuffer)
540 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx,
541 useStorageBuffer ? "opatomic_storage_buffer" : "opatomic",
542 "Test the OpAtomic* opcodes"));
543 const int numElements = 65535;
544 vector<OpAtomicCase> cases;
546 const StringTemplate shaderTemplate (
548 string("OpCapability Shader\n") +
549 (useStorageBuffer ? "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" : "") +
550 "OpMemoryModel Logical GLSL450\n"
551 "OpEntryPoint GLCompute %main \"main\" %id\n"
552 "OpExecutionMode %main LocalSize 1 1 1\n" +
554 "OpSource GLSL 430\n"
555 "OpName %main \"main\"\n"
556 "OpName %id \"gl_GlobalInvocationID\"\n"
558 "OpDecorate %id BuiltIn GlobalInvocationId\n"
560 "OpDecorate %buf ${BLOCK_DECORATION}\n"
561 "OpDecorate %indata DescriptorSet 0\n"
562 "OpDecorate %indata Binding 0\n"
563 "OpDecorate %i32arr ArrayStride 4\n"
564 "OpMemberDecorate %buf 0 Offset 0\n"
566 "OpDecorate %sumbuf ${BLOCK_DECORATION}\n"
567 "OpDecorate %sum DescriptorSet 0\n"
568 "OpDecorate %sum Binding 1\n"
569 "OpMemberDecorate %sumbuf 0 Coherent\n"
570 "OpMemberDecorate %sumbuf 0 Offset 0\n"
572 + getComputeAsmCommonTypes("${BLOCK_POINTER_TYPE}") +
574 "%buf = OpTypeStruct %i32arr\n"
575 "%bufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %buf\n"
576 "%indata = OpVariable %bufptr ${BLOCK_POINTER_TYPE}\n"
578 "%sumbuf = OpTypeStruct %i32arr\n"
579 "%sumbufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %sumbuf\n"
580 "%sum = OpVariable %sumbufptr ${BLOCK_POINTER_TYPE}\n"
582 "%id = OpVariable %uvec3ptr Input\n"
583 "%minusone = OpConstant %i32 -1\n"
584 "%zero = OpConstant %i32 0\n"
585 "%one = OpConstant %u32 1\n"
586 "%two = OpConstant %i32 2\n"
588 "%main = OpFunction %void None %voidf\n"
590 "%idval = OpLoad %uvec3 %id\n"
591 "%x = OpCompositeExtract %u32 %idval 0\n"
593 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
594 "%inval = OpLoad %i32 %inloc\n"
596 "%outloc = OpAccessChain %i32ptr %sum %zero ${INDEX}\n"
602 #define ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, NUM_OUTPUT_ELEMENTS) \
604 DE_STATIC_ASSERT((NUM_OUTPUT_ELEMENTS) == 1 || (NUM_OUTPUT_ELEMENTS) == numElements); \
605 cases.push_back(OpAtomicCase(#NAME, ASSEMBLY, OPATOMIC, NUM_OUTPUT_ELEMENTS)); \
606 } while (deGetFalse())
607 #define ADD_OPATOMIC_CASE_1(NAME, ASSEMBLY, OPATOMIC) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, 1)
608 #define ADD_OPATOMIC_CASE_N(NAME, ASSEMBLY, OPATOMIC) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, numElements)
610 ADD_OPATOMIC_CASE_1(iadd, "%unused = OpAtomicIAdd %i32 %outloc %one %zero %inval\n", OPATOMIC_IADD );
611 ADD_OPATOMIC_CASE_1(isub, "%unused = OpAtomicISub %i32 %outloc %one %zero %inval\n", OPATOMIC_ISUB );
612 ADD_OPATOMIC_CASE_1(iinc, "%unused = OpAtomicIIncrement %i32 %outloc %one %zero\n", OPATOMIC_IINC );
613 ADD_OPATOMIC_CASE_1(idec, "%unused = OpAtomicIDecrement %i32 %outloc %one %zero\n", OPATOMIC_IDEC );
614 ADD_OPATOMIC_CASE_N(load, "%inval2 = OpAtomicLoad %i32 %inloc %zero %zero\n"
615 " OpStore %outloc %inval2\n", OPATOMIC_LOAD );
616 ADD_OPATOMIC_CASE_N(store, " OpAtomicStore %outloc %zero %zero %inval\n", OPATOMIC_STORE );
617 ADD_OPATOMIC_CASE_N(compex, "%even = OpSMod %i32 %inval %two\n"
618 " OpStore %outloc %even\n"
619 "%unused = OpAtomicCompareExchange %i32 %outloc %one %zero %zero %minusone %zero\n", OPATOMIC_COMPEX );
621 #undef ADD_OPATOMIC_CASE
622 #undef ADD_OPATOMIC_CASE_1
623 #undef ADD_OPATOMIC_CASE_N
625 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
627 map<string, string> specializations;
628 ComputeShaderSpec spec;
629 vector<deInt32> inputInts (numElements, 0);
630 vector<deInt32> expected (cases[caseNdx].numOutputElements, -1);
632 specializations["INDEX"] = (cases[caseNdx].numOutputElements == 1) ? "%zero" : "%x";
633 specializations["INSTRUCTION"] = cases[caseNdx].assembly;
634 specializations["BLOCK_DECORATION"] = useStorageBuffer ? "Block" : "BufferBlock";
635 specializations["BLOCK_POINTER_TYPE"] = useStorageBuffer ? "StorageBuffer" : "Uniform";
636 spec.assembly = shaderTemplate.specialize(specializations);
638 if (useStorageBuffer)
639 spec.extensions.push_back("VK_KHR_storage_buffer_storage_class");
641 spec.inputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_INPUT)));
642 spec.outputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_EXPECTED)));
643 spec.numWorkGroups = IVec3(numElements, 1, 1);
644 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
647 return group.release();
650 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
652 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
653 ComputeShaderSpec spec;
654 de::Random rnd (deStringHash(group->getName()));
655 const int numElements = 100;
656 vector<float> positiveFloats (numElements, 0);
657 vector<float> negativeFloats (numElements, 0);
659 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
661 for (size_t ndx = 0; ndx < numElements; ++ndx)
662 negativeFloats[ndx] = -positiveFloats[ndx];
665 string(getComputeAsmShaderPreamble()) +
667 "%fname1 = OpString \"negateInputs.comp\"\n"
668 "%fname2 = OpString \"negateInputs\"\n"
670 "OpSource GLSL 430\n"
671 "OpName %main \"main\"\n"
672 "OpName %id \"gl_GlobalInvocationID\"\n"
674 "OpDecorate %id BuiltIn GlobalInvocationId\n"
676 + string(getComputeAsmInputOutputBufferTraits()) +
678 "OpLine %fname1 0 0\n" // At the earliest possible position
680 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
682 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
683 "OpLine %fname2 1 0\n" // Different filenames
684 "OpLine %fname1 1000 100000\n"
686 "%id = OpVariable %uvec3ptr Input\n"
687 "%zero = OpConstant %i32 0\n"
689 "OpLine %fname1 1 1\n" // Before a function
691 "%main = OpFunction %void None %voidf\n"
694 "OpLine %fname1 1 1\n" // In a function
696 "%idval = OpLoad %uvec3 %id\n"
697 "%x = OpCompositeExtract %u32 %idval 0\n"
698 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
699 "%inval = OpLoad %f32 %inloc\n"
700 "%neg = OpFNegate %f32 %inval\n"
701 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
702 " OpStore %outloc %neg\n"
705 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
706 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
707 spec.numWorkGroups = IVec3(numElements, 1, 1);
709 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
711 return group.release();
714 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
716 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
717 ComputeShaderSpec spec;
718 de::Random rnd (deStringHash(group->getName()));
719 const int numElements = 100;
720 vector<float> positiveFloats (numElements, 0);
721 vector<float> negativeFloats (numElements, 0);
723 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
725 for (size_t ndx = 0; ndx < numElements; ++ndx)
726 negativeFloats[ndx] = -positiveFloats[ndx];
729 string(getComputeAsmShaderPreamble()) +
731 "%fname = OpString \"negateInputs.comp\"\n"
733 "OpSource GLSL 430\n"
734 "OpName %main \"main\"\n"
735 "OpName %id \"gl_GlobalInvocationID\"\n"
737 "OpDecorate %id BuiltIn GlobalInvocationId\n"
739 + string(getComputeAsmInputOutputBufferTraits()) +
741 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
743 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
745 "OpLine %fname 0 1\n"
746 "OpNoLine\n" // Immediately following a preceding OpLine
748 "OpLine %fname 1000 1\n"
750 "%id = OpVariable %uvec3ptr Input\n"
751 "%zero = OpConstant %i32 0\n"
753 "OpNoLine\n" // Contents after the previous OpLine
755 "%main = OpFunction %void None %voidf\n"
757 "%idval = OpLoad %uvec3 %id\n"
758 "%x = OpCompositeExtract %u32 %idval 0\n"
760 "OpNoLine\n" // Multiple OpNoLine
764 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
765 "%inval = OpLoad %f32 %inloc\n"
766 "%neg = OpFNegate %f32 %inval\n"
767 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
768 " OpStore %outloc %neg\n"
771 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
772 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
773 spec.numWorkGroups = IVec3(numElements, 1, 1);
775 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
777 return group.release();
780 // Compare instruction for the contraction compute case.
781 // Returns true if the output is what is expected from the test case.
782 bool compareNoContractCase(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
784 if (outputAllocs.size() != 1)
787 // Only size is needed because we are not comparing the exact values.
788 size_t byteSize = expectedOutputs[0]->getByteSize();
790 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
792 for(size_t i = 0; i < byteSize / sizeof(float); ++i) {
793 if (outputAsFloat[i] != 0.f &&
794 outputAsFloat[i] != -ldexp(1, -24)) {
802 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
804 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
805 vector<CaseParameter> cases;
806 const int numElements = 100;
807 vector<float> inputFloats1 (numElements, 0);
808 vector<float> inputFloats2 (numElements, 0);
809 vector<float> outputFloats (numElements, 0);
810 const StringTemplate shaderTemplate (
811 string(getComputeAsmShaderPreamble()) +
813 "OpName %main \"main\"\n"
814 "OpName %id \"gl_GlobalInvocationID\"\n"
816 "OpDecorate %id BuiltIn GlobalInvocationId\n"
820 "OpDecorate %buf BufferBlock\n"
821 "OpDecorate %indata1 DescriptorSet 0\n"
822 "OpDecorate %indata1 Binding 0\n"
823 "OpDecorate %indata2 DescriptorSet 0\n"
824 "OpDecorate %indata2 Binding 1\n"
825 "OpDecorate %outdata DescriptorSet 0\n"
826 "OpDecorate %outdata Binding 2\n"
827 "OpDecorate %f32arr ArrayStride 4\n"
828 "OpMemberDecorate %buf 0 Offset 0\n"
830 + string(getComputeAsmCommonTypes()) +
832 "%buf = OpTypeStruct %f32arr\n"
833 "%bufptr = OpTypePointer Uniform %buf\n"
834 "%indata1 = OpVariable %bufptr Uniform\n"
835 "%indata2 = OpVariable %bufptr Uniform\n"
836 "%outdata = OpVariable %bufptr Uniform\n"
838 "%id = OpVariable %uvec3ptr Input\n"
839 "%zero = OpConstant %i32 0\n"
840 "%c_f_m1 = OpConstant %f32 -1.\n"
842 "%main = OpFunction %void None %voidf\n"
844 "%idval = OpLoad %uvec3 %id\n"
845 "%x = OpCompositeExtract %u32 %idval 0\n"
846 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
847 "%inval1 = OpLoad %f32 %inloc1\n"
848 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
849 "%inval2 = OpLoad %f32 %inloc2\n"
850 "%mul = OpFMul %f32 %inval1 %inval2\n"
851 "%add = OpFAdd %f32 %mul %c_f_m1\n"
852 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
853 " OpStore %outloc %add\n"
857 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
858 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
859 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
861 for (size_t ndx = 0; ndx < numElements; ++ndx)
863 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
864 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
865 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
866 // conducted separately and the result is rounded to 1, or 0x1.fffffcp-1
867 // So the final result will be 0.f or 0x1p-24.
868 // If the operation is combined into a precise fused multiply-add, then the result would be
869 // 2^-46 (0xa8800000).
870 outputFloats[ndx] = 0.f;
873 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
875 map<string, string> specializations;
876 ComputeShaderSpec spec;
878 specializations["DECORATION"] = cases[caseNdx].param;
879 spec.assembly = shaderTemplate.specialize(specializations);
880 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
881 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
882 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
883 spec.numWorkGroups = IVec3(numElements, 1, 1);
884 // Check against the two possible answers based on rounding mode.
885 spec.verifyIO = &compareNoContractCase;
887 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
889 return group.release();
892 bool compareFRem(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
894 if (outputAllocs.size() != 1)
897 vector<deUint8> expectedBytes;
898 expectedOutputs[0]->getBytes(expectedBytes);
900 const float* expectedOutputAsFloat = reinterpret_cast<const float*>(&expectedBytes.front());
901 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
903 for (size_t idx = 0; idx < expectedBytes.size() / sizeof(float); ++idx)
905 const float f0 = expectedOutputAsFloat[idx];
906 const float f1 = outputAsFloat[idx];
907 // \todo relative error needs to be fairly high because FRem may be implemented as
908 // (roughly) frac(a/b)*b, so LSB errors can be magnified. But this should be fine for now.
909 if (deFloatAbs((f1 - f0) / f0) > 0.02)
916 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
918 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
919 ComputeShaderSpec spec;
920 de::Random rnd (deStringHash(group->getName()));
921 const int numElements = 200;
922 vector<float> inputFloats1 (numElements, 0);
923 vector<float> inputFloats2 (numElements, 0);
924 vector<float> outputFloats (numElements, 0);
926 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
927 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
929 for (size_t ndx = 0; ndx < numElements; ++ndx)
931 // Guard against divisors near zero.
932 if (std::fabs(inputFloats2[ndx]) < 1e-3)
933 inputFloats2[ndx] = 8.f;
935 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
936 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
940 string(getComputeAsmShaderPreamble()) +
942 "OpName %main \"main\"\n"
943 "OpName %id \"gl_GlobalInvocationID\"\n"
945 "OpDecorate %id BuiltIn GlobalInvocationId\n"
947 "OpDecorate %buf BufferBlock\n"
948 "OpDecorate %indata1 DescriptorSet 0\n"
949 "OpDecorate %indata1 Binding 0\n"
950 "OpDecorate %indata2 DescriptorSet 0\n"
951 "OpDecorate %indata2 Binding 1\n"
952 "OpDecorate %outdata DescriptorSet 0\n"
953 "OpDecorate %outdata Binding 2\n"
954 "OpDecorate %f32arr ArrayStride 4\n"
955 "OpMemberDecorate %buf 0 Offset 0\n"
957 + string(getComputeAsmCommonTypes()) +
959 "%buf = OpTypeStruct %f32arr\n"
960 "%bufptr = OpTypePointer Uniform %buf\n"
961 "%indata1 = OpVariable %bufptr Uniform\n"
962 "%indata2 = OpVariable %bufptr Uniform\n"
963 "%outdata = OpVariable %bufptr Uniform\n"
965 "%id = OpVariable %uvec3ptr Input\n"
966 "%zero = OpConstant %i32 0\n"
968 "%main = OpFunction %void None %voidf\n"
970 "%idval = OpLoad %uvec3 %id\n"
971 "%x = OpCompositeExtract %u32 %idval 0\n"
972 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
973 "%inval1 = OpLoad %f32 %inloc1\n"
974 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
975 "%inval2 = OpLoad %f32 %inloc2\n"
976 "%rem = OpFRem %f32 %inval1 %inval2\n"
977 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
978 " OpStore %outloc %rem\n"
982 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
983 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
984 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
985 spec.numWorkGroups = IVec3(numElements, 1, 1);
986 spec.verifyIO = &compareFRem;
988 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
990 return group.release();
993 bool compareNMin (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
995 if (outputAllocs.size() != 1)
998 const BufferSp& expectedOutput (expectedOutputs[0]);
999 std::vector<deUint8> data;
1000 expectedOutput->getBytes(data);
1002 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1003 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1005 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float); ++idx)
1007 const float f0 = expectedOutputAsFloat[idx];
1008 const float f1 = outputAsFloat[idx];
1010 // For NMin, we accept NaN as output if both inputs were NaN.
1011 // Otherwise the NaN is the wrong choise, as on architectures that
1012 // do not handle NaN, those are huge values.
1013 if (!(tcu::Float32(f1).isNaN() && tcu::Float32(f0).isNaN()) && deFloatAbs(f1 - f0) > 0.00001f)
1020 tcu::TestCaseGroup* createOpNMinGroup (tcu::TestContext& testCtx)
1022 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnmin", "Test the OpNMin instruction"));
1023 ComputeShaderSpec spec;
1024 de::Random rnd (deStringHash(group->getName()));
1025 const int numElements = 200;
1026 vector<float> inputFloats1 (numElements, 0);
1027 vector<float> inputFloats2 (numElements, 0);
1028 vector<float> outputFloats (numElements, 0);
1030 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1031 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1033 // Make the first case a full-NAN case.
1034 inputFloats1[0] = TCU_NAN;
1035 inputFloats2[0] = TCU_NAN;
1037 for (size_t ndx = 0; ndx < numElements; ++ndx)
1039 // By default, pick the smallest
1040 outputFloats[ndx] = std::min(inputFloats1[ndx], inputFloats2[ndx]);
1042 // Make half of the cases NaN cases
1045 // Alternate between the NaN operand
1048 outputFloats[ndx] = inputFloats2[ndx];
1049 inputFloats1[ndx] = TCU_NAN;
1053 outputFloats[ndx] = inputFloats1[ndx];
1054 inputFloats2[ndx] = TCU_NAN;
1060 "OpCapability Shader\n"
1061 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1062 "OpMemoryModel Logical GLSL450\n"
1063 "OpEntryPoint GLCompute %main \"main\" %id\n"
1064 "OpExecutionMode %main LocalSize 1 1 1\n"
1066 "OpName %main \"main\"\n"
1067 "OpName %id \"gl_GlobalInvocationID\"\n"
1069 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1071 "OpDecorate %buf BufferBlock\n"
1072 "OpDecorate %indata1 DescriptorSet 0\n"
1073 "OpDecorate %indata1 Binding 0\n"
1074 "OpDecorate %indata2 DescriptorSet 0\n"
1075 "OpDecorate %indata2 Binding 1\n"
1076 "OpDecorate %outdata DescriptorSet 0\n"
1077 "OpDecorate %outdata Binding 2\n"
1078 "OpDecorate %f32arr ArrayStride 4\n"
1079 "OpMemberDecorate %buf 0 Offset 0\n"
1081 + string(getComputeAsmCommonTypes()) +
1083 "%buf = OpTypeStruct %f32arr\n"
1084 "%bufptr = OpTypePointer Uniform %buf\n"
1085 "%indata1 = OpVariable %bufptr Uniform\n"
1086 "%indata2 = OpVariable %bufptr Uniform\n"
1087 "%outdata = OpVariable %bufptr Uniform\n"
1089 "%id = OpVariable %uvec3ptr Input\n"
1090 "%zero = OpConstant %i32 0\n"
1092 "%main = OpFunction %void None %voidf\n"
1093 "%label = OpLabel\n"
1094 "%idval = OpLoad %uvec3 %id\n"
1095 "%x = OpCompositeExtract %u32 %idval 0\n"
1096 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1097 "%inval1 = OpLoad %f32 %inloc1\n"
1098 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1099 "%inval2 = OpLoad %f32 %inloc2\n"
1100 "%rem = OpExtInst %f32 %std450 NMin %inval1 %inval2\n"
1101 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1102 " OpStore %outloc %rem\n"
1106 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1107 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1108 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1109 spec.numWorkGroups = IVec3(numElements, 1, 1);
1110 spec.verifyIO = &compareNMin;
1112 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1114 return group.release();
1117 bool compareNMax (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1119 if (outputAllocs.size() != 1)
1122 const BufferSp& expectedOutput = expectedOutputs[0];
1123 std::vector<deUint8> data;
1124 expectedOutput->getBytes(data);
1126 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1127 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1129 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float); ++idx)
1131 const float f0 = expectedOutputAsFloat[idx];
1132 const float f1 = outputAsFloat[idx];
1134 // For NMax, NaN is considered acceptable result, since in
1135 // architectures that do not handle NaNs, those are huge values.
1136 if (!tcu::Float32(f1).isNaN() && deFloatAbs(f1 - f0) > 0.00001f)
1143 tcu::TestCaseGroup* createOpNMaxGroup (tcu::TestContext& testCtx)
1145 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, "opnmax", "Test the OpNMax instruction"));
1146 ComputeShaderSpec spec;
1147 de::Random rnd (deStringHash(group->getName()));
1148 const int numElements = 200;
1149 vector<float> inputFloats1 (numElements, 0);
1150 vector<float> inputFloats2 (numElements, 0);
1151 vector<float> outputFloats (numElements, 0);
1153 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1154 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1156 // Make the first case a full-NAN case.
1157 inputFloats1[0] = TCU_NAN;
1158 inputFloats2[0] = TCU_NAN;
1160 for (size_t ndx = 0; ndx < numElements; ++ndx)
1162 // By default, pick the biggest
1163 outputFloats[ndx] = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1165 // Make half of the cases NaN cases
1168 // Alternate between the NaN operand
1171 outputFloats[ndx] = inputFloats2[ndx];
1172 inputFloats1[ndx] = TCU_NAN;
1176 outputFloats[ndx] = inputFloats1[ndx];
1177 inputFloats2[ndx] = TCU_NAN;
1183 "OpCapability Shader\n"
1184 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1185 "OpMemoryModel Logical GLSL450\n"
1186 "OpEntryPoint GLCompute %main \"main\" %id\n"
1187 "OpExecutionMode %main LocalSize 1 1 1\n"
1189 "OpName %main \"main\"\n"
1190 "OpName %id \"gl_GlobalInvocationID\"\n"
1192 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1194 "OpDecorate %buf BufferBlock\n"
1195 "OpDecorate %indata1 DescriptorSet 0\n"
1196 "OpDecorate %indata1 Binding 0\n"
1197 "OpDecorate %indata2 DescriptorSet 0\n"
1198 "OpDecorate %indata2 Binding 1\n"
1199 "OpDecorate %outdata DescriptorSet 0\n"
1200 "OpDecorate %outdata Binding 2\n"
1201 "OpDecorate %f32arr ArrayStride 4\n"
1202 "OpMemberDecorate %buf 0 Offset 0\n"
1204 + string(getComputeAsmCommonTypes()) +
1206 "%buf = OpTypeStruct %f32arr\n"
1207 "%bufptr = OpTypePointer Uniform %buf\n"
1208 "%indata1 = OpVariable %bufptr Uniform\n"
1209 "%indata2 = OpVariable %bufptr Uniform\n"
1210 "%outdata = OpVariable %bufptr Uniform\n"
1212 "%id = OpVariable %uvec3ptr Input\n"
1213 "%zero = OpConstant %i32 0\n"
1215 "%main = OpFunction %void None %voidf\n"
1216 "%label = OpLabel\n"
1217 "%idval = OpLoad %uvec3 %id\n"
1218 "%x = OpCompositeExtract %u32 %idval 0\n"
1219 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1220 "%inval1 = OpLoad %f32 %inloc1\n"
1221 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1222 "%inval2 = OpLoad %f32 %inloc2\n"
1223 "%rem = OpExtInst %f32 %std450 NMax %inval1 %inval2\n"
1224 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1225 " OpStore %outloc %rem\n"
1229 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1230 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1231 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1232 spec.numWorkGroups = IVec3(numElements, 1, 1);
1233 spec.verifyIO = &compareNMax;
1235 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1237 return group.release();
1240 bool compareNClamp (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1242 if (outputAllocs.size() != 1)
1245 const BufferSp& expectedOutput = expectedOutputs[0];
1246 std::vector<deUint8> data;
1247 expectedOutput->getBytes(data);
1249 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1250 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1252 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float) / 2; ++idx)
1254 const float e0 = expectedOutputAsFloat[idx * 2];
1255 const float e1 = expectedOutputAsFloat[idx * 2 + 1];
1256 const float res = outputAsFloat[idx];
1258 // For NClamp, we have two possible outcomes based on
1259 // whether NaNs are handled or not.
1260 // If either min or max value is NaN, the result is undefined,
1261 // so this test doesn't stress those. If the clamped value is
1262 // NaN, and NaNs are handled, the result is min; if NaNs are not
1263 // handled, they are big values that result in max.
1264 // If all three parameters are NaN, the result should be NaN.
1265 if (!((tcu::Float32(e0).isNaN() && tcu::Float32(res).isNaN()) ||
1266 (deFloatAbs(e0 - res) < 0.00001f) ||
1267 (deFloatAbs(e1 - res) < 0.00001f)))
1274 tcu::TestCaseGroup* createOpNClampGroup (tcu::TestContext& testCtx)
1276 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnclamp", "Test the OpNClamp instruction"));
1277 ComputeShaderSpec spec;
1278 de::Random rnd (deStringHash(group->getName()));
1279 const int numElements = 200;
1280 vector<float> inputFloats1 (numElements, 0);
1281 vector<float> inputFloats2 (numElements, 0);
1282 vector<float> inputFloats3 (numElements, 0);
1283 vector<float> outputFloats (numElements * 2, 0);
1285 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1286 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1287 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats3[0], numElements);
1289 for (size_t ndx = 0; ndx < numElements; ++ndx)
1291 // Results are only defined if max value is bigger than min value.
1292 if (inputFloats2[ndx] > inputFloats3[ndx])
1294 float t = inputFloats2[ndx];
1295 inputFloats2[ndx] = inputFloats3[ndx];
1296 inputFloats3[ndx] = t;
1299 // By default, do the clamp, setting both possible answers
1300 float defaultRes = std::min(std::max(inputFloats1[ndx], inputFloats2[ndx]), inputFloats3[ndx]);
1302 float maxResA = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1303 float maxResB = maxResA;
1305 // Alternate between the NaN cases
1308 inputFloats1[ndx] = TCU_NAN;
1309 // If NaN is handled, the result should be same as the clamp minimum.
1310 // If NaN is not handled, the result should clamp to the clamp maximum.
1311 maxResA = inputFloats2[ndx];
1312 maxResB = inputFloats3[ndx];
1316 // Not a NaN case - only one legal result.
1317 maxResA = defaultRes;
1318 maxResB = defaultRes;
1321 outputFloats[ndx * 2] = maxResA;
1322 outputFloats[ndx * 2 + 1] = maxResB;
1325 // Make the first case a full-NAN case.
1326 inputFloats1[0] = TCU_NAN;
1327 inputFloats2[0] = TCU_NAN;
1328 inputFloats3[0] = TCU_NAN;
1329 outputFloats[0] = TCU_NAN;
1330 outputFloats[1] = TCU_NAN;
1333 "OpCapability Shader\n"
1334 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1335 "OpMemoryModel Logical GLSL450\n"
1336 "OpEntryPoint GLCompute %main \"main\" %id\n"
1337 "OpExecutionMode %main LocalSize 1 1 1\n"
1339 "OpName %main \"main\"\n"
1340 "OpName %id \"gl_GlobalInvocationID\"\n"
1342 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1344 "OpDecorate %buf BufferBlock\n"
1345 "OpDecorate %indata1 DescriptorSet 0\n"
1346 "OpDecorate %indata1 Binding 0\n"
1347 "OpDecorate %indata2 DescriptorSet 0\n"
1348 "OpDecorate %indata2 Binding 1\n"
1349 "OpDecorate %indata3 DescriptorSet 0\n"
1350 "OpDecorate %indata3 Binding 2\n"
1351 "OpDecorate %outdata DescriptorSet 0\n"
1352 "OpDecorate %outdata Binding 3\n"
1353 "OpDecorate %f32arr ArrayStride 4\n"
1354 "OpMemberDecorate %buf 0 Offset 0\n"
1356 + string(getComputeAsmCommonTypes()) +
1358 "%buf = OpTypeStruct %f32arr\n"
1359 "%bufptr = OpTypePointer Uniform %buf\n"
1360 "%indata1 = OpVariable %bufptr Uniform\n"
1361 "%indata2 = OpVariable %bufptr Uniform\n"
1362 "%indata3 = OpVariable %bufptr Uniform\n"
1363 "%outdata = OpVariable %bufptr Uniform\n"
1365 "%id = OpVariable %uvec3ptr Input\n"
1366 "%zero = OpConstant %i32 0\n"
1368 "%main = OpFunction %void None %voidf\n"
1369 "%label = OpLabel\n"
1370 "%idval = OpLoad %uvec3 %id\n"
1371 "%x = OpCompositeExtract %u32 %idval 0\n"
1372 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1373 "%inval1 = OpLoad %f32 %inloc1\n"
1374 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1375 "%inval2 = OpLoad %f32 %inloc2\n"
1376 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1377 "%inval3 = OpLoad %f32 %inloc3\n"
1378 "%rem = OpExtInst %f32 %std450 NClamp %inval1 %inval2 %inval3\n"
1379 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1380 " OpStore %outloc %rem\n"
1384 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1385 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1386 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1387 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1388 spec.numWorkGroups = IVec3(numElements, 1, 1);
1389 spec.verifyIO = &compareNClamp;
1391 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1393 return group.release();
1396 tcu::TestCaseGroup* createOpSRemComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1398 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem", "Test the OpSRem instruction"));
1399 de::Random rnd (deStringHash(group->getName()));
1400 const int numElements = 200;
1402 const struct CaseParams
1405 const char* failMessage; // customized status message
1406 qpTestResult failResult; // override status on failure
1407 int op1Min, op1Max; // operand ranges
1411 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1412 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1414 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1416 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1418 const CaseParams& params = cases[caseNdx];
1419 ComputeShaderSpec spec;
1420 vector<deInt32> inputInts1 (numElements, 0);
1421 vector<deInt32> inputInts2 (numElements, 0);
1422 vector<deInt32> outputInts (numElements, 0);
1424 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1425 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1427 for (int ndx = 0; ndx < numElements; ++ndx)
1429 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1430 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1434 string(getComputeAsmShaderPreamble()) +
1436 "OpName %main \"main\"\n"
1437 "OpName %id \"gl_GlobalInvocationID\"\n"
1439 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1441 "OpDecorate %buf BufferBlock\n"
1442 "OpDecorate %indata1 DescriptorSet 0\n"
1443 "OpDecorate %indata1 Binding 0\n"
1444 "OpDecorate %indata2 DescriptorSet 0\n"
1445 "OpDecorate %indata2 Binding 1\n"
1446 "OpDecorate %outdata DescriptorSet 0\n"
1447 "OpDecorate %outdata Binding 2\n"
1448 "OpDecorate %i32arr ArrayStride 4\n"
1449 "OpMemberDecorate %buf 0 Offset 0\n"
1451 + string(getComputeAsmCommonTypes()) +
1453 "%buf = OpTypeStruct %i32arr\n"
1454 "%bufptr = OpTypePointer Uniform %buf\n"
1455 "%indata1 = OpVariable %bufptr Uniform\n"
1456 "%indata2 = OpVariable %bufptr Uniform\n"
1457 "%outdata = OpVariable %bufptr Uniform\n"
1459 "%id = OpVariable %uvec3ptr Input\n"
1460 "%zero = OpConstant %i32 0\n"
1462 "%main = OpFunction %void None %voidf\n"
1463 "%label = OpLabel\n"
1464 "%idval = OpLoad %uvec3 %id\n"
1465 "%x = OpCompositeExtract %u32 %idval 0\n"
1466 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1467 "%inval1 = OpLoad %i32 %inloc1\n"
1468 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1469 "%inval2 = OpLoad %i32 %inloc2\n"
1470 "%rem = OpSRem %i32 %inval1 %inval2\n"
1471 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1472 " OpStore %outloc %rem\n"
1476 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1477 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1478 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1479 spec.numWorkGroups = IVec3(numElements, 1, 1);
1480 spec.failResult = params.failResult;
1481 spec.failMessage = params.failMessage;
1483 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1486 return group.release();
1489 tcu::TestCaseGroup* createOpSRemComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1491 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem64", "Test the 64-bit OpSRem instruction"));
1492 de::Random rnd (deStringHash(group->getName()));
1493 const int numElements = 200;
1495 const struct CaseParams
1498 const char* failMessage; // customized status message
1499 qpTestResult failResult; // override status on failure
1503 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1504 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1506 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1508 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1510 const CaseParams& params = cases[caseNdx];
1511 ComputeShaderSpec spec;
1512 vector<deInt64> inputInts1 (numElements, 0);
1513 vector<deInt64> inputInts2 (numElements, 0);
1514 vector<deInt64> outputInts (numElements, 0);
1516 if (params.positive)
1518 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1519 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1523 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1524 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1527 for (int ndx = 0; ndx < numElements; ++ndx)
1529 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1530 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1534 "OpCapability Int64\n"
1536 + string(getComputeAsmShaderPreamble()) +
1538 "OpName %main \"main\"\n"
1539 "OpName %id \"gl_GlobalInvocationID\"\n"
1541 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1543 "OpDecorate %buf BufferBlock\n"
1544 "OpDecorate %indata1 DescriptorSet 0\n"
1545 "OpDecorate %indata1 Binding 0\n"
1546 "OpDecorate %indata2 DescriptorSet 0\n"
1547 "OpDecorate %indata2 Binding 1\n"
1548 "OpDecorate %outdata DescriptorSet 0\n"
1549 "OpDecorate %outdata Binding 2\n"
1550 "OpDecorate %i64arr ArrayStride 8\n"
1551 "OpMemberDecorate %buf 0 Offset 0\n"
1553 + string(getComputeAsmCommonTypes())
1554 + string(getComputeAsmCommonInt64Types()) +
1556 "%buf = OpTypeStruct %i64arr\n"
1557 "%bufptr = OpTypePointer Uniform %buf\n"
1558 "%indata1 = OpVariable %bufptr Uniform\n"
1559 "%indata2 = OpVariable %bufptr Uniform\n"
1560 "%outdata = OpVariable %bufptr Uniform\n"
1562 "%id = OpVariable %uvec3ptr Input\n"
1563 "%zero = OpConstant %i64 0\n"
1565 "%main = OpFunction %void None %voidf\n"
1566 "%label = OpLabel\n"
1567 "%idval = OpLoad %uvec3 %id\n"
1568 "%x = OpCompositeExtract %u32 %idval 0\n"
1569 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1570 "%inval1 = OpLoad %i64 %inloc1\n"
1571 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1572 "%inval2 = OpLoad %i64 %inloc2\n"
1573 "%rem = OpSRem %i64 %inval1 %inval2\n"
1574 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1575 " OpStore %outloc %rem\n"
1579 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1580 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1581 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1582 spec.numWorkGroups = IVec3(numElements, 1, 1);
1583 spec.failResult = params.failResult;
1584 spec.failMessage = params.failMessage;
1586 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1589 return group.release();
1592 tcu::TestCaseGroup* createOpSModComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1594 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod", "Test the OpSMod instruction"));
1595 de::Random rnd (deStringHash(group->getName()));
1596 const int numElements = 200;
1598 const struct CaseParams
1601 const char* failMessage; // customized status message
1602 qpTestResult failResult; // override status on failure
1603 int op1Min, op1Max; // operand ranges
1607 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1608 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1610 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1612 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1614 const CaseParams& params = cases[caseNdx];
1616 ComputeShaderSpec spec;
1617 vector<deInt32> inputInts1 (numElements, 0);
1618 vector<deInt32> inputInts2 (numElements, 0);
1619 vector<deInt32> outputInts (numElements, 0);
1621 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1622 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1624 for (int ndx = 0; ndx < numElements; ++ndx)
1626 deInt32 rem = inputInts1[ndx] % inputInts2[ndx];
1629 outputInts[ndx] = 0;
1631 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1633 // They have the same sign
1634 outputInts[ndx] = rem;
1638 // They have opposite sign. The remainder operation takes the
1639 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1640 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1641 // the result has the correct sign and that it is still
1642 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1644 // See also http://mathforum.org/library/drmath/view/52343.html
1645 outputInts[ndx] = rem + inputInts2[ndx];
1650 string(getComputeAsmShaderPreamble()) +
1652 "OpName %main \"main\"\n"
1653 "OpName %id \"gl_GlobalInvocationID\"\n"
1655 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1657 "OpDecorate %buf BufferBlock\n"
1658 "OpDecorate %indata1 DescriptorSet 0\n"
1659 "OpDecorate %indata1 Binding 0\n"
1660 "OpDecorate %indata2 DescriptorSet 0\n"
1661 "OpDecorate %indata2 Binding 1\n"
1662 "OpDecorate %outdata DescriptorSet 0\n"
1663 "OpDecorate %outdata Binding 2\n"
1664 "OpDecorate %i32arr ArrayStride 4\n"
1665 "OpMemberDecorate %buf 0 Offset 0\n"
1667 + string(getComputeAsmCommonTypes()) +
1669 "%buf = OpTypeStruct %i32arr\n"
1670 "%bufptr = OpTypePointer Uniform %buf\n"
1671 "%indata1 = OpVariable %bufptr Uniform\n"
1672 "%indata2 = OpVariable %bufptr Uniform\n"
1673 "%outdata = OpVariable %bufptr Uniform\n"
1675 "%id = OpVariable %uvec3ptr Input\n"
1676 "%zero = OpConstant %i32 0\n"
1678 "%main = OpFunction %void None %voidf\n"
1679 "%label = OpLabel\n"
1680 "%idval = OpLoad %uvec3 %id\n"
1681 "%x = OpCompositeExtract %u32 %idval 0\n"
1682 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1683 "%inval1 = OpLoad %i32 %inloc1\n"
1684 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1685 "%inval2 = OpLoad %i32 %inloc2\n"
1686 "%rem = OpSMod %i32 %inval1 %inval2\n"
1687 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1688 " OpStore %outloc %rem\n"
1692 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1693 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1694 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1695 spec.numWorkGroups = IVec3(numElements, 1, 1);
1696 spec.failResult = params.failResult;
1697 spec.failMessage = params.failMessage;
1699 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1702 return group.release();
1705 tcu::TestCaseGroup* createOpSModComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1707 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod64", "Test the OpSMod instruction"));
1708 de::Random rnd (deStringHash(group->getName()));
1709 const int numElements = 200;
1711 const struct CaseParams
1714 const char* failMessage; // customized status message
1715 qpTestResult failResult; // override status on failure
1719 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1720 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1722 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1724 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1726 const CaseParams& params = cases[caseNdx];
1728 ComputeShaderSpec spec;
1729 vector<deInt64> inputInts1 (numElements, 0);
1730 vector<deInt64> inputInts2 (numElements, 0);
1731 vector<deInt64> outputInts (numElements, 0);
1734 if (params.positive)
1736 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1737 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1741 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1742 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1745 for (int ndx = 0; ndx < numElements; ++ndx)
1747 deInt64 rem = inputInts1[ndx] % inputInts2[ndx];
1750 outputInts[ndx] = 0;
1752 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1754 // They have the same sign
1755 outputInts[ndx] = rem;
1759 // They have opposite sign. The remainder operation takes the
1760 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1761 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1762 // the result has the correct sign and that it is still
1763 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1765 // See also http://mathforum.org/library/drmath/view/52343.html
1766 outputInts[ndx] = rem + inputInts2[ndx];
1771 "OpCapability Int64\n"
1773 + string(getComputeAsmShaderPreamble()) +
1775 "OpName %main \"main\"\n"
1776 "OpName %id \"gl_GlobalInvocationID\"\n"
1778 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1780 "OpDecorate %buf BufferBlock\n"
1781 "OpDecorate %indata1 DescriptorSet 0\n"
1782 "OpDecorate %indata1 Binding 0\n"
1783 "OpDecorate %indata2 DescriptorSet 0\n"
1784 "OpDecorate %indata2 Binding 1\n"
1785 "OpDecorate %outdata DescriptorSet 0\n"
1786 "OpDecorate %outdata Binding 2\n"
1787 "OpDecorate %i64arr ArrayStride 8\n"
1788 "OpMemberDecorate %buf 0 Offset 0\n"
1790 + string(getComputeAsmCommonTypes())
1791 + string(getComputeAsmCommonInt64Types()) +
1793 "%buf = OpTypeStruct %i64arr\n"
1794 "%bufptr = OpTypePointer Uniform %buf\n"
1795 "%indata1 = OpVariable %bufptr Uniform\n"
1796 "%indata2 = OpVariable %bufptr Uniform\n"
1797 "%outdata = OpVariable %bufptr Uniform\n"
1799 "%id = OpVariable %uvec3ptr Input\n"
1800 "%zero = OpConstant %i64 0\n"
1802 "%main = OpFunction %void None %voidf\n"
1803 "%label = OpLabel\n"
1804 "%idval = OpLoad %uvec3 %id\n"
1805 "%x = OpCompositeExtract %u32 %idval 0\n"
1806 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1807 "%inval1 = OpLoad %i64 %inloc1\n"
1808 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1809 "%inval2 = OpLoad %i64 %inloc2\n"
1810 "%rem = OpSMod %i64 %inval1 %inval2\n"
1811 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1812 " OpStore %outloc %rem\n"
1816 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1817 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1818 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1819 spec.numWorkGroups = IVec3(numElements, 1, 1);
1820 spec.failResult = params.failResult;
1821 spec.failMessage = params.failMessage;
1823 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1826 return group.release();
1829 // Copy contents in the input buffer to the output buffer.
1830 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
1832 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
1833 de::Random rnd (deStringHash(group->getName()));
1834 const int numElements = 100;
1836 // 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.
1837 ComputeShaderSpec spec1;
1838 vector<Vec4> inputFloats1 (numElements);
1839 vector<Vec4> outputFloats1 (numElements);
1841 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
1843 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1844 floorAll(inputFloats1);
1846 for (size_t ndx = 0; ndx < numElements; ++ndx)
1847 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
1850 string(getComputeAsmShaderPreamble()) +
1852 "OpName %main \"main\"\n"
1853 "OpName %id \"gl_GlobalInvocationID\"\n"
1855 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1856 "OpDecorate %vec4arr ArrayStride 16\n"
1858 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1860 "%vec4 = OpTypeVector %f32 4\n"
1861 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
1862 "%vec4ptr_f = OpTypePointer Function %vec4\n"
1863 "%vec4arr = OpTypeRuntimeArray %vec4\n"
1864 "%buf = OpTypeStruct %vec4arr\n"
1865 "%bufptr = OpTypePointer Uniform %buf\n"
1866 "%indata = OpVariable %bufptr Uniform\n"
1867 "%outdata = OpVariable %bufptr Uniform\n"
1869 "%id = OpVariable %uvec3ptr Input\n"
1870 "%zero = OpConstant %i32 0\n"
1871 "%c_f_0 = OpConstant %f32 0.\n"
1872 "%c_f_0_5 = OpConstant %f32 0.5\n"
1873 "%c_f_1_5 = OpConstant %f32 1.5\n"
1874 "%c_f_2_5 = OpConstant %f32 2.5\n"
1875 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
1877 "%main = OpFunction %void None %voidf\n"
1878 "%label = OpLabel\n"
1879 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
1880 "%idval = OpLoad %uvec3 %id\n"
1881 "%x = OpCompositeExtract %u32 %idval 0\n"
1882 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
1883 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
1884 " OpCopyMemory %v_vec4 %inloc\n"
1885 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
1886 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
1887 " OpStore %outloc %add\n"
1891 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
1892 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
1893 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1895 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
1897 // The following case copies a float[100] variable from the input buffer to the output buffer.
1898 ComputeShaderSpec spec2;
1899 vector<float> inputFloats2 (numElements);
1900 vector<float> outputFloats2 (numElements);
1902 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
1904 for (size_t ndx = 0; ndx < numElements; ++ndx)
1905 outputFloats2[ndx] = inputFloats2[ndx];
1908 string(getComputeAsmShaderPreamble()) +
1910 "OpName %main \"main\"\n"
1911 "OpName %id \"gl_GlobalInvocationID\"\n"
1913 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1914 "OpDecorate %f32arr100 ArrayStride 4\n"
1916 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1918 "%hundred = OpConstant %u32 100\n"
1919 "%f32arr100 = OpTypeArray %f32 %hundred\n"
1920 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
1921 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
1922 "%buf = OpTypeStruct %f32arr100\n"
1923 "%bufptr = OpTypePointer Uniform %buf\n"
1924 "%indata = OpVariable %bufptr Uniform\n"
1925 "%outdata = OpVariable %bufptr Uniform\n"
1927 "%id = OpVariable %uvec3ptr Input\n"
1928 "%zero = OpConstant %i32 0\n"
1930 "%main = OpFunction %void None %voidf\n"
1931 "%label = OpLabel\n"
1932 "%var = OpVariable %f32arr100ptr_f Function\n"
1933 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
1934 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
1935 " OpCopyMemory %var %inarr\n"
1936 " OpCopyMemory %outarr %var\n"
1940 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1941 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1942 spec2.numWorkGroups = IVec3(1, 1, 1);
1944 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
1946 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
1947 ComputeShaderSpec spec3;
1948 vector<float> inputFloats3 (16);
1949 vector<float> outputFloats3 (16);
1951 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
1953 for (size_t ndx = 0; ndx < 16; ++ndx)
1954 outputFloats3[ndx] = inputFloats3[ndx];
1957 string(getComputeAsmShaderPreamble()) +
1959 "OpName %main \"main\"\n"
1960 "OpName %id \"gl_GlobalInvocationID\"\n"
1962 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1963 "OpMemberDecorate %buf 0 Offset 0\n"
1964 "OpMemberDecorate %buf 1 Offset 16\n"
1965 "OpMemberDecorate %buf 2 Offset 32\n"
1966 "OpMemberDecorate %buf 3 Offset 48\n"
1968 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1970 "%vec4 = OpTypeVector %f32 4\n"
1971 "%buf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
1972 "%bufptr = OpTypePointer Uniform %buf\n"
1973 "%indata = OpVariable %bufptr Uniform\n"
1974 "%outdata = OpVariable %bufptr Uniform\n"
1975 "%vec4stptr = OpTypePointer Function %buf\n"
1977 "%id = OpVariable %uvec3ptr Input\n"
1978 "%zero = OpConstant %i32 0\n"
1980 "%main = OpFunction %void None %voidf\n"
1981 "%label = OpLabel\n"
1982 "%var = OpVariable %vec4stptr Function\n"
1983 " OpCopyMemory %var %indata\n"
1984 " OpCopyMemory %outdata %var\n"
1988 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1989 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
1990 spec3.numWorkGroups = IVec3(1, 1, 1);
1992 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
1994 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
1995 ComputeShaderSpec spec4;
1996 vector<float> inputFloats4 (numElements);
1997 vector<float> outputFloats4 (numElements);
1999 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
2001 for (size_t ndx = 0; ndx < numElements; ++ndx)
2002 outputFloats4[ndx] = -inputFloats4[ndx];
2005 string(getComputeAsmShaderPreamble()) +
2007 "OpName %main \"main\"\n"
2008 "OpName %id \"gl_GlobalInvocationID\"\n"
2010 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2012 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2014 "%f32ptr_f = OpTypePointer Function %f32\n"
2015 "%id = OpVariable %uvec3ptr Input\n"
2016 "%zero = OpConstant %i32 0\n"
2018 "%main = OpFunction %void None %voidf\n"
2019 "%label = OpLabel\n"
2020 "%var = OpVariable %f32ptr_f Function\n"
2021 "%idval = OpLoad %uvec3 %id\n"
2022 "%x = OpCompositeExtract %u32 %idval 0\n"
2023 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2024 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2025 " OpCopyMemory %var %inloc\n"
2026 "%val = OpLoad %f32 %var\n"
2027 "%neg = OpFNegate %f32 %val\n"
2028 " OpStore %outloc %neg\n"
2032 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2033 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
2034 spec4.numWorkGroups = IVec3(numElements, 1, 1);
2036 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
2038 return group.release();
2041 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
2043 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
2044 ComputeShaderSpec spec;
2045 de::Random rnd (deStringHash(group->getName()));
2046 const int numElements = 100;
2047 vector<float> inputFloats (numElements, 0);
2048 vector<float> outputFloats (numElements, 0);
2050 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
2052 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2053 floorAll(inputFloats);
2055 for (size_t ndx = 0; ndx < numElements; ++ndx)
2056 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
2059 string(getComputeAsmShaderPreamble()) +
2061 "OpName %main \"main\"\n"
2062 "OpName %id \"gl_GlobalInvocationID\"\n"
2064 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2066 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2068 "%fmat = OpTypeMatrix %fvec3 3\n"
2069 "%three = OpConstant %u32 3\n"
2070 "%farr = OpTypeArray %f32 %three\n"
2071 "%fst = OpTypeStruct %f32 %f32\n"
2073 + string(getComputeAsmInputOutputBuffer()) +
2075 "%id = OpVariable %uvec3ptr Input\n"
2076 "%zero = OpConstant %i32 0\n"
2077 "%c_f = OpConstant %f32 1.5\n"
2078 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
2079 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
2080 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
2081 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
2083 "%main = OpFunction %void None %voidf\n"
2084 "%label = OpLabel\n"
2085 "%c_f_copy = OpCopyObject %f32 %c_f\n"
2086 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
2087 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
2088 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
2089 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
2090 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
2091 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
2092 "%farr_elem = OpCompositeExtract %f32 %c_farr_copy 2\n"
2093 "%fst_elem = OpCompositeExtract %f32 %c_fst_copy 1\n"
2094 // Add up. 1.5 * 5 = 7.5.
2095 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
2096 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
2097 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
2098 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
2100 "%idval = OpLoad %uvec3 %id\n"
2101 "%x = OpCompositeExtract %u32 %idval 0\n"
2102 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2103 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2104 "%inval = OpLoad %f32 %inloc\n"
2105 "%add = OpFAdd %f32 %add4 %inval\n"
2106 " OpStore %outloc %add\n"
2109 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2110 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2111 spec.numWorkGroups = IVec3(numElements, 1, 1);
2113 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
2115 return group.release();
2117 // Assembly code used for testing OpUnreachable is based on GLSL source code:
2121 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2122 // float elements[];
2124 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2125 // float elements[];
2128 // void not_called_func() {
2129 // // place OpUnreachable here
2132 // uint modulo4(uint val) {
2133 // switch (val % uint(4)) {
2134 // case 0: return 3;
2135 // case 1: return 2;
2136 // case 2: return 1;
2137 // case 3: return 0;
2138 // default: return 100; // place OpUnreachable here
2144 // // place OpUnreachable here
2148 // uint x = gl_GlobalInvocationID.x;
2149 // if (const5() > modulo4(1000)) {
2150 // output_data.elements[x] = -input_data.elements[x];
2152 // // place OpUnreachable here
2153 // output_data.elements[x] = input_data.elements[x];
2157 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
2159 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
2160 ComputeShaderSpec spec;
2161 de::Random rnd (deStringHash(group->getName()));
2162 const int numElements = 100;
2163 vector<float> positiveFloats (numElements, 0);
2164 vector<float> negativeFloats (numElements, 0);
2166 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2168 for (size_t ndx = 0; ndx < numElements; ++ndx)
2169 negativeFloats[ndx] = -positiveFloats[ndx];
2172 string(getComputeAsmShaderPreamble()) +
2174 "OpSource GLSL 430\n"
2175 "OpName %main \"main\"\n"
2176 "OpName %func_not_called_func \"not_called_func(\"\n"
2177 "OpName %func_modulo4 \"modulo4(u1;\"\n"
2178 "OpName %func_const5 \"const5(\"\n"
2179 "OpName %id \"gl_GlobalInvocationID\"\n"
2181 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2183 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2185 "%u32ptr = OpTypePointer Function %u32\n"
2186 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
2187 "%unitf = OpTypeFunction %u32\n"
2189 "%id = OpVariable %uvec3ptr Input\n"
2190 "%zero = OpConstant %u32 0\n"
2191 "%one = OpConstant %u32 1\n"
2192 "%two = OpConstant %u32 2\n"
2193 "%three = OpConstant %u32 3\n"
2194 "%four = OpConstant %u32 4\n"
2195 "%five = OpConstant %u32 5\n"
2196 "%hundred = OpConstant %u32 100\n"
2197 "%thousand = OpConstant %u32 1000\n"
2199 + string(getComputeAsmInputOutputBuffer()) +
2202 "%main = OpFunction %void None %voidf\n"
2203 "%main_entry = OpLabel\n"
2204 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
2205 "%idval = OpLoad %uvec3 %id\n"
2206 "%x = OpCompositeExtract %u32 %idval 0\n"
2207 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2208 "%inval = OpLoad %f32 %inloc\n"
2209 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2210 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
2211 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
2212 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
2213 " OpSelectionMerge %if_end None\n"
2214 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2215 "%if_true = OpLabel\n"
2216 "%negate = OpFNegate %f32 %inval\n"
2217 " OpStore %outloc %negate\n"
2218 " OpBranch %if_end\n"
2219 "%if_false = OpLabel\n"
2220 " OpUnreachable\n" // Unreachable else branch for if statement
2221 "%if_end = OpLabel\n"
2225 // not_called_function()
2226 "%func_not_called_func = OpFunction %void None %voidf\n"
2227 "%not_called_func_entry = OpLabel\n"
2228 " OpUnreachable\n" // Unreachable entry block in not called static function
2232 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
2233 "%valptr = OpFunctionParameter %u32ptr\n"
2234 "%modulo4_entry = OpLabel\n"
2235 "%val = OpLoad %u32 %valptr\n"
2236 "%modulo = OpUMod %u32 %val %four\n"
2237 " OpSelectionMerge %switch_merge None\n"
2238 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
2239 "%case0 = OpLabel\n"
2240 " OpReturnValue %three\n"
2241 "%case1 = OpLabel\n"
2242 " OpReturnValue %two\n"
2243 "%case2 = OpLabel\n"
2244 " OpReturnValue %one\n"
2245 "%case3 = OpLabel\n"
2246 " OpReturnValue %zero\n"
2247 "%default = OpLabel\n"
2248 " OpUnreachable\n" // Unreachable default case for switch statement
2249 "%switch_merge = OpLabel\n"
2250 " OpUnreachable\n" // Unreachable merge block for switch statement
2254 "%func_const5 = OpFunction %u32 None %unitf\n"
2255 "%const5_entry = OpLabel\n"
2256 " OpReturnValue %five\n"
2257 "%unreachable = OpLabel\n"
2258 " OpUnreachable\n" // Unreachable block in function
2260 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2261 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2262 spec.numWorkGroups = IVec3(numElements, 1, 1);
2264 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
2266 return group.release();
2269 // Assembly code used for testing decoration group is based on GLSL source code:
2273 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
2274 // float elements[];
2276 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
2277 // float elements[];
2279 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
2280 // float elements[];
2282 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
2283 // float elements[];
2285 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
2286 // float elements[];
2288 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
2289 // float elements[];
2293 // uint x = gl_GlobalInvocationID.x;
2294 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
2296 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
2298 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
2299 ComputeShaderSpec spec;
2300 de::Random rnd (deStringHash(group->getName()));
2301 const int numElements = 100;
2302 vector<float> inputFloats0 (numElements, 0);
2303 vector<float> inputFloats1 (numElements, 0);
2304 vector<float> inputFloats2 (numElements, 0);
2305 vector<float> inputFloats3 (numElements, 0);
2306 vector<float> inputFloats4 (numElements, 0);
2307 vector<float> outputFloats (numElements, 0);
2309 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
2310 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
2311 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
2312 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
2313 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
2315 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2316 floorAll(inputFloats0);
2317 floorAll(inputFloats1);
2318 floorAll(inputFloats2);
2319 floorAll(inputFloats3);
2320 floorAll(inputFloats4);
2322 for (size_t ndx = 0; ndx < numElements; ++ndx)
2323 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
2326 string(getComputeAsmShaderPreamble()) +
2328 "OpSource GLSL 430\n"
2329 "OpName %main \"main\"\n"
2330 "OpName %id \"gl_GlobalInvocationID\"\n"
2332 // Not using group decoration on variable.
2333 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2334 // Not using group decoration on type.
2335 "OpDecorate %f32arr ArrayStride 4\n"
2337 "OpDecorate %groups BufferBlock\n"
2338 "OpDecorate %groupm Offset 0\n"
2339 "%groups = OpDecorationGroup\n"
2340 "%groupm = OpDecorationGroup\n"
2342 // Group decoration on multiple structs.
2343 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
2344 // Group decoration on multiple struct members.
2345 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
2347 "OpDecorate %group1 DescriptorSet 0\n"
2348 "OpDecorate %group3 DescriptorSet 0\n"
2349 "OpDecorate %group3 NonWritable\n"
2350 "OpDecorate %group3 Restrict\n"
2351 "%group0 = OpDecorationGroup\n"
2352 "%group1 = OpDecorationGroup\n"
2353 "%group3 = OpDecorationGroup\n"
2355 // Applying the same decoration group multiple times.
2356 "OpGroupDecorate %group1 %outdata\n"
2357 "OpGroupDecorate %group1 %outdata\n"
2358 "OpGroupDecorate %group1 %outdata\n"
2359 "OpDecorate %outdata DescriptorSet 0\n"
2360 "OpDecorate %outdata Binding 5\n"
2361 // Applying decoration group containing nothing.
2362 "OpGroupDecorate %group0 %indata0\n"
2363 "OpDecorate %indata0 DescriptorSet 0\n"
2364 "OpDecorate %indata0 Binding 0\n"
2365 // Applying decoration group containing one decoration.
2366 "OpGroupDecorate %group1 %indata1\n"
2367 "OpDecorate %indata1 Binding 1\n"
2368 // Applying decoration group containing multiple decorations.
2369 "OpGroupDecorate %group3 %indata2 %indata3\n"
2370 "OpDecorate %indata2 Binding 2\n"
2371 "OpDecorate %indata3 Binding 3\n"
2372 // Applying multiple decoration groups (with overlapping).
2373 "OpGroupDecorate %group0 %indata4\n"
2374 "OpGroupDecorate %group1 %indata4\n"
2375 "OpGroupDecorate %group3 %indata4\n"
2376 "OpDecorate %indata4 Binding 4\n"
2378 + string(getComputeAsmCommonTypes()) +
2380 "%id = OpVariable %uvec3ptr Input\n"
2381 "%zero = OpConstant %i32 0\n"
2383 "%outbuf = OpTypeStruct %f32arr\n"
2384 "%outbufptr = OpTypePointer Uniform %outbuf\n"
2385 "%outdata = OpVariable %outbufptr Uniform\n"
2386 "%inbuf0 = OpTypeStruct %f32arr\n"
2387 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
2388 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
2389 "%inbuf1 = OpTypeStruct %f32arr\n"
2390 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
2391 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
2392 "%inbuf2 = OpTypeStruct %f32arr\n"
2393 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
2394 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
2395 "%inbuf3 = OpTypeStruct %f32arr\n"
2396 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
2397 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
2398 "%inbuf4 = OpTypeStruct %f32arr\n"
2399 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
2400 "%indata4 = OpVariable %inbufptr Uniform\n"
2402 "%main = OpFunction %void None %voidf\n"
2403 "%label = OpLabel\n"
2404 "%idval = OpLoad %uvec3 %id\n"
2405 "%x = OpCompositeExtract %u32 %idval 0\n"
2406 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
2407 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
2408 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
2409 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
2410 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
2411 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2412 "%inval0 = OpLoad %f32 %inloc0\n"
2413 "%inval1 = OpLoad %f32 %inloc1\n"
2414 "%inval2 = OpLoad %f32 %inloc2\n"
2415 "%inval3 = OpLoad %f32 %inloc3\n"
2416 "%inval4 = OpLoad %f32 %inloc4\n"
2417 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
2418 "%add1 = OpFAdd %f32 %add0 %inval2\n"
2419 "%add2 = OpFAdd %f32 %add1 %inval3\n"
2420 "%add = OpFAdd %f32 %add2 %inval4\n"
2421 " OpStore %outloc %add\n"
2424 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
2425 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
2426 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2427 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2428 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2429 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2430 spec.numWorkGroups = IVec3(numElements, 1, 1);
2432 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
2434 return group.release();
2437 struct SpecConstantTwoIntCase
2439 const char* caseName;
2440 const char* scDefinition0;
2441 const char* scDefinition1;
2442 const char* scResultType;
2443 const char* scOperation;
2444 deInt32 scActualValue0;
2445 deInt32 scActualValue1;
2446 const char* resultOperation;
2447 vector<deInt32> expectedOutput;
2449 SpecConstantTwoIntCase (const char* name,
2450 const char* definition0,
2451 const char* definition1,
2452 const char* resultType,
2453 const char* operation,
2456 const char* resultOp,
2457 const vector<deInt32>& output)
2459 , scDefinition0 (definition0)
2460 , scDefinition1 (definition1)
2461 , scResultType (resultType)
2462 , scOperation (operation)
2463 , scActualValue0 (value0)
2464 , scActualValue1 (value1)
2465 , resultOperation (resultOp)
2466 , expectedOutput (output) {}
2469 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
2471 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
2472 vector<SpecConstantTwoIntCase> cases;
2473 de::Random rnd (deStringHash(group->getName()));
2474 const int numElements = 100;
2475 vector<deInt32> inputInts (numElements, 0);
2476 vector<deInt32> outputInts1 (numElements, 0);
2477 vector<deInt32> outputInts2 (numElements, 0);
2478 vector<deInt32> outputInts3 (numElements, 0);
2479 vector<deInt32> outputInts4 (numElements, 0);
2480 const StringTemplate shaderTemplate (
2481 "${CAPABILITIES:opt}"
2482 + string(getComputeAsmShaderPreamble()) +
2484 "OpName %main \"main\"\n"
2485 "OpName %id \"gl_GlobalInvocationID\"\n"
2487 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2488 "OpDecorate %sc_0 SpecId 0\n"
2489 "OpDecorate %sc_1 SpecId 1\n"
2490 "OpDecorate %i32arr ArrayStride 4\n"
2492 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2494 "${OPTYPE_DEFINITIONS:opt}"
2495 "%buf = OpTypeStruct %i32arr\n"
2496 "%bufptr = OpTypePointer Uniform %buf\n"
2497 "%indata = OpVariable %bufptr Uniform\n"
2498 "%outdata = OpVariable %bufptr Uniform\n"
2500 "%id = OpVariable %uvec3ptr Input\n"
2501 "%zero = OpConstant %i32 0\n"
2503 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
2504 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
2505 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
2507 "%main = OpFunction %void None %voidf\n"
2508 "%label = OpLabel\n"
2509 "${TYPE_CONVERT:opt}"
2510 "%idval = OpLoad %uvec3 %id\n"
2511 "%x = OpCompositeExtract %u32 %idval 0\n"
2512 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2513 "%inval = OpLoad %i32 %inloc\n"
2514 "%final = ${GEN_RESULT}\n"
2515 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2516 " OpStore %outloc %final\n"
2518 " OpFunctionEnd\n");
2520 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
2522 for (size_t ndx = 0; ndx < numElements; ++ndx)
2524 outputInts1[ndx] = inputInts[ndx] + 42;
2525 outputInts2[ndx] = inputInts[ndx];
2526 outputInts3[ndx] = inputInts[ndx] - 11200;
2527 outputInts4[ndx] = inputInts[ndx] + 1;
2530 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
2531 const char addSc32ToInput[] = "OpIAdd %i32 %inval %sc_final32";
2532 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
2533 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
2535 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
2536 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
2537 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
2538 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
2539 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
2540 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2541 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2542 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
2543 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
2544 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
2545 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
2546 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2547 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2548 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
2549 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
2550 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
2551 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2552 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
2553 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
2554 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
2555 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2556 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
2557 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
2558 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2559 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2560 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2561 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2562 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
2563 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
2564 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
2565 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
2566 cases.push_back(SpecConstantTwoIntCase("sconvert", " %i32 0", " %i32 0", "%i16", "SConvert %sc_0", -11200, 0, addSc32ToInput, outputInts3));
2567 // -969998336 stored as 32-bit two's complement is the binary representation of -11200 as IEEE-754 Float
2568 cases.push_back(SpecConstantTwoIntCase("fconvert", " %f32 0", " %f32 0", "%f64", "FConvert %sc_0", -969998336, 0, addSc32ToInput, outputInts3));
2570 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2572 map<string, string> specializations;
2573 ComputeShaderSpec spec;
2574 ComputeTestFeatures features = COMPUTE_TEST_USES_NONE;
2576 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
2577 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
2578 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
2579 specializations["SC_OP"] = cases[caseNdx].scOperation;
2580 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
2582 // Special SPIR-V code for SConvert-case
2583 if (strcmp(cases[caseNdx].caseName, "sconvert") == 0)
2585 features = COMPUTE_TEST_USES_INT16;
2586 specializations["CAPABILITIES"] = "OpCapability Int16\n"; // Adds 16-bit integer capability
2587 specializations["OPTYPE_DEFINITIONS"] = "%i16 = OpTypeInt 16 1\n"; // Adds 16-bit integer type
2588 specializations["TYPE_CONVERT"] = "%sc_final32 = OpSConvert %i32 %sc_final\n"; // Converts 16-bit integer to 32-bit integer
2591 // Special SPIR-V code for FConvert-case
2592 if (strcmp(cases[caseNdx].caseName, "fconvert") == 0)
2594 features = COMPUTE_TEST_USES_FLOAT64;
2595 specializations["CAPABILITIES"] = "OpCapability Float64\n"; // Adds 64-bit float capability
2596 specializations["OPTYPE_DEFINITIONS"] = "%f64 = OpTypeFloat 64\n"; // Adds 64-bit float type
2597 specializations["TYPE_CONVERT"] = "%sc_final32 = OpConvertFToS %i32 %sc_final\n"; // Converts 64-bit float to 32-bit integer
2600 spec.assembly = shaderTemplate.specialize(specializations);
2601 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2602 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
2603 spec.numWorkGroups = IVec3(numElements, 1, 1);
2604 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
2605 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
2607 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec, features));
2610 ComputeShaderSpec spec;
2613 string(getComputeAsmShaderPreamble()) +
2615 "OpName %main \"main\"\n"
2616 "OpName %id \"gl_GlobalInvocationID\"\n"
2618 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2619 "OpDecorate %sc_0 SpecId 0\n"
2620 "OpDecorate %sc_1 SpecId 1\n"
2621 "OpDecorate %sc_2 SpecId 2\n"
2622 "OpDecorate %i32arr ArrayStride 4\n"
2624 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2626 "%ivec3 = OpTypeVector %i32 3\n"
2627 "%buf = OpTypeStruct %i32arr\n"
2628 "%bufptr = OpTypePointer Uniform %buf\n"
2629 "%indata = OpVariable %bufptr Uniform\n"
2630 "%outdata = OpVariable %bufptr Uniform\n"
2632 "%id = OpVariable %uvec3ptr Input\n"
2633 "%zero = OpConstant %i32 0\n"
2634 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
2635 "%vec3_undef = OpUndef %ivec3\n"
2637 "%sc_0 = OpSpecConstant %i32 0\n"
2638 "%sc_1 = OpSpecConstant %i32 0\n"
2639 "%sc_2 = OpSpecConstant %i32 0\n"
2640 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
2641 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
2642 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
2643 "%sc_vec3_0_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
2644 "%sc_vec3_1_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
2645 "%sc_vec3_2_s = OpSpecConstantOp %ivec3 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
2646 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
2647 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
2648 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
2649 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
2650 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
2651 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
2652 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
2654 "%main = OpFunction %void None %voidf\n"
2655 "%label = OpLabel\n"
2656 "%idval = OpLoad %uvec3 %id\n"
2657 "%x = OpCompositeExtract %u32 %idval 0\n"
2658 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2659 "%inval = OpLoad %i32 %inloc\n"
2660 "%final = OpIAdd %i32 %inval %sc_final\n"
2661 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2662 " OpStore %outloc %final\n"
2665 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2666 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
2667 spec.numWorkGroups = IVec3(numElements, 1, 1);
2668 spec.specConstants.push_back(123);
2669 spec.specConstants.push_back(56);
2670 spec.specConstants.push_back(-77);
2672 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
2674 return group.release();
2677 void createOpPhiVartypeTests (de::MovePtr<tcu::TestCaseGroup>& group, tcu::TestContext& testCtx)
2679 ComputeShaderSpec specInt;
2680 ComputeShaderSpec specFloat;
2681 ComputeShaderSpec specVec3;
2682 ComputeShaderSpec specMat4;
2683 ComputeShaderSpec specArray;
2684 ComputeShaderSpec specStruct;
2685 de::Random rnd (deStringHash(group->getName()));
2686 const int numElements = 100;
2687 vector<float> inputFloats (numElements, 0);
2688 vector<float> outputFloats (numElements, 0);
2690 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
2692 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2693 floorAll(inputFloats);
2695 for (size_t ndx = 0; ndx < numElements; ++ndx)
2697 // Just check if the value is positive or not
2698 outputFloats[ndx] = (inputFloats[ndx] > 0) ? 1.0f : -1.0f;
2701 // All of the tests are of the form:
2705 // if (inputdata > 0)
2712 specFloat.assembly =
2713 string(getComputeAsmShaderPreamble()) +
2715 "OpSource GLSL 430\n"
2716 "OpName %main \"main\"\n"
2717 "OpName %id \"gl_GlobalInvocationID\"\n"
2719 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2721 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2723 "%id = OpVariable %uvec3ptr Input\n"
2724 "%zero = OpConstant %i32 0\n"
2725 "%float_0 = OpConstant %f32 0.0\n"
2726 "%float_1 = OpConstant %f32 1.0\n"
2727 "%float_n1 = OpConstant %f32 -1.0\n"
2729 "%main = OpFunction %void None %voidf\n"
2730 "%entry = OpLabel\n"
2731 "%idval = OpLoad %uvec3 %id\n"
2732 "%x = OpCompositeExtract %u32 %idval 0\n"
2733 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2734 "%inval = OpLoad %f32 %inloc\n"
2736 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2737 " OpSelectionMerge %cm None\n"
2738 " OpBranchConditional %comp %tb %fb\n"
2744 "%res = OpPhi %f32 %float_1 %tb %float_n1 %fb\n"
2746 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2747 " OpStore %outloc %res\n"
2751 specFloat.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2752 specFloat.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2753 specFloat.numWorkGroups = IVec3(numElements, 1, 1);
2756 string(getComputeAsmShaderPreamble()) +
2758 "OpSource GLSL 430\n"
2759 "OpName %main \"main\"\n"
2760 "OpName %id \"gl_GlobalInvocationID\"\n"
2762 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2764 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2766 "%id = OpVariable %uvec3ptr Input\n"
2767 "%v4f32 = OpTypeVector %f32 4\n"
2768 "%mat4v4f32 = OpTypeMatrix %v4f32 4\n"
2769 "%zero = OpConstant %i32 0\n"
2770 "%float_0 = OpConstant %f32 0.0\n"
2771 "%float_1 = OpConstant %f32 1.0\n"
2772 "%float_n1 = OpConstant %f32 -1.0\n"
2773 "%m11 = OpConstantComposite %v4f32 %float_1 %float_0 %float_0 %float_0\n"
2774 "%m12 = OpConstantComposite %v4f32 %float_0 %float_1 %float_0 %float_0\n"
2775 "%m13 = OpConstantComposite %v4f32 %float_0 %float_0 %float_1 %float_0\n"
2776 "%m14 = OpConstantComposite %v4f32 %float_0 %float_0 %float_0 %float_1\n"
2777 "%m1 = OpConstantComposite %mat4v4f32 %m11 %m12 %m13 %m14\n"
2778 "%m21 = OpConstantComposite %v4f32 %float_n1 %float_0 %float_0 %float_0\n"
2779 "%m22 = OpConstantComposite %v4f32 %float_0 %float_n1 %float_0 %float_0\n"
2780 "%m23 = OpConstantComposite %v4f32 %float_0 %float_0 %float_n1 %float_0\n"
2781 "%m24 = OpConstantComposite %v4f32 %float_0 %float_0 %float_0 %float_n1\n"
2782 "%m2 = OpConstantComposite %mat4v4f32 %m21 %m22 %m23 %m24\n"
2784 "%main = OpFunction %void None %voidf\n"
2785 "%entry = OpLabel\n"
2786 "%idval = OpLoad %uvec3 %id\n"
2787 "%x = OpCompositeExtract %u32 %idval 0\n"
2788 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2789 "%inval = OpLoad %f32 %inloc\n"
2791 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2792 " OpSelectionMerge %cm None\n"
2793 " OpBranchConditional %comp %tb %fb\n"
2799 "%mres = OpPhi %mat4v4f32 %m1 %tb %m2 %fb\n"
2800 "%res = OpCompositeExtract %f32 %mres 2 2\n"
2802 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2803 " OpStore %outloc %res\n"
2807 specMat4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2808 specMat4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2809 specMat4.numWorkGroups = IVec3(numElements, 1, 1);
2812 string(getComputeAsmShaderPreamble()) +
2814 "OpSource GLSL 430\n"
2815 "OpName %main \"main\"\n"
2816 "OpName %id \"gl_GlobalInvocationID\"\n"
2818 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2820 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2822 "%id = OpVariable %uvec3ptr Input\n"
2823 "%zero = OpConstant %i32 0\n"
2824 "%float_0 = OpConstant %f32 0.0\n"
2825 "%float_1 = OpConstant %f32 1.0\n"
2826 "%float_n1 = OpConstant %f32 -1.0\n"
2827 "%v1 = OpConstantComposite %fvec3 %float_1 %float_1 %float_1\n"
2828 "%v2 = OpConstantComposite %fvec3 %float_n1 %float_n1 %float_n1\n"
2830 "%main = OpFunction %void None %voidf\n"
2831 "%entry = OpLabel\n"
2832 "%idval = OpLoad %uvec3 %id\n"
2833 "%x = OpCompositeExtract %u32 %idval 0\n"
2834 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2835 "%inval = OpLoad %f32 %inloc\n"
2837 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2838 " OpSelectionMerge %cm None\n"
2839 " OpBranchConditional %comp %tb %fb\n"
2845 "%vres = OpPhi %fvec3 %v1 %tb %v2 %fb\n"
2846 "%res = OpCompositeExtract %f32 %vres 2\n"
2848 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2849 " OpStore %outloc %res\n"
2853 specVec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2854 specVec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2855 specVec3.numWorkGroups = IVec3(numElements, 1, 1);
2858 string(getComputeAsmShaderPreamble()) +
2860 "OpSource GLSL 430\n"
2861 "OpName %main \"main\"\n"
2862 "OpName %id \"gl_GlobalInvocationID\"\n"
2864 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2866 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2868 "%id = OpVariable %uvec3ptr Input\n"
2869 "%zero = OpConstant %i32 0\n"
2870 "%float_0 = OpConstant %f32 0.0\n"
2871 "%i1 = OpConstant %i32 1\n"
2872 "%i2 = OpConstant %i32 -1\n"
2874 "%main = OpFunction %void None %voidf\n"
2875 "%entry = OpLabel\n"
2876 "%idval = OpLoad %uvec3 %id\n"
2877 "%x = OpCompositeExtract %u32 %idval 0\n"
2878 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2879 "%inval = OpLoad %f32 %inloc\n"
2881 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2882 " OpSelectionMerge %cm None\n"
2883 " OpBranchConditional %comp %tb %fb\n"
2889 "%ires = OpPhi %i32 %i1 %tb %i2 %fb\n"
2890 "%res = OpConvertSToF %f32 %ires\n"
2892 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2893 " OpStore %outloc %res\n"
2897 specInt.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2898 specInt.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2899 specInt.numWorkGroups = IVec3(numElements, 1, 1);
2901 specArray.assembly =
2902 string(getComputeAsmShaderPreamble()) +
2904 "OpSource GLSL 430\n"
2905 "OpName %main \"main\"\n"
2906 "OpName %id \"gl_GlobalInvocationID\"\n"
2908 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2910 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2912 "%id = OpVariable %uvec3ptr Input\n"
2913 "%zero = OpConstant %i32 0\n"
2914 "%u7 = OpConstant %u32 7\n"
2915 "%float_0 = OpConstant %f32 0.0\n"
2916 "%float_1 = OpConstant %f32 1.0\n"
2917 "%float_n1 = OpConstant %f32 -1.0\n"
2918 "%f32a7 = OpTypeArray %f32 %u7\n"
2919 "%a1 = OpConstantComposite %f32a7 %float_1 %float_1 %float_1 %float_1 %float_1 %float_1 %float_1\n"
2920 "%a2 = OpConstantComposite %f32a7 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1\n"
2921 "%main = OpFunction %void None %voidf\n"
2922 "%entry = OpLabel\n"
2923 "%idval = OpLoad %uvec3 %id\n"
2924 "%x = OpCompositeExtract %u32 %idval 0\n"
2925 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2926 "%inval = OpLoad %f32 %inloc\n"
2928 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2929 " OpSelectionMerge %cm None\n"
2930 " OpBranchConditional %comp %tb %fb\n"
2936 "%ares = OpPhi %f32a7 %a1 %tb %a2 %fb\n"
2937 "%res = OpCompositeExtract %f32 %ares 5\n"
2939 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2940 " OpStore %outloc %res\n"
2944 specArray.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2945 specArray.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2946 specArray.numWorkGroups = IVec3(numElements, 1, 1);
2948 specStruct.assembly =
2949 string(getComputeAsmShaderPreamble()) +
2951 "OpSource GLSL 430\n"
2952 "OpName %main \"main\"\n"
2953 "OpName %id \"gl_GlobalInvocationID\"\n"
2955 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2957 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2959 "%id = OpVariable %uvec3ptr Input\n"
2960 "%zero = OpConstant %i32 0\n"
2961 "%float_0 = OpConstant %f32 0.0\n"
2962 "%float_1 = OpConstant %f32 1.0\n"
2963 "%float_n1 = OpConstant %f32 -1.0\n"
2965 "%v2f32 = OpTypeVector %f32 2\n"
2966 "%Data2 = OpTypeStruct %f32 %v2f32\n"
2967 "%Data = OpTypeStruct %Data2 %f32\n"
2969 "%in1a = OpConstantComposite %v2f32 %float_1 %float_1\n"
2970 "%in1b = OpConstantComposite %Data2 %float_1 %in1a\n"
2971 "%s1 = OpConstantComposite %Data %in1b %float_1\n"
2972 "%in2a = OpConstantComposite %v2f32 %float_n1 %float_n1\n"
2973 "%in2b = OpConstantComposite %Data2 %float_n1 %in2a\n"
2974 "%s2 = OpConstantComposite %Data %in2b %float_n1\n"
2976 "%main = OpFunction %void None %voidf\n"
2977 "%entry = OpLabel\n"
2978 "%idval = OpLoad %uvec3 %id\n"
2979 "%x = OpCompositeExtract %u32 %idval 0\n"
2980 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2981 "%inval = OpLoad %f32 %inloc\n"
2983 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2984 " OpSelectionMerge %cm None\n"
2985 " OpBranchConditional %comp %tb %fb\n"
2991 "%sres = OpPhi %Data %s1 %tb %s2 %fb\n"
2992 "%res = OpCompositeExtract %f32 %sres 0 0\n"
2994 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2995 " OpStore %outloc %res\n"
2999 specStruct.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3000 specStruct.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3001 specStruct.numWorkGroups = IVec3(numElements, 1, 1);
3003 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_int", "OpPhi with int variables", specInt));
3004 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_float", "OpPhi with float variables", specFloat));
3005 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_vec3", "OpPhi with vec3 variables", specVec3));
3006 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_mat4", "OpPhi with mat4 variables", specMat4));
3007 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_array", "OpPhi with array variables", specArray));
3008 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_struct", "OpPhi with struct variables", specStruct));
3011 string generateConstantDefinitions (int count)
3013 std::ostringstream r;
3014 for (int i = 0; i < count; i++)
3015 r << "%cf" << (i * 10 + 5) << " = OpConstant %f32 " <<(i * 10 + 5) << ".0\n";
3020 string generateSwitchCases (int count)
3022 std::ostringstream r;
3023 for (int i = 0; i < count; i++)
3024 r << " " << i << " %case" << i;
3029 string generateSwitchTargets (int count)
3031 std::ostringstream r;
3032 for (int i = 0; i < count; i++)
3033 r << "%case" << i << " = OpLabel\n OpBranch %phi\n";
3038 string generateOpPhiParams (int count)
3040 std::ostringstream r;
3041 for (int i = 0; i < count; i++)
3042 r << " %cf" << (i * 10 + 5) << " %case" << i;
3047 string generateIntWidth (int value)
3049 std::ostringstream r;
3054 // Expand input string by injecting "ABC" between the input
3055 // string characters. The acc/add/treshold parameters are used
3056 // to skip some of the injections to make the result less
3057 // uniform (and a lot shorter).
3058 string expandOpPhiCase5 (const string& s, int &acc, int add, int treshold)
3060 std::ostringstream res;
3061 const char* p = s.c_str();
3077 // Calculate expected result based on the code string
3078 float calcOpPhiCase5 (float val, const string& s)
3080 const char* p = s.c_str();
3083 const float tv[8] = { 0.5f, 1.5f, 3.5f, 7.5f, 15.5f, 31.5f, 63.5f, 127.5f };
3084 const float v = deFloatAbs(val);
3089 for (int i = 7; i >= 0; --i)
3090 x[i] = std::fmod((float)v, (float)(2 << i));
3091 for (int i = 7; i >= 0; --i)
3092 b[i] = x[i] > tv[i];
3099 if (skip == 0 && b[depth])
3110 if (b[depth] || skip)
3124 // In the code string, the letters represent the following:
3127 // if (certain bit is set)
3138 // AABCBC leads to if(){r++;if(){r++;}else{}}else{}
3139 // ABABCC leads to if(){r++;}else{if(){r++;}else{}}
3140 // ABCABC leads to if(){r++;}else{}if(){r++;}else{}
3142 // Code generation gets a bit complicated due to the else-branches,
3143 // which do not generate new values. Thus, the generator needs to
3144 // keep track of the previous variable change seen by the else
3146 string generateOpPhiCase5 (const string& s)
3148 std::stack<int> idStack;
3149 std::stack<std::string> value;
3150 std::stack<std::string> valueLabel;
3151 std::stack<std::string> mergeLeft;
3152 std::stack<std::string> mergeRight;
3153 std::ostringstream res;
3154 const char* p = s.c_str();
3160 value.push("%f32_0");
3161 valueLabel.push("%f32_0 %entry");
3169 idStack.push(currId);
3170 res << "\tOpSelectionMerge %m" << currId << " None\n";
3171 res << "\tOpBranchConditional %b" << depth << " %t" << currId << " %f" << currId << "\n";
3172 res << "%t" << currId << " = OpLabel\n";
3173 res << "%rt" << currId << " = OpFAdd %f32 " << value.top() << " %f32_1\n";
3174 std::ostringstream tag;
3175 tag << "%rt" << currId;
3176 value.push(tag.str());
3177 tag << " %t" << currId;
3178 valueLabel.push(tag.str());
3183 mergeLeft.push(valueLabel.top());
3186 res << "\tOpBranch %m" << currId << "\n";
3187 res << "%f" << currId << " = OpLabel\n";
3188 std::ostringstream tag;
3189 tag << value.top() << " %f" << currId;
3191 valueLabel.push(tag.str());
3196 mergeRight.push(valueLabel.top());
3197 res << "\tOpBranch %m" << currId << "\n";
3198 res << "%m" << currId << " = OpLabel\n";
3200 res << "%res"; // last result goes to %res
3202 res << "%rm" << currId;
3203 res << " = OpPhi %f32 " << mergeLeft.top() << " " << mergeRight.top() << "\n";
3204 std::ostringstream tag;
3205 tag << "%rm" << currId;
3207 value.push(tag.str());
3208 tag << " %m" << currId;
3210 valueLabel.push(tag.str());
3215 currId = idStack.top();
3223 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
3225 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
3226 ComputeShaderSpec spec1;
3227 ComputeShaderSpec spec2;
3228 ComputeShaderSpec spec3;
3229 ComputeShaderSpec spec4;
3230 ComputeShaderSpec spec5;
3231 de::Random rnd (deStringHash(group->getName()));
3232 const int numElements = 100;
3233 vector<float> inputFloats (numElements, 0);
3234 vector<float> outputFloats1 (numElements, 0);
3235 vector<float> outputFloats2 (numElements, 0);
3236 vector<float> outputFloats3 (numElements, 0);
3237 vector<float> outputFloats4 (numElements, 0);
3238 vector<float> outputFloats5 (numElements, 0);
3239 std::string codestring = "ABC";
3240 const int test4Width = 1024;
3242 // Build case 5 code string. Each iteration makes the hierarchy more complicated.
3243 // 9 iterations with (7, 24) parameters makes the hierarchy 8 deep with about 1500 lines of
3245 for (int i = 0, acc = 0; i < 9; i++)
3246 codestring = expandOpPhiCase5(codestring, acc, 7, 24);
3248 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
3250 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3251 floorAll(inputFloats);
3253 for (size_t ndx = 0; ndx < numElements; ++ndx)
3257 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
3258 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
3259 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
3262 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
3263 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
3265 int index4 = (int)deFloor(deAbs((float)ndx * inputFloats[ndx]));
3266 outputFloats4[ndx] = (float)(index4 % test4Width) * 10.0f + 5.0f;
3268 outputFloats5[ndx] = calcOpPhiCase5(inputFloats[ndx], codestring);
3272 string(getComputeAsmShaderPreamble()) +
3274 "OpSource GLSL 430\n"
3275 "OpName %main \"main\"\n"
3276 "OpName %id \"gl_GlobalInvocationID\"\n"
3278 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3280 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3282 "%id = OpVariable %uvec3ptr Input\n"
3283 "%zero = OpConstant %i32 0\n"
3284 "%three = OpConstant %u32 3\n"
3285 "%constf5p5 = OpConstant %f32 5.5\n"
3286 "%constf20p5 = OpConstant %f32 20.5\n"
3287 "%constf1p75 = OpConstant %f32 1.75\n"
3288 "%constf8p5 = OpConstant %f32 8.5\n"
3289 "%constf6p5 = OpConstant %f32 6.5\n"
3291 "%main = OpFunction %void None %voidf\n"
3292 "%entry = OpLabel\n"
3293 "%idval = OpLoad %uvec3 %id\n"
3294 "%x = OpCompositeExtract %u32 %idval 0\n"
3295 "%selector = OpUMod %u32 %x %three\n"
3296 " OpSelectionMerge %phi None\n"
3297 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
3299 // Case 1 before OpPhi.
3300 "%case1 = OpLabel\n"
3303 "%default = OpLabel\n"
3307 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
3308 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3309 "%inval = OpLoad %f32 %inloc\n"
3310 "%add = OpFAdd %f32 %inval %operand\n"
3311 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3312 " OpStore %outloc %add\n"
3315 // Case 0 after OpPhi.
3316 "%case0 = OpLabel\n"
3320 // Case 2 after OpPhi.
3321 "%case2 = OpLabel\n"
3325 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3326 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
3327 spec1.numWorkGroups = IVec3(numElements, 1, 1);
3329 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
3332 string(getComputeAsmShaderPreamble()) +
3334 "OpName %main \"main\"\n"
3335 "OpName %id \"gl_GlobalInvocationID\"\n"
3337 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3339 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3341 "%id = OpVariable %uvec3ptr Input\n"
3342 "%zero = OpConstant %i32 0\n"
3343 "%one = OpConstant %i32 1\n"
3344 "%three = OpConstant %i32 3\n"
3345 "%constf6p5 = OpConstant %f32 6.5\n"
3347 "%main = OpFunction %void None %voidf\n"
3348 "%entry = OpLabel\n"
3349 "%idval = OpLoad %uvec3 %id\n"
3350 "%x = OpCompositeExtract %u32 %idval 0\n"
3351 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3352 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3353 "%inval = OpLoad %f32 %inloc\n"
3357 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
3358 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
3359 "%step_next = OpIAdd %i32 %step %one\n"
3360 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
3361 "%still_loop = OpSLessThan %bool %step %three\n"
3362 " OpLoopMerge %exit %phi None\n"
3363 " OpBranchConditional %still_loop %phi %exit\n"
3366 " OpStore %outloc %accum\n"
3369 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3370 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
3371 spec2.numWorkGroups = IVec3(numElements, 1, 1);
3373 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
3376 string(getComputeAsmShaderPreamble()) +
3378 "OpName %main \"main\"\n"
3379 "OpName %id \"gl_GlobalInvocationID\"\n"
3381 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3383 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3385 "%f32ptr_f = OpTypePointer Function %f32\n"
3386 "%id = OpVariable %uvec3ptr Input\n"
3387 "%true = OpConstantTrue %bool\n"
3388 "%false = OpConstantFalse %bool\n"
3389 "%zero = OpConstant %i32 0\n"
3390 "%constf8p5 = OpConstant %f32 8.5\n"
3392 "%main = OpFunction %void None %voidf\n"
3393 "%entry = OpLabel\n"
3394 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
3395 "%idval = OpLoad %uvec3 %id\n"
3396 "%x = OpCompositeExtract %u32 %idval 0\n"
3397 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3398 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3399 "%a_init = OpLoad %f32 %inloc\n"
3400 "%b_init = OpLoad %f32 %b\n"
3404 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
3405 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
3406 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
3407 " OpLoopMerge %exit %phi None\n"
3408 " OpBranchConditional %still_loop %phi %exit\n"
3411 "%sub = OpFSub %f32 %a_next %b_next\n"
3412 " OpStore %outloc %sub\n"
3415 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3416 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
3417 spec3.numWorkGroups = IVec3(numElements, 1, 1);
3419 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
3422 "OpCapability Shader\n"
3423 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
3424 "OpMemoryModel Logical GLSL450\n"
3425 "OpEntryPoint GLCompute %main \"main\" %id\n"
3426 "OpExecutionMode %main LocalSize 1 1 1\n"
3428 "OpSource GLSL 430\n"
3429 "OpName %main \"main\"\n"
3430 "OpName %id \"gl_GlobalInvocationID\"\n"
3432 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3434 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3436 "%id = OpVariable %uvec3ptr Input\n"
3437 "%zero = OpConstant %i32 0\n"
3438 "%cimod = OpConstant %u32 " + generateIntWidth(test4Width) + "\n"
3440 + generateConstantDefinitions(test4Width) +
3442 "%main = OpFunction %void None %voidf\n"
3443 "%entry = OpLabel\n"
3444 "%idval = OpLoad %uvec3 %id\n"
3445 "%x = OpCompositeExtract %u32 %idval 0\n"
3446 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3447 "%inval = OpLoad %f32 %inloc\n"
3448 "%xf = OpConvertUToF %f32 %x\n"
3449 "%xm = OpFMul %f32 %xf %inval\n"
3450 "%xa = OpExtInst %f32 %ext FAbs %xm\n"
3451 "%xi = OpConvertFToU %u32 %xa\n"
3452 "%selector = OpUMod %u32 %xi %cimod\n"
3453 " OpSelectionMerge %phi None\n"
3454 " OpSwitch %selector %default "
3456 + generateSwitchCases(test4Width) +
3458 "%default = OpLabel\n"
3461 + generateSwitchTargets(test4Width) +
3464 "%result = OpPhi %f32"
3466 + generateOpPhiParams(test4Width) +
3468 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3469 " OpStore %outloc %result\n"
3473 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3474 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
3475 spec4.numWorkGroups = IVec3(numElements, 1, 1);
3477 group->addChild(new SpvAsmComputeShaderCase(testCtx, "wide", "OpPhi with a lot of parameters", spec4));
3480 "OpCapability Shader\n"
3481 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
3482 "OpMemoryModel Logical GLSL450\n"
3483 "OpEntryPoint GLCompute %main \"main\" %id\n"
3484 "OpExecutionMode %main LocalSize 1 1 1\n"
3485 "%code = OpString \"" + codestring + "\"\n"
3487 "OpSource GLSL 430\n"
3488 "OpName %main \"main\"\n"
3489 "OpName %id \"gl_GlobalInvocationID\"\n"
3491 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3493 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3495 "%id = OpVariable %uvec3ptr Input\n"
3496 "%zero = OpConstant %i32 0\n"
3497 "%f32_0 = OpConstant %f32 0.0\n"
3498 "%f32_0_5 = OpConstant %f32 0.5\n"
3499 "%f32_1 = OpConstant %f32 1.0\n"
3500 "%f32_1_5 = OpConstant %f32 1.5\n"
3501 "%f32_2 = OpConstant %f32 2.0\n"
3502 "%f32_3_5 = OpConstant %f32 3.5\n"
3503 "%f32_4 = OpConstant %f32 4.0\n"
3504 "%f32_7_5 = OpConstant %f32 7.5\n"
3505 "%f32_8 = OpConstant %f32 8.0\n"
3506 "%f32_15_5 = OpConstant %f32 15.5\n"
3507 "%f32_16 = OpConstant %f32 16.0\n"
3508 "%f32_31_5 = OpConstant %f32 31.5\n"
3509 "%f32_32 = OpConstant %f32 32.0\n"
3510 "%f32_63_5 = OpConstant %f32 63.5\n"
3511 "%f32_64 = OpConstant %f32 64.0\n"
3512 "%f32_127_5 = OpConstant %f32 127.5\n"
3513 "%f32_128 = OpConstant %f32 128.0\n"
3514 "%f32_256 = OpConstant %f32 256.0\n"
3516 "%main = OpFunction %void None %voidf\n"
3517 "%entry = OpLabel\n"
3518 "%idval = OpLoad %uvec3 %id\n"
3519 "%x = OpCompositeExtract %u32 %idval 0\n"
3520 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3521 "%inval = OpLoad %f32 %inloc\n"
3523 "%xabs = OpExtInst %f32 %ext FAbs %inval\n"
3524 "%x8 = OpFMod %f32 %xabs %f32_256\n"
3525 "%x7 = OpFMod %f32 %xabs %f32_128\n"
3526 "%x6 = OpFMod %f32 %xabs %f32_64\n"
3527 "%x5 = OpFMod %f32 %xabs %f32_32\n"
3528 "%x4 = OpFMod %f32 %xabs %f32_16\n"
3529 "%x3 = OpFMod %f32 %xabs %f32_8\n"
3530 "%x2 = OpFMod %f32 %xabs %f32_4\n"
3531 "%x1 = OpFMod %f32 %xabs %f32_2\n"
3533 "%b7 = OpFOrdGreaterThanEqual %bool %x8 %f32_127_5\n"
3534 "%b6 = OpFOrdGreaterThanEqual %bool %x7 %f32_63_5\n"
3535 "%b5 = OpFOrdGreaterThanEqual %bool %x6 %f32_31_5\n"
3536 "%b4 = OpFOrdGreaterThanEqual %bool %x5 %f32_15_5\n"
3537 "%b3 = OpFOrdGreaterThanEqual %bool %x4 %f32_7_5\n"
3538 "%b2 = OpFOrdGreaterThanEqual %bool %x3 %f32_3_5\n"
3539 "%b1 = OpFOrdGreaterThanEqual %bool %x2 %f32_1_5\n"
3540 "%b0 = OpFOrdGreaterThanEqual %bool %x1 %f32_0_5\n"
3542 + generateOpPhiCase5(codestring) +
3544 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3545 " OpStore %outloc %res\n"
3549 spec5.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3550 spec5.outputs.push_back(BufferSp(new Float32Buffer(outputFloats5)));
3551 spec5.numWorkGroups = IVec3(numElements, 1, 1);
3553 group->addChild(new SpvAsmComputeShaderCase(testCtx, "nested", "Stress OpPhi with a lot of nesting", spec5));
3555 createOpPhiVartypeTests(group, testCtx);
3557 return group.release();
3560 // Assembly code used for testing block order is based on GLSL source code:
3564 // layout(std140, set = 0, binding = 0) readonly buffer Input {
3565 // float elements[];
3567 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
3568 // float elements[];
3572 // uint x = gl_GlobalInvocationID.x;
3573 // output_data.elements[x] = input_data.elements[x];
3574 // if (x > uint(50)) {
3575 // switch (x % uint(3)) {
3576 // case 0: output_data.elements[x] += 1.5f; break;
3577 // case 1: output_data.elements[x] += 42.f; break;
3578 // case 2: output_data.elements[x] -= 27.f; break;
3582 // output_data.elements[x] = -input_data.elements[x];
3585 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
3587 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
3588 ComputeShaderSpec spec;
3589 de::Random rnd (deStringHash(group->getName()));
3590 const int numElements = 100;
3591 vector<float> inputFloats (numElements, 0);
3592 vector<float> outputFloats (numElements, 0);
3594 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3596 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3597 floorAll(inputFloats);
3599 for (size_t ndx = 0; ndx <= 50; ++ndx)
3600 outputFloats[ndx] = -inputFloats[ndx];
3602 for (size_t ndx = 51; ndx < numElements; ++ndx)
3606 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
3607 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
3608 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
3614 string(getComputeAsmShaderPreamble()) +
3616 "OpSource GLSL 430\n"
3617 "OpName %main \"main\"\n"
3618 "OpName %id \"gl_GlobalInvocationID\"\n"
3620 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3622 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3624 "%u32ptr = OpTypePointer Function %u32\n"
3625 "%u32ptr_input = OpTypePointer Input %u32\n"
3627 + string(getComputeAsmInputOutputBuffer()) +
3629 "%id = OpVariable %uvec3ptr Input\n"
3630 "%zero = OpConstant %i32 0\n"
3631 "%const3 = OpConstant %u32 3\n"
3632 "%const50 = OpConstant %u32 50\n"
3633 "%constf1p5 = OpConstant %f32 1.5\n"
3634 "%constf27 = OpConstant %f32 27.0\n"
3635 "%constf42 = OpConstant %f32 42.0\n"
3637 "%main = OpFunction %void None %voidf\n"
3640 "%entry = OpLabel\n"
3642 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
3643 "%xvar = OpVariable %u32ptr Function\n"
3644 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
3645 "%x = OpLoad %u32 %xptr\n"
3646 " OpStore %xvar %x\n"
3648 "%cmp = OpUGreaterThan %bool %x %const50\n"
3649 " OpSelectionMerge %if_merge None\n"
3650 " OpBranchConditional %cmp %if_true %if_false\n"
3652 // False branch for if-statement: placed in the middle of switch cases and before true branch.
3653 "%if_false = OpLabel\n"
3654 "%x_f = OpLoad %u32 %xvar\n"
3655 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
3656 "%inval_f = OpLoad %f32 %inloc_f\n"
3657 "%negate = OpFNegate %f32 %inval_f\n"
3658 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
3659 " OpStore %outloc_f %negate\n"
3660 " OpBranch %if_merge\n"
3662 // Merge block for if-statement: placed in the middle of true and false branch.
3663 "%if_merge = OpLabel\n"
3666 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
3667 "%if_true = OpLabel\n"
3668 "%xval_t = OpLoad %u32 %xvar\n"
3669 "%mod = OpUMod %u32 %xval_t %const3\n"
3670 " OpSelectionMerge %switch_merge None\n"
3671 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
3673 // Merge block for switch-statement: placed before the case
3674 // bodies. But it must follow OpSwitch which dominates it.
3675 "%switch_merge = OpLabel\n"
3676 " OpBranch %if_merge\n"
3678 // Case 1 for switch-statement: placed before case 0.
3679 // It must follow the OpSwitch that dominates it.
3680 "%case1 = OpLabel\n"
3681 "%x_1 = OpLoad %u32 %xvar\n"
3682 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
3683 "%inval_1 = OpLoad %f32 %inloc_1\n"
3684 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
3685 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
3686 " OpStore %outloc_1 %addf42\n"
3687 " OpBranch %switch_merge\n"
3689 // Case 2 for switch-statement.
3690 "%case2 = OpLabel\n"
3691 "%x_2 = OpLoad %u32 %xvar\n"
3692 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
3693 "%inval_2 = OpLoad %f32 %inloc_2\n"
3694 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
3695 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
3696 " OpStore %outloc_2 %subf27\n"
3697 " OpBranch %switch_merge\n"
3699 // Default case for switch-statement: placed in the middle of normal cases.
3700 "%default = OpLabel\n"
3701 " OpBranch %switch_merge\n"
3703 // Case 0 for switch-statement: out of order.
3704 "%case0 = OpLabel\n"
3705 "%x_0 = OpLoad %u32 %xvar\n"
3706 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
3707 "%inval_0 = OpLoad %f32 %inloc_0\n"
3708 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
3709 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
3710 " OpStore %outloc_0 %addf1p5\n"
3711 " OpBranch %switch_merge\n"
3714 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3715 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3716 spec.numWorkGroups = IVec3(numElements, 1, 1);
3718 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
3720 return group.release();
3723 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
3725 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
3726 ComputeShaderSpec spec1;
3727 ComputeShaderSpec spec2;
3728 de::Random rnd (deStringHash(group->getName()));
3729 const int numElements = 100;
3730 vector<float> inputFloats (numElements, 0);
3731 vector<float> outputFloats1 (numElements, 0);
3732 vector<float> outputFloats2 (numElements, 0);
3733 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
3735 for (size_t ndx = 0; ndx < numElements; ++ndx)
3737 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
3738 outputFloats2[ndx] = -inputFloats[ndx];
3741 const string assembly(
3742 "OpCapability Shader\n"
3743 "OpCapability ClipDistance\n"
3744 "OpMemoryModel Logical GLSL450\n"
3745 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
3746 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
3747 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
3748 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
3749 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
3750 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
3752 "OpName %comp_main1 \"entrypoint1\"\n"
3753 "OpName %comp_main2 \"entrypoint2\"\n"
3754 "OpName %vert_main \"entrypoint2\"\n"
3755 "OpName %id \"gl_GlobalInvocationID\"\n"
3756 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
3757 "OpName %vertexIndex \"gl_VertexIndex\"\n"
3758 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
3759 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
3760 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
3761 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
3763 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3764 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
3765 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
3766 "OpDecorate %vert_builtin_st Block\n"
3767 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
3768 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
3769 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
3771 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3773 "%zero = OpConstant %i32 0\n"
3774 "%one = OpConstant %u32 1\n"
3775 "%c_f32_1 = OpConstant %f32 1\n"
3777 "%i32inputptr = OpTypePointer Input %i32\n"
3778 "%vec4 = OpTypeVector %f32 4\n"
3779 "%vec4ptr = OpTypePointer Output %vec4\n"
3780 "%f32arr1 = OpTypeArray %f32 %one\n"
3781 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
3782 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
3783 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
3785 "%id = OpVariable %uvec3ptr Input\n"
3786 "%vertexIndex = OpVariable %i32inputptr Input\n"
3787 "%instanceIndex = OpVariable %i32inputptr Input\n"
3788 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
3790 // gl_Position = vec4(1.);
3791 "%vert_main = OpFunction %void None %voidf\n"
3792 "%vert_entry = OpLabel\n"
3793 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
3794 " OpStore %position %c_vec4_1\n"
3799 "%comp_main1 = OpFunction %void None %voidf\n"
3800 "%comp1_entry = OpLabel\n"
3801 "%idval1 = OpLoad %uvec3 %id\n"
3802 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
3803 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
3804 "%inval1 = OpLoad %f32 %inloc1\n"
3805 "%add = OpFAdd %f32 %inval1 %inval1\n"
3806 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
3807 " OpStore %outloc1 %add\n"
3812 "%comp_main2 = OpFunction %void None %voidf\n"
3813 "%comp2_entry = OpLabel\n"
3814 "%idval2 = OpLoad %uvec3 %id\n"
3815 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
3816 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
3817 "%inval2 = OpLoad %f32 %inloc2\n"
3818 "%neg = OpFNegate %f32 %inval2\n"
3819 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
3820 " OpStore %outloc2 %neg\n"
3822 " OpFunctionEnd\n");
3824 spec1.assembly = assembly;
3825 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3826 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
3827 spec1.numWorkGroups = IVec3(numElements, 1, 1);
3828 spec1.entryPoint = "entrypoint1";
3830 spec2.assembly = assembly;
3831 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3832 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
3833 spec2.numWorkGroups = IVec3(numElements, 1, 1);
3834 spec2.entryPoint = "entrypoint2";
3836 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
3837 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
3839 return group.release();
3842 inline std::string makeLongUTF8String (size_t num4ByteChars)
3844 // An example of a longest valid UTF-8 character. Be explicit about the
3845 // character type because Microsoft compilers can otherwise interpret the
3846 // character string as being over wide (16-bit) characters. Ideally, we
3847 // would just use a C++11 UTF-8 string literal, but we want to support older
3848 // Microsoft compilers.
3849 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
3850 std::string longString;
3851 longString.reserve(num4ByteChars * 4);
3852 for (size_t count = 0; count < num4ByteChars; count++)
3854 longString += earthAfrica;
3859 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
3861 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
3862 vector<CaseParameter> cases;
3863 de::Random rnd (deStringHash(group->getName()));
3864 const int numElements = 100;
3865 vector<float> positiveFloats (numElements, 0);
3866 vector<float> negativeFloats (numElements, 0);
3867 const StringTemplate shaderTemplate (
3868 "OpCapability Shader\n"
3869 "OpMemoryModel Logical GLSL450\n"
3871 "OpEntryPoint GLCompute %main \"main\" %id\n"
3872 "OpExecutionMode %main LocalSize 1 1 1\n"
3876 "OpName %main \"main\"\n"
3877 "OpName %id \"gl_GlobalInvocationID\"\n"
3879 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3881 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3883 "%id = OpVariable %uvec3ptr Input\n"
3884 "%zero = OpConstant %i32 0\n"
3886 "%main = OpFunction %void None %voidf\n"
3887 "%label = OpLabel\n"
3888 "%idval = OpLoad %uvec3 %id\n"
3889 "%x = OpCompositeExtract %u32 %idval 0\n"
3890 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3891 "%inval = OpLoad %f32 %inloc\n"
3892 "%neg = OpFNegate %f32 %inval\n"
3893 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3894 " OpStore %outloc %neg\n"
3896 " OpFunctionEnd\n");
3898 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
3899 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
3900 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
3901 "OpSource GLSL 430 %fname"));
3902 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
3903 "OpSource GLSL 430 %fname"));
3904 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
3905 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
3906 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
3907 "OpSource GLSL 430 %fname \"\""));
3908 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
3909 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
3910 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
3911 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
3912 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
3913 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
3914 "OpSourceContinued \"id main() {}\""));
3915 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
3916 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3917 "OpSourceContinued \"\""));
3918 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
3919 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3920 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
3921 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
3922 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3923 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
3924 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
3925 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
3926 "OpSourceContinued \"void\"\n"
3927 "OpSourceContinued \"main()\"\n"
3928 "OpSourceContinued \"{}\""));
3929 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
3930 "OpSource GLSL 430 %fname \"\"\n"
3931 "OpSourceContinued \"#version 430\nvoid main() {}\""));
3933 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3935 for (size_t ndx = 0; ndx < numElements; ++ndx)
3936 negativeFloats[ndx] = -positiveFloats[ndx];
3938 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3940 map<string, string> specializations;
3941 ComputeShaderSpec spec;
3943 specializations["SOURCE"] = cases[caseNdx].param;
3944 spec.assembly = shaderTemplate.specialize(specializations);
3945 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3946 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3947 spec.numWorkGroups = IVec3(numElements, 1, 1);
3949 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3952 return group.release();
3955 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
3957 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
3958 vector<CaseParameter> cases;
3959 de::Random rnd (deStringHash(group->getName()));
3960 const int numElements = 100;
3961 vector<float> inputFloats (numElements, 0);
3962 vector<float> outputFloats (numElements, 0);
3963 const StringTemplate shaderTemplate (
3964 string(getComputeAsmShaderPreamble()) +
3966 "OpSourceExtension \"${EXTENSION}\"\n"
3968 "OpName %main \"main\"\n"
3969 "OpName %id \"gl_GlobalInvocationID\"\n"
3971 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3973 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3975 "%id = OpVariable %uvec3ptr Input\n"
3976 "%zero = OpConstant %i32 0\n"
3978 "%main = OpFunction %void None %voidf\n"
3979 "%label = OpLabel\n"
3980 "%idval = OpLoad %uvec3 %id\n"
3981 "%x = OpCompositeExtract %u32 %idval 0\n"
3982 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3983 "%inval = OpLoad %f32 %inloc\n"
3984 "%neg = OpFNegate %f32 %inval\n"
3985 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3986 " OpStore %outloc %neg\n"
3988 " OpFunctionEnd\n");
3990 cases.push_back(CaseParameter("empty_extension", ""));
3991 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
3992 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
3993 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
3994 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
3996 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
3998 for (size_t ndx = 0; ndx < numElements; ++ndx)
3999 outputFloats[ndx] = -inputFloats[ndx];
4001 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4003 map<string, string> specializations;
4004 ComputeShaderSpec spec;
4006 specializations["EXTENSION"] = cases[caseNdx].param;
4007 spec.assembly = shaderTemplate.specialize(specializations);
4008 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4009 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4010 spec.numWorkGroups = IVec3(numElements, 1, 1);
4012 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4015 return group.release();
4018 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
4019 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
4021 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
4022 vector<CaseParameter> cases;
4023 de::Random rnd (deStringHash(group->getName()));
4024 const int numElements = 100;
4025 vector<float> positiveFloats (numElements, 0);
4026 vector<float> negativeFloats (numElements, 0);
4027 const StringTemplate shaderTemplate (
4028 string(getComputeAsmShaderPreamble()) +
4030 "OpSource GLSL 430\n"
4031 "OpName %main \"main\"\n"
4032 "OpName %id \"gl_GlobalInvocationID\"\n"
4034 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4036 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4037 "%uvec2 = OpTypeVector %u32 2\n"
4038 "%bvec3 = OpTypeVector %bool 3\n"
4039 "%fvec4 = OpTypeVector %f32 4\n"
4040 "%fmat33 = OpTypeMatrix %fvec3 3\n"
4041 "%const100 = OpConstant %u32 100\n"
4042 "%uarr100 = OpTypeArray %i32 %const100\n"
4043 "%struct = OpTypeStruct %f32 %i32 %u32\n"
4044 "%pointer = OpTypePointer Function %i32\n"
4045 + string(getComputeAsmInputOutputBuffer()) +
4047 "%null = OpConstantNull ${TYPE}\n"
4049 "%id = OpVariable %uvec3ptr Input\n"
4050 "%zero = OpConstant %i32 0\n"
4052 "%main = OpFunction %void None %voidf\n"
4053 "%label = OpLabel\n"
4054 "%idval = OpLoad %uvec3 %id\n"
4055 "%x = OpCompositeExtract %u32 %idval 0\n"
4056 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4057 "%inval = OpLoad %f32 %inloc\n"
4058 "%neg = OpFNegate %f32 %inval\n"
4059 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4060 " OpStore %outloc %neg\n"
4062 " OpFunctionEnd\n");
4064 cases.push_back(CaseParameter("bool", "%bool"));
4065 cases.push_back(CaseParameter("sint32", "%i32"));
4066 cases.push_back(CaseParameter("uint32", "%u32"));
4067 cases.push_back(CaseParameter("float32", "%f32"));
4068 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
4069 cases.push_back(CaseParameter("vec3bool", "%bvec3"));
4070 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
4071 cases.push_back(CaseParameter("matrix", "%fmat33"));
4072 cases.push_back(CaseParameter("array", "%uarr100"));
4073 cases.push_back(CaseParameter("struct", "%struct"));
4074 cases.push_back(CaseParameter("pointer", "%pointer"));
4076 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4078 for (size_t ndx = 0; ndx < numElements; ++ndx)
4079 negativeFloats[ndx] = -positiveFloats[ndx];
4081 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4083 map<string, string> specializations;
4084 ComputeShaderSpec spec;
4086 specializations["TYPE"] = cases[caseNdx].param;
4087 spec.assembly = shaderTemplate.specialize(specializations);
4088 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4089 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4090 spec.numWorkGroups = IVec3(numElements, 1, 1);
4092 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4095 return group.release();
4098 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
4099 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
4101 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
4102 vector<CaseParameter> cases;
4103 de::Random rnd (deStringHash(group->getName()));
4104 const int numElements = 100;
4105 vector<float> positiveFloats (numElements, 0);
4106 vector<float> negativeFloats (numElements, 0);
4107 const StringTemplate shaderTemplate (
4108 string(getComputeAsmShaderPreamble()) +
4110 "OpSource GLSL 430\n"
4111 "OpName %main \"main\"\n"
4112 "OpName %id \"gl_GlobalInvocationID\"\n"
4114 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4116 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4118 "%id = OpVariable %uvec3ptr Input\n"
4119 "%zero = OpConstant %i32 0\n"
4123 "%main = OpFunction %void None %voidf\n"
4124 "%label = OpLabel\n"
4125 "%idval = OpLoad %uvec3 %id\n"
4126 "%x = OpCompositeExtract %u32 %idval 0\n"
4127 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4128 "%inval = OpLoad %f32 %inloc\n"
4129 "%neg = OpFNegate %f32 %inval\n"
4130 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4131 " OpStore %outloc %neg\n"
4133 " OpFunctionEnd\n");
4135 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
4136 "%const = OpConstantComposite %uvec3 %five %zero %five"));
4137 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
4138 "%ten = OpConstant %f32 10.\n"
4139 "%fzero = OpConstant %f32 0.\n"
4140 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
4141 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
4142 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
4143 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
4144 "%fzero = OpConstant %f32 0.\n"
4145 "%one = OpConstant %f32 1.\n"
4146 "%point5 = OpConstant %f32 0.5\n"
4147 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
4148 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
4149 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
4150 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
4151 "%st2 = OpTypeStruct %i32 %i32\n"
4152 "%struct = OpTypeStruct %st1 %st2\n"
4153 "%point5 = OpConstant %f32 0.5\n"
4154 "%one = OpConstant %u32 1\n"
4155 "%ten = OpConstant %i32 10\n"
4156 "%st1val = OpConstantComposite %st1 %one %point5\n"
4157 "%st2val = OpConstantComposite %st2 %ten %ten\n"
4158 "%const = OpConstantComposite %struct %st1val %st2val"));
4160 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4162 for (size_t ndx = 0; ndx < numElements; ++ndx)
4163 negativeFloats[ndx] = -positiveFloats[ndx];
4165 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4167 map<string, string> specializations;
4168 ComputeShaderSpec spec;
4170 specializations["CONSTANT"] = cases[caseNdx].param;
4171 spec.assembly = shaderTemplate.specialize(specializations);
4172 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4173 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4174 spec.numWorkGroups = IVec3(numElements, 1, 1);
4176 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4179 return group.release();
4182 // Creates a floating point number with the given exponent, and significand
4183 // bits set. It can only create normalized numbers. Only the least significant
4184 // 24 bits of the significand will be examined. The final bit of the
4185 // significand will also be ignored. This allows alignment to be written
4186 // similarly to C99 hex-floats.
4187 // For example if you wanted to write 0x1.7f34p-12 you would call
4188 // constructNormalizedFloat(-12, 0x7f3400)
4189 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
4193 for (deInt32 idx = 0; idx < 23; ++idx)
4195 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
4199 return std::ldexp(f, exponent);
4202 // Compare instruction for the OpQuantizeF16 compute exact case.
4203 // Returns true if the output is what is expected from the test case.
4204 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
4206 if (outputAllocs.size() != 1)
4209 // Only size is needed because we cannot compare Nans.
4210 size_t byteSize = expectedOutputs[0]->getByteSize();
4212 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
4214 if (byteSize != 4*sizeof(float)) {
4218 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
4219 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
4224 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
4225 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
4230 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
4231 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
4236 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
4237 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
4244 // Checks that every output from a test-case is a float NaN.
4245 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
4247 if (outputAllocs.size() != 1)
4250 // Only size is needed because we cannot compare Nans.
4251 size_t byteSize = expectedOutputs[0]->getByteSize();
4253 const float* const output_as_float = static_cast<const float* const>(outputAllocs[0]->getHostPtr());
4255 for (size_t idx = 0; idx < byteSize / sizeof(float); ++idx)
4257 if (!deFloatIsNaN(output_as_float[idx]))
4266 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
4267 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
4269 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
4271 const std::string shader (
4272 string(getComputeAsmShaderPreamble()) +
4274 "OpSource GLSL 430\n"
4275 "OpName %main \"main\"\n"
4276 "OpName %id \"gl_GlobalInvocationID\"\n"
4278 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4280 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4282 "%id = OpVariable %uvec3ptr Input\n"
4283 "%zero = OpConstant %i32 0\n"
4285 "%main = OpFunction %void None %voidf\n"
4286 "%label = OpLabel\n"
4287 "%idval = OpLoad %uvec3 %id\n"
4288 "%x = OpCompositeExtract %u32 %idval 0\n"
4289 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4290 "%inval = OpLoad %f32 %inloc\n"
4291 "%quant = OpQuantizeToF16 %f32 %inval\n"
4292 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4293 " OpStore %outloc %quant\n"
4295 " OpFunctionEnd\n");
4298 ComputeShaderSpec spec;
4299 const deUint32 numElements = 100;
4300 vector<float> infinities;
4301 vector<float> results;
4303 infinities.reserve(numElements);
4304 results.reserve(numElements);
4306 for (size_t idx = 0; idx < numElements; ++idx)
4311 infinities.push_back(std::numeric_limits<float>::infinity());
4312 results.push_back(std::numeric_limits<float>::infinity());
4315 infinities.push_back(-std::numeric_limits<float>::infinity());
4316 results.push_back(-std::numeric_limits<float>::infinity());
4319 infinities.push_back(std::ldexp(1.0f, 16));
4320 results.push_back(std::numeric_limits<float>::infinity());
4323 infinities.push_back(std::ldexp(-1.0f, 32));
4324 results.push_back(-std::numeric_limits<float>::infinity());
4329 spec.assembly = shader;
4330 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
4331 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
4332 spec.numWorkGroups = IVec3(numElements, 1, 1);
4334 group->addChild(new SpvAsmComputeShaderCase(
4335 testCtx, "infinities", "Check that infinities propagated and created", spec));
4339 ComputeShaderSpec spec;
4341 const deUint32 numElements = 100;
4343 nans.reserve(numElements);
4345 for (size_t idx = 0; idx < numElements; ++idx)
4349 nans.push_back(std::numeric_limits<float>::quiet_NaN());
4353 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
4357 spec.assembly = shader;
4358 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
4359 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
4360 spec.numWorkGroups = IVec3(numElements, 1, 1);
4361 spec.verifyIO = &compareNan;
4363 group->addChild(new SpvAsmComputeShaderCase(
4364 testCtx, "propagated_nans", "Check that nans are propagated", spec));
4368 ComputeShaderSpec spec;
4369 vector<float> small;
4370 vector<float> zeros;
4371 const deUint32 numElements = 100;
4373 small.reserve(numElements);
4374 zeros.reserve(numElements);
4376 for (size_t idx = 0; idx < numElements; ++idx)
4381 small.push_back(0.f);
4382 zeros.push_back(0.f);
4385 small.push_back(-0.f);
4386 zeros.push_back(-0.f);
4389 small.push_back(std::ldexp(1.0f, -16));
4390 zeros.push_back(0.f);
4393 small.push_back(std::ldexp(-1.0f, -32));
4394 zeros.push_back(-0.f);
4397 small.push_back(std::ldexp(1.0f, -127));
4398 zeros.push_back(0.f);
4401 small.push_back(-std::ldexp(1.0f, -128));
4402 zeros.push_back(-0.f);
4407 spec.assembly = shader;
4408 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
4409 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
4410 spec.numWorkGroups = IVec3(numElements, 1, 1);
4412 group->addChild(new SpvAsmComputeShaderCase(
4413 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
4417 ComputeShaderSpec spec;
4418 vector<float> exact;
4419 const deUint32 numElements = 200;
4421 exact.reserve(numElements);
4423 for (size_t idx = 0; idx < numElements; ++idx)
4424 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
4426 spec.assembly = shader;
4427 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
4428 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
4429 spec.numWorkGroups = IVec3(numElements, 1, 1);
4431 group->addChild(new SpvAsmComputeShaderCase(
4432 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
4436 ComputeShaderSpec spec;
4437 vector<float> inputs;
4438 const deUint32 numElements = 4;
4440 inputs.push_back(constructNormalizedFloat(8, 0x300300));
4441 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
4442 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
4443 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
4445 spec.assembly = shader;
4446 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
4447 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4448 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
4449 spec.numWorkGroups = IVec3(numElements, 1, 1);
4451 group->addChild(new SpvAsmComputeShaderCase(
4452 testCtx, "rounded", "Check that are rounded when needed", spec));
4455 return group.release();
4458 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
4460 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
4462 const std::string shader (
4463 string(getComputeAsmShaderPreamble()) +
4465 "OpName %main \"main\"\n"
4466 "OpName %id \"gl_GlobalInvocationID\"\n"
4468 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4470 "OpDecorate %sc_0 SpecId 0\n"
4471 "OpDecorate %sc_1 SpecId 1\n"
4472 "OpDecorate %sc_2 SpecId 2\n"
4473 "OpDecorate %sc_3 SpecId 3\n"
4474 "OpDecorate %sc_4 SpecId 4\n"
4475 "OpDecorate %sc_5 SpecId 5\n"
4477 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4479 "%id = OpVariable %uvec3ptr Input\n"
4480 "%zero = OpConstant %i32 0\n"
4481 "%c_u32_6 = OpConstant %u32 6\n"
4483 "%sc_0 = OpSpecConstant %f32 0.\n"
4484 "%sc_1 = OpSpecConstant %f32 0.\n"
4485 "%sc_2 = OpSpecConstant %f32 0.\n"
4486 "%sc_3 = OpSpecConstant %f32 0.\n"
4487 "%sc_4 = OpSpecConstant %f32 0.\n"
4488 "%sc_5 = OpSpecConstant %f32 0.\n"
4490 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
4491 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
4492 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
4493 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
4494 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
4495 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
4497 "%main = OpFunction %void None %voidf\n"
4498 "%label = OpLabel\n"
4499 "%idval = OpLoad %uvec3 %id\n"
4500 "%x = OpCompositeExtract %u32 %idval 0\n"
4501 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4502 "%selector = OpUMod %u32 %x %c_u32_6\n"
4503 " OpSelectionMerge %exit None\n"
4504 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
4506 "%case0 = OpLabel\n"
4507 " OpStore %outloc %sc_0_quant\n"
4510 "%case1 = OpLabel\n"
4511 " OpStore %outloc %sc_1_quant\n"
4514 "%case2 = OpLabel\n"
4515 " OpStore %outloc %sc_2_quant\n"
4518 "%case3 = OpLabel\n"
4519 " OpStore %outloc %sc_3_quant\n"
4522 "%case4 = OpLabel\n"
4523 " OpStore %outloc %sc_4_quant\n"
4526 "%case5 = OpLabel\n"
4527 " OpStore %outloc %sc_5_quant\n"
4533 " OpFunctionEnd\n");
4536 ComputeShaderSpec spec;
4537 const deUint8 numCases = 4;
4538 vector<float> inputs (numCases, 0.f);
4539 vector<float> outputs;
4541 spec.assembly = shader;
4542 spec.numWorkGroups = IVec3(numCases, 1, 1);
4544 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
4545 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
4546 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
4547 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
4549 outputs.push_back(std::numeric_limits<float>::infinity());
4550 outputs.push_back(-std::numeric_limits<float>::infinity());
4551 outputs.push_back(std::numeric_limits<float>::infinity());
4552 outputs.push_back(-std::numeric_limits<float>::infinity());
4554 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4555 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4557 group->addChild(new SpvAsmComputeShaderCase(
4558 testCtx, "infinities", "Check that infinities propagated and created", spec));
4562 ComputeShaderSpec spec;
4563 const deUint8 numCases = 2;
4564 vector<float> inputs (numCases, 0.f);
4565 vector<float> outputs;
4567 spec.assembly = shader;
4568 spec.numWorkGroups = IVec3(numCases, 1, 1);
4569 spec.verifyIO = &compareNan;
4571 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
4572 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
4574 for (deUint8 idx = 0; idx < numCases; ++idx)
4575 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
4577 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4578 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4580 group->addChild(new SpvAsmComputeShaderCase(
4581 testCtx, "propagated_nans", "Check that nans are propagated", spec));
4585 ComputeShaderSpec spec;
4586 const deUint8 numCases = 6;
4587 vector<float> inputs (numCases, 0.f);
4588 vector<float> outputs;
4590 spec.assembly = shader;
4591 spec.numWorkGroups = IVec3(numCases, 1, 1);
4593 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
4594 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
4595 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
4596 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
4597 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
4598 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
4600 outputs.push_back(0.f);
4601 outputs.push_back(-0.f);
4602 outputs.push_back(0.f);
4603 outputs.push_back(-0.f);
4604 outputs.push_back(0.f);
4605 outputs.push_back(-0.f);
4607 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4608 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4610 group->addChild(new SpvAsmComputeShaderCase(
4611 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
4615 ComputeShaderSpec spec;
4616 const deUint8 numCases = 6;
4617 vector<float> inputs (numCases, 0.f);
4618 vector<float> outputs;
4620 spec.assembly = shader;
4621 spec.numWorkGroups = IVec3(numCases, 1, 1);
4623 for (deUint8 idx = 0; idx < 6; ++idx)
4625 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
4626 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
4627 outputs.push_back(f);
4630 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4631 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4633 group->addChild(new SpvAsmComputeShaderCase(
4634 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
4638 ComputeShaderSpec spec;
4639 const deUint8 numCases = 4;
4640 vector<float> inputs (numCases, 0.f);
4641 vector<float> outputs;
4643 spec.assembly = shader;
4644 spec.numWorkGroups = IVec3(numCases, 1, 1);
4645 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
4647 outputs.push_back(constructNormalizedFloat(8, 0x300300));
4648 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
4649 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
4650 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
4652 for (deUint8 idx = 0; idx < numCases; ++idx)
4653 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
4655 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4656 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4658 group->addChild(new SpvAsmComputeShaderCase(
4659 testCtx, "rounded", "Check that are rounded when needed", spec));
4662 return group.release();
4665 // Checks that constant null/composite values can be used in computation.
4666 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
4668 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
4669 ComputeShaderSpec spec;
4670 de::Random rnd (deStringHash(group->getName()));
4671 const int numElements = 100;
4672 vector<float> positiveFloats (numElements, 0);
4673 vector<float> negativeFloats (numElements, 0);
4675 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4677 for (size_t ndx = 0; ndx < numElements; ++ndx)
4678 negativeFloats[ndx] = -positiveFloats[ndx];
4681 "OpCapability Shader\n"
4682 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
4683 "OpMemoryModel Logical GLSL450\n"
4684 "OpEntryPoint GLCompute %main \"main\" %id\n"
4685 "OpExecutionMode %main LocalSize 1 1 1\n"
4687 "OpSource GLSL 430\n"
4688 "OpName %main \"main\"\n"
4689 "OpName %id \"gl_GlobalInvocationID\"\n"
4691 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4693 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4695 "%fmat = OpTypeMatrix %fvec3 3\n"
4696 "%ten = OpConstant %u32 10\n"
4697 "%f32arr10 = OpTypeArray %f32 %ten\n"
4698 "%fst = OpTypeStruct %f32 %f32\n"
4700 + string(getComputeAsmInputOutputBuffer()) +
4702 "%id = OpVariable %uvec3ptr Input\n"
4703 "%zero = OpConstant %i32 0\n"
4705 // Create a bunch of null values
4706 "%unull = OpConstantNull %u32\n"
4707 "%fnull = OpConstantNull %f32\n"
4708 "%vnull = OpConstantNull %fvec3\n"
4709 "%mnull = OpConstantNull %fmat\n"
4710 "%anull = OpConstantNull %f32arr10\n"
4711 "%snull = OpConstantComposite %fst %fnull %fnull\n"
4713 "%main = OpFunction %void None %voidf\n"
4714 "%label = OpLabel\n"
4715 "%idval = OpLoad %uvec3 %id\n"
4716 "%x = OpCompositeExtract %u32 %idval 0\n"
4717 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4718 "%inval = OpLoad %f32 %inloc\n"
4719 "%neg = OpFNegate %f32 %inval\n"
4721 // Get the abs() of (a certain element of) those null values
4722 "%unull_cov = OpConvertUToF %f32 %unull\n"
4723 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
4724 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
4725 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
4726 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
4727 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
4728 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
4729 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
4730 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
4731 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
4732 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
4735 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
4736 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
4737 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
4738 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
4739 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
4740 "%final = OpFAdd %f32 %add5 %snull_abs\n"
4742 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4743 " OpStore %outloc %final\n" // write to output
4746 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4747 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4748 spec.numWorkGroups = IVec3(numElements, 1, 1);
4750 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
4752 return group.release();
4755 // Assembly code used for testing loop control is based on GLSL source code:
4758 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4759 // float elements[];
4761 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4762 // float elements[];
4766 // uint x = gl_GlobalInvocationID.x;
4767 // output_data.elements[x] = input_data.elements[x];
4768 // for (uint i = 0; i < 4; ++i)
4769 // output_data.elements[x] += 1.f;
4771 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
4773 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
4774 vector<CaseParameter> cases;
4775 de::Random rnd (deStringHash(group->getName()));
4776 const int numElements = 100;
4777 vector<float> inputFloats (numElements, 0);
4778 vector<float> outputFloats (numElements, 0);
4779 const StringTemplate shaderTemplate (
4780 string(getComputeAsmShaderPreamble()) +
4782 "OpSource GLSL 430\n"
4783 "OpName %main \"main\"\n"
4784 "OpName %id \"gl_GlobalInvocationID\"\n"
4786 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4788 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4790 "%u32ptr = OpTypePointer Function %u32\n"
4792 "%id = OpVariable %uvec3ptr Input\n"
4793 "%zero = OpConstant %i32 0\n"
4794 "%uzero = OpConstant %u32 0\n"
4795 "%one = OpConstant %i32 1\n"
4796 "%constf1 = OpConstant %f32 1.0\n"
4797 "%four = OpConstant %u32 4\n"
4799 "%main = OpFunction %void None %voidf\n"
4800 "%entry = OpLabel\n"
4801 "%i = OpVariable %u32ptr Function\n"
4802 " OpStore %i %uzero\n"
4804 "%idval = OpLoad %uvec3 %id\n"
4805 "%x = OpCompositeExtract %u32 %idval 0\n"
4806 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4807 "%inval = OpLoad %f32 %inloc\n"
4808 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4809 " OpStore %outloc %inval\n"
4810 " OpBranch %loop_entry\n"
4812 "%loop_entry = OpLabel\n"
4813 "%i_val = OpLoad %u32 %i\n"
4814 "%cmp_lt = OpULessThan %bool %i_val %four\n"
4815 " OpLoopMerge %loop_merge %loop_body ${CONTROL}\n"
4816 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
4817 "%loop_body = OpLabel\n"
4818 "%outval = OpLoad %f32 %outloc\n"
4819 "%addf1 = OpFAdd %f32 %outval %constf1\n"
4820 " OpStore %outloc %addf1\n"
4821 "%new_i = OpIAdd %u32 %i_val %one\n"
4822 " OpStore %i %new_i\n"
4823 " OpBranch %loop_entry\n"
4824 "%loop_merge = OpLabel\n"
4826 " OpFunctionEnd\n");
4828 cases.push_back(CaseParameter("none", "None"));
4829 cases.push_back(CaseParameter("unroll", "Unroll"));
4830 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
4831 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
4833 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4835 for (size_t ndx = 0; ndx < numElements; ++ndx)
4836 outputFloats[ndx] = inputFloats[ndx] + 4.f;
4838 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4840 map<string, string> specializations;
4841 ComputeShaderSpec spec;
4843 specializations["CONTROL"] = cases[caseNdx].param;
4844 spec.assembly = shaderTemplate.specialize(specializations);
4845 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4846 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4847 spec.numWorkGroups = IVec3(numElements, 1, 1);
4849 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4852 return group.release();
4855 // Assembly code used for testing selection control is based on GLSL source code:
4858 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4859 // float elements[];
4861 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4862 // float elements[];
4866 // uint x = gl_GlobalInvocationID.x;
4867 // float val = input_data.elements[x];
4869 // output_data.elements[x] = val + 1.f;
4871 // output_data.elements[x] = val - 1.f;
4873 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
4875 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
4876 vector<CaseParameter> cases;
4877 de::Random rnd (deStringHash(group->getName()));
4878 const int numElements = 100;
4879 vector<float> inputFloats (numElements, 0);
4880 vector<float> outputFloats (numElements, 0);
4881 const StringTemplate shaderTemplate (
4882 string(getComputeAsmShaderPreamble()) +
4884 "OpSource GLSL 430\n"
4885 "OpName %main \"main\"\n"
4886 "OpName %id \"gl_GlobalInvocationID\"\n"
4888 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4890 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4892 "%id = OpVariable %uvec3ptr Input\n"
4893 "%zero = OpConstant %i32 0\n"
4894 "%constf1 = OpConstant %f32 1.0\n"
4895 "%constf10 = OpConstant %f32 10.0\n"
4897 "%main = OpFunction %void None %voidf\n"
4898 "%entry = OpLabel\n"
4899 "%idval = OpLoad %uvec3 %id\n"
4900 "%x = OpCompositeExtract %u32 %idval 0\n"
4901 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4902 "%inval = OpLoad %f32 %inloc\n"
4903 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4904 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
4906 " OpSelectionMerge %if_end ${CONTROL}\n"
4907 " OpBranchConditional %cmp_gt %if_true %if_false\n"
4908 "%if_true = OpLabel\n"
4909 "%addf1 = OpFAdd %f32 %inval %constf1\n"
4910 " OpStore %outloc %addf1\n"
4911 " OpBranch %if_end\n"
4912 "%if_false = OpLabel\n"
4913 "%subf1 = OpFSub %f32 %inval %constf1\n"
4914 " OpStore %outloc %subf1\n"
4915 " OpBranch %if_end\n"
4916 "%if_end = OpLabel\n"
4918 " OpFunctionEnd\n");
4920 cases.push_back(CaseParameter("none", "None"));
4921 cases.push_back(CaseParameter("flatten", "Flatten"));
4922 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
4923 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
4925 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4927 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4928 floorAll(inputFloats);
4930 for (size_t ndx = 0; ndx < numElements; ++ndx)
4931 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
4933 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4935 map<string, string> specializations;
4936 ComputeShaderSpec spec;
4938 specializations["CONTROL"] = cases[caseNdx].param;
4939 spec.assembly = shaderTemplate.specialize(specializations);
4940 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4941 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4942 spec.numWorkGroups = IVec3(numElements, 1, 1);
4944 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4947 return group.release();
4950 // Assembly code used for testing function control is based on GLSL source code:
4954 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4955 // float elements[];
4957 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4958 // float elements[];
4961 // float const10() { return 10.f; }
4964 // uint x = gl_GlobalInvocationID.x;
4965 // output_data.elements[x] = input_data.elements[x] + const10();
4967 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
4969 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
4970 vector<CaseParameter> cases;
4971 de::Random rnd (deStringHash(group->getName()));
4972 const int numElements = 100;
4973 vector<float> inputFloats (numElements, 0);
4974 vector<float> outputFloats (numElements, 0);
4975 const StringTemplate shaderTemplate (
4976 string(getComputeAsmShaderPreamble()) +
4978 "OpSource GLSL 430\n"
4979 "OpName %main \"main\"\n"
4980 "OpName %func_const10 \"const10(\"\n"
4981 "OpName %id \"gl_GlobalInvocationID\"\n"
4983 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4985 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4987 "%f32f = OpTypeFunction %f32\n"
4988 "%id = OpVariable %uvec3ptr Input\n"
4989 "%zero = OpConstant %i32 0\n"
4990 "%constf10 = OpConstant %f32 10.0\n"
4992 "%main = OpFunction %void None %voidf\n"
4993 "%entry = OpLabel\n"
4994 "%idval = OpLoad %uvec3 %id\n"
4995 "%x = OpCompositeExtract %u32 %idval 0\n"
4996 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4997 "%inval = OpLoad %f32 %inloc\n"
4998 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
4999 "%fadd = OpFAdd %f32 %inval %ret_10\n"
5000 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5001 " OpStore %outloc %fadd\n"
5005 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
5006 "%label = OpLabel\n"
5007 " OpReturnValue %constf10\n"
5008 " OpFunctionEnd\n");
5010 cases.push_back(CaseParameter("none", "None"));
5011 cases.push_back(CaseParameter("inline", "Inline"));
5012 cases.push_back(CaseParameter("dont_inline", "DontInline"));
5013 cases.push_back(CaseParameter("pure", "Pure"));
5014 cases.push_back(CaseParameter("const", "Const"));
5015 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
5016 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
5017 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
5018 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
5020 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5022 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
5023 floorAll(inputFloats);
5025 for (size_t ndx = 0; ndx < numElements; ++ndx)
5026 outputFloats[ndx] = inputFloats[ndx] + 10.f;
5028 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5030 map<string, string> specializations;
5031 ComputeShaderSpec spec;
5033 specializations["CONTROL"] = cases[caseNdx].param;
5034 spec.assembly = shaderTemplate.specialize(specializations);
5035 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5036 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5037 spec.numWorkGroups = IVec3(numElements, 1, 1);
5039 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5042 return group.release();
5045 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
5047 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
5048 vector<CaseParameter> cases;
5049 de::Random rnd (deStringHash(group->getName()));
5050 const int numElements = 100;
5051 vector<float> inputFloats (numElements, 0);
5052 vector<float> outputFloats (numElements, 0);
5053 const StringTemplate shaderTemplate (
5054 string(getComputeAsmShaderPreamble()) +
5056 "OpSource GLSL 430\n"
5057 "OpName %main \"main\"\n"
5058 "OpName %id \"gl_GlobalInvocationID\"\n"
5060 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5062 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5064 "%f32ptr_f = OpTypePointer Function %f32\n"
5066 "%id = OpVariable %uvec3ptr Input\n"
5067 "%zero = OpConstant %i32 0\n"
5068 "%four = OpConstant %i32 4\n"
5070 "%main = OpFunction %void None %voidf\n"
5071 "%label = OpLabel\n"
5072 "%copy = OpVariable %f32ptr_f Function\n"
5073 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
5074 "%x = OpCompositeExtract %u32 %idval 0\n"
5075 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5076 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5077 " OpCopyMemory %copy %inloc ${ACCESS}\n"
5078 "%val1 = OpLoad %f32 %copy\n"
5079 "%val2 = OpLoad %f32 %inloc\n"
5080 "%add = OpFAdd %f32 %val1 %val2\n"
5081 " OpStore %outloc %add ${ACCESS}\n"
5083 " OpFunctionEnd\n");
5085 cases.push_back(CaseParameter("null", ""));
5086 cases.push_back(CaseParameter("none", "None"));
5087 cases.push_back(CaseParameter("volatile", "Volatile"));
5088 cases.push_back(CaseParameter("aligned", "Aligned 4"));
5089 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
5090 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
5091 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
5093 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5095 for (size_t ndx = 0; ndx < numElements; ++ndx)
5096 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
5098 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5100 map<string, string> specializations;
5101 ComputeShaderSpec spec;
5103 specializations["ACCESS"] = cases[caseNdx].param;
5104 spec.assembly = shaderTemplate.specialize(specializations);
5105 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5106 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5107 spec.numWorkGroups = IVec3(numElements, 1, 1);
5109 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5112 return group.release();
5115 // Checks that we can get undefined values for various types, without exercising a computation with it.
5116 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
5118 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
5119 vector<CaseParameter> cases;
5120 de::Random rnd (deStringHash(group->getName()));
5121 const int numElements = 100;
5122 vector<float> positiveFloats (numElements, 0);
5123 vector<float> negativeFloats (numElements, 0);
5124 const StringTemplate shaderTemplate (
5125 string(getComputeAsmShaderPreamble()) +
5127 "OpSource GLSL 430\n"
5128 "OpName %main \"main\"\n"
5129 "OpName %id \"gl_GlobalInvocationID\"\n"
5131 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5133 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
5134 "%uvec2 = OpTypeVector %u32 2\n"
5135 "%fvec4 = OpTypeVector %f32 4\n"
5136 "%fmat33 = OpTypeMatrix %fvec3 3\n"
5137 "%image = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
5138 "%sampler = OpTypeSampler\n"
5139 "%simage = OpTypeSampledImage %image\n"
5140 "%const100 = OpConstant %u32 100\n"
5141 "%uarr100 = OpTypeArray %i32 %const100\n"
5142 "%struct = OpTypeStruct %f32 %i32 %u32\n"
5143 "%pointer = OpTypePointer Function %i32\n"
5144 + string(getComputeAsmInputOutputBuffer()) +
5146 "%id = OpVariable %uvec3ptr Input\n"
5147 "%zero = OpConstant %i32 0\n"
5149 "%main = OpFunction %void None %voidf\n"
5150 "%label = OpLabel\n"
5152 "%undef = OpUndef ${TYPE}\n"
5154 "%idval = OpLoad %uvec3 %id\n"
5155 "%x = OpCompositeExtract %u32 %idval 0\n"
5157 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5158 "%inval = OpLoad %f32 %inloc\n"
5159 "%neg = OpFNegate %f32 %inval\n"
5160 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5161 " OpStore %outloc %neg\n"
5163 " OpFunctionEnd\n");
5165 cases.push_back(CaseParameter("bool", "%bool"));
5166 cases.push_back(CaseParameter("sint32", "%i32"));
5167 cases.push_back(CaseParameter("uint32", "%u32"));
5168 cases.push_back(CaseParameter("float32", "%f32"));
5169 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
5170 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
5171 cases.push_back(CaseParameter("matrix", "%fmat33"));
5172 cases.push_back(CaseParameter("image", "%image"));
5173 cases.push_back(CaseParameter("sampler", "%sampler"));
5174 cases.push_back(CaseParameter("sampledimage", "%simage"));
5175 cases.push_back(CaseParameter("array", "%uarr100"));
5176 cases.push_back(CaseParameter("runtimearray", "%f32arr"));
5177 cases.push_back(CaseParameter("struct", "%struct"));
5178 cases.push_back(CaseParameter("pointer", "%pointer"));
5180 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
5182 for (size_t ndx = 0; ndx < numElements; ++ndx)
5183 negativeFloats[ndx] = -positiveFloats[ndx];
5185 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5187 map<string, string> specializations;
5188 ComputeShaderSpec spec;
5190 specializations["TYPE"] = cases[caseNdx].param;
5191 spec.assembly = shaderTemplate.specialize(specializations);
5192 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
5193 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
5194 spec.numWorkGroups = IVec3(numElements, 1, 1);
5196 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5199 return group.release();
5204 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
5206 struct NameCodePair { string name, code; };
5207 RGBA defaultColors[4];
5208 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
5209 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
5210 map<string, string> fragments = passthruFragments();
5211 const NameCodePair tests[] =
5213 {"unknown", "OpSource Unknown 321"},
5214 {"essl", "OpSource ESSL 310"},
5215 {"glsl", "OpSource GLSL 450"},
5216 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
5217 {"opencl_c", "OpSource OpenCL_C 120"},
5218 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
5219 {"file", opsourceGLSLWithFile},
5220 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
5221 // Longest possible source string: SPIR-V limits instructions to 65535
5222 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
5223 // contain 65530 UTF8 characters (one word each) plus one last word
5224 // containing 3 ASCII characters and \0.
5225 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
5228 getDefaultColors(defaultColors);
5229 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5231 fragments["debug"] = tests[testNdx].code;
5232 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5235 return opSourceTests.release();
5238 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
5240 struct NameCodePair { string name, code; };
5241 RGBA defaultColors[4];
5242 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
5243 map<string, string> fragments = passthruFragments();
5244 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
5245 const NameCodePair tests[] =
5247 {"empty", opsource + "OpSourceContinued \"\""},
5248 {"short", opsource + "OpSourceContinued \"abcde\""},
5249 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
5250 // Longest possible source string: SPIR-V limits instructions to 65535
5251 // words, of which the first one is OpSourceContinued/length; the rest
5252 // will contain 65533 UTF8 characters (one word each) plus one last word
5253 // containing 3 ASCII characters and \0.
5254 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
5257 getDefaultColors(defaultColors);
5258 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5260 fragments["debug"] = tests[testNdx].code;
5261 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5264 return opSourceTests.release();
5267 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
5269 RGBA defaultColors[4];
5270 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
5271 map<string, string> fragments;
5272 getDefaultColors(defaultColors);
5273 fragments["debug"] =
5274 "%name = OpString \"name\"\n";
5276 fragments["pre_main"] =
5279 "OpLine %name 1 1\n"
5281 "OpLine %name 1 1\n"
5282 "OpLine %name 1 1\n"
5283 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5285 "OpLine %name 1 1\n"
5287 "OpLine %name 1 1\n"
5288 "OpLine %name 1 1\n"
5289 "%second_param1 = OpFunctionParameter %v4f32\n"
5292 "%label_secondfunction = OpLabel\n"
5294 "OpReturnValue %second_param1\n"
5299 fragments["testfun"] =
5300 // A %test_code function that returns its argument unchanged.
5303 "OpLine %name 1 1\n"
5304 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5306 "%param1 = OpFunctionParameter %v4f32\n"
5309 "%label_testfun = OpLabel\n"
5311 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5312 "OpReturnValue %val1\n"
5314 "OpLine %name 1 1\n"
5317 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
5319 return opLineTests.release();
5323 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
5325 RGBA defaultColors[4];
5326 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
5327 map<string, string> fragments;
5328 std::vector<std::pair<std::string, std::string> > problemStrings;
5330 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
5331 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
5332 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
5333 getDefaultColors(defaultColors);
5335 fragments["debug"] =
5336 "%other_name = OpString \"other_name\"\n";
5338 fragments["pre_main"] =
5339 "OpLine %file_name 32 0\n"
5340 "OpLine %file_name 32 32\n"
5341 "OpLine %file_name 32 40\n"
5342 "OpLine %other_name 32 40\n"
5343 "OpLine %other_name 0 100\n"
5344 "OpLine %other_name 0 4294967295\n"
5345 "OpLine %other_name 4294967295 0\n"
5346 "OpLine %other_name 32 40\n"
5347 "OpLine %file_name 0 0\n"
5348 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5349 "OpLine %file_name 1 0\n"
5350 "%second_param1 = OpFunctionParameter %v4f32\n"
5351 "OpLine %file_name 1 3\n"
5352 "OpLine %file_name 1 2\n"
5353 "%label_secondfunction = OpLabel\n"
5354 "OpLine %file_name 0 2\n"
5355 "OpReturnValue %second_param1\n"
5357 "OpLine %file_name 0 2\n"
5358 "OpLine %file_name 0 2\n";
5360 fragments["testfun"] =
5361 // A %test_code function that returns its argument unchanged.
5362 "OpLine %file_name 1 0\n"
5363 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5364 "OpLine %file_name 16 330\n"
5365 "%param1 = OpFunctionParameter %v4f32\n"
5366 "OpLine %file_name 14 442\n"
5367 "%label_testfun = OpLabel\n"
5368 "OpLine %file_name 11 1024\n"
5369 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5370 "OpLine %file_name 2 97\n"
5371 "OpReturnValue %val1\n"
5373 "OpLine %file_name 5 32\n";
5375 for (size_t i = 0; i < problemStrings.size(); ++i)
5377 map<string, string> testFragments = fragments;
5378 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
5379 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
5382 return opLineTests.release();
5385 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
5387 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
5391 const char functionStart[] =
5392 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5393 "%param1 = OpFunctionParameter %v4f32\n"
5396 const char functionEnd[] =
5397 "OpReturnValue %transformed_param\n"
5400 struct NameConstantsCode
5407 NameConstantsCode tests[] =
5411 "%cnull = OpConstantNull %v4f32\n",
5412 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
5416 "%cnull = OpConstantNull %f32\n",
5417 "%vp = OpVariable %fp_v4f32 Function\n"
5418 "%v = OpLoad %v4f32 %vp\n"
5419 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
5420 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
5421 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
5422 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
5423 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
5427 "%cnull = OpConstantNull %bool\n",
5428 "%v = OpVariable %fp_v4f32 Function\n"
5429 " OpStore %v %param1\n"
5430 " OpSelectionMerge %false_label None\n"
5431 " OpBranchConditional %cnull %true_label %false_label\n"
5432 "%true_label = OpLabel\n"
5433 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
5434 " OpBranch %false_label\n"
5435 "%false_label = OpLabel\n"
5436 "%transformed_param = OpLoad %v4f32 %v\n"
5440 "%cnull = OpConstantNull %i32\n",
5441 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
5442 "%b = OpIEqual %bool %cnull %c_i32_0\n"
5443 " OpSelectionMerge %false_label None\n"
5444 " OpBranchConditional %b %true_label %false_label\n"
5445 "%true_label = OpLabel\n"
5446 " OpStore %v %param1\n"
5447 " OpBranch %false_label\n"
5448 "%false_label = OpLabel\n"
5449 "%transformed_param = OpLoad %v4f32 %v\n"
5453 "%stype = OpTypeStruct %f32 %v4f32\n"
5454 "%fp_stype = OpTypePointer Function %stype\n"
5455 "%cnull = OpConstantNull %stype\n",
5456 "%v = OpVariable %fp_stype Function %cnull\n"
5457 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
5458 "%f_val = OpLoad %v4f32 %f\n"
5459 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
5463 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
5464 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
5465 "%cnull = OpConstantNull %a4_v4f32\n",
5466 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
5467 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5468 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
5469 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
5470 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
5471 "%f_val = OpLoad %v4f32 %f\n"
5472 "%f1_val = OpLoad %v4f32 %f1\n"
5473 "%f2_val = OpLoad %v4f32 %f2\n"
5474 "%f3_val = OpLoad %v4f32 %f3\n"
5475 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
5476 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
5477 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
5478 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
5482 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5483 "%cnull = OpConstantNull %mat4x4_f32\n",
5484 // Our null matrix * any vector should result in a zero vector.
5485 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
5486 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
5490 getHalfColorsFullAlpha(colors);
5492 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5494 map<string, string> fragments;
5495 fragments["pre_main"] = tests[testNdx].constants;
5496 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5497 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
5499 return opConstantNullTests.release();
5501 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
5503 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
5504 RGBA inputColors[4];
5505 RGBA outputColors[4];
5508 const char functionStart[] =
5509 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5510 "%param1 = OpFunctionParameter %v4f32\n"
5513 const char functionEnd[] =
5514 "OpReturnValue %transformed_param\n"
5517 struct NameConstantsCode
5524 NameConstantsCode tests[] =
5529 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
5530 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
5535 "%stype = OpTypeStruct %v4f32 %f32\n"
5536 "%fp_stype = OpTypePointer Function %stype\n"
5537 "%f32_n_1 = OpConstant %f32 -1.0\n"
5538 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5539 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
5540 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
5542 "%v = OpVariable %fp_stype Function %cval\n"
5543 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5544 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
5545 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
5546 "%f32_val = OpLoad %f32 %f32_ptr\n"
5547 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
5548 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
5549 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
5552 // [1|0|0|0.5] [x] = x + 0.5
5553 // [0|1|0|0.5] [y] = y + 0.5
5554 // [0|0|1|0.5] [z] = z + 0.5
5555 // [0|0|0|1 ] [1] = 1
5558 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5559 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
5560 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
5561 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
5562 "%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"
5563 "%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",
5565 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
5570 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5571 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5572 "%f32_n_1 = OpConstant %f32 -1.0\n"
5573 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5574 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
5576 "%v = OpVariable %fp_a4f32 Function %carr\n"
5577 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
5578 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
5579 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5580 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
5581 "%f_val = OpLoad %f32 %f\n"
5582 "%f1_val = OpLoad %f32 %f1\n"
5583 "%f2_val = OpLoad %f32 %f2\n"
5584 "%f3_val = OpLoad %f32 %f3\n"
5585 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
5586 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
5587 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
5588 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
5589 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5596 // [ 1.0, 1.0, 1.0, 1.0]
5600 // [ 0.0, 0.5, 0.0, 0.0]
5604 // [ 1.0, 1.0, 1.0, 1.0]
5607 "array_of_struct_of_array",
5609 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5610 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5611 "%stype = OpTypeStruct %f32 %a4f32\n"
5612 "%a3stype = OpTypeArray %stype %c_u32_3\n"
5613 "%fp_a3stype = OpTypePointer Function %a3stype\n"
5614 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
5615 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5616 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
5617 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
5618 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
5620 "%v = OpVariable %fp_a3stype Function %carr\n"
5621 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
5622 "%f_l = OpLoad %f32 %f\n"
5623 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
5624 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5628 getHalfColorsFullAlpha(inputColors);
5629 outputColors[0] = RGBA(255, 255, 255, 255);
5630 outputColors[1] = RGBA(255, 127, 127, 255);
5631 outputColors[2] = RGBA(127, 255, 127, 255);
5632 outputColors[3] = RGBA(127, 127, 255, 255);
5634 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5636 map<string, string> fragments;
5637 fragments["pre_main"] = tests[testNdx].constants;
5638 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5639 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
5641 return opConstantCompositeTests.release();
5644 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
5646 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
5647 RGBA inputColors[4];
5648 RGBA outputColors[4];
5649 map<string, string> fragments;
5651 // vec4 test_code(vec4 param) {
5652 // vec4 result = param;
5653 // for (int i = 0; i < 4; ++i) {
5654 // if (i == 0) result[i] = 0.;
5655 // else result[i] = 1. - result[i];
5659 const char function[] =
5660 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5661 "%param1 = OpFunctionParameter %v4f32\n"
5663 "%iptr = OpVariable %fp_i32 Function\n"
5664 "%result = OpVariable %fp_v4f32 Function\n"
5665 " OpStore %iptr %c_i32_0\n"
5666 " OpStore %result %param1\n"
5669 // Loop entry block.
5671 "%ival = OpLoad %i32 %iptr\n"
5672 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5673 " OpLoopMerge %exit %if_entry None\n"
5674 " OpBranchConditional %lt_4 %if_entry %exit\n"
5676 // Merge block for loop.
5678 "%ret = OpLoad %v4f32 %result\n"
5679 " OpReturnValue %ret\n"
5681 // If-statement entry block.
5682 "%if_entry = OpLabel\n"
5683 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5684 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
5685 " OpSelectionMerge %if_exit None\n"
5686 " OpBranchConditional %eq_0 %if_true %if_false\n"
5688 // False branch for if-statement.
5689 "%if_false = OpLabel\n"
5690 "%val = OpLoad %f32 %loc\n"
5691 "%sub = OpFSub %f32 %c_f32_1 %val\n"
5692 " OpStore %loc %sub\n"
5693 " OpBranch %if_exit\n"
5695 // Merge block for if-statement.
5696 "%if_exit = OpLabel\n"
5697 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5698 " OpStore %iptr %ival_next\n"
5701 // True branch for if-statement.
5702 "%if_true = OpLabel\n"
5703 " OpStore %loc %c_f32_0\n"
5704 " OpBranch %if_exit\n"
5708 fragments["testfun"] = function;
5710 inputColors[0] = RGBA(127, 127, 127, 0);
5711 inputColors[1] = RGBA(127, 0, 0, 0);
5712 inputColors[2] = RGBA(0, 127, 0, 0);
5713 inputColors[3] = RGBA(0, 0, 127, 0);
5715 outputColors[0] = RGBA(0, 128, 128, 255);
5716 outputColors[1] = RGBA(0, 255, 255, 255);
5717 outputColors[2] = RGBA(0, 128, 255, 255);
5718 outputColors[3] = RGBA(0, 255, 128, 255);
5720 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5722 return group.release();
5725 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
5727 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
5728 RGBA inputColors[4];
5729 RGBA outputColors[4];
5730 map<string, string> fragments;
5732 const char typesAndConstants[] =
5733 "%c_f32_p2 = OpConstant %f32 0.2\n"
5734 "%c_f32_p4 = OpConstant %f32 0.4\n"
5735 "%c_f32_p6 = OpConstant %f32 0.6\n"
5736 "%c_f32_p8 = OpConstant %f32 0.8\n";
5738 // vec4 test_code(vec4 param) {
5739 // vec4 result = param;
5740 // for (int i = 0; i < 4; ++i) {
5742 // case 0: result[i] += .2; break;
5743 // case 1: result[i] += .6; break;
5744 // case 2: result[i] += .4; break;
5745 // case 3: result[i] += .8; break;
5746 // default: break; // unreachable
5751 const char function[] =
5752 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5753 "%param1 = OpFunctionParameter %v4f32\n"
5755 "%iptr = OpVariable %fp_i32 Function\n"
5756 "%result = OpVariable %fp_v4f32 Function\n"
5757 " OpStore %iptr %c_i32_0\n"
5758 " OpStore %result %param1\n"
5761 // Loop entry block.
5763 "%ival = OpLoad %i32 %iptr\n"
5764 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5765 " OpLoopMerge %exit %switch_exit None\n"
5766 " OpBranchConditional %lt_4 %switch_entry %exit\n"
5768 // Merge block for loop.
5770 "%ret = OpLoad %v4f32 %result\n"
5771 " OpReturnValue %ret\n"
5773 // Switch-statement entry block.
5774 "%switch_entry = OpLabel\n"
5775 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5776 "%val = OpLoad %f32 %loc\n"
5777 " OpSelectionMerge %switch_exit None\n"
5778 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5780 "%case2 = OpLabel\n"
5781 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
5782 " OpStore %loc %addp4\n"
5783 " OpBranch %switch_exit\n"
5785 "%switch_default = OpLabel\n"
5788 "%case3 = OpLabel\n"
5789 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
5790 " OpStore %loc %addp8\n"
5791 " OpBranch %switch_exit\n"
5793 "%case0 = OpLabel\n"
5794 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
5795 " OpStore %loc %addp2\n"
5796 " OpBranch %switch_exit\n"
5798 // Merge block for switch-statement.
5799 "%switch_exit = OpLabel\n"
5800 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5801 " OpStore %iptr %ival_next\n"
5804 "%case1 = OpLabel\n"
5805 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
5806 " OpStore %loc %addp6\n"
5807 " OpBranch %switch_exit\n"
5811 fragments["pre_main"] = typesAndConstants;
5812 fragments["testfun"] = function;
5814 inputColors[0] = RGBA(127, 27, 127, 51);
5815 inputColors[1] = RGBA(127, 0, 0, 51);
5816 inputColors[2] = RGBA(0, 27, 0, 51);
5817 inputColors[3] = RGBA(0, 0, 127, 51);
5819 outputColors[0] = RGBA(178, 180, 229, 255);
5820 outputColors[1] = RGBA(178, 153, 102, 255);
5821 outputColors[2] = RGBA(51, 180, 102, 255);
5822 outputColors[3] = RGBA(51, 153, 229, 255);
5824 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5826 return group.release();
5829 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
5831 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
5832 RGBA inputColors[4];
5833 RGBA outputColors[4];
5834 map<string, string> fragments;
5836 const char decorations[] =
5837 "OpDecorate %array_group ArrayStride 4\n"
5838 "OpDecorate %struct_member_group Offset 0\n"
5839 "%array_group = OpDecorationGroup\n"
5840 "%struct_member_group = OpDecorationGroup\n"
5842 "OpDecorate %group1 RelaxedPrecision\n"
5843 "OpDecorate %group3 RelaxedPrecision\n"
5844 "OpDecorate %group3 Invariant\n"
5845 "OpDecorate %group3 Restrict\n"
5846 "%group0 = OpDecorationGroup\n"
5847 "%group1 = OpDecorationGroup\n"
5848 "%group3 = OpDecorationGroup\n";
5850 const char typesAndConstants[] =
5851 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
5852 "%struct1 = OpTypeStruct %a3f32\n"
5853 "%struct2 = OpTypeStruct %a3f32\n"
5854 "%fp_struct1 = OpTypePointer Function %struct1\n"
5855 "%fp_struct2 = OpTypePointer Function %struct2\n"
5856 "%c_f32_2 = OpConstant %f32 2.\n"
5857 "%c_f32_n2 = OpConstant %f32 -2.\n"
5859 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
5860 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
5861 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
5862 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
5864 const char function[] =
5865 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5866 "%param = OpFunctionParameter %v4f32\n"
5867 "%entry = OpLabel\n"
5868 "%result = OpVariable %fp_v4f32 Function\n"
5869 "%v_struct1 = OpVariable %fp_struct1 Function\n"
5870 "%v_struct2 = OpVariable %fp_struct2 Function\n"
5871 " OpStore %result %param\n"
5872 " OpStore %v_struct1 %c_struct1\n"
5873 " OpStore %v_struct2 %c_struct2\n"
5874 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
5875 "%val1 = OpLoad %f32 %ptr1\n"
5876 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
5877 "%val2 = OpLoad %f32 %ptr2\n"
5878 "%addvalues = OpFAdd %f32 %val1 %val2\n"
5879 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
5880 "%val = OpLoad %f32 %ptr\n"
5881 "%addresult = OpFAdd %f32 %addvalues %val\n"
5882 " OpStore %ptr %addresult\n"
5883 "%ret = OpLoad %v4f32 %result\n"
5884 " OpReturnValue %ret\n"
5887 struct CaseNameDecoration
5893 CaseNameDecoration tests[] =
5896 "same_decoration_group_on_multiple_types",
5897 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
5900 "empty_decoration_group",
5901 "OpGroupDecorate %group0 %a3f32\n"
5902 "OpGroupDecorate %group0 %result\n"
5905 "one_element_decoration_group",
5906 "OpGroupDecorate %array_group %a3f32\n"
5909 "multiple_elements_decoration_group",
5910 "OpGroupDecorate %group3 %v_struct1\n"
5913 "multiple_decoration_groups_on_same_variable",
5914 "OpGroupDecorate %group0 %v_struct2\n"
5915 "OpGroupDecorate %group1 %v_struct2\n"
5916 "OpGroupDecorate %group3 %v_struct2\n"
5919 "same_decoration_group_multiple_times",
5920 "OpGroupDecorate %group1 %addvalues\n"
5921 "OpGroupDecorate %group1 %addvalues\n"
5922 "OpGroupDecorate %group1 %addvalues\n"
5927 getHalfColorsFullAlpha(inputColors);
5928 getHalfColorsFullAlpha(outputColors);
5930 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
5932 fragments["decoration"] = decorations + tests[idx].decoration;
5933 fragments["pre_main"] = typesAndConstants;
5934 fragments["testfun"] = function;
5936 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
5939 return group.release();
5942 struct SpecConstantTwoIntGraphicsCase
5944 const char* caseName;
5945 const char* scDefinition0;
5946 const char* scDefinition1;
5947 const char* scResultType;
5948 const char* scOperation;
5949 deInt32 scActualValue0;
5950 deInt32 scActualValue1;
5951 const char* resultOperation;
5952 RGBA expectedColors[4];
5954 SpecConstantTwoIntGraphicsCase (const char* name,
5955 const char* definition0,
5956 const char* definition1,
5957 const char* resultType,
5958 const char* operation,
5961 const char* resultOp,
5962 const RGBA (&output)[4])
5964 , scDefinition0 (definition0)
5965 , scDefinition1 (definition1)
5966 , scResultType (resultType)
5967 , scOperation (operation)
5968 , scActualValue0 (value0)
5969 , scActualValue1 (value1)
5970 , resultOperation (resultOp)
5972 expectedColors[0] = output[0];
5973 expectedColors[1] = output[1];
5974 expectedColors[2] = output[2];
5975 expectedColors[3] = output[3];
5979 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
5981 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
5982 vector<SpecConstantTwoIntGraphicsCase> cases;
5983 RGBA inputColors[4];
5984 RGBA outputColors0[4];
5985 RGBA outputColors1[4];
5986 RGBA outputColors2[4];
5988 const char decorations1[] =
5989 "OpDecorate %sc_0 SpecId 0\n"
5990 "OpDecorate %sc_1 SpecId 1\n";
5992 const char typesAndConstants1[] =
5993 "${OPTYPE_DEFINITIONS:opt}"
5994 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
5995 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
5996 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
5998 const char function1[] =
5999 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6000 "%param = OpFunctionParameter %v4f32\n"
6001 "%label = OpLabel\n"
6002 "${TYPE_CONVERT:opt}"
6003 "%result = OpVariable %fp_v4f32 Function\n"
6004 " OpStore %result %param\n"
6005 "%gen = ${GEN_RESULT}\n"
6006 "%index = OpIAdd %i32 %gen %c_i32_1\n"
6007 "%loc = OpAccessChain %fp_f32 %result %index\n"
6008 "%val = OpLoad %f32 %loc\n"
6009 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6010 " OpStore %loc %add\n"
6011 "%ret = OpLoad %v4f32 %result\n"
6012 " OpReturnValue %ret\n"
6015 inputColors[0] = RGBA(127, 127, 127, 255);
6016 inputColors[1] = RGBA(127, 0, 0, 255);
6017 inputColors[2] = RGBA(0, 127, 0, 255);
6018 inputColors[3] = RGBA(0, 0, 127, 255);
6020 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
6021 outputColors0[0] = RGBA(255, 127, 127, 255);
6022 outputColors0[1] = RGBA(255, 0, 0, 255);
6023 outputColors0[2] = RGBA(128, 127, 0, 255);
6024 outputColors0[3] = RGBA(128, 0, 127, 255);
6026 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
6027 outputColors1[0] = RGBA(127, 255, 127, 255);
6028 outputColors1[1] = RGBA(127, 128, 0, 255);
6029 outputColors1[2] = RGBA(0, 255, 0, 255);
6030 outputColors1[3] = RGBA(0, 128, 127, 255);
6032 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
6033 outputColors2[0] = RGBA(127, 127, 255, 255);
6034 outputColors2[1] = RGBA(127, 0, 128, 255);
6035 outputColors2[2] = RGBA(0, 127, 128, 255);
6036 outputColors2[3] = RGBA(0, 0, 255, 255);
6038 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
6039 const char addZeroToSc32[] = "OpIAdd %i32 %c_i32_0 %sc_op32";
6040 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
6041 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
6043 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
6044 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
6045 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
6046 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
6047 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
6048 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6049 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6050 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
6051 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
6052 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
6053 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
6054 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
6055 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
6056 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
6057 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
6058 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
6059 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6060 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
6061 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
6062 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
6063 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6064 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
6065 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
6066 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6067 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6068 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6069 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6070 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
6071 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
6072 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
6073 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
6074 cases.push_back(SpecConstantTwoIntGraphicsCase("sconvert", " %i32 0", " %i32 0", "%i16", "SConvert %sc_0", -1, 0, addZeroToSc32, outputColors0));
6075 // -1082130432 stored as 32-bit two's complement is the binary representation of -1 as IEEE-754 Float
6076 cases.push_back(SpecConstantTwoIntGraphicsCase("fconvert", " %f32 0", " %f32 0", "%f64", "FConvert %sc_0", -1082130432, 0, addZeroToSc32, outputColors0));
6077 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
6079 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
6081 map<string, string> specializations;
6082 map<string, string> fragments;
6083 vector<deInt32> specConstants;
6084 vector<string> features;
6085 PushConstants noPushConstants;
6086 GraphicsResources noResources;
6087 GraphicsInterfaces noInterfaces;
6088 std::vector<std::string> noExtensions;
6090 // Special SPIR-V code for SConvert-case
6091 if (strcmp(cases[caseNdx].caseName, "sconvert") == 0)
6093 features.push_back("shaderInt16");
6094 fragments["capability"] = "OpCapability Int16\n"; // Adds 16-bit integer capability
6095 specializations["OPTYPE_DEFINITIONS"] = "%i16 = OpTypeInt 16 1\n"; // Adds 16-bit integer type
6096 specializations["TYPE_CONVERT"] = "%sc_op32 = OpSConvert %i32 %sc_op\n"; // Converts 16-bit integer to 32-bit integer
6099 // Special SPIR-V code for FConvert-case
6100 if (strcmp(cases[caseNdx].caseName, "fconvert") == 0)
6102 features.push_back("shaderFloat64");
6103 fragments["capability"] = "OpCapability Float64\n"; // Adds 64-bit float capability
6104 specializations["OPTYPE_DEFINITIONS"] = "%f64 = OpTypeFloat 64\n"; // Adds 64-bit float type
6105 specializations["TYPE_CONVERT"] = "%sc_op32 = OpConvertFToS %i32 %sc_op\n"; // Converts 64-bit float to 32-bit integer
6108 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
6109 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
6110 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
6111 specializations["SC_OP"] = cases[caseNdx].scOperation;
6112 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
6114 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
6115 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
6116 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
6118 specConstants.push_back(cases[caseNdx].scActualValue0);
6119 specConstants.push_back(cases[caseNdx].scActualValue1);
6121 createTestsForAllStages(
6122 cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants,
6123 noPushConstants, noResources, noInterfaces, noExtensions, features, VulkanFeatures(), group.get());
6126 const char decorations2[] =
6127 "OpDecorate %sc_0 SpecId 0\n"
6128 "OpDecorate %sc_1 SpecId 1\n"
6129 "OpDecorate %sc_2 SpecId 2\n";
6131 const char typesAndConstants2[] =
6132 "%v3i32 = OpTypeVector %i32 3\n"
6133 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
6134 "%vec3_undef = OpUndef %v3i32\n"
6136 "%sc_0 = OpSpecConstant %i32 0\n"
6137 "%sc_1 = OpSpecConstant %i32 0\n"
6138 "%sc_2 = OpSpecConstant %i32 0\n"
6139 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
6140 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
6141 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
6142 "%sc_vec3_0_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
6143 "%sc_vec3_1_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
6144 "%sc_vec3_2_s = OpSpecConstantOp %v3i32 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
6145 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
6146 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
6147 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
6148 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
6149 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
6150 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
6151 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
6153 const char function2[] =
6154 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6155 "%param = OpFunctionParameter %v4f32\n"
6156 "%label = OpLabel\n"
6157 "%result = OpVariable %fp_v4f32 Function\n"
6158 " OpStore %result %param\n"
6159 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
6160 "%val = OpLoad %f32 %loc\n"
6161 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6162 " OpStore %loc %add\n"
6163 "%ret = OpLoad %v4f32 %result\n"
6164 " OpReturnValue %ret\n"
6167 map<string, string> fragments;
6168 vector<deInt32> specConstants;
6170 fragments["decoration"] = decorations2;
6171 fragments["pre_main"] = typesAndConstants2;
6172 fragments["testfun"] = function2;
6174 specConstants.push_back(56789);
6175 specConstants.push_back(-2);
6176 specConstants.push_back(56788);
6178 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
6180 return group.release();
6183 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
6185 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
6186 RGBA inputColors[4];
6187 RGBA outputColors1[4];
6188 RGBA outputColors2[4];
6189 RGBA outputColors3[4];
6190 map<string, string> fragments1;
6191 map<string, string> fragments2;
6192 map<string, string> fragments3;
6194 const char typesAndConstants1[] =
6195 "%c_f32_p2 = OpConstant %f32 0.2\n"
6196 "%c_f32_p4 = OpConstant %f32 0.4\n"
6197 "%c_f32_p5 = OpConstant %f32 0.5\n"
6198 "%c_f32_p8 = OpConstant %f32 0.8\n";
6200 // vec4 test_code(vec4 param) {
6201 // vec4 result = param;
6202 // for (int i = 0; i < 4; ++i) {
6205 // case 0: operand = .2; break;
6206 // case 1: operand = .5; break;
6207 // case 2: operand = .4; break;
6208 // case 3: operand = .0; break;
6209 // default: break; // unreachable
6211 // result[i] += operand;
6215 const char function1[] =
6216 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6217 "%param1 = OpFunctionParameter %v4f32\n"
6219 "%iptr = OpVariable %fp_i32 Function\n"
6220 "%result = OpVariable %fp_v4f32 Function\n"
6221 " OpStore %iptr %c_i32_0\n"
6222 " OpStore %result %param1\n"
6226 "%ival = OpLoad %i32 %iptr\n"
6227 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6228 " OpLoopMerge %exit %phi None\n"
6229 " OpBranchConditional %lt_4 %entry %exit\n"
6231 "%entry = OpLabel\n"
6232 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6233 "%val = OpLoad %f32 %loc\n"
6234 " OpSelectionMerge %phi None\n"
6235 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6237 "%case0 = OpLabel\n"
6239 "%case1 = OpLabel\n"
6241 "%case2 = OpLabel\n"
6243 "%case3 = OpLabel\n"
6246 "%default = OpLabel\n"
6250 "%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
6251 "%add = OpFAdd %f32 %val %operand\n"
6252 " OpStore %loc %add\n"
6253 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6254 " OpStore %iptr %ival_next\n"
6258 "%ret = OpLoad %v4f32 %result\n"
6259 " OpReturnValue %ret\n"
6263 fragments1["pre_main"] = typesAndConstants1;
6264 fragments1["testfun"] = function1;
6266 getHalfColorsFullAlpha(inputColors);
6268 outputColors1[0] = RGBA(178, 255, 229, 255);
6269 outputColors1[1] = RGBA(178, 127, 102, 255);
6270 outputColors1[2] = RGBA(51, 255, 102, 255);
6271 outputColors1[3] = RGBA(51, 127, 229, 255);
6273 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
6275 const char typesAndConstants2[] =
6276 "%c_f32_p2 = OpConstant %f32 0.2\n";
6278 // Add .4 to the second element of the given parameter.
6279 const char function2[] =
6280 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6281 "%param = OpFunctionParameter %v4f32\n"
6282 "%entry = OpLabel\n"
6283 "%result = OpVariable %fp_v4f32 Function\n"
6284 " OpStore %result %param\n"
6285 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6286 "%val = OpLoad %f32 %loc\n"
6290 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
6291 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
6292 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
6293 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
6294 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
6295 " OpLoopMerge %exit %phi None\n"
6296 " OpBranchConditional %still_loop %phi %exit\n"
6299 " OpStore %loc %accum\n"
6300 "%ret = OpLoad %v4f32 %result\n"
6301 " OpReturnValue %ret\n"
6305 fragments2["pre_main"] = typesAndConstants2;
6306 fragments2["testfun"] = function2;
6308 outputColors2[0] = RGBA(127, 229, 127, 255);
6309 outputColors2[1] = RGBA(127, 102, 0, 255);
6310 outputColors2[2] = RGBA(0, 229, 0, 255);
6311 outputColors2[3] = RGBA(0, 102, 127, 255);
6313 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
6315 const char typesAndConstants3[] =
6316 "%true = OpConstantTrue %bool\n"
6317 "%false = OpConstantFalse %bool\n"
6318 "%c_f32_p2 = OpConstant %f32 0.2\n";
6320 // Swap the second and the third element of the given parameter.
6321 const char function3[] =
6322 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6323 "%param = OpFunctionParameter %v4f32\n"
6324 "%entry = OpLabel\n"
6325 "%result = OpVariable %fp_v4f32 Function\n"
6326 " OpStore %result %param\n"
6327 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6328 "%a_init = OpLoad %f32 %a_loc\n"
6329 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
6330 "%b_init = OpLoad %f32 %b_loc\n"
6334 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
6335 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
6336 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
6337 " OpLoopMerge %exit %phi None\n"
6338 " OpBranchConditional %still_loop %phi %exit\n"
6341 " OpStore %a_loc %a_next\n"
6342 " OpStore %b_loc %b_next\n"
6343 "%ret = OpLoad %v4f32 %result\n"
6344 " OpReturnValue %ret\n"
6348 fragments3["pre_main"] = typesAndConstants3;
6349 fragments3["testfun"] = function3;
6351 outputColors3[0] = RGBA(127, 127, 127, 255);
6352 outputColors3[1] = RGBA(127, 0, 0, 255);
6353 outputColors3[2] = RGBA(0, 0, 127, 255);
6354 outputColors3[3] = RGBA(0, 127, 0, 255);
6356 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
6358 return group.release();
6361 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
6363 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
6364 RGBA inputColors[4];
6365 RGBA outputColors[4];
6367 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
6368 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
6369 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
6370 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
6371 const char constantsAndTypes[] =
6372 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
6373 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
6374 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
6375 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
6376 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n";
6378 const char function[] =
6379 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6380 "%param = OpFunctionParameter %v4f32\n"
6381 "%label = OpLabel\n"
6382 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
6383 "%var2 = OpVariable %fp_f32 Function\n"
6384 "%red = OpCompositeExtract %f32 %param 0\n"
6385 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
6386 " OpStore %var2 %plus_red\n"
6387 "%val1 = OpLoad %f32 %var1\n"
6388 "%val2 = OpLoad %f32 %var2\n"
6389 "%mul = OpFMul %f32 %val1 %val2\n"
6390 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
6391 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
6392 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
6393 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
6394 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
6395 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
6396 " OpReturnValue %ret\n"
6399 struct CaseNameDecoration
6406 CaseNameDecoration tests[] = {
6407 {"multiplication", "OpDecorate %mul NoContraction"},
6408 {"addition", "OpDecorate %add NoContraction"},
6409 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
6412 getHalfColorsFullAlpha(inputColors);
6414 for (deUint8 idx = 0; idx < 4; ++idx)
6416 inputColors[idx].setRed(0);
6417 outputColors[idx] = RGBA(0, 0, 0, 255);
6420 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
6422 map<string, string> fragments;
6424 fragments["decoration"] = tests[testNdx].decoration;
6425 fragments["pre_main"] = constantsAndTypes;
6426 fragments["testfun"] = function;
6428 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
6431 return group.release();
6434 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
6436 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
6439 const char constantsAndTypes[] =
6440 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
6441 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
6442 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
6443 "%fp_stype = OpTypePointer Function %stype\n";
6445 const char function[] =
6446 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6447 "%param1 = OpFunctionParameter %v4f32\n"
6449 "%v1 = OpVariable %fp_v4f32 Function\n"
6450 "%v2 = OpVariable %fp_a2f32 Function\n"
6451 "%v3 = OpVariable %fp_f32 Function\n"
6452 "%v = OpVariable %fp_stype Function\n"
6453 "%vv = OpVariable %fp_stype Function\n"
6454 "%vvv = OpVariable %fp_f32 Function\n"
6456 " OpStore %v1 %c_v4f32_1_1_1_1\n"
6457 " OpStore %v2 %c_a2f32_1\n"
6458 " OpStore %v3 %c_f32_1\n"
6460 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
6461 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
6462 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
6463 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
6464 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
6465 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
6467 " OpStore %p_v4f32 %v1_v ${access_type}\n"
6468 " OpStore %p_a2f32 %v2_v ${access_type}\n"
6469 " OpStore %p_f32 %v3_v ${access_type}\n"
6471 " OpCopyMemory %vv %v ${access_type}\n"
6472 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
6474 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
6475 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
6476 "%v_f32_3 = OpLoad %f32 %vvv\n"
6478 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
6479 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
6480 " OpReturnValue %ret2\n"
6483 struct NameMemoryAccess
6490 NameMemoryAccess tests[] =
6493 { "volatile", "Volatile" },
6494 { "aligned", "Aligned 1" },
6495 { "volatile_aligned", "Volatile|Aligned 1" },
6496 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
6497 { "volatile_nontemporal", "Volatile|Nontemporal" },
6498 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
6501 getHalfColorsFullAlpha(colors);
6503 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
6505 map<string, string> fragments;
6506 map<string, string> memoryAccess;
6507 memoryAccess["access_type"] = tests[testNdx].accessType;
6509 fragments["pre_main"] = constantsAndTypes;
6510 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
6511 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
6513 return memoryAccessTests.release();
6515 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
6517 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
6518 RGBA defaultColors[4];
6519 map<string, string> fragments;
6520 getDefaultColors(defaultColors);
6522 // First, simple cases that don't do anything with the OpUndef result.
6523 struct NameCodePair { string name, decl, type; };
6524 const NameCodePair tests[] =
6526 {"bool", "", "%bool"},
6527 {"vec2uint32", "", "%v2u32"},
6528 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown", "%type"},
6529 {"sampler", "%type = OpTypeSampler", "%type"},
6530 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img", "%type"},
6531 {"pointer", "", "%fp_i32"},
6532 {"runtimearray", "%type = OpTypeRuntimeArray %f32", "%type"},
6533 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100", "%type"},
6534 {"struct", "%type = OpTypeStruct %f32 %i32 %u32", "%type"}};
6535 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
6537 fragments["undef_type"] = tests[testNdx].type;
6538 fragments["testfun"] = StringTemplate(
6539 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6540 "%param1 = OpFunctionParameter %v4f32\n"
6541 "%label_testfun = OpLabel\n"
6542 "%undef = OpUndef ${undef_type}\n"
6543 "OpReturnValue %param1\n"
6544 "OpFunctionEnd\n").specialize(fragments);
6545 fragments["pre_main"] = tests[testNdx].decl;
6546 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
6550 fragments["testfun"] =
6551 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6552 "%param1 = OpFunctionParameter %v4f32\n"
6553 "%label_testfun = OpLabel\n"
6554 "%undef = OpUndef %f32\n"
6555 "%zero = OpFMul %f32 %undef %c_f32_0\n"
6556 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
6557 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
6558 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6559 "%b = OpFAdd %f32 %a %actually_zero\n"
6560 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
6561 "OpReturnValue %ret\n"
6564 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6566 fragments["testfun"] =
6567 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6568 "%param1 = OpFunctionParameter %v4f32\n"
6569 "%label_testfun = OpLabel\n"
6570 "%undef = OpUndef %i32\n"
6571 "%zero = OpIMul %i32 %undef %c_i32_0\n"
6572 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6573 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6574 "OpReturnValue %ret\n"
6577 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6579 fragments["testfun"] =
6580 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6581 "%param1 = OpFunctionParameter %v4f32\n"
6582 "%label_testfun = OpLabel\n"
6583 "%undef = OpUndef %u32\n"
6584 "%zero = OpIMul %u32 %undef %c_i32_0\n"
6585 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6586 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6587 "OpReturnValue %ret\n"
6590 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6592 fragments["testfun"] =
6593 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6594 "%param1 = OpFunctionParameter %v4f32\n"
6595 "%label_testfun = OpLabel\n"
6596 "%undef = OpUndef %v4f32\n"
6597 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
6598 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
6599 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
6600 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
6601 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
6602 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6603 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6604 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6605 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6606 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6607 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
6608 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
6609 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
6610 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6611 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6612 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6613 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6614 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6615 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6616 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6617 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6618 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6619 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6620 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6621 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6622 "OpReturnValue %ret\n"
6625 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6627 fragments["pre_main"] =
6628 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
6629 fragments["testfun"] =
6630 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6631 "%param1 = OpFunctionParameter %v4f32\n"
6632 "%label_testfun = OpLabel\n"
6633 "%undef = OpUndef %m2x2f32\n"
6634 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
6635 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
6636 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
6637 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
6638 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
6639 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6640 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6641 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6642 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6643 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6644 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
6645 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
6646 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
6647 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6648 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6649 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6650 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6651 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6652 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6653 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6654 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6655 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6656 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6657 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6658 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6659 "OpReturnValue %ret\n"
6662 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
6664 return opUndefTests.release();
6667 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
6669 const RGBA inputColors[4] =
6672 RGBA(0, 0, 255, 255),
6673 RGBA(0, 255, 0, 255),
6674 RGBA(0, 255, 255, 255)
6677 const RGBA expectedColors[4] =
6679 RGBA(255, 0, 0, 255),
6680 RGBA(255, 0, 0, 255),
6681 RGBA(255, 0, 0, 255),
6682 RGBA(255, 0, 0, 255)
6685 const struct SingleFP16Possibility
6688 const char* constant; // Value to assign to %test_constant.
6690 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
6696 -constructNormalizedFloat(1, 0x300000),
6697 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6702 constructNormalizedFloat(7, 0x000000),
6703 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6705 // SPIR-V requires that OpQuantizeToF16 flushes
6706 // any numbers that would end up denormalized in F16 to zero.
6710 std::ldexp(1.5f, -140),
6711 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6716 -std::ldexp(1.5f, -140),
6717 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6722 std::ldexp(1.0f, -16),
6723 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6724 }, // too small positive
6726 "negative_too_small",
6728 -std::ldexp(1.0f, -32),
6729 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6730 }, // too small negative
6734 -std::ldexp(1.0f, 128),
6736 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6737 "%inf = OpIsInf %bool %c\n"
6738 "%cond = OpLogicalAnd %bool %gz %inf\n"
6743 std::ldexp(1.0f, 128),
6745 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6746 "%inf = OpIsInf %bool %c\n"
6747 "%cond = OpLogicalAnd %bool %gz %inf\n"
6750 "round_to_negative_inf",
6752 -std::ldexp(1.0f, 32),
6754 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6755 "%inf = OpIsInf %bool %c\n"
6756 "%cond = OpLogicalAnd %bool %gz %inf\n"
6761 std::ldexp(1.0f, 16),
6763 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6764 "%inf = OpIsInf %bool %c\n"
6765 "%cond = OpLogicalAnd %bool %gz %inf\n"
6770 std::numeric_limits<float>::quiet_NaN(),
6772 // Test for any NaN value, as NaNs are not preserved
6773 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6774 "%cond = OpIsNan %bool %direct_quant\n"
6779 std::numeric_limits<float>::quiet_NaN(),
6781 // Test for any NaN value, as NaNs are not preserved
6782 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6783 "%cond = OpIsNan %bool %direct_quant\n"
6786 const char* constants =
6787 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
6789 StringTemplate function (
6790 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6791 "%param1 = OpFunctionParameter %v4f32\n"
6792 "%label_testfun = OpLabel\n"
6793 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6794 "%b = OpFAdd %f32 %test_constant %a\n"
6795 "%c = OpQuantizeToF16 %f32 %b\n"
6797 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6798 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
6799 " OpReturnValue %retval\n"
6803 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
6804 const char* specConstants =
6805 "%test_constant = OpSpecConstant %f32 0.\n"
6806 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
6808 StringTemplate specConstantFunction(
6809 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6810 "%param1 = OpFunctionParameter %v4f32\n"
6811 "%label_testfun = OpLabel\n"
6813 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6814 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
6815 " OpReturnValue %retval\n"
6819 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6821 map<string, string> codeSpecialization;
6822 map<string, string> fragments;
6823 codeSpecialization["condition"] = tests[idx].condition;
6824 fragments["testfun"] = function.specialize(codeSpecialization);
6825 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
6826 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6829 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6831 map<string, string> codeSpecialization;
6832 map<string, string> fragments;
6833 vector<deInt32> passConstants;
6834 deInt32 specConstant;
6836 codeSpecialization["condition"] = tests[idx].condition;
6837 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
6838 fragments["decoration"] = specDecorations;
6839 fragments["pre_main"] = specConstants;
6841 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
6842 passConstants.push_back(specConstant);
6844 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6848 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
6850 RGBA inputColors[4] = {
6852 RGBA(0, 0, 255, 255),
6853 RGBA(0, 255, 0, 255),
6854 RGBA(0, 255, 255, 255)
6857 RGBA expectedColors[4] =
6859 RGBA(255, 0, 0, 255),
6860 RGBA(255, 0, 0, 255),
6861 RGBA(255, 0, 0, 255),
6862 RGBA(255, 0, 0, 255)
6865 struct DualFP16Possibility
6870 const char* possibleOutput1;
6871 const char* possibleOutput2;
6874 "positive_round_up_or_round_down",
6876 constructNormalizedFloat(8, 0x300300),
6881 "negative_round_up_or_round_down",
6883 -constructNormalizedFloat(-7, 0x600800),
6890 constructNormalizedFloat(2, 0x01e000),
6895 "carry_to_exponent",
6897 constructNormalizedFloat(1, 0xffe000),
6902 StringTemplate constants (
6903 "%input_const = OpConstant %f32 ${input}\n"
6904 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6905 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6908 StringTemplate specConstants (
6909 "%input_const = OpSpecConstant %f32 0.\n"
6910 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6911 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6914 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
6916 const char* function =
6917 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6918 "%param1 = OpFunctionParameter %v4f32\n"
6919 "%label_testfun = OpLabel\n"
6920 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6921 // For the purposes of this test we assume that 0.f will always get
6922 // faithfully passed through the pipeline stages.
6923 "%b = OpFAdd %f32 %input_const %a\n"
6924 "%c = OpQuantizeToF16 %f32 %b\n"
6925 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
6926 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
6927 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
6928 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6929 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
6930 " OpReturnValue %retval\n"
6933 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6934 map<string, string> fragments;
6935 map<string, string> constantSpecialization;
6937 constantSpecialization["input"] = tests[idx].input;
6938 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6939 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6940 fragments["testfun"] = function;
6941 fragments["pre_main"] = constants.specialize(constantSpecialization);
6942 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6945 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6946 map<string, string> fragments;
6947 map<string, string> constantSpecialization;
6948 vector<deInt32> passConstants;
6949 deInt32 specConstant;
6951 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6952 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6953 fragments["testfun"] = function;
6954 fragments["decoration"] = specDecorations;
6955 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
6957 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
6958 passConstants.push_back(specConstant);
6960 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6964 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
6966 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
6967 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
6968 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
6969 return opQuantizeTests.release();
6972 struct ShaderPermutation
6974 deUint8 vertexPermutation;
6975 deUint8 geometryPermutation;
6976 deUint8 tesscPermutation;
6977 deUint8 tessePermutation;
6978 deUint8 fragmentPermutation;
6981 ShaderPermutation getShaderPermutation(deUint8 inputValue)
6983 ShaderPermutation permutation =
6985 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
6986 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
6987 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
6988 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
6989 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
6994 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
6996 RGBA defaultColors[4];
6997 RGBA invertedColors[4];
6998 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
7000 const ShaderElement combinedPipeline[] =
7002 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
7003 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
7004 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7005 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7006 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
7009 getDefaultColors(defaultColors);
7010 getInvertedDefaultColors(invertedColors);
7011 addFunctionCaseWithPrograms<InstanceContext>(
7012 moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline,
7013 createInstanceContext(combinedPipeline, map<string, string>()));
7015 const char* numbers[] =
7020 for (deInt8 idx = 0; idx < 32; ++idx)
7022 ShaderPermutation permutation = getShaderPermutation(idx);
7023 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
7024 const ShaderElement pipeline[] =
7026 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
7027 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
7028 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7029 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7030 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
7033 // If there are an even number of swaps, then it should be no-op.
7034 // If there are an odd number, the color should be flipped.
7035 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
7037 addFunctionCaseWithPrograms<InstanceContext>(
7038 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
7039 createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
7043 addFunctionCaseWithPrograms<InstanceContext>(
7044 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
7045 createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
7048 return moduleTests.release();
7051 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
7053 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
7054 RGBA defaultColors[4];
7055 getDefaultColors(defaultColors);
7056 map<string, string> fragments;
7057 fragments["pre_main"] =
7058 "%c_f32_5 = OpConstant %f32 5.\n";
7060 // A loop with a single block. The Continue Target is the loop block
7061 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
7062 // -- the "continue construct" forms the entire loop.
7063 fragments["testfun"] =
7064 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7065 "%param1 = OpFunctionParameter %v4f32\n"
7067 "%entry = OpLabel\n"
7068 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7071 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7073 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7074 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
7075 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7076 "%val = OpFAdd %f32 %val1 %delta\n"
7077 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
7078 "%count__ = OpISub %i32 %count %c_i32_1\n"
7079 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7080 "OpLoopMerge %exit %loop None\n"
7081 "OpBranchConditional %again %loop %exit\n"
7084 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7085 "OpReturnValue %result\n"
7089 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
7091 // Body comprised of multiple basic blocks.
7092 const StringTemplate multiBlock(
7093 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7094 "%param1 = OpFunctionParameter %v4f32\n"
7096 "%entry = OpLabel\n"
7097 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7100 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7102 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
7103 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
7104 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
7105 // There are several possibilities for the Continue Target below. Each
7106 // will be specialized into a separate test case.
7107 "OpLoopMerge %exit ${continue_target} None\n"
7111 ";delta_next = (delta > 0) ? -1 : 1;\n"
7112 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
7113 "OpSelectionMerge %gather DontFlatten\n"
7114 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
7117 "OpBranch %gather\n"
7120 "OpBranch %gather\n"
7122 "%gather = OpLabel\n"
7123 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
7124 "%val = OpFAdd %f32 %val1 %delta\n"
7125 "%count__ = OpISub %i32 %count %c_i32_1\n"
7126 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7127 "OpBranchConditional %again %loop %exit\n"
7130 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7131 "OpReturnValue %result\n"
7135 map<string, string> continue_target;
7137 // The Continue Target is the loop block itself.
7138 continue_target["continue_target"] = "%loop";
7139 fragments["testfun"] = multiBlock.specialize(continue_target);
7140 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
7142 // The Continue Target is at the end of the loop.
7143 continue_target["continue_target"] = "%gather";
7144 fragments["testfun"] = multiBlock.specialize(continue_target);
7145 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
7147 // A loop with continue statement.
7148 fragments["testfun"] =
7149 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7150 "%param1 = OpFunctionParameter %v4f32\n"
7152 "%entry = OpLabel\n"
7153 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7156 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
7158 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7159 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
7160 "OpLoopMerge %exit %continue None\n"
7164 ";skip if %count==2\n"
7165 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
7166 "OpSelectionMerge %continue DontFlatten\n"
7167 "OpBranchConditional %eq2 %continue %body\n"
7170 "%fcount = OpConvertSToF %f32 %count\n"
7171 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7172 "OpBranch %continue\n"
7174 "%continue = OpLabel\n"
7175 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
7176 "%count__ = OpISub %i32 %count %c_i32_1\n"
7177 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7178 "OpBranchConditional %again %loop %exit\n"
7181 "%same = OpFSub %f32 %val %c_f32_8\n"
7182 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7183 "OpReturnValue %result\n"
7185 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
7187 // A loop with break.
7188 fragments["testfun"] =
7189 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7190 "%param1 = OpFunctionParameter %v4f32\n"
7192 "%entry = OpLabel\n"
7193 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7194 "%dot = OpDot %f32 %param1 %param1\n"
7195 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7196 "%zero = OpConvertFToU %u32 %div\n"
7197 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7198 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7201 ";adds 4 and 3 to %val0 (exits early)\n"
7203 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7204 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7205 "OpLoopMerge %exit %continue None\n"
7209 ";end loop if %count==%two\n"
7210 "%above2 = OpSGreaterThan %bool %count %two\n"
7211 "OpSelectionMerge %continue DontFlatten\n"
7212 "OpBranchConditional %above2 %body %exit\n"
7215 "%fcount = OpConvertSToF %f32 %count\n"
7216 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7217 "OpBranch %continue\n"
7219 "%continue = OpLabel\n"
7220 "%count__ = OpISub %i32 %count %c_i32_1\n"
7221 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7222 "OpBranchConditional %again %loop %exit\n"
7225 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
7226 "%same = OpFSub %f32 %val_post %c_f32_7\n"
7227 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7228 "OpReturnValue %result\n"
7230 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
7232 // A loop with return.
7233 fragments["testfun"] =
7234 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7235 "%param1 = OpFunctionParameter %v4f32\n"
7237 "%entry = OpLabel\n"
7238 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7239 "%dot = OpDot %f32 %param1 %param1\n"
7240 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7241 "%zero = OpConvertFToU %u32 %div\n"
7242 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7243 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7246 ";returns early without modifying %param1\n"
7248 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7249 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7250 "OpLoopMerge %exit %continue None\n"
7254 ";return if %count==%two\n"
7255 "%above2 = OpSGreaterThan %bool %count %two\n"
7256 "OpSelectionMerge %continue DontFlatten\n"
7257 "OpBranchConditional %above2 %body %early_exit\n"
7259 "%early_exit = OpLabel\n"
7260 "OpReturnValue %param1\n"
7263 "%fcount = OpConvertSToF %f32 %count\n"
7264 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7265 "OpBranch %continue\n"
7267 "%continue = OpLabel\n"
7268 "%count__ = OpISub %i32 %count %c_i32_1\n"
7269 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7270 "OpBranchConditional %again %loop %exit\n"
7273 ";should never get here, so return an incorrect result\n"
7274 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
7275 "OpReturnValue %result\n"
7277 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
7279 return testGroup.release();
7282 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
7283 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
7285 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
7286 map<string, string> fragments;
7288 // A barrier inside a function body.
7289 fragments["pre_main"] =
7290 "%Workgroup = OpConstant %i32 2\n"
7291 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
7292 fragments["testfun"] =
7293 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7294 "%param1 = OpFunctionParameter %v4f32\n"
7295 "%label_testfun = OpLabel\n"
7296 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7297 "OpReturnValue %param1\n"
7299 addTessCtrlTest(testGroup.get(), "in_function", fragments);
7301 // Common setup code for the following tests.
7302 fragments["pre_main"] =
7303 "%Workgroup = OpConstant %i32 2\n"
7304 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7305 "%c_f32_5 = OpConstant %f32 5.\n";
7306 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
7307 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7308 "%param1 = OpFunctionParameter %v4f32\n"
7309 "%entry = OpLabel\n"
7310 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7311 "%dot = OpDot %f32 %param1 %param1\n"
7312 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7313 "%zero = OpConvertFToU %u32 %div\n";
7315 // Barriers inside OpSwitch branches.
7316 fragments["testfun"] =
7318 "OpSelectionMerge %switch_exit None\n"
7319 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
7321 "%case1 = OpLabel\n"
7322 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7323 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7324 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7325 "OpBranch %switch_exit\n"
7327 "%switch_default = OpLabel\n"
7328 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7329 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7330 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7331 "OpBranch %switch_exit\n"
7333 "%case0 = OpLabel\n"
7334 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7335 "OpBranch %switch_exit\n"
7337 "%switch_exit = OpLabel\n"
7338 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
7339 "OpReturnValue %ret\n"
7341 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
7343 // Barriers inside if-then-else.
7344 fragments["testfun"] =
7346 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
7347 "OpSelectionMerge %exit DontFlatten\n"
7348 "OpBranchConditional %eq0 %then %else\n"
7351 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7352 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7353 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7357 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7361 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
7362 "OpReturnValue %ret\n"
7364 addTessCtrlTest(testGroup.get(), "in_if", fragments);
7366 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
7367 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
7368 fragments["testfun"] =
7370 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
7371 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
7372 "OpSelectionMerge %exit DontFlatten\n"
7373 "OpBranchConditional %thread0 %then %else\n"
7376 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7380 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
7384 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
7385 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7386 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
7387 "OpReturnValue %ret\n"
7389 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
7391 // A barrier inside a loop.
7392 fragments["pre_main"] =
7393 "%Workgroup = OpConstant %i32 2\n"
7394 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7395 "%c_f32_10 = OpConstant %f32 10.\n";
7396 fragments["testfun"] =
7397 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7398 "%param1 = OpFunctionParameter %v4f32\n"
7399 "%entry = OpLabel\n"
7400 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7403 ";adds 4, 3, 2, and 1 to %val0\n"
7405 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7406 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7407 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7408 "%fcount = OpConvertSToF %f32 %count\n"
7409 "%val = OpFAdd %f32 %val1 %fcount\n"
7410 "%count__ = OpISub %i32 %count %c_i32_1\n"
7411 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7412 "OpLoopMerge %exit %loop None\n"
7413 "OpBranchConditional %again %loop %exit\n"
7416 "%same = OpFSub %f32 %val %c_f32_10\n"
7417 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7418 "OpReturnValue %ret\n"
7420 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
7422 return testGroup.release();
7425 // Test for the OpFRem instruction.
7426 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
7428 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
7429 map<string, string> fragments;
7430 RGBA inputColors[4];
7431 RGBA outputColors[4];
7433 fragments["pre_main"] =
7434 "%c_f32_3 = OpConstant %f32 3.0\n"
7435 "%c_f32_n3 = OpConstant %f32 -3.0\n"
7436 "%c_f32_4 = OpConstant %f32 4.0\n"
7437 "%c_f32_p75 = OpConstant %f32 0.75\n"
7438 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
7439 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
7440 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
7442 // The test does the following.
7443 // vec4 result = (param1 * 8.0) - 4.0;
7444 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
7445 fragments["testfun"] =
7446 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7447 "%param1 = OpFunctionParameter %v4f32\n"
7448 "%label_testfun = OpLabel\n"
7449 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
7450 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
7451 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
7452 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
7453 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
7454 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
7455 "OpReturnValue %xy_0_1\n"
7459 inputColors[0] = RGBA(16, 16, 0, 255);
7460 inputColors[1] = RGBA(232, 232, 0, 255);
7461 inputColors[2] = RGBA(232, 16, 0, 255);
7462 inputColors[3] = RGBA(16, 232, 0, 255);
7464 outputColors[0] = RGBA(64, 64, 0, 255);
7465 outputColors[1] = RGBA(255, 255, 0, 255);
7466 outputColors[2] = RGBA(255, 64, 0, 255);
7467 outputColors[3] = RGBA(64, 255, 0, 255);
7469 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
7470 return testGroup.release();
7473 // Test for the OpSRem instruction.
7474 tcu::TestCaseGroup* createOpSRemGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
7476 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "srem", "OpSRem"));
7477 map<string, string> fragments;
7479 fragments["pre_main"] =
7480 "%c_f32_255 = OpConstant %f32 255.0\n"
7481 "%c_i32_128 = OpConstant %i32 128\n"
7482 "%c_i32_255 = OpConstant %i32 255\n"
7483 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
7484 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
7485 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
7487 // The test does the following.
7488 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
7489 // ivec4 result = ivec4(srem(ints.x, ints.y), srem(ints.y, ints.z), srem(ints.z, ints.x), 255);
7490 // return float(result + 128) / 255.0;
7491 fragments["testfun"] =
7492 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7493 "%param1 = OpFunctionParameter %v4f32\n"
7494 "%label_testfun = OpLabel\n"
7495 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
7496 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
7497 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
7498 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
7499 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
7500 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
7501 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
7502 "%x_out = OpSRem %i32 %x_in %y_in\n"
7503 "%y_out = OpSRem %i32 %y_in %z_in\n"
7504 "%z_out = OpSRem %i32 %z_in %x_in\n"
7505 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
7506 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
7507 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
7508 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
7509 "OpReturnValue %float_out\n"
7512 const struct CaseParams
7515 const char* failMessageTemplate; // customized status message
7516 qpTestResult failResult; // override status on failure
7517 int operands[4][3]; // four (x, y, z) vectors of operands
7518 int results[4][3]; // four (x, y, z) vectors of results
7524 QP_TEST_RESULT_FAIL,
7525 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
7526 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
7530 "Inconsistent results, but within specification: ${reason}",
7531 negFailResult, // negative operands, not required by the spec
7532 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
7533 { { 5, 12, -2 }, { 0, -5, 2 }, { 3, 8, -6 }, { 25, -60, 0 } }, // results
7536 // If either operand is negative the result is undefined. Some implementations may still return correct values.
7538 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
7540 const CaseParams& params = cases[caseNdx];
7541 RGBA inputColors[4];
7542 RGBA outputColors[4];
7544 for (int i = 0; i < 4; ++i)
7546 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
7547 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
7550 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
7553 return testGroup.release();
7556 // Test for the OpSMod instruction.
7557 tcu::TestCaseGroup* createOpSModGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
7559 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "smod", "OpSMod"));
7560 map<string, string> fragments;
7562 fragments["pre_main"] =
7563 "%c_f32_255 = OpConstant %f32 255.0\n"
7564 "%c_i32_128 = OpConstant %i32 128\n"
7565 "%c_i32_255 = OpConstant %i32 255\n"
7566 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
7567 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
7568 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
7570 // The test does the following.
7571 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
7572 // ivec4 result = ivec4(smod(ints.x, ints.y), smod(ints.y, ints.z), smod(ints.z, ints.x), 255);
7573 // return float(result + 128) / 255.0;
7574 fragments["testfun"] =
7575 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7576 "%param1 = OpFunctionParameter %v4f32\n"
7577 "%label_testfun = OpLabel\n"
7578 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
7579 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
7580 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
7581 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
7582 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
7583 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
7584 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
7585 "%x_out = OpSMod %i32 %x_in %y_in\n"
7586 "%y_out = OpSMod %i32 %y_in %z_in\n"
7587 "%z_out = OpSMod %i32 %z_in %x_in\n"
7588 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
7589 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
7590 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
7591 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
7592 "OpReturnValue %float_out\n"
7595 const struct CaseParams
7598 const char* failMessageTemplate; // customized status message
7599 qpTestResult failResult; // override status on failure
7600 int operands[4][3]; // four (x, y, z) vectors of operands
7601 int results[4][3]; // four (x, y, z) vectors of results
7607 QP_TEST_RESULT_FAIL,
7608 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
7609 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
7613 "Inconsistent results, but within specification: ${reason}",
7614 negFailResult, // negative operands, not required by the spec
7615 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
7616 { { 5, -5, 3 }, { 0, 2, -3 }, { 3, -73, 69 }, { -35, 40, 0 } }, // results
7619 // If either operand is negative the result is undefined. Some implementations may still return correct values.
7621 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
7623 const CaseParams& params = cases[caseNdx];
7624 RGBA inputColors[4];
7625 RGBA outputColors[4];
7627 for (int i = 0; i < 4; ++i)
7629 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
7630 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
7633 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
7636 return testGroup.release();
7641 INTEGER_TYPE_SIGNED_16,
7642 INTEGER_TYPE_SIGNED_32,
7643 INTEGER_TYPE_SIGNED_64,
7645 INTEGER_TYPE_UNSIGNED_16,
7646 INTEGER_TYPE_UNSIGNED_32,
7647 INTEGER_TYPE_UNSIGNED_64,
7650 const string getBitWidthStr (IntegerType type)
7654 case INTEGER_TYPE_SIGNED_16:
7655 case INTEGER_TYPE_UNSIGNED_16: return "16";
7657 case INTEGER_TYPE_SIGNED_32:
7658 case INTEGER_TYPE_UNSIGNED_32: return "32";
7660 case INTEGER_TYPE_SIGNED_64:
7661 case INTEGER_TYPE_UNSIGNED_64: return "64";
7663 default: DE_ASSERT(false);
7668 const string getByteWidthStr (IntegerType type)
7672 case INTEGER_TYPE_SIGNED_16:
7673 case INTEGER_TYPE_UNSIGNED_16: return "2";
7675 case INTEGER_TYPE_SIGNED_32:
7676 case INTEGER_TYPE_UNSIGNED_32: return "4";
7678 case INTEGER_TYPE_SIGNED_64:
7679 case INTEGER_TYPE_UNSIGNED_64: return "8";
7681 default: DE_ASSERT(false);
7686 bool isSigned (IntegerType type)
7688 return (type <= INTEGER_TYPE_SIGNED_64);
7691 const string getTypeName (IntegerType type)
7693 string prefix = isSigned(type) ? "" : "u";
7694 return prefix + "int" + getBitWidthStr(type);
7697 const string getTestName (IntegerType from, IntegerType to)
7699 return getTypeName(from) + "_to_" + getTypeName(to);
7702 const string getAsmTypeDeclaration (IntegerType type)
7704 string sign = isSigned(type) ? " 1" : " 0";
7705 return "OpTypeInt " + getBitWidthStr(type) + sign;
7708 const string getAsmTypeName (IntegerType type)
7710 const string prefix = isSigned(type) ? "%i" : "%u";
7711 return prefix + getBitWidthStr(type);
7714 template<typename T>
7715 BufferSp getSpecializedBuffer (deInt64 number)
7717 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
7720 BufferSp getBuffer (IntegerType type, deInt64 number)
7724 case INTEGER_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
7725 case INTEGER_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
7726 case INTEGER_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
7728 case INTEGER_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
7729 case INTEGER_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
7730 case INTEGER_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
7732 default: DE_ASSERT(false);
7733 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
7737 bool usesInt16 (IntegerType from, IntegerType to)
7739 return (from == INTEGER_TYPE_SIGNED_16 || from == INTEGER_TYPE_UNSIGNED_16
7740 || to == INTEGER_TYPE_SIGNED_16 || to == INTEGER_TYPE_UNSIGNED_16);
7743 bool usesInt64 (IntegerType from, IntegerType to)
7745 return (from == INTEGER_TYPE_SIGNED_64 || from == INTEGER_TYPE_UNSIGNED_64
7746 || to == INTEGER_TYPE_SIGNED_64 || to == INTEGER_TYPE_UNSIGNED_64);
7749 ComputeTestFeatures getConversionUsedFeatures (IntegerType from, IntegerType to)
7751 if (usesInt16(from, to))
7753 if (usesInt64(from, to))
7755 return COMPUTE_TEST_USES_INT16_INT64;
7759 return COMPUTE_TEST_USES_INT16;
7764 return COMPUTE_TEST_USES_INT64;
7770 ConvertCase (IntegerType from, IntegerType to, deInt64 number)
7773 , m_features (getConversionUsedFeatures(from, to))
7774 , m_name (getTestName(from, to))
7775 , m_inputBuffer (getBuffer(from, number))
7776 , m_outputBuffer (getBuffer(to, number))
7778 m_asmTypes["inputType"] = getAsmTypeName(from);
7779 m_asmTypes["outputType"] = getAsmTypeName(to);
7781 if (m_features == COMPUTE_TEST_USES_INT16)
7783 m_asmTypes["int_capabilities"] = "OpCapability Int16\n"
7784 "OpCapability StorageUniformBufferBlock16\n";
7785 m_asmTypes["int_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
7786 "%u16 = OpTypeInt 16 0\n";
7787 m_asmTypes["int_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
7789 else if (m_features == COMPUTE_TEST_USES_INT64)
7791 m_asmTypes["int_capabilities"] = "OpCapability Int64\n";
7792 m_asmTypes["int_additional_decl"] = "%i64 = OpTypeInt 64 1\n"
7793 "%u64 = OpTypeInt 64 0\n";
7794 m_asmTypes["int_extensions"] = "";
7796 else if (m_features == COMPUTE_TEST_USES_INT16_INT64)
7798 m_asmTypes["int_capabilities"] = "OpCapability Int16\n"
7799 "OpCapability StorageUniformBufferBlock16\n"
7800 "OpCapability Int64\n";
7801 m_asmTypes["int_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
7802 "%u16 = OpTypeInt 16 0\n"
7803 "%i64 = OpTypeInt 64 1\n"
7804 "%u64 = OpTypeInt 64 0\n";
7805 m_asmTypes["int_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
7813 IntegerType m_fromType;
7814 IntegerType m_toType;
7815 ComputeTestFeatures m_features;
7817 map<string, string> m_asmTypes;
7818 BufferSp m_inputBuffer;
7819 BufferSp m_outputBuffer;
7822 const string getConvertCaseShaderStr (const string& instruction, const ConvertCase& convertCase)
7824 map<string, string> params = convertCase.m_asmTypes;
7826 params["instruction"] = instruction;
7828 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
7829 params["outDecorator"] = getByteWidthStr(convertCase.m_toType);
7831 const StringTemplate shader (
7832 "OpCapability Shader\n"
7833 "${int_capabilities}"
7835 "OpMemoryModel Logical GLSL450\n"
7836 "OpEntryPoint GLCompute %main \"main\" %id\n"
7837 "OpExecutionMode %main LocalSize 1 1 1\n"
7838 "OpSource GLSL 430\n"
7839 "OpName %main \"main\"\n"
7840 "OpName %id \"gl_GlobalInvocationID\"\n"
7842 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7843 "OpDecorate %indata DescriptorSet 0\n"
7844 "OpDecorate %indata Binding 0\n"
7845 "OpDecorate %outdata DescriptorSet 0\n"
7846 "OpDecorate %outdata Binding 1\n"
7847 "OpDecorate %in_arr ArrayStride ${inDecorator}\n"
7848 "OpDecorate %out_arr ArrayStride ${outDecorator}\n"
7849 "OpDecorate %in_buf BufferBlock\n"
7850 "OpDecorate %out_buf BufferBlock\n"
7851 "OpMemberDecorate %in_buf 0 Offset 0\n"
7852 "OpMemberDecorate %out_buf 0 Offset 0\n"
7854 "%void = OpTypeVoid\n"
7855 "%voidf = OpTypeFunction %void\n"
7856 "%u32 = OpTypeInt 32 0\n"
7857 "%i32 = OpTypeInt 32 1\n"
7858 "${int_additional_decl}"
7859 "%uvec3 = OpTypeVector %u32 3\n"
7860 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7862 "%in_ptr = OpTypePointer Uniform ${inputType}\n"
7863 "%out_ptr = OpTypePointer Uniform ${outputType}\n"
7864 "%in_arr = OpTypeRuntimeArray ${inputType}\n"
7865 "%out_arr = OpTypeRuntimeArray ${outputType}\n"
7866 "%in_buf = OpTypeStruct %in_arr\n"
7867 "%out_buf = OpTypeStruct %out_arr\n"
7868 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
7869 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
7870 "%indata = OpVariable %in_bufptr Uniform\n"
7871 "%outdata = OpVariable %out_bufptr Uniform\n"
7872 "%inputptr = OpTypePointer Input ${inputType}\n"
7873 "%id = OpVariable %uvec3ptr Input\n"
7875 "%zero = OpConstant %i32 0\n"
7877 "%main = OpFunction %void None %voidf\n"
7878 "%label = OpLabel\n"
7879 "%idval = OpLoad %uvec3 %id\n"
7880 "%x = OpCompositeExtract %u32 %idval 0\n"
7881 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
7882 "%outloc = OpAccessChain %out_ptr %outdata %zero %x\n"
7883 "%inval = OpLoad ${inputType} %inloc\n"
7884 "%conv = ${instruction} ${outputType} %inval\n"
7885 " OpStore %outloc %conv\n"
7890 return shader.specialize(params);
7893 void createSConvertCases (vector<ConvertCase>& testCases)
7895 // Convert int to int
7896 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_32, 14669));
7897 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_64, 3341));
7899 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_SIGNED_64, 973610259));
7901 // Convert int to unsigned int
7902 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_32, 9288));
7903 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_64, 15460));
7905 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_UNSIGNED_64, 346213461));
7908 // Test for the OpSConvert instruction.
7909 tcu::TestCaseGroup* createSConvertTests (tcu::TestContext& testCtx)
7911 const string instruction ("OpSConvert");
7912 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "sconvert", "OpSConvert"));
7913 vector<ConvertCase> testCases;
7914 createSConvertCases(testCases);
7916 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7918 ComputeShaderSpec spec;
7920 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7921 spec.inputs.push_back(test->m_inputBuffer);
7922 spec.outputs.push_back(test->m_outputBuffer);
7923 spec.numWorkGroups = IVec3(1, 1, 1);
7925 if (test->m_features == COMPUTE_TEST_USES_INT16 || test->m_features == COMPUTE_TEST_USES_INT16_INT64)
7927 spec.extensions.push_back("VK_KHR_16bit_storage");
7930 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpSConvert.", spec, test->m_features));
7933 return group.release();
7936 void createUConvertCases (vector<ConvertCase>& testCases)
7938 // Convert unsigned int to unsigned int
7939 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_32, 60653));
7940 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_64, 17991));
7942 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_UNSIGNED_64, 904256275));
7944 // Convert unsigned int to int
7945 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_32, 38002));
7946 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_64, 64921));
7948 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_SIGNED_64, 4294956295ll));
7951 // Test for the OpUConvert instruction.
7952 tcu::TestCaseGroup* createUConvertTests (tcu::TestContext& testCtx)
7954 const string instruction ("OpUConvert");
7955 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "uconvert", "OpUConvert"));
7956 vector<ConvertCase> testCases;
7957 createUConvertCases(testCases);
7959 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7961 ComputeShaderSpec spec;
7963 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7964 spec.inputs.push_back(test->m_inputBuffer);
7965 spec.outputs.push_back(test->m_outputBuffer);
7966 spec.numWorkGroups = IVec3(1, 1, 1);
7968 if (test->m_features == COMPUTE_TEST_USES_INT16 || test->m_features == COMPUTE_TEST_USES_INT16_INT64)
7970 spec.extensions.push_back("VK_KHR_16bit_storage");
7973 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpUConvert.", spec, test->m_features));
7975 return group.release();
7978 const string getNumberTypeName (const NumberType type)
7980 if (type == NUMBERTYPE_INT32)
7984 else if (type == NUMBERTYPE_UINT32)
7988 else if (type == NUMBERTYPE_FLOAT32)
7999 deInt32 getInt(de::Random& rnd)
8001 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
8004 const string repeatString (const string& str, int times)
8007 for (int i = 0; i < times; ++i)
8014 const string getRandomConstantString (const NumberType type, de::Random& rnd)
8016 if (type == NUMBERTYPE_INT32)
8018 return numberToString<deInt32>(getInt(rnd));
8020 else if (type == NUMBERTYPE_UINT32)
8022 return numberToString<deUint32>(rnd.getUint32());
8024 else if (type == NUMBERTYPE_FLOAT32)
8026 return numberToString<float>(rnd.getFloat());
8035 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8037 map<string, string> params;
8040 for (int width = 2; width <= 4; ++width)
8042 const string randomConst = numberToString(getInt(rnd));
8043 const string widthStr = numberToString(width);
8044 const string composite_type = "${customType}vec" + widthStr;
8045 const int index = rnd.getInt(0, width-1);
8047 params["type"] = "vec";
8048 params["name"] = params["type"] + "_" + widthStr;
8049 params["compositeDecl"] = composite_type + " = OpTypeVector ${customType} " + widthStr +"\n";
8050 params["compositeType"] = composite_type;
8051 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8052 params["compositeConstruct"] = "%instance = OpCompositeConstruct " + composite_type + repeatString(" %filler", width) + "\n";
8053 params["indexes"] = numberToString(index);
8054 testCases.push_back(params);
8058 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8060 const int limit = 10;
8061 map<string, string> params;
8063 for (int width = 2; width <= limit; ++width)
8065 string randomConst = numberToString(getInt(rnd));
8066 string widthStr = numberToString(width);
8067 int index = rnd.getInt(0, width-1);
8069 params["type"] = "array";
8070 params["name"] = params["type"] + "_" + widthStr;
8071 params["compositeDecl"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
8072 + "%composite = OpTypeArray ${customType} %arraywidth\n";
8073 params["compositeType"] = "%composite";
8074 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8075 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8076 params["indexes"] = numberToString(index);
8077 testCases.push_back(params);
8081 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8083 const int limit = 10;
8084 map<string, string> params;
8086 for (int width = 2; width <= limit; ++width)
8088 string randomConst = numberToString(getInt(rnd));
8089 int index = rnd.getInt(0, width-1);
8091 params["type"] = "struct";
8092 params["name"] = params["type"] + "_" + numberToString(width);
8093 params["compositeDecl"] = "%composite = OpTypeStruct" + repeatString(" ${customType}", width) + "\n";
8094 params["compositeType"] = "%composite";
8095 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8096 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8097 params["indexes"] = numberToString(index);
8098 testCases.push_back(params);
8102 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8104 map<string, string> params;
8107 for (int width = 2; width <= 4; ++width)
8109 string widthStr = numberToString(width);
8111 for (int column = 2 ; column <= 4; ++column)
8113 int index_0 = rnd.getInt(0, column-1);
8114 int index_1 = rnd.getInt(0, width-1);
8115 string columnStr = numberToString(column);
8117 params["type"] = "matrix";
8118 params["name"] = params["type"] + "_" + widthStr + "x" + columnStr;
8119 params["compositeDecl"] = string("%vectype = OpTypeVector ${customType} " + widthStr + "\n")
8120 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
8121 params["compositeType"] = "%composite";
8123 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n"
8124 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
8126 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
8127 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
8128 testCases.push_back(params);
8133 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8135 createVectorCompositeCases(testCases, rnd, type);
8136 createArrayCompositeCases(testCases, rnd, type);
8137 createStructCompositeCases(testCases, rnd, type);
8138 // Matrix only supports float types
8139 if (type == NUMBERTYPE_FLOAT32)
8141 createMatrixCompositeCases(testCases, rnd, type);
8145 const string getAssemblyTypeDeclaration (const NumberType type)
8149 case NUMBERTYPE_INT32: return "OpTypeInt 32 1";
8150 case NUMBERTYPE_UINT32: return "OpTypeInt 32 0";
8151 case NUMBERTYPE_FLOAT32: return "OpTypeFloat 32";
8152 default: DE_ASSERT(false); return "";
8156 const string getAssemblyTypeName (const NumberType type)
8160 case NUMBERTYPE_INT32: return "%i32";
8161 case NUMBERTYPE_UINT32: return "%u32";
8162 case NUMBERTYPE_FLOAT32: return "%f32";
8163 default: DE_ASSERT(false); return "";
8167 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
8169 map<string, string> parameters(params);
8171 const string customType = getAssemblyTypeName(type);
8172 map<string, string> substCustomType;
8173 substCustomType["customType"] = customType;
8174 parameters["compositeDecl"] = StringTemplate(parameters.at("compositeDecl")).specialize(substCustomType);
8175 parameters["compositeType"] = StringTemplate(parameters.at("compositeType")).specialize(substCustomType);
8176 parameters["compositeConstruct"] = StringTemplate(parameters.at("compositeConstruct")).specialize(substCustomType);
8177 parameters["filler"] = StringTemplate(parameters.at("filler")).specialize(substCustomType);
8178 parameters["customType"] = customType;
8179 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
8181 if (parameters.at("compositeType") != "%u32vec3")
8183 parameters["u32vec3Decl"] = "%u32vec3 = OpTypeVector %u32 3\n";
8186 return StringTemplate(
8187 "OpCapability Shader\n"
8188 "OpCapability Matrix\n"
8189 "OpMemoryModel Logical GLSL450\n"
8190 "OpEntryPoint GLCompute %main \"main\" %id\n"
8191 "OpExecutionMode %main LocalSize 1 1 1\n"
8193 "OpSource GLSL 430\n"
8194 "OpName %main \"main\"\n"
8195 "OpName %id \"gl_GlobalInvocationID\"\n"
8198 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8199 "OpDecorate %buf BufferBlock\n"
8200 "OpDecorate %indata DescriptorSet 0\n"
8201 "OpDecorate %indata Binding 0\n"
8202 "OpDecorate %outdata DescriptorSet 0\n"
8203 "OpDecorate %outdata Binding 1\n"
8204 "OpDecorate %customarr ArrayStride 4\n"
8205 "${compositeDecorator}"
8206 "OpMemberDecorate %buf 0 Offset 0\n"
8209 "%void = OpTypeVoid\n"
8210 "%voidf = OpTypeFunction %void\n"
8211 "%u32 = OpTypeInt 32 0\n"
8212 "%i32 = OpTypeInt 32 1\n"
8213 "%f32 = OpTypeFloat 32\n"
8215 // Composite declaration
8221 "${u32vec3Decl:opt}"
8222 "%uvec3ptr = OpTypePointer Input %u32vec3\n"
8224 // Inherited from custom
8225 "%customptr = OpTypePointer Uniform ${customType}\n"
8226 "%customarr = OpTypeRuntimeArray ${customType}\n"
8227 "%buf = OpTypeStruct %customarr\n"
8228 "%bufptr = OpTypePointer Uniform %buf\n"
8230 "%indata = OpVariable %bufptr Uniform\n"
8231 "%outdata = OpVariable %bufptr Uniform\n"
8233 "%id = OpVariable %uvec3ptr Input\n"
8234 "%zero = OpConstant %i32 0\n"
8236 "%main = OpFunction %void None %voidf\n"
8237 "%label = OpLabel\n"
8238 "%idval = OpLoad %u32vec3 %id\n"
8239 "%x = OpCompositeExtract %u32 %idval 0\n"
8241 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
8242 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
8243 // Read the input value
8244 "%inval = OpLoad ${customType} %inloc\n"
8245 // Create the composite and fill it
8246 "${compositeConstruct}"
8247 // Insert the input value to a place
8248 "%instance2 = OpCompositeInsert ${compositeType} %inval %instance ${indexes}\n"
8249 // Read back the value from the position
8250 "%out_val = OpCompositeExtract ${customType} %instance2 ${indexes}\n"
8251 // Store it in the output position
8252 " OpStore %outloc %out_val\n"
8255 ).specialize(parameters);
8258 template<typename T>
8259 BufferSp createCompositeBuffer(T number)
8261 return BufferSp(new Buffer<T>(vector<T>(1, number)));
8264 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
8266 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
8267 de::Random rnd (deStringHash(group->getName()));
8269 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8271 NumberType numberType = NumberType(type);
8272 const string typeName = getNumberTypeName(numberType);
8273 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
8274 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8275 vector<map<string, string> > testCases;
8277 createCompositeCases(testCases, rnd, numberType);
8279 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8281 ComputeShaderSpec spec;
8283 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
8287 case NUMBERTYPE_INT32:
8289 deInt32 number = getInt(rnd);
8290 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8291 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8294 case NUMBERTYPE_UINT32:
8296 deUint32 number = rnd.getUint32();
8297 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8298 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8301 case NUMBERTYPE_FLOAT32:
8303 float number = rnd.getFloat();
8304 spec.inputs.push_back(createCompositeBuffer<float>(number));
8305 spec.outputs.push_back(createCompositeBuffer<float>(number));
8312 spec.numWorkGroups = IVec3(1, 1, 1);
8313 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
8315 group->addChild(subGroup.release());
8317 return group.release();
8320 struct AssemblyStructInfo
8322 AssemblyStructInfo (const deUint32 comp, const deUint32 idx)
8327 deUint32 components;
8331 const string specializeInBoundsShaderTemplate (const NumberType type, const AssemblyStructInfo& structInfo, const map<string, string>& params)
8333 // Create the full index string
8334 string fullIndex = numberToString(structInfo.index) + " " + params.at("indexes");
8335 // Convert it to list of indexes
8336 vector<string> indexes = de::splitString(fullIndex, ' ');
8338 map<string, string> parameters (params);
8339 parameters["structType"] = repeatString(" ${compositeType}", structInfo.components);
8340 parameters["structConstruct"] = repeatString(" %instance", structInfo.components);
8341 parameters["insertIndexes"] = fullIndex;
8343 // In matrix cases the last two index is the CompositeExtract indexes
8344 const deUint32 extractIndexes = (parameters["type"] == "matrix") ? 2 : 1;
8346 // Construct the extractIndex
8347 for (vector<string>::const_iterator index = indexes.end() - extractIndexes; index != indexes.end(); ++index)
8349 parameters["extractIndexes"] += " " + *index;
8352 // Remove the last 1 or 2 element depends on matrix case or not
8353 indexes.erase(indexes.end() - extractIndexes, indexes.end());
8356 // Generate AccessChain index expressions (except for the last one, because we use ptr to the composite)
8357 for (vector<string>::const_iterator index = indexes.begin(); index != indexes.end(); ++index)
8359 string indexId = "%index_" + numberToString(id++);
8360 parameters["accessChainConstDeclaration"] += indexId + " = OpConstant %u32 " + *index + "\n";
8361 parameters["accessChainIndexes"] += " " + indexId;
8364 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
8366 const string customType = getAssemblyTypeName(type);
8367 map<string, string> substCustomType;
8368 substCustomType["customType"] = customType;
8369 parameters["compositeDecl"] = StringTemplate(parameters.at("compositeDecl")).specialize(substCustomType);
8370 parameters["compositeType"] = StringTemplate(parameters.at("compositeType")).specialize(substCustomType);
8371 parameters["compositeConstruct"] = StringTemplate(parameters.at("compositeConstruct")).specialize(substCustomType);
8372 parameters["filler"] = StringTemplate(parameters.at("filler")).specialize(substCustomType);
8373 parameters["customType"] = customType;
8375 const string compositeType = parameters.at("compositeType");
8376 map<string, string> substCompositeType;
8377 substCompositeType["compositeType"] = compositeType;
8378 parameters["structType"] = StringTemplate(parameters.at("structType")).specialize(substCompositeType);
8379 if (compositeType != "%u32vec3")
8381 parameters["u32vec3Decl"] = "%u32vec3 = OpTypeVector %u32 3\n";
8384 return StringTemplate(
8385 "OpCapability Shader\n"
8386 "OpCapability Matrix\n"
8387 "OpMemoryModel Logical GLSL450\n"
8388 "OpEntryPoint GLCompute %main \"main\" %id\n"
8389 "OpExecutionMode %main LocalSize 1 1 1\n"
8391 "OpSource GLSL 430\n"
8392 "OpName %main \"main\"\n"
8393 "OpName %id \"gl_GlobalInvocationID\"\n"
8395 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8396 "OpDecorate %buf BufferBlock\n"
8397 "OpDecorate %indata DescriptorSet 0\n"
8398 "OpDecorate %indata Binding 0\n"
8399 "OpDecorate %outdata DescriptorSet 0\n"
8400 "OpDecorate %outdata Binding 1\n"
8401 "OpDecorate %customarr ArrayStride 4\n"
8402 "${compositeDecorator}"
8403 "OpMemberDecorate %buf 0 Offset 0\n"
8405 "%void = OpTypeVoid\n"
8406 "%voidf = OpTypeFunction %void\n"
8407 "%i32 = OpTypeInt 32 1\n"
8408 "%u32 = OpTypeInt 32 0\n"
8409 "%f32 = OpTypeFloat 32\n"
8412 // %u32vec3 if not already declared in ${compositeDecl}
8413 "${u32vec3Decl:opt}"
8414 "%uvec3ptr = OpTypePointer Input %u32vec3\n"
8415 // Inherited from composite
8416 "%composite_p = OpTypePointer Function ${compositeType}\n"
8417 "%struct_t = OpTypeStruct${structType}\n"
8418 "%struct_p = OpTypePointer Function %struct_t\n"
8421 "${accessChainConstDeclaration}"
8422 // Inherited from custom
8423 "%customptr = OpTypePointer Uniform ${customType}\n"
8424 "%customarr = OpTypeRuntimeArray ${customType}\n"
8425 "%buf = OpTypeStruct %customarr\n"
8426 "%bufptr = OpTypePointer Uniform %buf\n"
8427 "%indata = OpVariable %bufptr Uniform\n"
8428 "%outdata = OpVariable %bufptr Uniform\n"
8430 "%id = OpVariable %uvec3ptr Input\n"
8431 "%zero = OpConstant %u32 0\n"
8432 "%main = OpFunction %void None %voidf\n"
8433 "%label = OpLabel\n"
8434 "%struct_v = OpVariable %struct_p Function\n"
8435 "%idval = OpLoad %u32vec3 %id\n"
8436 "%x = OpCompositeExtract %u32 %idval 0\n"
8437 // Create the input/output type
8438 "%inloc = OpInBoundsAccessChain %customptr %indata %zero %x\n"
8439 "%outloc = OpInBoundsAccessChain %customptr %outdata %zero %x\n"
8440 // Read the input value
8441 "%inval = OpLoad ${customType} %inloc\n"
8442 // Create the composite and fill it
8443 "${compositeConstruct}"
8444 // Create the struct and fill it with the composite
8445 "%struct = OpCompositeConstruct %struct_t${structConstruct}\n"
8447 "%comp_obj = OpCompositeInsert %struct_t %inval %struct ${insertIndexes}\n"
8449 " OpStore %struct_v %comp_obj\n"
8450 // Get deepest possible composite pointer
8451 "%inner_ptr = OpInBoundsAccessChain %composite_p %struct_v${accessChainIndexes}\n"
8452 "%read_obj = OpLoad ${compositeType} %inner_ptr\n"
8453 // Read back the stored value
8454 "%read_val = OpCompositeExtract ${customType} %read_obj${extractIndexes}\n"
8455 " OpStore %outloc %read_val\n"
8458 ).specialize(parameters);
8461 tcu::TestCaseGroup* createOpInBoundsAccessChainGroup (tcu::TestContext& testCtx)
8463 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opinboundsaccesschain", "Test the OpInBoundsAccessChain instruction"));
8464 de::Random rnd (deStringHash(group->getName()));
8466 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8468 NumberType numberType = NumberType(type);
8469 const string typeName = getNumberTypeName(numberType);
8470 const string description = "Test the OpInBoundsAccessChain instruction with " + typeName + "s";
8471 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8473 vector<map<string, string> > testCases;
8474 createCompositeCases(testCases, rnd, numberType);
8476 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8478 ComputeShaderSpec spec;
8480 // Number of components inside of a struct
8481 deUint32 structComponents = rnd.getInt(2, 8);
8482 // Component index value
8483 deUint32 structIndex = rnd.getInt(0, structComponents - 1);
8484 AssemblyStructInfo structInfo(structComponents, structIndex);
8486 spec.assembly = specializeInBoundsShaderTemplate(numberType, structInfo, *test);
8490 case NUMBERTYPE_INT32:
8492 deInt32 number = getInt(rnd);
8493 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8494 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8497 case NUMBERTYPE_UINT32:
8499 deUint32 number = rnd.getUint32();
8500 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8501 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8504 case NUMBERTYPE_FLOAT32:
8506 float number = rnd.getFloat();
8507 spec.inputs.push_back(createCompositeBuffer<float>(number));
8508 spec.outputs.push_back(createCompositeBuffer<float>(number));
8514 spec.numWorkGroups = IVec3(1, 1, 1);
8515 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpInBoundsAccessChain test", spec));
8517 group->addChild(subGroup.release());
8519 return group.release();
8522 // If the params missing, uninitialized case
8523 const string specializeDefaultOutputShaderTemplate (const NumberType type, const map<string, string>& params = map<string, string>())
8525 map<string, string> parameters(params);
8527 parameters["customType"] = getAssemblyTypeName(type);
8529 // Declare the const value, and use it in the initializer
8530 if (params.find("constValue") != params.end())
8532 parameters["variableInitializer"] = " %const";
8534 // Uninitialized case
8537 parameters["commentDecl"] = ";";
8540 return StringTemplate(
8541 "OpCapability Shader\n"
8542 "OpMemoryModel Logical GLSL450\n"
8543 "OpEntryPoint GLCompute %main \"main\" %id\n"
8544 "OpExecutionMode %main LocalSize 1 1 1\n"
8545 "OpSource GLSL 430\n"
8546 "OpName %main \"main\"\n"
8547 "OpName %id \"gl_GlobalInvocationID\"\n"
8549 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8550 "OpDecorate %indata DescriptorSet 0\n"
8551 "OpDecorate %indata Binding 0\n"
8552 "OpDecorate %outdata DescriptorSet 0\n"
8553 "OpDecorate %outdata Binding 1\n"
8554 "OpDecorate %in_arr ArrayStride 4\n"
8555 "OpDecorate %in_buf BufferBlock\n"
8556 "OpMemberDecorate %in_buf 0 Offset 0\n"
8558 "%void = OpTypeVoid\n"
8559 "%voidf = OpTypeFunction %void\n"
8560 "%u32 = OpTypeInt 32 0\n"
8561 "%i32 = OpTypeInt 32 1\n"
8562 "%f32 = OpTypeFloat 32\n"
8563 "%uvec3 = OpTypeVector %u32 3\n"
8564 "%uvec3ptr = OpTypePointer Input %uvec3\n"
8565 "${commentDecl:opt}%const = OpConstant ${customType} ${constValue:opt}\n"
8567 "%in_ptr = OpTypePointer Uniform ${customType}\n"
8568 "%in_arr = OpTypeRuntimeArray ${customType}\n"
8569 "%in_buf = OpTypeStruct %in_arr\n"
8570 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
8571 "%indata = OpVariable %in_bufptr Uniform\n"
8572 "%outdata = OpVariable %in_bufptr Uniform\n"
8573 "%id = OpVariable %uvec3ptr Input\n"
8574 "%var_ptr = OpTypePointer Function ${customType}\n"
8576 "%zero = OpConstant %i32 0\n"
8578 "%main = OpFunction %void None %voidf\n"
8579 "%label = OpLabel\n"
8580 "%out_var = OpVariable %var_ptr Function${variableInitializer:opt}\n"
8581 "%idval = OpLoad %uvec3 %id\n"
8582 "%x = OpCompositeExtract %u32 %idval 0\n"
8583 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
8584 "%outloc = OpAccessChain %in_ptr %outdata %zero %x\n"
8586 "%outval = OpLoad ${customType} %out_var\n"
8587 " OpStore %outloc %outval\n"
8590 ).specialize(parameters);
8593 bool compareFloats (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
8595 DE_ASSERT(outputAllocs.size() != 0);
8596 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
8598 // Use custom epsilon because of the float->string conversion
8599 const float epsilon = 0.00001f;
8601 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
8603 vector<deUint8> expectedBytes;
8607 expectedOutputs[outputNdx]->getBytes(expectedBytes);
8608 memcpy(&expected, &expectedBytes.front(), expectedBytes.size());
8609 memcpy(&actual, outputAllocs[outputNdx]->getHostPtr(), expectedBytes.size());
8611 // Test with epsilon
8612 if (fabs(expected - actual) > epsilon)
8614 log << TestLog::Message << "Error: The actual and expected values not matching."
8615 << " Expected: " << expected << " Actual: " << actual << " Epsilon: " << epsilon << TestLog::EndMessage;
8622 // Checks if the driver crash with uninitialized cases
8623 bool passthruVerify (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
8625 DE_ASSERT(outputAllocs.size() != 0);
8626 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
8628 // Copy and discard the result.
8629 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
8631 vector<deUint8> expectedBytes;
8632 expectedOutputs[outputNdx]->getBytes(expectedBytes);
8634 const size_t width = expectedBytes.size();
8635 vector<char> data (width);
8637 memcpy(&data[0], outputAllocs[outputNdx]->getHostPtr(), width);
8642 tcu::TestCaseGroup* createShaderDefaultOutputGroup (tcu::TestContext& testCtx)
8644 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "shader_default_output", "Test shader default output."));
8645 de::Random rnd (deStringHash(group->getName()));
8647 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8649 NumberType numberType = NumberType(type);
8650 const string typeName = getNumberTypeName(numberType);
8651 const string description = "Test the OpVariable initializer with " + typeName + ".";
8652 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8654 // 2 similar subcases (initialized and uninitialized)
8655 for (int subCase = 0; subCase < 2; ++subCase)
8657 ComputeShaderSpec spec;
8658 spec.numWorkGroups = IVec3(1, 1, 1);
8660 map<string, string> params;
8664 case NUMBERTYPE_INT32:
8666 deInt32 number = getInt(rnd);
8667 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8668 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8669 params["constValue"] = numberToString(number);
8672 case NUMBERTYPE_UINT32:
8674 deUint32 number = rnd.getUint32();
8675 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8676 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8677 params["constValue"] = numberToString(number);
8680 case NUMBERTYPE_FLOAT32:
8682 float number = rnd.getFloat();
8683 spec.inputs.push_back(createCompositeBuffer<float>(number));
8684 spec.outputs.push_back(createCompositeBuffer<float>(number));
8685 spec.verifyIO = &compareFloats;
8686 params["constValue"] = numberToString(number);
8693 // Initialized subcase
8696 spec.assembly = specializeDefaultOutputShaderTemplate(numberType, params);
8697 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "initialized", "OpVariable initializer tests.", spec));
8699 // Uninitialized subcase
8702 spec.assembly = specializeDefaultOutputShaderTemplate(numberType);
8703 spec.verifyIO = &passthruVerify;
8704 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "uninitialized", "OpVariable initializer tests.", spec));
8707 group->addChild(subGroup.release());
8709 return group.release();
8712 tcu::TestCaseGroup* createOpNopTests (tcu::TestContext& testCtx)
8714 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
8715 RGBA defaultColors[4];
8716 map<string, string> opNopFragments;
8718 getDefaultColors(defaultColors);
8720 opNopFragments["testfun"] =
8721 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
8722 "%param1 = OpFunctionParameter %v4f32\n"
8723 "%label_testfun = OpLabel\n"
8732 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
8733 "%b = OpFAdd %f32 %a %a\n"
8735 "%c = OpFSub %f32 %b %a\n"
8736 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
8739 "OpReturnValue %ret\n"
8742 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, testGroup.get());
8744 return testGroup.release();
8747 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
8749 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
8750 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
8751 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
8753 computeTests->addChild(createLocalSizeGroup(testCtx));
8754 computeTests->addChild(createOpNopGroup(testCtx));
8755 computeTests->addChild(createOpFUnordGroup(testCtx));
8756 computeTests->addChild(createOpAtomicGroup(testCtx, false));
8757 computeTests->addChild(createOpAtomicGroup(testCtx, true)); // Using new StorageBuffer decoration
8758 computeTests->addChild(createOpLineGroup(testCtx));
8759 computeTests->addChild(createOpNoLineGroup(testCtx));
8760 computeTests->addChild(createOpConstantNullGroup(testCtx));
8761 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
8762 computeTests->addChild(createOpConstantUsageGroup(testCtx));
8763 computeTests->addChild(createSpecConstantGroup(testCtx));
8764 computeTests->addChild(createOpSourceGroup(testCtx));
8765 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
8766 computeTests->addChild(createDecorationGroupGroup(testCtx));
8767 computeTests->addChild(createOpPhiGroup(testCtx));
8768 computeTests->addChild(createLoopControlGroup(testCtx));
8769 computeTests->addChild(createFunctionControlGroup(testCtx));
8770 computeTests->addChild(createSelectionControlGroup(testCtx));
8771 computeTests->addChild(createBlockOrderGroup(testCtx));
8772 computeTests->addChild(createMultipleShaderGroup(testCtx));
8773 computeTests->addChild(createMemoryAccessGroup(testCtx));
8774 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
8775 computeTests->addChild(createOpCopyObjectGroup(testCtx));
8776 computeTests->addChild(createNoContractionGroup(testCtx));
8777 computeTests->addChild(createOpUndefGroup(testCtx));
8778 computeTests->addChild(createOpUnreachableGroup(testCtx));
8779 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
8780 computeTests ->addChild(createOpFRemGroup(testCtx));
8781 computeTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_PASS));
8782 computeTests->addChild(createOpSRemComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
8783 computeTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_PASS));
8784 computeTests->addChild(createOpSModComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
8785 computeTests->addChild(createSConvertTests(testCtx));
8786 computeTests->addChild(createUConvertTests(testCtx));
8787 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
8788 computeTests->addChild(createOpInBoundsAccessChainGroup(testCtx));
8789 computeTests->addChild(createShaderDefaultOutputGroup(testCtx));
8790 computeTests->addChild(createOpNMinGroup(testCtx));
8791 computeTests->addChild(createOpNMaxGroup(testCtx));
8792 computeTests->addChild(createOpNClampGroup(testCtx));
8794 de::MovePtr<tcu::TestCaseGroup> computeAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
8796 computeAndroidTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
8797 computeAndroidTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
8799 computeTests->addChild(computeAndroidTests.release());
8802 computeTests->addChild(create16BitStorageComputeGroup(testCtx));
8803 computeTests->addChild(createUboMatrixPaddingComputeGroup(testCtx));
8804 computeTests->addChild(createConditionalBranchComputeGroup(testCtx));
8805 computeTests->addChild(createIndexingComputeGroup(testCtx));
8806 computeTests->addChild(createVariablePointersComputeGroup(testCtx));
8807 computeTests->addChild(createImageSamplerComputeGroup(testCtx));
8808 graphicsTests->addChild(createOpNopTests(testCtx));
8809 graphicsTests->addChild(createOpSourceTests(testCtx));
8810 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
8811 graphicsTests->addChild(createOpLineTests(testCtx));
8812 graphicsTests->addChild(createOpNoLineTests(testCtx));
8813 graphicsTests->addChild(createOpConstantNullTests(testCtx));
8814 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
8815 graphicsTests->addChild(createMemoryAccessTests(testCtx));
8816 graphicsTests->addChild(createOpUndefTests(testCtx));
8817 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
8818 graphicsTests->addChild(createModuleTests(testCtx));
8819 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
8820 graphicsTests->addChild(createOpPhiTests(testCtx));
8821 graphicsTests->addChild(createNoContractionTests(testCtx));
8822 graphicsTests->addChild(createOpQuantizeTests(testCtx));
8823 graphicsTests->addChild(createLoopTests(testCtx));
8824 graphicsTests->addChild(createSpecConstantTests(testCtx));
8825 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
8826 graphicsTests->addChild(createBarrierTests(testCtx));
8827 graphicsTests->addChild(createDecorationGroupTests(testCtx));
8828 graphicsTests->addChild(createFRemTests(testCtx));
8829 graphicsTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
8830 graphicsTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
8833 de::MovePtr<tcu::TestCaseGroup> graphicsAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
8835 graphicsAndroidTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
8836 graphicsAndroidTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
8838 graphicsTests->addChild(graphicsAndroidTests.release());
8841 graphicsTests->addChild(create16BitStorageGraphicsGroup(testCtx));
8842 graphicsTests->addChild(createUboMatrixPaddingGraphicsGroup(testCtx));
8843 graphicsTests->addChild(createConditionalBranchGraphicsGroup(testCtx));
8844 graphicsTests->addChild(createIndexingGraphicsGroup(testCtx));
8845 graphicsTests->addChild(createVariablePointersGraphicsGroup(testCtx));
8846 graphicsTests->addChild(createImageSamplerGraphicsGroup(testCtx));
8848 instructionTests->addChild(computeTests.release());
8849 instructionTests->addChild(graphicsTests.release());
8851 return instructionTests.release();