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 "vktSpvAsmSpirvVersionTests.hpp"
62 #include "vktTestCaseUtil.hpp"
63 #include "vktSpvAsmLoopDepLenTests.hpp"
64 #include "vktSpvAsmLoopDepInfTests.hpp"
76 namespace SpirVAssembly
90 using tcu::TestStatus;
93 using tcu::StringTemplate;
97 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0)
99 T* const typedPtr = (T*)dst;
100 for (int ndx = 0; ndx < numValues; ndx++)
101 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue);
104 // Filter is a function that returns true if a value should pass, false otherwise.
105 template<typename T, typename FilterT>
106 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, FilterT filter, int offset = 0)
108 T* const typedPtr = (T*)dst;
110 for (int ndx = 0; ndx < numValues; ndx++)
113 value = randomScalar<T>(rnd, minValue, maxValue);
114 while (!filter(value));
116 typedPtr[offset + ndx] = value;
120 // Gets a 64-bit integer with a more logarithmic distribution
121 deInt64 randomInt64LogDistributed (de::Random& rnd)
123 deInt64 val = rnd.getUint64();
124 val &= (1ull << rnd.getInt(1, 63)) - 1;
130 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues)
132 for (int ndx = 0; ndx < numValues; ndx++)
133 dst[ndx] = randomInt64LogDistributed(rnd);
136 template<typename FilterT>
137 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues, FilterT filter)
139 for (int ndx = 0; ndx < numValues; ndx++)
143 value = randomInt64LogDistributed(rnd);
144 } while (!filter(value));
149 inline bool filterNonNegative (const deInt64 value)
154 inline bool filterPositive (const deInt64 value)
159 inline bool filterNotZero (const deInt64 value)
164 static void floorAll (vector<float>& values)
166 for (size_t i = 0; i < values.size(); i++)
167 values[i] = deFloatFloor(values[i]);
170 static void floorAll (vector<Vec4>& values)
172 for (size_t i = 0; i < values.size(); i++)
173 values[i] = floor(values[i]);
181 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {}
184 // Assembly code used for testing LocalSize, OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code:
188 // layout(std140, set = 0, binding = 0) readonly buffer Input {
191 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
195 // layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
198 // uint x = gl_GlobalInvocationID.x;
199 // output_data.elements[x] = -input_data.elements[x];
202 static string getAsmForLocalSizeTest(bool useLiteralLocalSize, bool useSpecConstantWorkgroupSize, IVec3 workGroupSize, deUint32 ndx)
204 std::ostringstream out;
205 out << getComputeAsmShaderPreambleWithoutLocalSize();
207 if (useLiteralLocalSize)
209 out << "OpExecutionMode %main LocalSize "
210 << workGroupSize.x() << " " << workGroupSize.y() << " " << workGroupSize.z() << "\n";
213 out << "OpSource GLSL 430\n"
214 "OpName %main \"main\"\n"
215 "OpName %id \"gl_GlobalInvocationID\"\n"
216 "OpDecorate %id BuiltIn GlobalInvocationId\n";
218 if (useSpecConstantWorkgroupSize)
220 out << "OpDecorate %spec_0 SpecId 100\n"
221 << "OpDecorate %spec_1 SpecId 101\n"
222 << "OpDecorate %spec_2 SpecId 102\n"
223 << "OpDecorate %gl_WorkGroupSize BuiltIn WorkgroupSize\n";
226 out << getComputeAsmInputOutputBufferTraits()
227 << getComputeAsmCommonTypes()
228 << getComputeAsmInputOutputBuffer()
229 << "%id = OpVariable %uvec3ptr Input\n"
230 << "%zero = OpConstant %i32 0 \n";
232 if (useSpecConstantWorkgroupSize)
234 out << "%spec_0 = OpSpecConstant %u32 "<< workGroupSize.x() << "\n"
235 << "%spec_1 = OpSpecConstant %u32 "<< workGroupSize.y() << "\n"
236 << "%spec_2 = OpSpecConstant %u32 "<< workGroupSize.z() << "\n"
237 << "%gl_WorkGroupSize = OpSpecConstantComposite %uvec3 %spec_0 %spec_1 %spec_2\n";
240 out << "%main = OpFunction %void None %voidf\n"
241 << "%label = OpLabel\n"
242 << "%idval = OpLoad %uvec3 %id\n"
243 << "%ndx = OpCompositeExtract %u32 %idval " << ndx << "\n"
245 "%inloc = OpAccessChain %f32ptr %indata %zero %ndx\n"
246 "%inval = OpLoad %f32 %inloc\n"
247 "%neg = OpFNegate %f32 %inval\n"
248 "%outloc = OpAccessChain %f32ptr %outdata %zero %ndx\n"
249 " OpStore %outloc %neg\n"
255 tcu::TestCaseGroup* createLocalSizeGroup (tcu::TestContext& testCtx)
257 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "localsize", ""));
258 ComputeShaderSpec spec;
259 de::Random rnd (deStringHash(group->getName()));
260 const deUint32 numElements = 64u;
261 vector<float> positiveFloats (numElements, 0);
262 vector<float> negativeFloats (numElements, 0);
264 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
266 for (size_t ndx = 0; ndx < numElements; ++ndx)
267 negativeFloats[ndx] = -positiveFloats[ndx];
269 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
270 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
272 spec.numWorkGroups = IVec3(numElements, 1, 1);
274 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, 1, 1), 0u);
275 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize", "", spec));
277 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, 1, 1), 0u);
278 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize", "", spec));
280 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, 1, 1), 0u);
281 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize", "", spec));
283 spec.numWorkGroups = IVec3(1, 1, 1);
285 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(numElements, 1, 1), 0u);
286 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_x", "", spec));
288 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(numElements, 1, 1), 0u);
289 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_x", "", spec));
291 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(numElements, 1, 1), 0u);
292 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_x", "", spec));
294 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, numElements, 1), 1u);
295 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_y", "", spec));
297 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, numElements, 1), 1u);
298 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_y", "", spec));
300 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, numElements, 1), 1u);
301 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_y", "", spec));
303 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, 1, numElements), 2u);
304 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_z", "", spec));
306 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, 1, numElements), 2u);
307 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_z", "", spec));
309 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, 1, numElements), 2u);
310 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_z", "", spec));
312 return group.release();
315 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
317 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
318 ComputeShaderSpec spec;
319 de::Random rnd (deStringHash(group->getName()));
320 const int numElements = 100;
321 vector<float> positiveFloats (numElements, 0);
322 vector<float> negativeFloats (numElements, 0);
324 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
326 for (size_t ndx = 0; ndx < numElements; ++ndx)
327 negativeFloats[ndx] = -positiveFloats[ndx];
330 string(getComputeAsmShaderPreamble()) +
332 "OpSource GLSL 430\n"
333 "OpName %main \"main\"\n"
334 "OpName %id \"gl_GlobalInvocationID\"\n"
336 "OpDecorate %id BuiltIn GlobalInvocationId\n"
338 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes())
340 + string(getComputeAsmInputOutputBuffer()) +
342 "%id = OpVariable %uvec3ptr Input\n"
343 "%zero = OpConstant %i32 0\n"
345 "%main = OpFunction %void None %voidf\n"
347 "%idval = OpLoad %uvec3 %id\n"
348 "%x = OpCompositeExtract %u32 %idval 0\n"
350 " OpNop\n" // Inside a function body
352 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
353 "%inval = OpLoad %f32 %inloc\n"
354 "%neg = OpFNegate %f32 %inval\n"
355 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
356 " OpStore %outloc %neg\n"
359 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
360 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
361 spec.numWorkGroups = IVec3(numElements, 1, 1);
363 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
365 return group.release();
368 bool compareFUnord (const std::vector<BufferSp>& inputs, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
370 if (outputAllocs.size() != 1)
373 vector<deUint8> input1Bytes;
374 vector<deUint8> input2Bytes;
375 vector<deUint8> expectedBytes;
377 inputs[0]->getBytes(input1Bytes);
378 inputs[1]->getBytes(input2Bytes);
379 expectedOutputs[0]->getBytes(expectedBytes);
381 const deInt32* const expectedOutputAsInt = reinterpret_cast<const deInt32* const>(&expectedBytes.front());
382 const deInt32* const outputAsInt = static_cast<const deInt32* const>(outputAllocs[0]->getHostPtr());
383 const float* const input1AsFloat = reinterpret_cast<const float* const>(&input1Bytes.front());
384 const float* const input2AsFloat = reinterpret_cast<const float* const>(&input2Bytes.front());
385 bool returnValue = true;
387 for (size_t idx = 0; idx < expectedBytes.size() / sizeof(deInt32); ++idx)
389 if (outputAsInt[idx] != expectedOutputAsInt[idx])
391 log << TestLog::Message << "ERROR: Sub-case failed. inputs: " << input1AsFloat[idx] << "," << input2AsFloat[idx] << " output: " << outputAsInt[idx]<< " expected output: " << expectedOutputAsInt[idx] << TestLog::EndMessage;
398 typedef VkBool32 (*compareFuncType) (float, float);
404 compareFuncType compareFunc;
406 OpFUnordCase (const char* _name, const char* _opCode, compareFuncType _compareFunc)
409 , compareFunc (_compareFunc) {}
412 #define ADD_OPFUNORD_CASE(NAME, OPCODE, OPERATOR) \
414 struct compare_##NAME { static VkBool32 compare(float x, float y) { return (x OPERATOR y) ? VK_TRUE : VK_FALSE; } }; \
415 cases.push_back(OpFUnordCase(#NAME, OPCODE, compare_##NAME::compare)); \
416 } while (deGetFalse())
418 tcu::TestCaseGroup* createOpFUnordGroup (tcu::TestContext& testCtx)
420 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfunord", "Test the OpFUnord* opcodes"));
421 de::Random rnd (deStringHash(group->getName()));
422 const int numElements = 100;
423 vector<OpFUnordCase> cases;
425 const StringTemplate shaderTemplate (
427 string(getComputeAsmShaderPreamble()) +
429 "OpSource GLSL 430\n"
430 "OpName %main \"main\"\n"
431 "OpName %id \"gl_GlobalInvocationID\"\n"
433 "OpDecorate %id BuiltIn GlobalInvocationId\n"
435 "OpDecorate %buf BufferBlock\n"
436 "OpDecorate %buf2 BufferBlock\n"
437 "OpDecorate %indata1 DescriptorSet 0\n"
438 "OpDecorate %indata1 Binding 0\n"
439 "OpDecorate %indata2 DescriptorSet 0\n"
440 "OpDecorate %indata2 Binding 1\n"
441 "OpDecorate %outdata DescriptorSet 0\n"
442 "OpDecorate %outdata Binding 2\n"
443 "OpDecorate %f32arr ArrayStride 4\n"
444 "OpDecorate %i32arr ArrayStride 4\n"
445 "OpMemberDecorate %buf 0 Offset 0\n"
446 "OpMemberDecorate %buf2 0 Offset 0\n"
448 + string(getComputeAsmCommonTypes()) +
450 "%buf = OpTypeStruct %f32arr\n"
451 "%bufptr = OpTypePointer Uniform %buf\n"
452 "%indata1 = OpVariable %bufptr Uniform\n"
453 "%indata2 = OpVariable %bufptr Uniform\n"
455 "%buf2 = OpTypeStruct %i32arr\n"
456 "%buf2ptr = OpTypePointer Uniform %buf2\n"
457 "%outdata = OpVariable %buf2ptr Uniform\n"
459 "%id = OpVariable %uvec3ptr Input\n"
460 "%zero = OpConstant %i32 0\n"
461 "%consti1 = OpConstant %i32 1\n"
462 "%constf1 = OpConstant %f32 1.0\n"
464 "%main = OpFunction %void None %voidf\n"
466 "%idval = OpLoad %uvec3 %id\n"
467 "%x = OpCompositeExtract %u32 %idval 0\n"
469 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
470 "%inval1 = OpLoad %f32 %inloc1\n"
471 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
472 "%inval2 = OpLoad %f32 %inloc2\n"
473 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
475 "%result = ${OPCODE} %bool %inval1 %inval2\n"
476 "%int_res = OpSelect %i32 %result %consti1 %zero\n"
477 " OpStore %outloc %int_res\n"
482 ADD_OPFUNORD_CASE(equal, "OpFUnordEqual", ==);
483 ADD_OPFUNORD_CASE(less, "OpFUnordLessThan", <);
484 ADD_OPFUNORD_CASE(lessequal, "OpFUnordLessThanEqual", <=);
485 ADD_OPFUNORD_CASE(greater, "OpFUnordGreaterThan", >);
486 ADD_OPFUNORD_CASE(greaterequal, "OpFUnordGreaterThanEqual", >=);
487 ADD_OPFUNORD_CASE(notequal, "OpFUnordNotEqual", !=);
489 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
491 map<string, string> specializations;
492 ComputeShaderSpec spec;
493 const float NaN = std::numeric_limits<float>::quiet_NaN();
494 vector<float> inputFloats1 (numElements, 0);
495 vector<float> inputFloats2 (numElements, 0);
496 vector<deInt32> expectedInts (numElements, 0);
498 specializations["OPCODE"] = cases[caseNdx].opCode;
499 spec.assembly = shaderTemplate.specialize(specializations);
501 fillRandomScalars(rnd, 1.f, 100.f, &inputFloats1[0], numElements);
502 for (size_t ndx = 0; ndx < numElements; ++ndx)
506 case 0: inputFloats2[ndx] = inputFloats1[ndx] + 1.0f; break;
507 case 1: inputFloats2[ndx] = inputFloats1[ndx] - 1.0f; break;
508 case 2: inputFloats2[ndx] = inputFloats1[ndx]; break;
509 case 3: inputFloats2[ndx] = NaN; break;
510 case 4: inputFloats2[ndx] = inputFloats1[ndx]; inputFloats1[ndx] = NaN; break;
511 case 5: inputFloats2[ndx] = NaN; inputFloats1[ndx] = NaN; break;
513 expectedInts[ndx] = tcu::Float32(inputFloats1[ndx]).isNaN() || tcu::Float32(inputFloats2[ndx]).isNaN() || cases[caseNdx].compareFunc(inputFloats1[ndx], inputFloats2[ndx]);
516 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
517 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
518 spec.outputs.push_back(BufferSp(new Int32Buffer(expectedInts)));
519 spec.numWorkGroups = IVec3(numElements, 1, 1);
520 spec.verifyIO = &compareFUnord;
521 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
524 return group.release();
530 const char* assembly;
531 OpAtomicType opAtomic;
532 deInt32 numOutputElements;
534 OpAtomicCase (const char* _name, const char* _assembly, OpAtomicType _opAtomic, deInt32 _numOutputElements)
536 , assembly (_assembly)
537 , opAtomic (_opAtomic)
538 , numOutputElements (_numOutputElements) {}
541 tcu::TestCaseGroup* createOpAtomicGroup (tcu::TestContext& testCtx, bool useStorageBuffer)
543 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx,
544 useStorageBuffer ? "opatomic_storage_buffer" : "opatomic",
545 "Test the OpAtomic* opcodes"));
546 const int numElements = 65535;
547 vector<OpAtomicCase> cases;
549 const StringTemplate shaderTemplate (
551 string("OpCapability Shader\n") +
552 (useStorageBuffer ? "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" : "") +
553 "OpMemoryModel Logical GLSL450\n"
554 "OpEntryPoint GLCompute %main \"main\" %id\n"
555 "OpExecutionMode %main LocalSize 1 1 1\n" +
557 "OpSource GLSL 430\n"
558 "OpName %main \"main\"\n"
559 "OpName %id \"gl_GlobalInvocationID\"\n"
561 "OpDecorate %id BuiltIn GlobalInvocationId\n"
563 "OpDecorate %buf ${BLOCK_DECORATION}\n"
564 "OpDecorate %indata DescriptorSet 0\n"
565 "OpDecorate %indata Binding 0\n"
566 "OpDecorate %i32arr ArrayStride 4\n"
567 "OpMemberDecorate %buf 0 Offset 0\n"
569 "OpDecorate %sumbuf ${BLOCK_DECORATION}\n"
570 "OpDecorate %sum DescriptorSet 0\n"
571 "OpDecorate %sum Binding 1\n"
572 "OpMemberDecorate %sumbuf 0 Coherent\n"
573 "OpMemberDecorate %sumbuf 0 Offset 0\n"
575 + getComputeAsmCommonTypes("${BLOCK_POINTER_TYPE}") +
577 "%buf = OpTypeStruct %i32arr\n"
578 "%bufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %buf\n"
579 "%indata = OpVariable %bufptr ${BLOCK_POINTER_TYPE}\n"
581 "%sumbuf = OpTypeStruct %i32arr\n"
582 "%sumbufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %sumbuf\n"
583 "%sum = OpVariable %sumbufptr ${BLOCK_POINTER_TYPE}\n"
585 "%id = OpVariable %uvec3ptr Input\n"
586 "%minusone = OpConstant %i32 -1\n"
587 "%zero = OpConstant %i32 0\n"
588 "%one = OpConstant %u32 1\n"
589 "%two = OpConstant %i32 2\n"
591 "%main = OpFunction %void None %voidf\n"
593 "%idval = OpLoad %uvec3 %id\n"
594 "%x = OpCompositeExtract %u32 %idval 0\n"
596 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
597 "%inval = OpLoad %i32 %inloc\n"
599 "%outloc = OpAccessChain %i32ptr %sum %zero ${INDEX}\n"
605 #define ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, NUM_OUTPUT_ELEMENTS) \
607 DE_STATIC_ASSERT((NUM_OUTPUT_ELEMENTS) == 1 || (NUM_OUTPUT_ELEMENTS) == numElements); \
608 cases.push_back(OpAtomicCase(#NAME, ASSEMBLY, OPATOMIC, NUM_OUTPUT_ELEMENTS)); \
609 } while (deGetFalse())
610 #define ADD_OPATOMIC_CASE_1(NAME, ASSEMBLY, OPATOMIC) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, 1)
611 #define ADD_OPATOMIC_CASE_N(NAME, ASSEMBLY, OPATOMIC) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, numElements)
613 ADD_OPATOMIC_CASE_1(iadd, "%unused = OpAtomicIAdd %i32 %outloc %one %zero %inval\n", OPATOMIC_IADD );
614 ADD_OPATOMIC_CASE_1(isub, "%unused = OpAtomicISub %i32 %outloc %one %zero %inval\n", OPATOMIC_ISUB );
615 ADD_OPATOMIC_CASE_1(iinc, "%unused = OpAtomicIIncrement %i32 %outloc %one %zero\n", OPATOMIC_IINC );
616 ADD_OPATOMIC_CASE_1(idec, "%unused = OpAtomicIDecrement %i32 %outloc %one %zero\n", OPATOMIC_IDEC );
617 ADD_OPATOMIC_CASE_N(load, "%inval2 = OpAtomicLoad %i32 %inloc %zero %zero\n"
618 " OpStore %outloc %inval2\n", OPATOMIC_LOAD );
619 ADD_OPATOMIC_CASE_N(store, " OpAtomicStore %outloc %zero %zero %inval\n", OPATOMIC_STORE );
620 ADD_OPATOMIC_CASE_N(compex, "%even = OpSMod %i32 %inval %two\n"
621 " OpStore %outloc %even\n"
622 "%unused = OpAtomicCompareExchange %i32 %outloc %one %zero %zero %minusone %zero\n", OPATOMIC_COMPEX );
624 #undef ADD_OPATOMIC_CASE
625 #undef ADD_OPATOMIC_CASE_1
626 #undef ADD_OPATOMIC_CASE_N
628 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
630 map<string, string> specializations;
631 ComputeShaderSpec spec;
632 vector<deInt32> inputInts (numElements, 0);
633 vector<deInt32> expected (cases[caseNdx].numOutputElements, -1);
635 specializations["INDEX"] = (cases[caseNdx].numOutputElements == 1) ? "%zero" : "%x";
636 specializations["INSTRUCTION"] = cases[caseNdx].assembly;
637 specializations["BLOCK_DECORATION"] = useStorageBuffer ? "Block" : "BufferBlock";
638 specializations["BLOCK_POINTER_TYPE"] = useStorageBuffer ? "StorageBuffer" : "Uniform";
639 spec.assembly = shaderTemplate.specialize(specializations);
641 if (useStorageBuffer)
642 spec.extensions.push_back("VK_KHR_storage_buffer_storage_class");
644 spec.inputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_INPUT)));
645 spec.outputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_EXPECTED)));
646 spec.numWorkGroups = IVec3(numElements, 1, 1);
647 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
650 return group.release();
653 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
655 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
656 ComputeShaderSpec spec;
657 de::Random rnd (deStringHash(group->getName()));
658 const int numElements = 100;
659 vector<float> positiveFloats (numElements, 0);
660 vector<float> negativeFloats (numElements, 0);
662 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
664 for (size_t ndx = 0; ndx < numElements; ++ndx)
665 negativeFloats[ndx] = -positiveFloats[ndx];
668 string(getComputeAsmShaderPreamble()) +
670 "%fname1 = OpString \"negateInputs.comp\"\n"
671 "%fname2 = OpString \"negateInputs\"\n"
673 "OpSource GLSL 430\n"
674 "OpName %main \"main\"\n"
675 "OpName %id \"gl_GlobalInvocationID\"\n"
677 "OpDecorate %id BuiltIn GlobalInvocationId\n"
679 + string(getComputeAsmInputOutputBufferTraits()) +
681 "OpLine %fname1 0 0\n" // At the earliest possible position
683 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
685 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
686 "OpLine %fname2 1 0\n" // Different filenames
687 "OpLine %fname1 1000 100000\n"
689 "%id = OpVariable %uvec3ptr Input\n"
690 "%zero = OpConstant %i32 0\n"
692 "OpLine %fname1 1 1\n" // Before a function
694 "%main = OpFunction %void None %voidf\n"
697 "OpLine %fname1 1 1\n" // In a function
699 "%idval = OpLoad %uvec3 %id\n"
700 "%x = OpCompositeExtract %u32 %idval 0\n"
701 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
702 "%inval = OpLoad %f32 %inloc\n"
703 "%neg = OpFNegate %f32 %inval\n"
704 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
705 " OpStore %outloc %neg\n"
708 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
709 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
710 spec.numWorkGroups = IVec3(numElements, 1, 1);
712 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
714 return group.release();
717 bool veryfiBinaryShader (const ProgramBinary& binary)
719 const size_t paternCount = 3u;
720 bool paternsCheck[paternCount] =
724 const string patersns[paternCount] =
730 size_t paternNdx = 0u;
732 for (size_t ndx = 0u; ndx < binary.getSize(); ++ndx)
734 if (false == paternsCheck[paternNdx] &&
735 patersns[paternNdx][0] == static_cast<char>(binary.getBinary()[ndx]) &&
736 deMemoryEqual((const char*)&binary.getBinary()[ndx], &patersns[paternNdx][0], patersns[paternNdx].length()))
738 paternsCheck[paternNdx]= true;
740 if (paternNdx == paternCount)
745 for (size_t ndx = 0u; ndx < paternCount; ++ndx)
747 if (!paternsCheck[ndx])
754 tcu::TestCaseGroup* createOpModuleProcessedGroup (tcu::TestContext& testCtx)
756 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opmoduleprocessed", "Test the OpModuleProcessed instruction"));
757 ComputeShaderSpec spec;
758 de::Random rnd (deStringHash(group->getName()));
759 const int numElements = 10;
760 vector<float> positiveFloats (numElements, 0);
761 vector<float> negativeFloats (numElements, 0);
763 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
765 for (size_t ndx = 0; ndx < numElements; ++ndx)
766 negativeFloats[ndx] = -positiveFloats[ndx];
769 string(getComputeAsmShaderPreamble()) +
770 "%fname = OpString \"negateInputs.comp\"\n"
772 "OpSource GLSL 430\n"
773 "OpName %main \"main\"\n"
774 "OpName %id \"gl_GlobalInvocationID\"\n"
775 "OpModuleProcessed \"VULKAN CTS\"\n" //OpModuleProcessed;
776 "OpModuleProcessed \"Negative values\"\n"
777 "OpModuleProcessed \"Date: 2017/09/21\"\n"
778 "OpDecorate %id BuiltIn GlobalInvocationId\n"
780 + string(getComputeAsmInputOutputBufferTraits())
782 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
784 "OpLine %fname 0 1\n"
786 "OpLine %fname 1000 1\n"
788 "%id = OpVariable %uvec3ptr Input\n"
789 "%zero = OpConstant %i32 0\n"
790 "%main = OpFunction %void None %voidf\n"
793 "%idval = OpLoad %uvec3 %id\n"
794 "%x = OpCompositeExtract %u32 %idval 0\n"
796 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
797 "%inval = OpLoad %f32 %inloc\n"
798 "%neg = OpFNegate %f32 %inval\n"
799 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
800 " OpStore %outloc %neg\n"
803 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
804 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
805 spec.numWorkGroups = IVec3(numElements, 1, 1);
806 spec.verifyBinary = veryfiBinaryShader;
807 spec.spirvVersion = SPIRV_VERSION_1_3;
809 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpModuleProcessed Tests", spec));
811 return group.release();
814 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
816 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
817 ComputeShaderSpec spec;
818 de::Random rnd (deStringHash(group->getName()));
819 const int numElements = 100;
820 vector<float> positiveFloats (numElements, 0);
821 vector<float> negativeFloats (numElements, 0);
823 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
825 for (size_t ndx = 0; ndx < numElements; ++ndx)
826 negativeFloats[ndx] = -positiveFloats[ndx];
829 string(getComputeAsmShaderPreamble()) +
831 "%fname = OpString \"negateInputs.comp\"\n"
833 "OpSource GLSL 430\n"
834 "OpName %main \"main\"\n"
835 "OpName %id \"gl_GlobalInvocationID\"\n"
837 "OpDecorate %id BuiltIn GlobalInvocationId\n"
839 + string(getComputeAsmInputOutputBufferTraits()) +
841 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
843 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
845 "OpLine %fname 0 1\n"
846 "OpNoLine\n" // Immediately following a preceding OpLine
848 "OpLine %fname 1000 1\n"
850 "%id = OpVariable %uvec3ptr Input\n"
851 "%zero = OpConstant %i32 0\n"
853 "OpNoLine\n" // Contents after the previous OpLine
855 "%main = OpFunction %void None %voidf\n"
857 "%idval = OpLoad %uvec3 %id\n"
858 "%x = OpCompositeExtract %u32 %idval 0\n"
860 "OpNoLine\n" // Multiple OpNoLine
864 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
865 "%inval = OpLoad %f32 %inloc\n"
866 "%neg = OpFNegate %f32 %inval\n"
867 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
868 " OpStore %outloc %neg\n"
871 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
872 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
873 spec.numWorkGroups = IVec3(numElements, 1, 1);
875 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
877 return group.release();
880 // Compare instruction for the contraction compute case.
881 // Returns true if the output is what is expected from the test case.
882 bool compareNoContractCase(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
884 if (outputAllocs.size() != 1)
887 // Only size is needed because we are not comparing the exact values.
888 size_t byteSize = expectedOutputs[0]->getByteSize();
890 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
892 for(size_t i = 0; i < byteSize / sizeof(float); ++i) {
893 if (outputAsFloat[i] != 0.f &&
894 outputAsFloat[i] != -ldexp(1, -24)) {
902 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
904 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
905 vector<CaseParameter> cases;
906 const int numElements = 100;
907 vector<float> inputFloats1 (numElements, 0);
908 vector<float> inputFloats2 (numElements, 0);
909 vector<float> outputFloats (numElements, 0);
910 const StringTemplate shaderTemplate (
911 string(getComputeAsmShaderPreamble()) +
913 "OpName %main \"main\"\n"
914 "OpName %id \"gl_GlobalInvocationID\"\n"
916 "OpDecorate %id BuiltIn GlobalInvocationId\n"
920 "OpDecorate %buf BufferBlock\n"
921 "OpDecorate %indata1 DescriptorSet 0\n"
922 "OpDecorate %indata1 Binding 0\n"
923 "OpDecorate %indata2 DescriptorSet 0\n"
924 "OpDecorate %indata2 Binding 1\n"
925 "OpDecorate %outdata DescriptorSet 0\n"
926 "OpDecorate %outdata Binding 2\n"
927 "OpDecorate %f32arr ArrayStride 4\n"
928 "OpMemberDecorate %buf 0 Offset 0\n"
930 + string(getComputeAsmCommonTypes()) +
932 "%buf = OpTypeStruct %f32arr\n"
933 "%bufptr = OpTypePointer Uniform %buf\n"
934 "%indata1 = OpVariable %bufptr Uniform\n"
935 "%indata2 = OpVariable %bufptr Uniform\n"
936 "%outdata = OpVariable %bufptr Uniform\n"
938 "%id = OpVariable %uvec3ptr Input\n"
939 "%zero = OpConstant %i32 0\n"
940 "%c_f_m1 = OpConstant %f32 -1.\n"
942 "%main = OpFunction %void None %voidf\n"
944 "%idval = OpLoad %uvec3 %id\n"
945 "%x = OpCompositeExtract %u32 %idval 0\n"
946 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
947 "%inval1 = OpLoad %f32 %inloc1\n"
948 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
949 "%inval2 = OpLoad %f32 %inloc2\n"
950 "%mul = OpFMul %f32 %inval1 %inval2\n"
951 "%add = OpFAdd %f32 %mul %c_f_m1\n"
952 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
953 " OpStore %outloc %add\n"
957 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
958 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
959 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
961 for (size_t ndx = 0; ndx < numElements; ++ndx)
963 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
964 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
965 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
966 // conducted separately and the result is rounded to 1, or 0x1.fffffcp-1
967 // So the final result will be 0.f or 0x1p-24.
968 // If the operation is combined into a precise fused multiply-add, then the result would be
969 // 2^-46 (0xa8800000).
970 outputFloats[ndx] = 0.f;
973 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
975 map<string, string> specializations;
976 ComputeShaderSpec spec;
978 specializations["DECORATION"] = cases[caseNdx].param;
979 spec.assembly = shaderTemplate.specialize(specializations);
980 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
981 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
982 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
983 spec.numWorkGroups = IVec3(numElements, 1, 1);
984 // Check against the two possible answers based on rounding mode.
985 spec.verifyIO = &compareNoContractCase;
987 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
989 return group.release();
992 bool compareFRem(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
994 if (outputAllocs.size() != 1)
997 vector<deUint8> expectedBytes;
998 expectedOutputs[0]->getBytes(expectedBytes);
1000 const float* expectedOutputAsFloat = reinterpret_cast<const float*>(&expectedBytes.front());
1001 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1003 for (size_t idx = 0; idx < expectedBytes.size() / sizeof(float); ++idx)
1005 const float f0 = expectedOutputAsFloat[idx];
1006 const float f1 = outputAsFloat[idx];
1007 // \todo relative error needs to be fairly high because FRem may be implemented as
1008 // (roughly) frac(a/b)*b, so LSB errors can be magnified. But this should be fine for now.
1009 if (deFloatAbs((f1 - f0) / f0) > 0.02)
1016 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
1018 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
1019 ComputeShaderSpec spec;
1020 de::Random rnd (deStringHash(group->getName()));
1021 const int numElements = 200;
1022 vector<float> inputFloats1 (numElements, 0);
1023 vector<float> inputFloats2 (numElements, 0);
1024 vector<float> outputFloats (numElements, 0);
1026 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1027 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
1029 for (size_t ndx = 0; ndx < numElements; ++ndx)
1031 // Guard against divisors near zero.
1032 if (std::fabs(inputFloats2[ndx]) < 1e-3)
1033 inputFloats2[ndx] = 8.f;
1035 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1036 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
1040 string(getComputeAsmShaderPreamble()) +
1042 "OpName %main \"main\"\n"
1043 "OpName %id \"gl_GlobalInvocationID\"\n"
1045 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1047 "OpDecorate %buf BufferBlock\n"
1048 "OpDecorate %indata1 DescriptorSet 0\n"
1049 "OpDecorate %indata1 Binding 0\n"
1050 "OpDecorate %indata2 DescriptorSet 0\n"
1051 "OpDecorate %indata2 Binding 1\n"
1052 "OpDecorate %outdata DescriptorSet 0\n"
1053 "OpDecorate %outdata Binding 2\n"
1054 "OpDecorate %f32arr ArrayStride 4\n"
1055 "OpMemberDecorate %buf 0 Offset 0\n"
1057 + string(getComputeAsmCommonTypes()) +
1059 "%buf = OpTypeStruct %f32arr\n"
1060 "%bufptr = OpTypePointer Uniform %buf\n"
1061 "%indata1 = OpVariable %bufptr Uniform\n"
1062 "%indata2 = OpVariable %bufptr Uniform\n"
1063 "%outdata = OpVariable %bufptr Uniform\n"
1065 "%id = OpVariable %uvec3ptr Input\n"
1066 "%zero = OpConstant %i32 0\n"
1068 "%main = OpFunction %void None %voidf\n"
1069 "%label = OpLabel\n"
1070 "%idval = OpLoad %uvec3 %id\n"
1071 "%x = OpCompositeExtract %u32 %idval 0\n"
1072 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1073 "%inval1 = OpLoad %f32 %inloc1\n"
1074 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1075 "%inval2 = OpLoad %f32 %inloc2\n"
1076 "%rem = OpFRem %f32 %inval1 %inval2\n"
1077 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1078 " OpStore %outloc %rem\n"
1082 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1083 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1084 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1085 spec.numWorkGroups = IVec3(numElements, 1, 1);
1086 spec.verifyIO = &compareFRem;
1088 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1090 return group.release();
1093 bool compareNMin (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1095 if (outputAllocs.size() != 1)
1098 const BufferSp& expectedOutput (expectedOutputs[0]);
1099 std::vector<deUint8> data;
1100 expectedOutput->getBytes(data);
1102 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1103 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1105 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float); ++idx)
1107 const float f0 = expectedOutputAsFloat[idx];
1108 const float f1 = outputAsFloat[idx];
1110 // For NMin, we accept NaN as output if both inputs were NaN.
1111 // Otherwise the NaN is the wrong choise, as on architectures that
1112 // do not handle NaN, those are huge values.
1113 if (!(tcu::Float32(f1).isNaN() && tcu::Float32(f0).isNaN()) && deFloatAbs(f1 - f0) > 0.00001f)
1120 tcu::TestCaseGroup* createOpNMinGroup (tcu::TestContext& testCtx)
1122 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnmin", "Test the OpNMin instruction"));
1123 ComputeShaderSpec spec;
1124 de::Random rnd (deStringHash(group->getName()));
1125 const int numElements = 200;
1126 vector<float> inputFloats1 (numElements, 0);
1127 vector<float> inputFloats2 (numElements, 0);
1128 vector<float> outputFloats (numElements, 0);
1130 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1131 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1133 // Make the first case a full-NAN case.
1134 inputFloats1[0] = TCU_NAN;
1135 inputFloats2[0] = TCU_NAN;
1137 for (size_t ndx = 0; ndx < numElements; ++ndx)
1139 // By default, pick the smallest
1140 outputFloats[ndx] = std::min(inputFloats1[ndx], inputFloats2[ndx]);
1142 // Make half of the cases NaN cases
1145 // Alternate between the NaN operand
1148 outputFloats[ndx] = inputFloats2[ndx];
1149 inputFloats1[ndx] = TCU_NAN;
1153 outputFloats[ndx] = inputFloats1[ndx];
1154 inputFloats2[ndx] = TCU_NAN;
1160 "OpCapability Shader\n"
1161 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1162 "OpMemoryModel Logical GLSL450\n"
1163 "OpEntryPoint GLCompute %main \"main\" %id\n"
1164 "OpExecutionMode %main LocalSize 1 1 1\n"
1166 "OpName %main \"main\"\n"
1167 "OpName %id \"gl_GlobalInvocationID\"\n"
1169 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1171 "OpDecorate %buf BufferBlock\n"
1172 "OpDecorate %indata1 DescriptorSet 0\n"
1173 "OpDecorate %indata1 Binding 0\n"
1174 "OpDecorate %indata2 DescriptorSet 0\n"
1175 "OpDecorate %indata2 Binding 1\n"
1176 "OpDecorate %outdata DescriptorSet 0\n"
1177 "OpDecorate %outdata Binding 2\n"
1178 "OpDecorate %f32arr ArrayStride 4\n"
1179 "OpMemberDecorate %buf 0 Offset 0\n"
1181 + string(getComputeAsmCommonTypes()) +
1183 "%buf = OpTypeStruct %f32arr\n"
1184 "%bufptr = OpTypePointer Uniform %buf\n"
1185 "%indata1 = OpVariable %bufptr Uniform\n"
1186 "%indata2 = OpVariable %bufptr Uniform\n"
1187 "%outdata = OpVariable %bufptr Uniform\n"
1189 "%id = OpVariable %uvec3ptr Input\n"
1190 "%zero = OpConstant %i32 0\n"
1192 "%main = OpFunction %void None %voidf\n"
1193 "%label = OpLabel\n"
1194 "%idval = OpLoad %uvec3 %id\n"
1195 "%x = OpCompositeExtract %u32 %idval 0\n"
1196 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1197 "%inval1 = OpLoad %f32 %inloc1\n"
1198 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1199 "%inval2 = OpLoad %f32 %inloc2\n"
1200 "%rem = OpExtInst %f32 %std450 NMin %inval1 %inval2\n"
1201 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1202 " OpStore %outloc %rem\n"
1206 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1207 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1208 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1209 spec.numWorkGroups = IVec3(numElements, 1, 1);
1210 spec.verifyIO = &compareNMin;
1212 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1214 return group.release();
1217 bool compareNMax (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1219 if (outputAllocs.size() != 1)
1222 const BufferSp& expectedOutput = expectedOutputs[0];
1223 std::vector<deUint8> data;
1224 expectedOutput->getBytes(data);
1226 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1227 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1229 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float); ++idx)
1231 const float f0 = expectedOutputAsFloat[idx];
1232 const float f1 = outputAsFloat[idx];
1234 // For NMax, NaN is considered acceptable result, since in
1235 // architectures that do not handle NaNs, those are huge values.
1236 if (!tcu::Float32(f1).isNaN() && deFloatAbs(f1 - f0) > 0.00001f)
1243 tcu::TestCaseGroup* createOpNMaxGroup (tcu::TestContext& testCtx)
1245 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, "opnmax", "Test the OpNMax instruction"));
1246 ComputeShaderSpec spec;
1247 de::Random rnd (deStringHash(group->getName()));
1248 const int numElements = 200;
1249 vector<float> inputFloats1 (numElements, 0);
1250 vector<float> inputFloats2 (numElements, 0);
1251 vector<float> outputFloats (numElements, 0);
1253 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1254 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1256 // Make the first case a full-NAN case.
1257 inputFloats1[0] = TCU_NAN;
1258 inputFloats2[0] = TCU_NAN;
1260 for (size_t ndx = 0; ndx < numElements; ++ndx)
1262 // By default, pick the biggest
1263 outputFloats[ndx] = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1265 // Make half of the cases NaN cases
1268 // Alternate between the NaN operand
1271 outputFloats[ndx] = inputFloats2[ndx];
1272 inputFloats1[ndx] = TCU_NAN;
1276 outputFloats[ndx] = inputFloats1[ndx];
1277 inputFloats2[ndx] = TCU_NAN;
1283 "OpCapability Shader\n"
1284 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1285 "OpMemoryModel Logical GLSL450\n"
1286 "OpEntryPoint GLCompute %main \"main\" %id\n"
1287 "OpExecutionMode %main LocalSize 1 1 1\n"
1289 "OpName %main \"main\"\n"
1290 "OpName %id \"gl_GlobalInvocationID\"\n"
1292 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1294 "OpDecorate %buf BufferBlock\n"
1295 "OpDecorate %indata1 DescriptorSet 0\n"
1296 "OpDecorate %indata1 Binding 0\n"
1297 "OpDecorate %indata2 DescriptorSet 0\n"
1298 "OpDecorate %indata2 Binding 1\n"
1299 "OpDecorate %outdata DescriptorSet 0\n"
1300 "OpDecorate %outdata Binding 2\n"
1301 "OpDecorate %f32arr ArrayStride 4\n"
1302 "OpMemberDecorate %buf 0 Offset 0\n"
1304 + string(getComputeAsmCommonTypes()) +
1306 "%buf = OpTypeStruct %f32arr\n"
1307 "%bufptr = OpTypePointer Uniform %buf\n"
1308 "%indata1 = OpVariable %bufptr Uniform\n"
1309 "%indata2 = OpVariable %bufptr Uniform\n"
1310 "%outdata = OpVariable %bufptr Uniform\n"
1312 "%id = OpVariable %uvec3ptr Input\n"
1313 "%zero = OpConstant %i32 0\n"
1315 "%main = OpFunction %void None %voidf\n"
1316 "%label = OpLabel\n"
1317 "%idval = OpLoad %uvec3 %id\n"
1318 "%x = OpCompositeExtract %u32 %idval 0\n"
1319 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1320 "%inval1 = OpLoad %f32 %inloc1\n"
1321 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1322 "%inval2 = OpLoad %f32 %inloc2\n"
1323 "%rem = OpExtInst %f32 %std450 NMax %inval1 %inval2\n"
1324 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1325 " OpStore %outloc %rem\n"
1329 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1330 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1331 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1332 spec.numWorkGroups = IVec3(numElements, 1, 1);
1333 spec.verifyIO = &compareNMax;
1335 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1337 return group.release();
1340 bool compareNClamp (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1342 if (outputAllocs.size() != 1)
1345 const BufferSp& expectedOutput = expectedOutputs[0];
1346 std::vector<deUint8> data;
1347 expectedOutput->getBytes(data);
1349 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1350 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1352 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float) / 2; ++idx)
1354 const float e0 = expectedOutputAsFloat[idx * 2];
1355 const float e1 = expectedOutputAsFloat[idx * 2 + 1];
1356 const float res = outputAsFloat[idx];
1358 // For NClamp, we have two possible outcomes based on
1359 // whether NaNs are handled or not.
1360 // If either min or max value is NaN, the result is undefined,
1361 // so this test doesn't stress those. If the clamped value is
1362 // NaN, and NaNs are handled, the result is min; if NaNs are not
1363 // handled, they are big values that result in max.
1364 // If all three parameters are NaN, the result should be NaN.
1365 if (!((tcu::Float32(e0).isNaN() && tcu::Float32(res).isNaN()) ||
1366 (deFloatAbs(e0 - res) < 0.00001f) ||
1367 (deFloatAbs(e1 - res) < 0.00001f)))
1374 tcu::TestCaseGroup* createOpNClampGroup (tcu::TestContext& testCtx)
1376 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnclamp", "Test the OpNClamp instruction"));
1377 ComputeShaderSpec spec;
1378 de::Random rnd (deStringHash(group->getName()));
1379 const int numElements = 200;
1380 vector<float> inputFloats1 (numElements, 0);
1381 vector<float> inputFloats2 (numElements, 0);
1382 vector<float> inputFloats3 (numElements, 0);
1383 vector<float> outputFloats (numElements * 2, 0);
1385 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1386 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1387 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats3[0], numElements);
1389 for (size_t ndx = 0; ndx < numElements; ++ndx)
1391 // Results are only defined if max value is bigger than min value.
1392 if (inputFloats2[ndx] > inputFloats3[ndx])
1394 float t = inputFloats2[ndx];
1395 inputFloats2[ndx] = inputFloats3[ndx];
1396 inputFloats3[ndx] = t;
1399 // By default, do the clamp, setting both possible answers
1400 float defaultRes = std::min(std::max(inputFloats1[ndx], inputFloats2[ndx]), inputFloats3[ndx]);
1402 float maxResA = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1403 float maxResB = maxResA;
1405 // Alternate between the NaN cases
1408 inputFloats1[ndx] = TCU_NAN;
1409 // If NaN is handled, the result should be same as the clamp minimum.
1410 // If NaN is not handled, the result should clamp to the clamp maximum.
1411 maxResA = inputFloats2[ndx];
1412 maxResB = inputFloats3[ndx];
1416 // Not a NaN case - only one legal result.
1417 maxResA = defaultRes;
1418 maxResB = defaultRes;
1421 outputFloats[ndx * 2] = maxResA;
1422 outputFloats[ndx * 2 + 1] = maxResB;
1425 // Make the first case a full-NAN case.
1426 inputFloats1[0] = TCU_NAN;
1427 inputFloats2[0] = TCU_NAN;
1428 inputFloats3[0] = TCU_NAN;
1429 outputFloats[0] = TCU_NAN;
1430 outputFloats[1] = TCU_NAN;
1433 "OpCapability Shader\n"
1434 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1435 "OpMemoryModel Logical GLSL450\n"
1436 "OpEntryPoint GLCompute %main \"main\" %id\n"
1437 "OpExecutionMode %main LocalSize 1 1 1\n"
1439 "OpName %main \"main\"\n"
1440 "OpName %id \"gl_GlobalInvocationID\"\n"
1442 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1444 "OpDecorate %buf BufferBlock\n"
1445 "OpDecorate %indata1 DescriptorSet 0\n"
1446 "OpDecorate %indata1 Binding 0\n"
1447 "OpDecorate %indata2 DescriptorSet 0\n"
1448 "OpDecorate %indata2 Binding 1\n"
1449 "OpDecorate %indata3 DescriptorSet 0\n"
1450 "OpDecorate %indata3 Binding 2\n"
1451 "OpDecorate %outdata DescriptorSet 0\n"
1452 "OpDecorate %outdata Binding 3\n"
1453 "OpDecorate %f32arr ArrayStride 4\n"
1454 "OpMemberDecorate %buf 0 Offset 0\n"
1456 + string(getComputeAsmCommonTypes()) +
1458 "%buf = OpTypeStruct %f32arr\n"
1459 "%bufptr = OpTypePointer Uniform %buf\n"
1460 "%indata1 = OpVariable %bufptr Uniform\n"
1461 "%indata2 = OpVariable %bufptr Uniform\n"
1462 "%indata3 = OpVariable %bufptr Uniform\n"
1463 "%outdata = OpVariable %bufptr Uniform\n"
1465 "%id = OpVariable %uvec3ptr Input\n"
1466 "%zero = OpConstant %i32 0\n"
1468 "%main = OpFunction %void None %voidf\n"
1469 "%label = OpLabel\n"
1470 "%idval = OpLoad %uvec3 %id\n"
1471 "%x = OpCompositeExtract %u32 %idval 0\n"
1472 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1473 "%inval1 = OpLoad %f32 %inloc1\n"
1474 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1475 "%inval2 = OpLoad %f32 %inloc2\n"
1476 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1477 "%inval3 = OpLoad %f32 %inloc3\n"
1478 "%rem = OpExtInst %f32 %std450 NClamp %inval1 %inval2 %inval3\n"
1479 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1480 " OpStore %outloc %rem\n"
1484 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1485 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1486 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1487 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1488 spec.numWorkGroups = IVec3(numElements, 1, 1);
1489 spec.verifyIO = &compareNClamp;
1491 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1493 return group.release();
1496 tcu::TestCaseGroup* createOpSRemComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1498 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem", "Test the OpSRem instruction"));
1499 de::Random rnd (deStringHash(group->getName()));
1500 const int numElements = 200;
1502 const struct CaseParams
1505 const char* failMessage; // customized status message
1506 qpTestResult failResult; // override status on failure
1507 int op1Min, op1Max; // operand ranges
1511 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1512 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1514 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1516 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1518 const CaseParams& params = cases[caseNdx];
1519 ComputeShaderSpec spec;
1520 vector<deInt32> inputInts1 (numElements, 0);
1521 vector<deInt32> inputInts2 (numElements, 0);
1522 vector<deInt32> outputInts (numElements, 0);
1524 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1525 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], 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 string(getComputeAsmShaderPreamble()) +
1536 "OpName %main \"main\"\n"
1537 "OpName %id \"gl_GlobalInvocationID\"\n"
1539 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1541 "OpDecorate %buf BufferBlock\n"
1542 "OpDecorate %indata1 DescriptorSet 0\n"
1543 "OpDecorate %indata1 Binding 0\n"
1544 "OpDecorate %indata2 DescriptorSet 0\n"
1545 "OpDecorate %indata2 Binding 1\n"
1546 "OpDecorate %outdata DescriptorSet 0\n"
1547 "OpDecorate %outdata Binding 2\n"
1548 "OpDecorate %i32arr ArrayStride 4\n"
1549 "OpMemberDecorate %buf 0 Offset 0\n"
1551 + string(getComputeAsmCommonTypes()) +
1553 "%buf = OpTypeStruct %i32arr\n"
1554 "%bufptr = OpTypePointer Uniform %buf\n"
1555 "%indata1 = OpVariable %bufptr Uniform\n"
1556 "%indata2 = OpVariable %bufptr Uniform\n"
1557 "%outdata = OpVariable %bufptr Uniform\n"
1559 "%id = OpVariable %uvec3ptr Input\n"
1560 "%zero = OpConstant %i32 0\n"
1562 "%main = OpFunction %void None %voidf\n"
1563 "%label = OpLabel\n"
1564 "%idval = OpLoad %uvec3 %id\n"
1565 "%x = OpCompositeExtract %u32 %idval 0\n"
1566 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1567 "%inval1 = OpLoad %i32 %inloc1\n"
1568 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1569 "%inval2 = OpLoad %i32 %inloc2\n"
1570 "%rem = OpSRem %i32 %inval1 %inval2\n"
1571 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1572 " OpStore %outloc %rem\n"
1576 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1577 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1578 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1579 spec.numWorkGroups = IVec3(numElements, 1, 1);
1580 spec.failResult = params.failResult;
1581 spec.failMessage = params.failMessage;
1583 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1586 return group.release();
1589 tcu::TestCaseGroup* createOpSRemComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1591 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem64", "Test the 64-bit OpSRem instruction"));
1592 de::Random rnd (deStringHash(group->getName()));
1593 const int numElements = 200;
1595 const struct CaseParams
1598 const char* failMessage; // customized status message
1599 qpTestResult failResult; // override status on failure
1603 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1604 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1606 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1608 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1610 const CaseParams& params = cases[caseNdx];
1611 ComputeShaderSpec spec;
1612 vector<deInt64> inputInts1 (numElements, 0);
1613 vector<deInt64> inputInts2 (numElements, 0);
1614 vector<deInt64> outputInts (numElements, 0);
1616 if (params.positive)
1618 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1619 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1623 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1624 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1627 for (int ndx = 0; ndx < numElements; ++ndx)
1629 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1630 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1634 "OpCapability Int64\n"
1636 + string(getComputeAsmShaderPreamble()) +
1638 "OpName %main \"main\"\n"
1639 "OpName %id \"gl_GlobalInvocationID\"\n"
1641 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1643 "OpDecorate %buf BufferBlock\n"
1644 "OpDecorate %indata1 DescriptorSet 0\n"
1645 "OpDecorate %indata1 Binding 0\n"
1646 "OpDecorate %indata2 DescriptorSet 0\n"
1647 "OpDecorate %indata2 Binding 1\n"
1648 "OpDecorate %outdata DescriptorSet 0\n"
1649 "OpDecorate %outdata Binding 2\n"
1650 "OpDecorate %i64arr ArrayStride 8\n"
1651 "OpMemberDecorate %buf 0 Offset 0\n"
1653 + string(getComputeAsmCommonTypes())
1654 + string(getComputeAsmCommonInt64Types()) +
1656 "%buf = OpTypeStruct %i64arr\n"
1657 "%bufptr = OpTypePointer Uniform %buf\n"
1658 "%indata1 = OpVariable %bufptr Uniform\n"
1659 "%indata2 = OpVariable %bufptr Uniform\n"
1660 "%outdata = OpVariable %bufptr Uniform\n"
1662 "%id = OpVariable %uvec3ptr Input\n"
1663 "%zero = OpConstant %i64 0\n"
1665 "%main = OpFunction %void None %voidf\n"
1666 "%label = OpLabel\n"
1667 "%idval = OpLoad %uvec3 %id\n"
1668 "%x = OpCompositeExtract %u32 %idval 0\n"
1669 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1670 "%inval1 = OpLoad %i64 %inloc1\n"
1671 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1672 "%inval2 = OpLoad %i64 %inloc2\n"
1673 "%rem = OpSRem %i64 %inval1 %inval2\n"
1674 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1675 " OpStore %outloc %rem\n"
1679 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1680 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1681 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1682 spec.numWorkGroups = IVec3(numElements, 1, 1);
1683 spec.failResult = params.failResult;
1684 spec.failMessage = params.failMessage;
1686 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1689 return group.release();
1692 tcu::TestCaseGroup* createOpSModComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1694 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod", "Test the OpSMod instruction"));
1695 de::Random rnd (deStringHash(group->getName()));
1696 const int numElements = 200;
1698 const struct CaseParams
1701 const char* failMessage; // customized status message
1702 qpTestResult failResult; // override status on failure
1703 int op1Min, op1Max; // operand ranges
1707 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1708 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1710 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1712 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1714 const CaseParams& params = cases[caseNdx];
1716 ComputeShaderSpec spec;
1717 vector<deInt32> inputInts1 (numElements, 0);
1718 vector<deInt32> inputInts2 (numElements, 0);
1719 vector<deInt32> outputInts (numElements, 0);
1721 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1722 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1724 for (int ndx = 0; ndx < numElements; ++ndx)
1726 deInt32 rem = inputInts1[ndx] % inputInts2[ndx];
1729 outputInts[ndx] = 0;
1731 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1733 // They have the same sign
1734 outputInts[ndx] = rem;
1738 // They have opposite sign. The remainder operation takes the
1739 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1740 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1741 // the result has the correct sign and that it is still
1742 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1744 // See also http://mathforum.org/library/drmath/view/52343.html
1745 outputInts[ndx] = rem + inputInts2[ndx];
1750 string(getComputeAsmShaderPreamble()) +
1752 "OpName %main \"main\"\n"
1753 "OpName %id \"gl_GlobalInvocationID\"\n"
1755 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1757 "OpDecorate %buf BufferBlock\n"
1758 "OpDecorate %indata1 DescriptorSet 0\n"
1759 "OpDecorate %indata1 Binding 0\n"
1760 "OpDecorate %indata2 DescriptorSet 0\n"
1761 "OpDecorate %indata2 Binding 1\n"
1762 "OpDecorate %outdata DescriptorSet 0\n"
1763 "OpDecorate %outdata Binding 2\n"
1764 "OpDecorate %i32arr ArrayStride 4\n"
1765 "OpMemberDecorate %buf 0 Offset 0\n"
1767 + string(getComputeAsmCommonTypes()) +
1769 "%buf = OpTypeStruct %i32arr\n"
1770 "%bufptr = OpTypePointer Uniform %buf\n"
1771 "%indata1 = OpVariable %bufptr Uniform\n"
1772 "%indata2 = OpVariable %bufptr Uniform\n"
1773 "%outdata = OpVariable %bufptr Uniform\n"
1775 "%id = OpVariable %uvec3ptr Input\n"
1776 "%zero = OpConstant %i32 0\n"
1778 "%main = OpFunction %void None %voidf\n"
1779 "%label = OpLabel\n"
1780 "%idval = OpLoad %uvec3 %id\n"
1781 "%x = OpCompositeExtract %u32 %idval 0\n"
1782 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1783 "%inval1 = OpLoad %i32 %inloc1\n"
1784 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1785 "%inval2 = OpLoad %i32 %inloc2\n"
1786 "%rem = OpSMod %i32 %inval1 %inval2\n"
1787 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1788 " OpStore %outloc %rem\n"
1792 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1793 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1794 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1795 spec.numWorkGroups = IVec3(numElements, 1, 1);
1796 spec.failResult = params.failResult;
1797 spec.failMessage = params.failMessage;
1799 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1802 return group.release();
1805 tcu::TestCaseGroup* createOpSModComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1807 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod64", "Test the OpSMod instruction"));
1808 de::Random rnd (deStringHash(group->getName()));
1809 const int numElements = 200;
1811 const struct CaseParams
1814 const char* failMessage; // customized status message
1815 qpTestResult failResult; // override status on failure
1819 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1820 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1822 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1824 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1826 const CaseParams& params = cases[caseNdx];
1828 ComputeShaderSpec spec;
1829 vector<deInt64> inputInts1 (numElements, 0);
1830 vector<deInt64> inputInts2 (numElements, 0);
1831 vector<deInt64> outputInts (numElements, 0);
1834 if (params.positive)
1836 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1837 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1841 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1842 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1845 for (int ndx = 0; ndx < numElements; ++ndx)
1847 deInt64 rem = inputInts1[ndx] % inputInts2[ndx];
1850 outputInts[ndx] = 0;
1852 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1854 // They have the same sign
1855 outputInts[ndx] = rem;
1859 // They have opposite sign. The remainder operation takes the
1860 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1861 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1862 // the result has the correct sign and that it is still
1863 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1865 // See also http://mathforum.org/library/drmath/view/52343.html
1866 outputInts[ndx] = rem + inputInts2[ndx];
1871 "OpCapability Int64\n"
1873 + string(getComputeAsmShaderPreamble()) +
1875 "OpName %main \"main\"\n"
1876 "OpName %id \"gl_GlobalInvocationID\"\n"
1878 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1880 "OpDecorate %buf BufferBlock\n"
1881 "OpDecorate %indata1 DescriptorSet 0\n"
1882 "OpDecorate %indata1 Binding 0\n"
1883 "OpDecorate %indata2 DescriptorSet 0\n"
1884 "OpDecorate %indata2 Binding 1\n"
1885 "OpDecorate %outdata DescriptorSet 0\n"
1886 "OpDecorate %outdata Binding 2\n"
1887 "OpDecorate %i64arr ArrayStride 8\n"
1888 "OpMemberDecorate %buf 0 Offset 0\n"
1890 + string(getComputeAsmCommonTypes())
1891 + string(getComputeAsmCommonInt64Types()) +
1893 "%buf = OpTypeStruct %i64arr\n"
1894 "%bufptr = OpTypePointer Uniform %buf\n"
1895 "%indata1 = OpVariable %bufptr Uniform\n"
1896 "%indata2 = OpVariable %bufptr Uniform\n"
1897 "%outdata = OpVariable %bufptr Uniform\n"
1899 "%id = OpVariable %uvec3ptr Input\n"
1900 "%zero = OpConstant %i64 0\n"
1902 "%main = OpFunction %void None %voidf\n"
1903 "%label = OpLabel\n"
1904 "%idval = OpLoad %uvec3 %id\n"
1905 "%x = OpCompositeExtract %u32 %idval 0\n"
1906 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1907 "%inval1 = OpLoad %i64 %inloc1\n"
1908 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1909 "%inval2 = OpLoad %i64 %inloc2\n"
1910 "%rem = OpSMod %i64 %inval1 %inval2\n"
1911 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1912 " OpStore %outloc %rem\n"
1916 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1917 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1918 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1919 spec.numWorkGroups = IVec3(numElements, 1, 1);
1920 spec.failResult = params.failResult;
1921 spec.failMessage = params.failMessage;
1923 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1926 return group.release();
1929 // Copy contents in the input buffer to the output buffer.
1930 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
1932 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
1933 de::Random rnd (deStringHash(group->getName()));
1934 const int numElements = 100;
1936 // 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.
1937 ComputeShaderSpec spec1;
1938 vector<Vec4> inputFloats1 (numElements);
1939 vector<Vec4> outputFloats1 (numElements);
1941 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
1943 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1944 floorAll(inputFloats1);
1946 for (size_t ndx = 0; ndx < numElements; ++ndx)
1947 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
1950 string(getComputeAsmShaderPreamble()) +
1952 "OpName %main \"main\"\n"
1953 "OpName %id \"gl_GlobalInvocationID\"\n"
1955 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1956 "OpDecorate %vec4arr ArrayStride 16\n"
1958 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1960 "%vec4 = OpTypeVector %f32 4\n"
1961 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
1962 "%vec4ptr_f = OpTypePointer Function %vec4\n"
1963 "%vec4arr = OpTypeRuntimeArray %vec4\n"
1964 "%buf = OpTypeStruct %vec4arr\n"
1965 "%bufptr = OpTypePointer Uniform %buf\n"
1966 "%indata = OpVariable %bufptr Uniform\n"
1967 "%outdata = OpVariable %bufptr Uniform\n"
1969 "%id = OpVariable %uvec3ptr Input\n"
1970 "%zero = OpConstant %i32 0\n"
1971 "%c_f_0 = OpConstant %f32 0.\n"
1972 "%c_f_0_5 = OpConstant %f32 0.5\n"
1973 "%c_f_1_5 = OpConstant %f32 1.5\n"
1974 "%c_f_2_5 = OpConstant %f32 2.5\n"
1975 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
1977 "%main = OpFunction %void None %voidf\n"
1978 "%label = OpLabel\n"
1979 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
1980 "%idval = OpLoad %uvec3 %id\n"
1981 "%x = OpCompositeExtract %u32 %idval 0\n"
1982 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
1983 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
1984 " OpCopyMemory %v_vec4 %inloc\n"
1985 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
1986 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
1987 " OpStore %outloc %add\n"
1991 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
1992 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
1993 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1995 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
1997 // The following case copies a float[100] variable from the input buffer to the output buffer.
1998 ComputeShaderSpec spec2;
1999 vector<float> inputFloats2 (numElements);
2000 vector<float> outputFloats2 (numElements);
2002 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
2004 for (size_t ndx = 0; ndx < numElements; ++ndx)
2005 outputFloats2[ndx] = inputFloats2[ndx];
2008 string(getComputeAsmShaderPreamble()) +
2010 "OpName %main \"main\"\n"
2011 "OpName %id \"gl_GlobalInvocationID\"\n"
2013 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2014 "OpDecorate %f32arr100 ArrayStride 4\n"
2016 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2018 "%hundred = OpConstant %u32 100\n"
2019 "%f32arr100 = OpTypeArray %f32 %hundred\n"
2020 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
2021 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
2022 "%buf = OpTypeStruct %f32arr100\n"
2023 "%bufptr = OpTypePointer Uniform %buf\n"
2024 "%indata = OpVariable %bufptr Uniform\n"
2025 "%outdata = OpVariable %bufptr Uniform\n"
2027 "%id = OpVariable %uvec3ptr Input\n"
2028 "%zero = OpConstant %i32 0\n"
2030 "%main = OpFunction %void None %voidf\n"
2031 "%label = OpLabel\n"
2032 "%var = OpVariable %f32arr100ptr_f Function\n"
2033 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
2034 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
2035 " OpCopyMemory %var %inarr\n"
2036 " OpCopyMemory %outarr %var\n"
2040 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2041 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2042 spec2.numWorkGroups = IVec3(1, 1, 1);
2044 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
2046 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
2047 ComputeShaderSpec spec3;
2048 vector<float> inputFloats3 (16);
2049 vector<float> outputFloats3 (16);
2051 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
2053 for (size_t ndx = 0; ndx < 16; ++ndx)
2054 outputFloats3[ndx] = inputFloats3[ndx];
2057 string(getComputeAsmShaderPreamble()) +
2059 "OpName %main \"main\"\n"
2060 "OpName %id \"gl_GlobalInvocationID\"\n"
2062 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2063 "OpMemberDecorate %buf 0 Offset 0\n"
2064 "OpMemberDecorate %buf 1 Offset 16\n"
2065 "OpMemberDecorate %buf 2 Offset 32\n"
2066 "OpMemberDecorate %buf 3 Offset 48\n"
2068 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2070 "%vec4 = OpTypeVector %f32 4\n"
2071 "%buf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
2072 "%bufptr = OpTypePointer Uniform %buf\n"
2073 "%indata = OpVariable %bufptr Uniform\n"
2074 "%outdata = OpVariable %bufptr Uniform\n"
2075 "%vec4stptr = OpTypePointer Function %buf\n"
2077 "%id = OpVariable %uvec3ptr Input\n"
2078 "%zero = OpConstant %i32 0\n"
2080 "%main = OpFunction %void None %voidf\n"
2081 "%label = OpLabel\n"
2082 "%var = OpVariable %vec4stptr Function\n"
2083 " OpCopyMemory %var %indata\n"
2084 " OpCopyMemory %outdata %var\n"
2088 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2089 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
2090 spec3.numWorkGroups = IVec3(1, 1, 1);
2092 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
2094 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
2095 ComputeShaderSpec spec4;
2096 vector<float> inputFloats4 (numElements);
2097 vector<float> outputFloats4 (numElements);
2099 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
2101 for (size_t ndx = 0; ndx < numElements; ++ndx)
2102 outputFloats4[ndx] = -inputFloats4[ndx];
2105 string(getComputeAsmShaderPreamble()) +
2107 "OpName %main \"main\"\n"
2108 "OpName %id \"gl_GlobalInvocationID\"\n"
2110 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2112 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2114 "%f32ptr_f = OpTypePointer Function %f32\n"
2115 "%id = OpVariable %uvec3ptr Input\n"
2116 "%zero = OpConstant %i32 0\n"
2118 "%main = OpFunction %void None %voidf\n"
2119 "%label = OpLabel\n"
2120 "%var = OpVariable %f32ptr_f Function\n"
2121 "%idval = OpLoad %uvec3 %id\n"
2122 "%x = OpCompositeExtract %u32 %idval 0\n"
2123 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2124 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2125 " OpCopyMemory %var %inloc\n"
2126 "%val = OpLoad %f32 %var\n"
2127 "%neg = OpFNegate %f32 %val\n"
2128 " OpStore %outloc %neg\n"
2132 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2133 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
2134 spec4.numWorkGroups = IVec3(numElements, 1, 1);
2136 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
2138 return group.release();
2141 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
2143 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
2144 ComputeShaderSpec spec;
2145 de::Random rnd (deStringHash(group->getName()));
2146 const int numElements = 100;
2147 vector<float> inputFloats (numElements, 0);
2148 vector<float> outputFloats (numElements, 0);
2150 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
2152 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2153 floorAll(inputFloats);
2155 for (size_t ndx = 0; ndx < numElements; ++ndx)
2156 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
2159 string(getComputeAsmShaderPreamble()) +
2161 "OpName %main \"main\"\n"
2162 "OpName %id \"gl_GlobalInvocationID\"\n"
2164 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2166 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2168 "%fmat = OpTypeMatrix %fvec3 3\n"
2169 "%three = OpConstant %u32 3\n"
2170 "%farr = OpTypeArray %f32 %three\n"
2171 "%fst = OpTypeStruct %f32 %f32\n"
2173 + string(getComputeAsmInputOutputBuffer()) +
2175 "%id = OpVariable %uvec3ptr Input\n"
2176 "%zero = OpConstant %i32 0\n"
2177 "%c_f = OpConstant %f32 1.5\n"
2178 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
2179 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
2180 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
2181 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
2183 "%main = OpFunction %void None %voidf\n"
2184 "%label = OpLabel\n"
2185 "%c_f_copy = OpCopyObject %f32 %c_f\n"
2186 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
2187 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
2188 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
2189 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
2190 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
2191 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
2192 "%farr_elem = OpCompositeExtract %f32 %c_farr_copy 2\n"
2193 "%fst_elem = OpCompositeExtract %f32 %c_fst_copy 1\n"
2194 // Add up. 1.5 * 5 = 7.5.
2195 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
2196 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
2197 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
2198 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
2200 "%idval = OpLoad %uvec3 %id\n"
2201 "%x = OpCompositeExtract %u32 %idval 0\n"
2202 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2203 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2204 "%inval = OpLoad %f32 %inloc\n"
2205 "%add = OpFAdd %f32 %add4 %inval\n"
2206 " OpStore %outloc %add\n"
2209 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2210 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2211 spec.numWorkGroups = IVec3(numElements, 1, 1);
2213 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
2215 return group.release();
2217 // Assembly code used for testing OpUnreachable is based on GLSL source code:
2221 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2222 // float elements[];
2224 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2225 // float elements[];
2228 // void not_called_func() {
2229 // // place OpUnreachable here
2232 // uint modulo4(uint val) {
2233 // switch (val % uint(4)) {
2234 // case 0: return 3;
2235 // case 1: return 2;
2236 // case 2: return 1;
2237 // case 3: return 0;
2238 // default: return 100; // place OpUnreachable here
2244 // // place OpUnreachable here
2248 // uint x = gl_GlobalInvocationID.x;
2249 // if (const5() > modulo4(1000)) {
2250 // output_data.elements[x] = -input_data.elements[x];
2252 // // place OpUnreachable here
2253 // output_data.elements[x] = input_data.elements[x];
2257 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
2259 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
2260 ComputeShaderSpec spec;
2261 de::Random rnd (deStringHash(group->getName()));
2262 const int numElements = 100;
2263 vector<float> positiveFloats (numElements, 0);
2264 vector<float> negativeFloats (numElements, 0);
2266 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2268 for (size_t ndx = 0; ndx < numElements; ++ndx)
2269 negativeFloats[ndx] = -positiveFloats[ndx];
2272 string(getComputeAsmShaderPreamble()) +
2274 "OpSource GLSL 430\n"
2275 "OpName %main \"main\"\n"
2276 "OpName %func_not_called_func \"not_called_func(\"\n"
2277 "OpName %func_modulo4 \"modulo4(u1;\"\n"
2278 "OpName %func_const5 \"const5(\"\n"
2279 "OpName %id \"gl_GlobalInvocationID\"\n"
2281 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2283 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2285 "%u32ptr = OpTypePointer Function %u32\n"
2286 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
2287 "%unitf = OpTypeFunction %u32\n"
2289 "%id = OpVariable %uvec3ptr Input\n"
2290 "%zero = OpConstant %u32 0\n"
2291 "%one = OpConstant %u32 1\n"
2292 "%two = OpConstant %u32 2\n"
2293 "%three = OpConstant %u32 3\n"
2294 "%four = OpConstant %u32 4\n"
2295 "%five = OpConstant %u32 5\n"
2296 "%hundred = OpConstant %u32 100\n"
2297 "%thousand = OpConstant %u32 1000\n"
2299 + string(getComputeAsmInputOutputBuffer()) +
2302 "%main = OpFunction %void None %voidf\n"
2303 "%main_entry = OpLabel\n"
2304 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
2305 "%idval = OpLoad %uvec3 %id\n"
2306 "%x = OpCompositeExtract %u32 %idval 0\n"
2307 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2308 "%inval = OpLoad %f32 %inloc\n"
2309 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2310 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
2311 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
2312 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
2313 " OpSelectionMerge %if_end None\n"
2314 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2315 "%if_true = OpLabel\n"
2316 "%negate = OpFNegate %f32 %inval\n"
2317 " OpStore %outloc %negate\n"
2318 " OpBranch %if_end\n"
2319 "%if_false = OpLabel\n"
2320 " OpUnreachable\n" // Unreachable else branch for if statement
2321 "%if_end = OpLabel\n"
2325 // not_called_function()
2326 "%func_not_called_func = OpFunction %void None %voidf\n"
2327 "%not_called_func_entry = OpLabel\n"
2328 " OpUnreachable\n" // Unreachable entry block in not called static function
2332 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
2333 "%valptr = OpFunctionParameter %u32ptr\n"
2334 "%modulo4_entry = OpLabel\n"
2335 "%val = OpLoad %u32 %valptr\n"
2336 "%modulo = OpUMod %u32 %val %four\n"
2337 " OpSelectionMerge %switch_merge None\n"
2338 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
2339 "%case0 = OpLabel\n"
2340 " OpReturnValue %three\n"
2341 "%case1 = OpLabel\n"
2342 " OpReturnValue %two\n"
2343 "%case2 = OpLabel\n"
2344 " OpReturnValue %one\n"
2345 "%case3 = OpLabel\n"
2346 " OpReturnValue %zero\n"
2347 "%default = OpLabel\n"
2348 " OpUnreachable\n" // Unreachable default case for switch statement
2349 "%switch_merge = OpLabel\n"
2350 " OpUnreachable\n" // Unreachable merge block for switch statement
2354 "%func_const5 = OpFunction %u32 None %unitf\n"
2355 "%const5_entry = OpLabel\n"
2356 " OpReturnValue %five\n"
2357 "%unreachable = OpLabel\n"
2358 " OpUnreachable\n" // Unreachable block in function
2360 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2361 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2362 spec.numWorkGroups = IVec3(numElements, 1, 1);
2364 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
2366 return group.release();
2369 // Assembly code used for testing decoration group is based on GLSL source code:
2373 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
2374 // float elements[];
2376 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
2377 // float elements[];
2379 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
2380 // float elements[];
2382 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
2383 // float elements[];
2385 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
2386 // float elements[];
2388 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
2389 // float elements[];
2393 // uint x = gl_GlobalInvocationID.x;
2394 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
2396 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
2398 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
2399 ComputeShaderSpec spec;
2400 de::Random rnd (deStringHash(group->getName()));
2401 const int numElements = 100;
2402 vector<float> inputFloats0 (numElements, 0);
2403 vector<float> inputFloats1 (numElements, 0);
2404 vector<float> inputFloats2 (numElements, 0);
2405 vector<float> inputFloats3 (numElements, 0);
2406 vector<float> inputFloats4 (numElements, 0);
2407 vector<float> outputFloats (numElements, 0);
2409 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
2410 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
2411 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
2412 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
2413 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
2415 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2416 floorAll(inputFloats0);
2417 floorAll(inputFloats1);
2418 floorAll(inputFloats2);
2419 floorAll(inputFloats3);
2420 floorAll(inputFloats4);
2422 for (size_t ndx = 0; ndx < numElements; ++ndx)
2423 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
2426 string(getComputeAsmShaderPreamble()) +
2428 "OpSource GLSL 430\n"
2429 "OpName %main \"main\"\n"
2430 "OpName %id \"gl_GlobalInvocationID\"\n"
2432 // Not using group decoration on variable.
2433 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2434 // Not using group decoration on type.
2435 "OpDecorate %f32arr ArrayStride 4\n"
2437 "OpDecorate %groups BufferBlock\n"
2438 "OpDecorate %groupm Offset 0\n"
2439 "%groups = OpDecorationGroup\n"
2440 "%groupm = OpDecorationGroup\n"
2442 // Group decoration on multiple structs.
2443 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
2444 // Group decoration on multiple struct members.
2445 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
2447 "OpDecorate %group1 DescriptorSet 0\n"
2448 "OpDecorate %group3 DescriptorSet 0\n"
2449 "OpDecorate %group3 NonWritable\n"
2450 "OpDecorate %group3 Restrict\n"
2451 "%group0 = OpDecorationGroup\n"
2452 "%group1 = OpDecorationGroup\n"
2453 "%group3 = OpDecorationGroup\n"
2455 // Applying the same decoration group multiple times.
2456 "OpGroupDecorate %group1 %outdata\n"
2457 "OpGroupDecorate %group1 %outdata\n"
2458 "OpGroupDecorate %group1 %outdata\n"
2459 "OpDecorate %outdata DescriptorSet 0\n"
2460 "OpDecorate %outdata Binding 5\n"
2461 // Applying decoration group containing nothing.
2462 "OpGroupDecorate %group0 %indata0\n"
2463 "OpDecorate %indata0 DescriptorSet 0\n"
2464 "OpDecorate %indata0 Binding 0\n"
2465 // Applying decoration group containing one decoration.
2466 "OpGroupDecorate %group1 %indata1\n"
2467 "OpDecorate %indata1 Binding 1\n"
2468 // Applying decoration group containing multiple decorations.
2469 "OpGroupDecorate %group3 %indata2 %indata3\n"
2470 "OpDecorate %indata2 Binding 2\n"
2471 "OpDecorate %indata3 Binding 3\n"
2472 // Applying multiple decoration groups (with overlapping).
2473 "OpGroupDecorate %group0 %indata4\n"
2474 "OpGroupDecorate %group1 %indata4\n"
2475 "OpGroupDecorate %group3 %indata4\n"
2476 "OpDecorate %indata4 Binding 4\n"
2478 + string(getComputeAsmCommonTypes()) +
2480 "%id = OpVariable %uvec3ptr Input\n"
2481 "%zero = OpConstant %i32 0\n"
2483 "%outbuf = OpTypeStruct %f32arr\n"
2484 "%outbufptr = OpTypePointer Uniform %outbuf\n"
2485 "%outdata = OpVariable %outbufptr Uniform\n"
2486 "%inbuf0 = OpTypeStruct %f32arr\n"
2487 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
2488 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
2489 "%inbuf1 = OpTypeStruct %f32arr\n"
2490 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
2491 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
2492 "%inbuf2 = OpTypeStruct %f32arr\n"
2493 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
2494 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
2495 "%inbuf3 = OpTypeStruct %f32arr\n"
2496 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
2497 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
2498 "%inbuf4 = OpTypeStruct %f32arr\n"
2499 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
2500 "%indata4 = OpVariable %inbufptr Uniform\n"
2502 "%main = OpFunction %void None %voidf\n"
2503 "%label = OpLabel\n"
2504 "%idval = OpLoad %uvec3 %id\n"
2505 "%x = OpCompositeExtract %u32 %idval 0\n"
2506 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
2507 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
2508 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
2509 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
2510 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
2511 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2512 "%inval0 = OpLoad %f32 %inloc0\n"
2513 "%inval1 = OpLoad %f32 %inloc1\n"
2514 "%inval2 = OpLoad %f32 %inloc2\n"
2515 "%inval3 = OpLoad %f32 %inloc3\n"
2516 "%inval4 = OpLoad %f32 %inloc4\n"
2517 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
2518 "%add1 = OpFAdd %f32 %add0 %inval2\n"
2519 "%add2 = OpFAdd %f32 %add1 %inval3\n"
2520 "%add = OpFAdd %f32 %add2 %inval4\n"
2521 " OpStore %outloc %add\n"
2524 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
2525 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
2526 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2527 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2528 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2529 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2530 spec.numWorkGroups = IVec3(numElements, 1, 1);
2532 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
2534 return group.release();
2537 struct SpecConstantTwoIntCase
2539 const char* caseName;
2540 const char* scDefinition0;
2541 const char* scDefinition1;
2542 const char* scResultType;
2543 const char* scOperation;
2544 deInt32 scActualValue0;
2545 deInt32 scActualValue1;
2546 const char* resultOperation;
2547 vector<deInt32> expectedOutput;
2549 SpecConstantTwoIntCase (const char* name,
2550 const char* definition0,
2551 const char* definition1,
2552 const char* resultType,
2553 const char* operation,
2556 const char* resultOp,
2557 const vector<deInt32>& output)
2559 , scDefinition0 (definition0)
2560 , scDefinition1 (definition1)
2561 , scResultType (resultType)
2562 , scOperation (operation)
2563 , scActualValue0 (value0)
2564 , scActualValue1 (value1)
2565 , resultOperation (resultOp)
2566 , expectedOutput (output) {}
2569 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
2571 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
2572 vector<SpecConstantTwoIntCase> cases;
2573 de::Random rnd (deStringHash(group->getName()));
2574 const int numElements = 100;
2575 vector<deInt32> inputInts (numElements, 0);
2576 vector<deInt32> outputInts1 (numElements, 0);
2577 vector<deInt32> outputInts2 (numElements, 0);
2578 vector<deInt32> outputInts3 (numElements, 0);
2579 vector<deInt32> outputInts4 (numElements, 0);
2580 const StringTemplate shaderTemplate (
2581 "${CAPABILITIES:opt}"
2582 + string(getComputeAsmShaderPreamble()) +
2584 "OpName %main \"main\"\n"
2585 "OpName %id \"gl_GlobalInvocationID\"\n"
2587 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2588 "OpDecorate %sc_0 SpecId 0\n"
2589 "OpDecorate %sc_1 SpecId 1\n"
2590 "OpDecorate %i32arr ArrayStride 4\n"
2592 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2594 "${OPTYPE_DEFINITIONS:opt}"
2595 "%buf = OpTypeStruct %i32arr\n"
2596 "%bufptr = OpTypePointer Uniform %buf\n"
2597 "%indata = OpVariable %bufptr Uniform\n"
2598 "%outdata = OpVariable %bufptr Uniform\n"
2600 "%id = OpVariable %uvec3ptr Input\n"
2601 "%zero = OpConstant %i32 0\n"
2603 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
2604 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
2605 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
2607 "%main = OpFunction %void None %voidf\n"
2608 "%label = OpLabel\n"
2609 "${TYPE_CONVERT:opt}"
2610 "%idval = OpLoad %uvec3 %id\n"
2611 "%x = OpCompositeExtract %u32 %idval 0\n"
2612 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2613 "%inval = OpLoad %i32 %inloc\n"
2614 "%final = ${GEN_RESULT}\n"
2615 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2616 " OpStore %outloc %final\n"
2618 " OpFunctionEnd\n");
2620 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
2622 for (size_t ndx = 0; ndx < numElements; ++ndx)
2624 outputInts1[ndx] = inputInts[ndx] + 42;
2625 outputInts2[ndx] = inputInts[ndx];
2626 outputInts3[ndx] = inputInts[ndx] - 11200;
2627 outputInts4[ndx] = inputInts[ndx] + 1;
2630 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
2631 const char addSc32ToInput[] = "OpIAdd %i32 %inval %sc_final32";
2632 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
2633 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
2635 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
2636 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
2637 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
2638 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
2639 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
2640 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2641 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2642 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
2643 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
2644 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
2645 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
2646 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2647 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2648 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
2649 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
2650 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
2651 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2652 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
2653 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
2654 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
2655 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2656 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
2657 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
2658 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2659 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2660 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2661 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2662 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
2663 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
2664 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
2665 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
2666 cases.push_back(SpecConstantTwoIntCase("sconvert", " %i32 0", " %i32 0", "%i16", "SConvert %sc_0", -11200, 0, addSc32ToInput, outputInts3));
2667 // -969998336 stored as 32-bit two's complement is the binary representation of -11200 as IEEE-754 Float
2668 cases.push_back(SpecConstantTwoIntCase("fconvert", " %f32 0", " %f32 0", "%f64", "FConvert %sc_0", -969998336, 0, addSc32ToInput, outputInts3));
2670 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2672 map<string, string> specializations;
2673 ComputeShaderSpec spec;
2674 ComputeTestFeatures features = COMPUTE_TEST_USES_NONE;
2676 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
2677 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
2678 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
2679 specializations["SC_OP"] = cases[caseNdx].scOperation;
2680 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
2682 // Special SPIR-V code for SConvert-case
2683 if (strcmp(cases[caseNdx].caseName, "sconvert") == 0)
2685 features = COMPUTE_TEST_USES_INT16;
2686 specializations["CAPABILITIES"] = "OpCapability Int16\n"; // Adds 16-bit integer capability
2687 specializations["OPTYPE_DEFINITIONS"] = "%i16 = OpTypeInt 16 1\n"; // Adds 16-bit integer type
2688 specializations["TYPE_CONVERT"] = "%sc_final32 = OpSConvert %i32 %sc_final\n"; // Converts 16-bit integer to 32-bit integer
2691 // Special SPIR-V code for FConvert-case
2692 if (strcmp(cases[caseNdx].caseName, "fconvert") == 0)
2694 features = COMPUTE_TEST_USES_FLOAT64;
2695 specializations["CAPABILITIES"] = "OpCapability Float64\n"; // Adds 64-bit float capability
2696 specializations["OPTYPE_DEFINITIONS"] = "%f64 = OpTypeFloat 64\n"; // Adds 64-bit float type
2697 specializations["TYPE_CONVERT"] = "%sc_final32 = OpConvertFToS %i32 %sc_final\n"; // Converts 64-bit float to 32-bit integer
2700 spec.assembly = shaderTemplate.specialize(specializations);
2701 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2702 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
2703 spec.numWorkGroups = IVec3(numElements, 1, 1);
2704 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
2705 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
2707 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec, features));
2710 ComputeShaderSpec spec;
2713 string(getComputeAsmShaderPreamble()) +
2715 "OpName %main \"main\"\n"
2716 "OpName %id \"gl_GlobalInvocationID\"\n"
2718 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2719 "OpDecorate %sc_0 SpecId 0\n"
2720 "OpDecorate %sc_1 SpecId 1\n"
2721 "OpDecorate %sc_2 SpecId 2\n"
2722 "OpDecorate %i32arr ArrayStride 4\n"
2724 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2726 "%ivec3 = OpTypeVector %i32 3\n"
2727 "%buf = OpTypeStruct %i32arr\n"
2728 "%bufptr = OpTypePointer Uniform %buf\n"
2729 "%indata = OpVariable %bufptr Uniform\n"
2730 "%outdata = OpVariable %bufptr Uniform\n"
2732 "%id = OpVariable %uvec3ptr Input\n"
2733 "%zero = OpConstant %i32 0\n"
2734 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
2735 "%vec3_undef = OpUndef %ivec3\n"
2737 "%sc_0 = OpSpecConstant %i32 0\n"
2738 "%sc_1 = OpSpecConstant %i32 0\n"
2739 "%sc_2 = OpSpecConstant %i32 0\n"
2740 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
2741 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
2742 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
2743 "%sc_vec3_0_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
2744 "%sc_vec3_1_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
2745 "%sc_vec3_2_s = OpSpecConstantOp %ivec3 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
2746 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
2747 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
2748 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
2749 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
2750 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
2751 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
2752 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
2754 "%main = OpFunction %void None %voidf\n"
2755 "%label = OpLabel\n"
2756 "%idval = OpLoad %uvec3 %id\n"
2757 "%x = OpCompositeExtract %u32 %idval 0\n"
2758 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2759 "%inval = OpLoad %i32 %inloc\n"
2760 "%final = OpIAdd %i32 %inval %sc_final\n"
2761 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2762 " OpStore %outloc %final\n"
2765 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2766 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
2767 spec.numWorkGroups = IVec3(numElements, 1, 1);
2768 spec.specConstants.push_back(123);
2769 spec.specConstants.push_back(56);
2770 spec.specConstants.push_back(-77);
2772 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
2774 return group.release();
2777 void createOpPhiVartypeTests (de::MovePtr<tcu::TestCaseGroup>& group, tcu::TestContext& testCtx)
2779 ComputeShaderSpec specInt;
2780 ComputeShaderSpec specFloat;
2781 ComputeShaderSpec specVec3;
2782 ComputeShaderSpec specMat4;
2783 ComputeShaderSpec specArray;
2784 ComputeShaderSpec specStruct;
2785 de::Random rnd (deStringHash(group->getName()));
2786 const int numElements = 100;
2787 vector<float> inputFloats (numElements, 0);
2788 vector<float> outputFloats (numElements, 0);
2790 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
2792 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2793 floorAll(inputFloats);
2795 for (size_t ndx = 0; ndx < numElements; ++ndx)
2797 // Just check if the value is positive or not
2798 outputFloats[ndx] = (inputFloats[ndx] > 0) ? 1.0f : -1.0f;
2801 // All of the tests are of the form:
2805 // if (inputdata > 0)
2812 specFloat.assembly =
2813 string(getComputeAsmShaderPreamble()) +
2815 "OpSource GLSL 430\n"
2816 "OpName %main \"main\"\n"
2817 "OpName %id \"gl_GlobalInvocationID\"\n"
2819 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2821 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2823 "%id = OpVariable %uvec3ptr Input\n"
2824 "%zero = OpConstant %i32 0\n"
2825 "%float_0 = OpConstant %f32 0.0\n"
2826 "%float_1 = OpConstant %f32 1.0\n"
2827 "%float_n1 = OpConstant %f32 -1.0\n"
2829 "%main = OpFunction %void None %voidf\n"
2830 "%entry = OpLabel\n"
2831 "%idval = OpLoad %uvec3 %id\n"
2832 "%x = OpCompositeExtract %u32 %idval 0\n"
2833 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2834 "%inval = OpLoad %f32 %inloc\n"
2836 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2837 " OpSelectionMerge %cm None\n"
2838 " OpBranchConditional %comp %tb %fb\n"
2844 "%res = OpPhi %f32 %float_1 %tb %float_n1 %fb\n"
2846 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2847 " OpStore %outloc %res\n"
2851 specFloat.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2852 specFloat.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2853 specFloat.numWorkGroups = IVec3(numElements, 1, 1);
2856 string(getComputeAsmShaderPreamble()) +
2858 "OpSource GLSL 430\n"
2859 "OpName %main \"main\"\n"
2860 "OpName %id \"gl_GlobalInvocationID\"\n"
2862 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2864 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2866 "%id = OpVariable %uvec3ptr Input\n"
2867 "%v4f32 = OpTypeVector %f32 4\n"
2868 "%mat4v4f32 = OpTypeMatrix %v4f32 4\n"
2869 "%zero = OpConstant %i32 0\n"
2870 "%float_0 = OpConstant %f32 0.0\n"
2871 "%float_1 = OpConstant %f32 1.0\n"
2872 "%float_n1 = OpConstant %f32 -1.0\n"
2873 "%m11 = OpConstantComposite %v4f32 %float_1 %float_0 %float_0 %float_0\n"
2874 "%m12 = OpConstantComposite %v4f32 %float_0 %float_1 %float_0 %float_0\n"
2875 "%m13 = OpConstantComposite %v4f32 %float_0 %float_0 %float_1 %float_0\n"
2876 "%m14 = OpConstantComposite %v4f32 %float_0 %float_0 %float_0 %float_1\n"
2877 "%m1 = OpConstantComposite %mat4v4f32 %m11 %m12 %m13 %m14\n"
2878 "%m21 = OpConstantComposite %v4f32 %float_n1 %float_0 %float_0 %float_0\n"
2879 "%m22 = OpConstantComposite %v4f32 %float_0 %float_n1 %float_0 %float_0\n"
2880 "%m23 = OpConstantComposite %v4f32 %float_0 %float_0 %float_n1 %float_0\n"
2881 "%m24 = OpConstantComposite %v4f32 %float_0 %float_0 %float_0 %float_n1\n"
2882 "%m2 = OpConstantComposite %mat4v4f32 %m21 %m22 %m23 %m24\n"
2884 "%main = OpFunction %void None %voidf\n"
2885 "%entry = OpLabel\n"
2886 "%idval = OpLoad %uvec3 %id\n"
2887 "%x = OpCompositeExtract %u32 %idval 0\n"
2888 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2889 "%inval = OpLoad %f32 %inloc\n"
2891 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2892 " OpSelectionMerge %cm None\n"
2893 " OpBranchConditional %comp %tb %fb\n"
2899 "%mres = OpPhi %mat4v4f32 %m1 %tb %m2 %fb\n"
2900 "%res = OpCompositeExtract %f32 %mres 2 2\n"
2902 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2903 " OpStore %outloc %res\n"
2907 specMat4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2908 specMat4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2909 specMat4.numWorkGroups = IVec3(numElements, 1, 1);
2912 string(getComputeAsmShaderPreamble()) +
2914 "OpSource GLSL 430\n"
2915 "OpName %main \"main\"\n"
2916 "OpName %id \"gl_GlobalInvocationID\"\n"
2918 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2920 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2922 "%id = OpVariable %uvec3ptr Input\n"
2923 "%zero = OpConstant %i32 0\n"
2924 "%float_0 = OpConstant %f32 0.0\n"
2925 "%float_1 = OpConstant %f32 1.0\n"
2926 "%float_n1 = OpConstant %f32 -1.0\n"
2927 "%v1 = OpConstantComposite %fvec3 %float_1 %float_1 %float_1\n"
2928 "%v2 = OpConstantComposite %fvec3 %float_n1 %float_n1 %float_n1\n"
2930 "%main = OpFunction %void None %voidf\n"
2931 "%entry = OpLabel\n"
2932 "%idval = OpLoad %uvec3 %id\n"
2933 "%x = OpCompositeExtract %u32 %idval 0\n"
2934 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2935 "%inval = OpLoad %f32 %inloc\n"
2937 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2938 " OpSelectionMerge %cm None\n"
2939 " OpBranchConditional %comp %tb %fb\n"
2945 "%vres = OpPhi %fvec3 %v1 %tb %v2 %fb\n"
2946 "%res = OpCompositeExtract %f32 %vres 2\n"
2948 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2949 " OpStore %outloc %res\n"
2953 specVec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2954 specVec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2955 specVec3.numWorkGroups = IVec3(numElements, 1, 1);
2958 string(getComputeAsmShaderPreamble()) +
2960 "OpSource GLSL 430\n"
2961 "OpName %main \"main\"\n"
2962 "OpName %id \"gl_GlobalInvocationID\"\n"
2964 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2966 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2968 "%id = OpVariable %uvec3ptr Input\n"
2969 "%zero = OpConstant %i32 0\n"
2970 "%float_0 = OpConstant %f32 0.0\n"
2971 "%i1 = OpConstant %i32 1\n"
2972 "%i2 = OpConstant %i32 -1\n"
2974 "%main = OpFunction %void None %voidf\n"
2975 "%entry = OpLabel\n"
2976 "%idval = OpLoad %uvec3 %id\n"
2977 "%x = OpCompositeExtract %u32 %idval 0\n"
2978 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2979 "%inval = OpLoad %f32 %inloc\n"
2981 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2982 " OpSelectionMerge %cm None\n"
2983 " OpBranchConditional %comp %tb %fb\n"
2989 "%ires = OpPhi %i32 %i1 %tb %i2 %fb\n"
2990 "%res = OpConvertSToF %f32 %ires\n"
2992 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2993 " OpStore %outloc %res\n"
2997 specInt.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2998 specInt.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2999 specInt.numWorkGroups = IVec3(numElements, 1, 1);
3001 specArray.assembly =
3002 string(getComputeAsmShaderPreamble()) +
3004 "OpSource GLSL 430\n"
3005 "OpName %main \"main\"\n"
3006 "OpName %id \"gl_GlobalInvocationID\"\n"
3008 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3010 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3012 "%id = OpVariable %uvec3ptr Input\n"
3013 "%zero = OpConstant %i32 0\n"
3014 "%u7 = OpConstant %u32 7\n"
3015 "%float_0 = OpConstant %f32 0.0\n"
3016 "%float_1 = OpConstant %f32 1.0\n"
3017 "%float_n1 = OpConstant %f32 -1.0\n"
3018 "%f32a7 = OpTypeArray %f32 %u7\n"
3019 "%a1 = OpConstantComposite %f32a7 %float_1 %float_1 %float_1 %float_1 %float_1 %float_1 %float_1\n"
3020 "%a2 = OpConstantComposite %f32a7 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1\n"
3021 "%main = OpFunction %void None %voidf\n"
3022 "%entry = OpLabel\n"
3023 "%idval = OpLoad %uvec3 %id\n"
3024 "%x = OpCompositeExtract %u32 %idval 0\n"
3025 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3026 "%inval = OpLoad %f32 %inloc\n"
3028 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3029 " OpSelectionMerge %cm None\n"
3030 " OpBranchConditional %comp %tb %fb\n"
3036 "%ares = OpPhi %f32a7 %a1 %tb %a2 %fb\n"
3037 "%res = OpCompositeExtract %f32 %ares 5\n"
3039 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3040 " OpStore %outloc %res\n"
3044 specArray.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3045 specArray.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3046 specArray.numWorkGroups = IVec3(numElements, 1, 1);
3048 specStruct.assembly =
3049 string(getComputeAsmShaderPreamble()) +
3051 "OpSource GLSL 430\n"
3052 "OpName %main \"main\"\n"
3053 "OpName %id \"gl_GlobalInvocationID\"\n"
3055 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3057 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3059 "%id = OpVariable %uvec3ptr Input\n"
3060 "%zero = OpConstant %i32 0\n"
3061 "%float_0 = OpConstant %f32 0.0\n"
3062 "%float_1 = OpConstant %f32 1.0\n"
3063 "%float_n1 = OpConstant %f32 -1.0\n"
3065 "%v2f32 = OpTypeVector %f32 2\n"
3066 "%Data2 = OpTypeStruct %f32 %v2f32\n"
3067 "%Data = OpTypeStruct %Data2 %f32\n"
3069 "%in1a = OpConstantComposite %v2f32 %float_1 %float_1\n"
3070 "%in1b = OpConstantComposite %Data2 %float_1 %in1a\n"
3071 "%s1 = OpConstantComposite %Data %in1b %float_1\n"
3072 "%in2a = OpConstantComposite %v2f32 %float_n1 %float_n1\n"
3073 "%in2b = OpConstantComposite %Data2 %float_n1 %in2a\n"
3074 "%s2 = OpConstantComposite %Data %in2b %float_n1\n"
3076 "%main = OpFunction %void None %voidf\n"
3077 "%entry = OpLabel\n"
3078 "%idval = OpLoad %uvec3 %id\n"
3079 "%x = OpCompositeExtract %u32 %idval 0\n"
3080 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3081 "%inval = OpLoad %f32 %inloc\n"
3083 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3084 " OpSelectionMerge %cm None\n"
3085 " OpBranchConditional %comp %tb %fb\n"
3091 "%sres = OpPhi %Data %s1 %tb %s2 %fb\n"
3092 "%res = OpCompositeExtract %f32 %sres 0 0\n"
3094 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3095 " OpStore %outloc %res\n"
3099 specStruct.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3100 specStruct.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3101 specStruct.numWorkGroups = IVec3(numElements, 1, 1);
3103 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_int", "OpPhi with int variables", specInt));
3104 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_float", "OpPhi with float variables", specFloat));
3105 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_vec3", "OpPhi with vec3 variables", specVec3));
3106 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_mat4", "OpPhi with mat4 variables", specMat4));
3107 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_array", "OpPhi with array variables", specArray));
3108 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_struct", "OpPhi with struct variables", specStruct));
3111 string generateConstantDefinitions (int count)
3113 std::ostringstream r;
3114 for (int i = 0; i < count; i++)
3115 r << "%cf" << (i * 10 + 5) << " = OpConstant %f32 " <<(i * 10 + 5) << ".0\n";
3120 string generateSwitchCases (int count)
3122 std::ostringstream r;
3123 for (int i = 0; i < count; i++)
3124 r << " " << i << " %case" << i;
3129 string generateSwitchTargets (int count)
3131 std::ostringstream r;
3132 for (int i = 0; i < count; i++)
3133 r << "%case" << i << " = OpLabel\n OpBranch %phi\n";
3138 string generateOpPhiParams (int count)
3140 std::ostringstream r;
3141 for (int i = 0; i < count; i++)
3142 r << " %cf" << (i * 10 + 5) << " %case" << i;
3147 string generateIntWidth (int value)
3149 std::ostringstream r;
3154 // Expand input string by injecting "ABC" between the input
3155 // string characters. The acc/add/treshold parameters are used
3156 // to skip some of the injections to make the result less
3157 // uniform (and a lot shorter).
3158 string expandOpPhiCase5 (const string& s, int &acc, int add, int treshold)
3160 std::ostringstream res;
3161 const char* p = s.c_str();
3177 // Calculate expected result based on the code string
3178 float calcOpPhiCase5 (float val, const string& s)
3180 const char* p = s.c_str();
3183 const float tv[8] = { 0.5f, 1.5f, 3.5f, 7.5f, 15.5f, 31.5f, 63.5f, 127.5f };
3184 const float v = deFloatAbs(val);
3189 for (int i = 7; i >= 0; --i)
3190 x[i] = std::fmod((float)v, (float)(2 << i));
3191 for (int i = 7; i >= 0; --i)
3192 b[i] = x[i] > tv[i];
3199 if (skip == 0 && b[depth])
3210 if (b[depth] || skip)
3224 // In the code string, the letters represent the following:
3227 // if (certain bit is set)
3238 // AABCBC leads to if(){r++;if(){r++;}else{}}else{}
3239 // ABABCC leads to if(){r++;}else{if(){r++;}else{}}
3240 // ABCABC leads to if(){r++;}else{}if(){r++;}else{}
3242 // Code generation gets a bit complicated due to the else-branches,
3243 // which do not generate new values. Thus, the generator needs to
3244 // keep track of the previous variable change seen by the else
3246 string generateOpPhiCase5 (const string& s)
3248 std::stack<int> idStack;
3249 std::stack<std::string> value;
3250 std::stack<std::string> valueLabel;
3251 std::stack<std::string> mergeLeft;
3252 std::stack<std::string> mergeRight;
3253 std::ostringstream res;
3254 const char* p = s.c_str();
3260 value.push("%f32_0");
3261 valueLabel.push("%f32_0 %entry");
3269 idStack.push(currId);
3270 res << "\tOpSelectionMerge %m" << currId << " None\n";
3271 res << "\tOpBranchConditional %b" << depth << " %t" << currId << " %f" << currId << "\n";
3272 res << "%t" << currId << " = OpLabel\n";
3273 res << "%rt" << currId << " = OpFAdd %f32 " << value.top() << " %f32_1\n";
3274 std::ostringstream tag;
3275 tag << "%rt" << currId;
3276 value.push(tag.str());
3277 tag << " %t" << currId;
3278 valueLabel.push(tag.str());
3283 mergeLeft.push(valueLabel.top());
3286 res << "\tOpBranch %m" << currId << "\n";
3287 res << "%f" << currId << " = OpLabel\n";
3288 std::ostringstream tag;
3289 tag << value.top() << " %f" << currId;
3291 valueLabel.push(tag.str());
3296 mergeRight.push(valueLabel.top());
3297 res << "\tOpBranch %m" << currId << "\n";
3298 res << "%m" << currId << " = OpLabel\n";
3300 res << "%res"; // last result goes to %res
3302 res << "%rm" << currId;
3303 res << " = OpPhi %f32 " << mergeLeft.top() << " " << mergeRight.top() << "\n";
3304 std::ostringstream tag;
3305 tag << "%rm" << currId;
3307 value.push(tag.str());
3308 tag << " %m" << currId;
3310 valueLabel.push(tag.str());
3315 currId = idStack.top();
3323 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
3325 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
3326 ComputeShaderSpec spec1;
3327 ComputeShaderSpec spec2;
3328 ComputeShaderSpec spec3;
3329 ComputeShaderSpec spec4;
3330 ComputeShaderSpec spec5;
3331 de::Random rnd (deStringHash(group->getName()));
3332 const int numElements = 100;
3333 vector<float> inputFloats (numElements, 0);
3334 vector<float> outputFloats1 (numElements, 0);
3335 vector<float> outputFloats2 (numElements, 0);
3336 vector<float> outputFloats3 (numElements, 0);
3337 vector<float> outputFloats4 (numElements, 0);
3338 vector<float> outputFloats5 (numElements, 0);
3339 std::string codestring = "ABC";
3340 const int test4Width = 1024;
3342 // Build case 5 code string. Each iteration makes the hierarchy more complicated.
3343 // 9 iterations with (7, 24) parameters makes the hierarchy 8 deep with about 1500 lines of
3345 for (int i = 0, acc = 0; i < 9; i++)
3346 codestring = expandOpPhiCase5(codestring, acc, 7, 24);
3348 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
3350 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3351 floorAll(inputFloats);
3353 for (size_t ndx = 0; ndx < numElements; ++ndx)
3357 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
3358 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
3359 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
3362 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
3363 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
3365 int index4 = (int)deFloor(deAbs((float)ndx * inputFloats[ndx]));
3366 outputFloats4[ndx] = (float)(index4 % test4Width) * 10.0f + 5.0f;
3368 outputFloats5[ndx] = calcOpPhiCase5(inputFloats[ndx], codestring);
3372 string(getComputeAsmShaderPreamble()) +
3374 "OpSource GLSL 430\n"
3375 "OpName %main \"main\"\n"
3376 "OpName %id \"gl_GlobalInvocationID\"\n"
3378 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3380 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3382 "%id = OpVariable %uvec3ptr Input\n"
3383 "%zero = OpConstant %i32 0\n"
3384 "%three = OpConstant %u32 3\n"
3385 "%constf5p5 = OpConstant %f32 5.5\n"
3386 "%constf20p5 = OpConstant %f32 20.5\n"
3387 "%constf1p75 = OpConstant %f32 1.75\n"
3388 "%constf8p5 = OpConstant %f32 8.5\n"
3389 "%constf6p5 = OpConstant %f32 6.5\n"
3391 "%main = OpFunction %void None %voidf\n"
3392 "%entry = OpLabel\n"
3393 "%idval = OpLoad %uvec3 %id\n"
3394 "%x = OpCompositeExtract %u32 %idval 0\n"
3395 "%selector = OpUMod %u32 %x %three\n"
3396 " OpSelectionMerge %phi None\n"
3397 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
3399 // Case 1 before OpPhi.
3400 "%case1 = OpLabel\n"
3403 "%default = OpLabel\n"
3407 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
3408 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3409 "%inval = OpLoad %f32 %inloc\n"
3410 "%add = OpFAdd %f32 %inval %operand\n"
3411 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3412 " OpStore %outloc %add\n"
3415 // Case 0 after OpPhi.
3416 "%case0 = OpLabel\n"
3420 // Case 2 after OpPhi.
3421 "%case2 = OpLabel\n"
3425 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3426 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
3427 spec1.numWorkGroups = IVec3(numElements, 1, 1);
3429 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
3432 string(getComputeAsmShaderPreamble()) +
3434 "OpName %main \"main\"\n"
3435 "OpName %id \"gl_GlobalInvocationID\"\n"
3437 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3439 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3441 "%id = OpVariable %uvec3ptr Input\n"
3442 "%zero = OpConstant %i32 0\n"
3443 "%one = OpConstant %i32 1\n"
3444 "%three = OpConstant %i32 3\n"
3445 "%constf6p5 = OpConstant %f32 6.5\n"
3447 "%main = OpFunction %void None %voidf\n"
3448 "%entry = OpLabel\n"
3449 "%idval = OpLoad %uvec3 %id\n"
3450 "%x = OpCompositeExtract %u32 %idval 0\n"
3451 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3452 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3453 "%inval = OpLoad %f32 %inloc\n"
3457 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
3458 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
3459 "%step_next = OpIAdd %i32 %step %one\n"
3460 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
3461 "%still_loop = OpSLessThan %bool %step %three\n"
3462 " OpLoopMerge %exit %phi None\n"
3463 " OpBranchConditional %still_loop %phi %exit\n"
3466 " OpStore %outloc %accum\n"
3469 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3470 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
3471 spec2.numWorkGroups = IVec3(numElements, 1, 1);
3473 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
3476 string(getComputeAsmShaderPreamble()) +
3478 "OpName %main \"main\"\n"
3479 "OpName %id \"gl_GlobalInvocationID\"\n"
3481 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3483 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3485 "%f32ptr_f = OpTypePointer Function %f32\n"
3486 "%id = OpVariable %uvec3ptr Input\n"
3487 "%true = OpConstantTrue %bool\n"
3488 "%false = OpConstantFalse %bool\n"
3489 "%zero = OpConstant %i32 0\n"
3490 "%constf8p5 = OpConstant %f32 8.5\n"
3492 "%main = OpFunction %void None %voidf\n"
3493 "%entry = OpLabel\n"
3494 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
3495 "%idval = OpLoad %uvec3 %id\n"
3496 "%x = OpCompositeExtract %u32 %idval 0\n"
3497 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3498 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3499 "%a_init = OpLoad %f32 %inloc\n"
3500 "%b_init = OpLoad %f32 %b\n"
3504 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
3505 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
3506 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
3507 " OpLoopMerge %exit %phi None\n"
3508 " OpBranchConditional %still_loop %phi %exit\n"
3511 "%sub = OpFSub %f32 %a_next %b_next\n"
3512 " OpStore %outloc %sub\n"
3515 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3516 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
3517 spec3.numWorkGroups = IVec3(numElements, 1, 1);
3519 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
3522 "OpCapability Shader\n"
3523 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
3524 "OpMemoryModel Logical GLSL450\n"
3525 "OpEntryPoint GLCompute %main \"main\" %id\n"
3526 "OpExecutionMode %main LocalSize 1 1 1\n"
3528 "OpSource GLSL 430\n"
3529 "OpName %main \"main\"\n"
3530 "OpName %id \"gl_GlobalInvocationID\"\n"
3532 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3534 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3536 "%id = OpVariable %uvec3ptr Input\n"
3537 "%zero = OpConstant %i32 0\n"
3538 "%cimod = OpConstant %u32 " + generateIntWidth(test4Width) + "\n"
3540 + generateConstantDefinitions(test4Width) +
3542 "%main = OpFunction %void None %voidf\n"
3543 "%entry = OpLabel\n"
3544 "%idval = OpLoad %uvec3 %id\n"
3545 "%x = OpCompositeExtract %u32 %idval 0\n"
3546 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3547 "%inval = OpLoad %f32 %inloc\n"
3548 "%xf = OpConvertUToF %f32 %x\n"
3549 "%xm = OpFMul %f32 %xf %inval\n"
3550 "%xa = OpExtInst %f32 %ext FAbs %xm\n"
3551 "%xi = OpConvertFToU %u32 %xa\n"
3552 "%selector = OpUMod %u32 %xi %cimod\n"
3553 " OpSelectionMerge %phi None\n"
3554 " OpSwitch %selector %default "
3556 + generateSwitchCases(test4Width) +
3558 "%default = OpLabel\n"
3561 + generateSwitchTargets(test4Width) +
3564 "%result = OpPhi %f32"
3566 + generateOpPhiParams(test4Width) +
3568 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3569 " OpStore %outloc %result\n"
3573 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3574 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
3575 spec4.numWorkGroups = IVec3(numElements, 1, 1);
3577 group->addChild(new SpvAsmComputeShaderCase(testCtx, "wide", "OpPhi with a lot of parameters", spec4));
3580 "OpCapability Shader\n"
3581 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
3582 "OpMemoryModel Logical GLSL450\n"
3583 "OpEntryPoint GLCompute %main \"main\" %id\n"
3584 "OpExecutionMode %main LocalSize 1 1 1\n"
3585 "%code = OpString \"" + codestring + "\"\n"
3587 "OpSource GLSL 430\n"
3588 "OpName %main \"main\"\n"
3589 "OpName %id \"gl_GlobalInvocationID\"\n"
3591 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3593 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3595 "%id = OpVariable %uvec3ptr Input\n"
3596 "%zero = OpConstant %i32 0\n"
3597 "%f32_0 = OpConstant %f32 0.0\n"
3598 "%f32_0_5 = OpConstant %f32 0.5\n"
3599 "%f32_1 = OpConstant %f32 1.0\n"
3600 "%f32_1_5 = OpConstant %f32 1.5\n"
3601 "%f32_2 = OpConstant %f32 2.0\n"
3602 "%f32_3_5 = OpConstant %f32 3.5\n"
3603 "%f32_4 = OpConstant %f32 4.0\n"
3604 "%f32_7_5 = OpConstant %f32 7.5\n"
3605 "%f32_8 = OpConstant %f32 8.0\n"
3606 "%f32_15_5 = OpConstant %f32 15.5\n"
3607 "%f32_16 = OpConstant %f32 16.0\n"
3608 "%f32_31_5 = OpConstant %f32 31.5\n"
3609 "%f32_32 = OpConstant %f32 32.0\n"
3610 "%f32_63_5 = OpConstant %f32 63.5\n"
3611 "%f32_64 = OpConstant %f32 64.0\n"
3612 "%f32_127_5 = OpConstant %f32 127.5\n"
3613 "%f32_128 = OpConstant %f32 128.0\n"
3614 "%f32_256 = OpConstant %f32 256.0\n"
3616 "%main = OpFunction %void None %voidf\n"
3617 "%entry = OpLabel\n"
3618 "%idval = OpLoad %uvec3 %id\n"
3619 "%x = OpCompositeExtract %u32 %idval 0\n"
3620 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3621 "%inval = OpLoad %f32 %inloc\n"
3623 "%xabs = OpExtInst %f32 %ext FAbs %inval\n"
3624 "%x8 = OpFMod %f32 %xabs %f32_256\n"
3625 "%x7 = OpFMod %f32 %xabs %f32_128\n"
3626 "%x6 = OpFMod %f32 %xabs %f32_64\n"
3627 "%x5 = OpFMod %f32 %xabs %f32_32\n"
3628 "%x4 = OpFMod %f32 %xabs %f32_16\n"
3629 "%x3 = OpFMod %f32 %xabs %f32_8\n"
3630 "%x2 = OpFMod %f32 %xabs %f32_4\n"
3631 "%x1 = OpFMod %f32 %xabs %f32_2\n"
3633 "%b7 = OpFOrdGreaterThanEqual %bool %x8 %f32_127_5\n"
3634 "%b6 = OpFOrdGreaterThanEqual %bool %x7 %f32_63_5\n"
3635 "%b5 = OpFOrdGreaterThanEqual %bool %x6 %f32_31_5\n"
3636 "%b4 = OpFOrdGreaterThanEqual %bool %x5 %f32_15_5\n"
3637 "%b3 = OpFOrdGreaterThanEqual %bool %x4 %f32_7_5\n"
3638 "%b2 = OpFOrdGreaterThanEqual %bool %x3 %f32_3_5\n"
3639 "%b1 = OpFOrdGreaterThanEqual %bool %x2 %f32_1_5\n"
3640 "%b0 = OpFOrdGreaterThanEqual %bool %x1 %f32_0_5\n"
3642 + generateOpPhiCase5(codestring) +
3644 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3645 " OpStore %outloc %res\n"
3649 spec5.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3650 spec5.outputs.push_back(BufferSp(new Float32Buffer(outputFloats5)));
3651 spec5.numWorkGroups = IVec3(numElements, 1, 1);
3653 group->addChild(new SpvAsmComputeShaderCase(testCtx, "nested", "Stress OpPhi with a lot of nesting", spec5));
3655 createOpPhiVartypeTests(group, testCtx);
3657 return group.release();
3660 // Assembly code used for testing block order is based on GLSL source code:
3664 // layout(std140, set = 0, binding = 0) readonly buffer Input {
3665 // float elements[];
3667 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
3668 // float elements[];
3672 // uint x = gl_GlobalInvocationID.x;
3673 // output_data.elements[x] = input_data.elements[x];
3674 // if (x > uint(50)) {
3675 // switch (x % uint(3)) {
3676 // case 0: output_data.elements[x] += 1.5f; break;
3677 // case 1: output_data.elements[x] += 42.f; break;
3678 // case 2: output_data.elements[x] -= 27.f; break;
3682 // output_data.elements[x] = -input_data.elements[x];
3685 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
3687 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
3688 ComputeShaderSpec spec;
3689 de::Random rnd (deStringHash(group->getName()));
3690 const int numElements = 100;
3691 vector<float> inputFloats (numElements, 0);
3692 vector<float> outputFloats (numElements, 0);
3694 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3696 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3697 floorAll(inputFloats);
3699 for (size_t ndx = 0; ndx <= 50; ++ndx)
3700 outputFloats[ndx] = -inputFloats[ndx];
3702 for (size_t ndx = 51; ndx < numElements; ++ndx)
3706 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
3707 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
3708 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
3714 string(getComputeAsmShaderPreamble()) +
3716 "OpSource GLSL 430\n"
3717 "OpName %main \"main\"\n"
3718 "OpName %id \"gl_GlobalInvocationID\"\n"
3720 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3722 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3724 "%u32ptr = OpTypePointer Function %u32\n"
3725 "%u32ptr_input = OpTypePointer Input %u32\n"
3727 + string(getComputeAsmInputOutputBuffer()) +
3729 "%id = OpVariable %uvec3ptr Input\n"
3730 "%zero = OpConstant %i32 0\n"
3731 "%const3 = OpConstant %u32 3\n"
3732 "%const50 = OpConstant %u32 50\n"
3733 "%constf1p5 = OpConstant %f32 1.5\n"
3734 "%constf27 = OpConstant %f32 27.0\n"
3735 "%constf42 = OpConstant %f32 42.0\n"
3737 "%main = OpFunction %void None %voidf\n"
3740 "%entry = OpLabel\n"
3742 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
3743 "%xvar = OpVariable %u32ptr Function\n"
3744 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
3745 "%x = OpLoad %u32 %xptr\n"
3746 " OpStore %xvar %x\n"
3748 "%cmp = OpUGreaterThan %bool %x %const50\n"
3749 " OpSelectionMerge %if_merge None\n"
3750 " OpBranchConditional %cmp %if_true %if_false\n"
3752 // False branch for if-statement: placed in the middle of switch cases and before true branch.
3753 "%if_false = OpLabel\n"
3754 "%x_f = OpLoad %u32 %xvar\n"
3755 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
3756 "%inval_f = OpLoad %f32 %inloc_f\n"
3757 "%negate = OpFNegate %f32 %inval_f\n"
3758 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
3759 " OpStore %outloc_f %negate\n"
3760 " OpBranch %if_merge\n"
3762 // Merge block for if-statement: placed in the middle of true and false branch.
3763 "%if_merge = OpLabel\n"
3766 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
3767 "%if_true = OpLabel\n"
3768 "%xval_t = OpLoad %u32 %xvar\n"
3769 "%mod = OpUMod %u32 %xval_t %const3\n"
3770 " OpSelectionMerge %switch_merge None\n"
3771 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
3773 // Merge block for switch-statement: placed before the case
3774 // bodies. But it must follow OpSwitch which dominates it.
3775 "%switch_merge = OpLabel\n"
3776 " OpBranch %if_merge\n"
3778 // Case 1 for switch-statement: placed before case 0.
3779 // It must follow the OpSwitch that dominates it.
3780 "%case1 = OpLabel\n"
3781 "%x_1 = OpLoad %u32 %xvar\n"
3782 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
3783 "%inval_1 = OpLoad %f32 %inloc_1\n"
3784 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
3785 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
3786 " OpStore %outloc_1 %addf42\n"
3787 " OpBranch %switch_merge\n"
3789 // Case 2 for switch-statement.
3790 "%case2 = OpLabel\n"
3791 "%x_2 = OpLoad %u32 %xvar\n"
3792 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
3793 "%inval_2 = OpLoad %f32 %inloc_2\n"
3794 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
3795 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
3796 " OpStore %outloc_2 %subf27\n"
3797 " OpBranch %switch_merge\n"
3799 // Default case for switch-statement: placed in the middle of normal cases.
3800 "%default = OpLabel\n"
3801 " OpBranch %switch_merge\n"
3803 // Case 0 for switch-statement: out of order.
3804 "%case0 = OpLabel\n"
3805 "%x_0 = OpLoad %u32 %xvar\n"
3806 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
3807 "%inval_0 = OpLoad %f32 %inloc_0\n"
3808 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
3809 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
3810 " OpStore %outloc_0 %addf1p5\n"
3811 " OpBranch %switch_merge\n"
3814 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3815 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3816 spec.numWorkGroups = IVec3(numElements, 1, 1);
3818 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
3820 return group.release();
3823 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
3825 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
3826 ComputeShaderSpec spec1;
3827 ComputeShaderSpec spec2;
3828 de::Random rnd (deStringHash(group->getName()));
3829 const int numElements = 100;
3830 vector<float> inputFloats (numElements, 0);
3831 vector<float> outputFloats1 (numElements, 0);
3832 vector<float> outputFloats2 (numElements, 0);
3833 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
3835 for (size_t ndx = 0; ndx < numElements; ++ndx)
3837 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
3838 outputFloats2[ndx] = -inputFloats[ndx];
3841 const string assembly(
3842 "OpCapability Shader\n"
3843 "OpCapability ClipDistance\n"
3844 "OpMemoryModel Logical GLSL450\n"
3845 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
3846 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
3847 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
3848 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
3849 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
3850 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
3852 "OpName %comp_main1 \"entrypoint1\"\n"
3853 "OpName %comp_main2 \"entrypoint2\"\n"
3854 "OpName %vert_main \"entrypoint2\"\n"
3855 "OpName %id \"gl_GlobalInvocationID\"\n"
3856 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
3857 "OpName %vertexIndex \"gl_VertexIndex\"\n"
3858 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
3859 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
3860 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
3861 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
3863 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3864 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
3865 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
3866 "OpDecorate %vert_builtin_st Block\n"
3867 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
3868 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
3869 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
3871 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3873 "%zero = OpConstant %i32 0\n"
3874 "%one = OpConstant %u32 1\n"
3875 "%c_f32_1 = OpConstant %f32 1\n"
3877 "%i32inputptr = OpTypePointer Input %i32\n"
3878 "%vec4 = OpTypeVector %f32 4\n"
3879 "%vec4ptr = OpTypePointer Output %vec4\n"
3880 "%f32arr1 = OpTypeArray %f32 %one\n"
3881 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
3882 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
3883 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
3885 "%id = OpVariable %uvec3ptr Input\n"
3886 "%vertexIndex = OpVariable %i32inputptr Input\n"
3887 "%instanceIndex = OpVariable %i32inputptr Input\n"
3888 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
3890 // gl_Position = vec4(1.);
3891 "%vert_main = OpFunction %void None %voidf\n"
3892 "%vert_entry = OpLabel\n"
3893 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
3894 " OpStore %position %c_vec4_1\n"
3899 "%comp_main1 = OpFunction %void None %voidf\n"
3900 "%comp1_entry = OpLabel\n"
3901 "%idval1 = OpLoad %uvec3 %id\n"
3902 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
3903 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
3904 "%inval1 = OpLoad %f32 %inloc1\n"
3905 "%add = OpFAdd %f32 %inval1 %inval1\n"
3906 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
3907 " OpStore %outloc1 %add\n"
3912 "%comp_main2 = OpFunction %void None %voidf\n"
3913 "%comp2_entry = OpLabel\n"
3914 "%idval2 = OpLoad %uvec3 %id\n"
3915 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
3916 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
3917 "%inval2 = OpLoad %f32 %inloc2\n"
3918 "%neg = OpFNegate %f32 %inval2\n"
3919 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
3920 " OpStore %outloc2 %neg\n"
3922 " OpFunctionEnd\n");
3924 spec1.assembly = assembly;
3925 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3926 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
3927 spec1.numWorkGroups = IVec3(numElements, 1, 1);
3928 spec1.entryPoint = "entrypoint1";
3930 spec2.assembly = assembly;
3931 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3932 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
3933 spec2.numWorkGroups = IVec3(numElements, 1, 1);
3934 spec2.entryPoint = "entrypoint2";
3936 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
3937 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
3939 return group.release();
3942 inline std::string makeLongUTF8String (size_t num4ByteChars)
3944 // An example of a longest valid UTF-8 character. Be explicit about the
3945 // character type because Microsoft compilers can otherwise interpret the
3946 // character string as being over wide (16-bit) characters. Ideally, we
3947 // would just use a C++11 UTF-8 string literal, but we want to support older
3948 // Microsoft compilers.
3949 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
3950 std::string longString;
3951 longString.reserve(num4ByteChars * 4);
3952 for (size_t count = 0; count < num4ByteChars; count++)
3954 longString += earthAfrica;
3959 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
3961 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
3962 vector<CaseParameter> cases;
3963 de::Random rnd (deStringHash(group->getName()));
3964 const int numElements = 100;
3965 vector<float> positiveFloats (numElements, 0);
3966 vector<float> negativeFloats (numElements, 0);
3967 const StringTemplate shaderTemplate (
3968 "OpCapability Shader\n"
3969 "OpMemoryModel Logical GLSL450\n"
3971 "OpEntryPoint GLCompute %main \"main\" %id\n"
3972 "OpExecutionMode %main LocalSize 1 1 1\n"
3976 "OpName %main \"main\"\n"
3977 "OpName %id \"gl_GlobalInvocationID\"\n"
3979 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3981 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3983 "%id = OpVariable %uvec3ptr Input\n"
3984 "%zero = OpConstant %i32 0\n"
3986 "%main = OpFunction %void None %voidf\n"
3987 "%label = OpLabel\n"
3988 "%idval = OpLoad %uvec3 %id\n"
3989 "%x = OpCompositeExtract %u32 %idval 0\n"
3990 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3991 "%inval = OpLoad %f32 %inloc\n"
3992 "%neg = OpFNegate %f32 %inval\n"
3993 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3994 " OpStore %outloc %neg\n"
3996 " OpFunctionEnd\n");
3998 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
3999 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
4000 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
4001 "OpSource GLSL 430 %fname"));
4002 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
4003 "OpSource GLSL 430 %fname"));
4004 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
4005 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
4006 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
4007 "OpSource GLSL 430 %fname \"\""));
4008 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
4009 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
4010 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
4011 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
4012 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
4013 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
4014 "OpSourceContinued \"id main() {}\""));
4015 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
4016 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4017 "OpSourceContinued \"\""));
4018 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
4019 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4020 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
4021 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
4022 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4023 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
4024 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
4025 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
4026 "OpSourceContinued \"void\"\n"
4027 "OpSourceContinued \"main()\"\n"
4028 "OpSourceContinued \"{}\""));
4029 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
4030 "OpSource GLSL 430 %fname \"\"\n"
4031 "OpSourceContinued \"#version 430\nvoid main() {}\""));
4033 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4035 for (size_t ndx = 0; ndx < numElements; ++ndx)
4036 negativeFloats[ndx] = -positiveFloats[ndx];
4038 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4040 map<string, string> specializations;
4041 ComputeShaderSpec spec;
4043 specializations["SOURCE"] = cases[caseNdx].param;
4044 spec.assembly = shaderTemplate.specialize(specializations);
4045 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4046 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4047 spec.numWorkGroups = IVec3(numElements, 1, 1);
4049 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4052 return group.release();
4055 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
4057 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
4058 vector<CaseParameter> cases;
4059 de::Random rnd (deStringHash(group->getName()));
4060 const int numElements = 100;
4061 vector<float> inputFloats (numElements, 0);
4062 vector<float> outputFloats (numElements, 0);
4063 const StringTemplate shaderTemplate (
4064 string(getComputeAsmShaderPreamble()) +
4066 "OpSourceExtension \"${EXTENSION}\"\n"
4068 "OpName %main \"main\"\n"
4069 "OpName %id \"gl_GlobalInvocationID\"\n"
4071 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4073 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4075 "%id = OpVariable %uvec3ptr Input\n"
4076 "%zero = OpConstant %i32 0\n"
4078 "%main = OpFunction %void None %voidf\n"
4079 "%label = OpLabel\n"
4080 "%idval = OpLoad %uvec3 %id\n"
4081 "%x = OpCompositeExtract %u32 %idval 0\n"
4082 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4083 "%inval = OpLoad %f32 %inloc\n"
4084 "%neg = OpFNegate %f32 %inval\n"
4085 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4086 " OpStore %outloc %neg\n"
4088 " OpFunctionEnd\n");
4090 cases.push_back(CaseParameter("empty_extension", ""));
4091 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
4092 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
4093 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
4094 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
4096 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
4098 for (size_t ndx = 0; ndx < numElements; ++ndx)
4099 outputFloats[ndx] = -inputFloats[ndx];
4101 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4103 map<string, string> specializations;
4104 ComputeShaderSpec spec;
4106 specializations["EXTENSION"] = cases[caseNdx].param;
4107 spec.assembly = shaderTemplate.specialize(specializations);
4108 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4109 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4110 spec.numWorkGroups = IVec3(numElements, 1, 1);
4112 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4115 return group.release();
4118 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
4119 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
4121 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
4122 vector<CaseParameter> cases;
4123 de::Random rnd (deStringHash(group->getName()));
4124 const int numElements = 100;
4125 vector<float> positiveFloats (numElements, 0);
4126 vector<float> negativeFloats (numElements, 0);
4127 const StringTemplate shaderTemplate (
4128 string(getComputeAsmShaderPreamble()) +
4130 "OpSource GLSL 430\n"
4131 "OpName %main \"main\"\n"
4132 "OpName %id \"gl_GlobalInvocationID\"\n"
4134 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4136 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4137 "%uvec2 = OpTypeVector %u32 2\n"
4138 "%bvec3 = OpTypeVector %bool 3\n"
4139 "%fvec4 = OpTypeVector %f32 4\n"
4140 "%fmat33 = OpTypeMatrix %fvec3 3\n"
4141 "%const100 = OpConstant %u32 100\n"
4142 "%uarr100 = OpTypeArray %i32 %const100\n"
4143 "%struct = OpTypeStruct %f32 %i32 %u32\n"
4144 "%pointer = OpTypePointer Function %i32\n"
4145 + string(getComputeAsmInputOutputBuffer()) +
4147 "%null = OpConstantNull ${TYPE}\n"
4149 "%id = OpVariable %uvec3ptr Input\n"
4150 "%zero = OpConstant %i32 0\n"
4152 "%main = OpFunction %void None %voidf\n"
4153 "%label = OpLabel\n"
4154 "%idval = OpLoad %uvec3 %id\n"
4155 "%x = OpCompositeExtract %u32 %idval 0\n"
4156 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4157 "%inval = OpLoad %f32 %inloc\n"
4158 "%neg = OpFNegate %f32 %inval\n"
4159 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4160 " OpStore %outloc %neg\n"
4162 " OpFunctionEnd\n");
4164 cases.push_back(CaseParameter("bool", "%bool"));
4165 cases.push_back(CaseParameter("sint32", "%i32"));
4166 cases.push_back(CaseParameter("uint32", "%u32"));
4167 cases.push_back(CaseParameter("float32", "%f32"));
4168 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
4169 cases.push_back(CaseParameter("vec3bool", "%bvec3"));
4170 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
4171 cases.push_back(CaseParameter("matrix", "%fmat33"));
4172 cases.push_back(CaseParameter("array", "%uarr100"));
4173 cases.push_back(CaseParameter("struct", "%struct"));
4174 cases.push_back(CaseParameter("pointer", "%pointer"));
4176 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4178 for (size_t ndx = 0; ndx < numElements; ++ndx)
4179 negativeFloats[ndx] = -positiveFloats[ndx];
4181 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4183 map<string, string> specializations;
4184 ComputeShaderSpec spec;
4186 specializations["TYPE"] = cases[caseNdx].param;
4187 spec.assembly = shaderTemplate.specialize(specializations);
4188 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4189 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4190 spec.numWorkGroups = IVec3(numElements, 1, 1);
4192 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4195 return group.release();
4198 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
4199 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
4201 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
4202 vector<CaseParameter> cases;
4203 de::Random rnd (deStringHash(group->getName()));
4204 const int numElements = 100;
4205 vector<float> positiveFloats (numElements, 0);
4206 vector<float> negativeFloats (numElements, 0);
4207 const StringTemplate shaderTemplate (
4208 string(getComputeAsmShaderPreamble()) +
4210 "OpSource GLSL 430\n"
4211 "OpName %main \"main\"\n"
4212 "OpName %id \"gl_GlobalInvocationID\"\n"
4214 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4216 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4218 "%id = OpVariable %uvec3ptr Input\n"
4219 "%zero = OpConstant %i32 0\n"
4223 "%main = OpFunction %void None %voidf\n"
4224 "%label = OpLabel\n"
4225 "%idval = OpLoad %uvec3 %id\n"
4226 "%x = OpCompositeExtract %u32 %idval 0\n"
4227 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4228 "%inval = OpLoad %f32 %inloc\n"
4229 "%neg = OpFNegate %f32 %inval\n"
4230 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4231 " OpStore %outloc %neg\n"
4233 " OpFunctionEnd\n");
4235 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
4236 "%const = OpConstantComposite %uvec3 %five %zero %five"));
4237 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
4238 "%ten = OpConstant %f32 10.\n"
4239 "%fzero = OpConstant %f32 0.\n"
4240 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
4241 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
4242 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
4243 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
4244 "%fzero = OpConstant %f32 0.\n"
4245 "%one = OpConstant %f32 1.\n"
4246 "%point5 = OpConstant %f32 0.5\n"
4247 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
4248 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
4249 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
4250 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
4251 "%st2 = OpTypeStruct %i32 %i32\n"
4252 "%struct = OpTypeStruct %st1 %st2\n"
4253 "%point5 = OpConstant %f32 0.5\n"
4254 "%one = OpConstant %u32 1\n"
4255 "%ten = OpConstant %i32 10\n"
4256 "%st1val = OpConstantComposite %st1 %one %point5\n"
4257 "%st2val = OpConstantComposite %st2 %ten %ten\n"
4258 "%const = OpConstantComposite %struct %st1val %st2val"));
4260 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4262 for (size_t ndx = 0; ndx < numElements; ++ndx)
4263 negativeFloats[ndx] = -positiveFloats[ndx];
4265 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4267 map<string, string> specializations;
4268 ComputeShaderSpec spec;
4270 specializations["CONSTANT"] = cases[caseNdx].param;
4271 spec.assembly = shaderTemplate.specialize(specializations);
4272 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4273 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4274 spec.numWorkGroups = IVec3(numElements, 1, 1);
4276 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4279 return group.release();
4282 // Creates a floating point number with the given exponent, and significand
4283 // bits set. It can only create normalized numbers. Only the least significant
4284 // 24 bits of the significand will be examined. The final bit of the
4285 // significand will also be ignored. This allows alignment to be written
4286 // similarly to C99 hex-floats.
4287 // For example if you wanted to write 0x1.7f34p-12 you would call
4288 // constructNormalizedFloat(-12, 0x7f3400)
4289 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
4293 for (deInt32 idx = 0; idx < 23; ++idx)
4295 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
4299 return std::ldexp(f, exponent);
4302 // Compare instruction for the OpQuantizeF16 compute exact case.
4303 // Returns true if the output is what is expected from the test case.
4304 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
4306 if (outputAllocs.size() != 1)
4309 // Only size is needed because we cannot compare Nans.
4310 size_t byteSize = expectedOutputs[0]->getByteSize();
4312 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
4314 if (byteSize != 4*sizeof(float)) {
4318 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
4319 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
4324 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
4325 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
4330 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
4331 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
4336 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
4337 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
4344 // Checks that every output from a test-case is a float NaN.
4345 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
4347 if (outputAllocs.size() != 1)
4350 // Only size is needed because we cannot compare Nans.
4351 size_t byteSize = expectedOutputs[0]->getByteSize();
4353 const float* const output_as_float = static_cast<const float* const>(outputAllocs[0]->getHostPtr());
4355 for (size_t idx = 0; idx < byteSize / sizeof(float); ++idx)
4357 if (!deFloatIsNaN(output_as_float[idx]))
4366 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
4367 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
4369 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
4371 const std::string shader (
4372 string(getComputeAsmShaderPreamble()) +
4374 "OpSource GLSL 430\n"
4375 "OpName %main \"main\"\n"
4376 "OpName %id \"gl_GlobalInvocationID\"\n"
4378 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4380 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4382 "%id = OpVariable %uvec3ptr Input\n"
4383 "%zero = OpConstant %i32 0\n"
4385 "%main = OpFunction %void None %voidf\n"
4386 "%label = OpLabel\n"
4387 "%idval = OpLoad %uvec3 %id\n"
4388 "%x = OpCompositeExtract %u32 %idval 0\n"
4389 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4390 "%inval = OpLoad %f32 %inloc\n"
4391 "%quant = OpQuantizeToF16 %f32 %inval\n"
4392 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4393 " OpStore %outloc %quant\n"
4395 " OpFunctionEnd\n");
4398 ComputeShaderSpec spec;
4399 const deUint32 numElements = 100;
4400 vector<float> infinities;
4401 vector<float> results;
4403 infinities.reserve(numElements);
4404 results.reserve(numElements);
4406 for (size_t idx = 0; idx < numElements; ++idx)
4411 infinities.push_back(std::numeric_limits<float>::infinity());
4412 results.push_back(std::numeric_limits<float>::infinity());
4415 infinities.push_back(-std::numeric_limits<float>::infinity());
4416 results.push_back(-std::numeric_limits<float>::infinity());
4419 infinities.push_back(std::ldexp(1.0f, 16));
4420 results.push_back(std::numeric_limits<float>::infinity());
4423 infinities.push_back(std::ldexp(-1.0f, 32));
4424 results.push_back(-std::numeric_limits<float>::infinity());
4429 spec.assembly = shader;
4430 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
4431 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
4432 spec.numWorkGroups = IVec3(numElements, 1, 1);
4434 group->addChild(new SpvAsmComputeShaderCase(
4435 testCtx, "infinities", "Check that infinities propagated and created", spec));
4439 ComputeShaderSpec spec;
4441 const deUint32 numElements = 100;
4443 nans.reserve(numElements);
4445 for (size_t idx = 0; idx < numElements; ++idx)
4449 nans.push_back(std::numeric_limits<float>::quiet_NaN());
4453 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
4457 spec.assembly = shader;
4458 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
4459 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
4460 spec.numWorkGroups = IVec3(numElements, 1, 1);
4461 spec.verifyIO = &compareNan;
4463 group->addChild(new SpvAsmComputeShaderCase(
4464 testCtx, "propagated_nans", "Check that nans are propagated", spec));
4468 ComputeShaderSpec spec;
4469 vector<float> small;
4470 vector<float> zeros;
4471 const deUint32 numElements = 100;
4473 small.reserve(numElements);
4474 zeros.reserve(numElements);
4476 for (size_t idx = 0; idx < numElements; ++idx)
4481 small.push_back(0.f);
4482 zeros.push_back(0.f);
4485 small.push_back(-0.f);
4486 zeros.push_back(-0.f);
4489 small.push_back(std::ldexp(1.0f, -16));
4490 zeros.push_back(0.f);
4493 small.push_back(std::ldexp(-1.0f, -32));
4494 zeros.push_back(-0.f);
4497 small.push_back(std::ldexp(1.0f, -127));
4498 zeros.push_back(0.f);
4501 small.push_back(-std::ldexp(1.0f, -128));
4502 zeros.push_back(-0.f);
4507 spec.assembly = shader;
4508 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
4509 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
4510 spec.numWorkGroups = IVec3(numElements, 1, 1);
4512 group->addChild(new SpvAsmComputeShaderCase(
4513 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
4517 ComputeShaderSpec spec;
4518 vector<float> exact;
4519 const deUint32 numElements = 200;
4521 exact.reserve(numElements);
4523 for (size_t idx = 0; idx < numElements; ++idx)
4524 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
4526 spec.assembly = shader;
4527 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
4528 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
4529 spec.numWorkGroups = IVec3(numElements, 1, 1);
4531 group->addChild(new SpvAsmComputeShaderCase(
4532 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
4536 ComputeShaderSpec spec;
4537 vector<float> inputs;
4538 const deUint32 numElements = 4;
4540 inputs.push_back(constructNormalizedFloat(8, 0x300300));
4541 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
4542 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
4543 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
4545 spec.assembly = shader;
4546 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
4547 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4548 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
4549 spec.numWorkGroups = IVec3(numElements, 1, 1);
4551 group->addChild(new SpvAsmComputeShaderCase(
4552 testCtx, "rounded", "Check that are rounded when needed", spec));
4555 return group.release();
4558 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
4560 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
4562 const std::string shader (
4563 string(getComputeAsmShaderPreamble()) +
4565 "OpName %main \"main\"\n"
4566 "OpName %id \"gl_GlobalInvocationID\"\n"
4568 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4570 "OpDecorate %sc_0 SpecId 0\n"
4571 "OpDecorate %sc_1 SpecId 1\n"
4572 "OpDecorate %sc_2 SpecId 2\n"
4573 "OpDecorate %sc_3 SpecId 3\n"
4574 "OpDecorate %sc_4 SpecId 4\n"
4575 "OpDecorate %sc_5 SpecId 5\n"
4577 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4579 "%id = OpVariable %uvec3ptr Input\n"
4580 "%zero = OpConstant %i32 0\n"
4581 "%c_u32_6 = OpConstant %u32 6\n"
4583 "%sc_0 = OpSpecConstant %f32 0.\n"
4584 "%sc_1 = OpSpecConstant %f32 0.\n"
4585 "%sc_2 = OpSpecConstant %f32 0.\n"
4586 "%sc_3 = OpSpecConstant %f32 0.\n"
4587 "%sc_4 = OpSpecConstant %f32 0.\n"
4588 "%sc_5 = OpSpecConstant %f32 0.\n"
4590 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
4591 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
4592 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
4593 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
4594 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
4595 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
4597 "%main = OpFunction %void None %voidf\n"
4598 "%label = OpLabel\n"
4599 "%idval = OpLoad %uvec3 %id\n"
4600 "%x = OpCompositeExtract %u32 %idval 0\n"
4601 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4602 "%selector = OpUMod %u32 %x %c_u32_6\n"
4603 " OpSelectionMerge %exit None\n"
4604 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
4606 "%case0 = OpLabel\n"
4607 " OpStore %outloc %sc_0_quant\n"
4610 "%case1 = OpLabel\n"
4611 " OpStore %outloc %sc_1_quant\n"
4614 "%case2 = OpLabel\n"
4615 " OpStore %outloc %sc_2_quant\n"
4618 "%case3 = OpLabel\n"
4619 " OpStore %outloc %sc_3_quant\n"
4622 "%case4 = OpLabel\n"
4623 " OpStore %outloc %sc_4_quant\n"
4626 "%case5 = OpLabel\n"
4627 " OpStore %outloc %sc_5_quant\n"
4633 " OpFunctionEnd\n");
4636 ComputeShaderSpec spec;
4637 const deUint8 numCases = 4;
4638 vector<float> inputs (numCases, 0.f);
4639 vector<float> outputs;
4641 spec.assembly = shader;
4642 spec.numWorkGroups = IVec3(numCases, 1, 1);
4644 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
4645 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
4646 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
4647 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
4649 outputs.push_back(std::numeric_limits<float>::infinity());
4650 outputs.push_back(-std::numeric_limits<float>::infinity());
4651 outputs.push_back(std::numeric_limits<float>::infinity());
4652 outputs.push_back(-std::numeric_limits<float>::infinity());
4654 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4655 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4657 group->addChild(new SpvAsmComputeShaderCase(
4658 testCtx, "infinities", "Check that infinities propagated and created", spec));
4662 ComputeShaderSpec spec;
4663 const deUint8 numCases = 2;
4664 vector<float> inputs (numCases, 0.f);
4665 vector<float> outputs;
4667 spec.assembly = shader;
4668 spec.numWorkGroups = IVec3(numCases, 1, 1);
4669 spec.verifyIO = &compareNan;
4671 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
4672 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
4674 for (deUint8 idx = 0; idx < numCases; ++idx)
4675 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
4677 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4678 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4680 group->addChild(new SpvAsmComputeShaderCase(
4681 testCtx, "propagated_nans", "Check that nans are propagated", spec));
4685 ComputeShaderSpec spec;
4686 const deUint8 numCases = 6;
4687 vector<float> inputs (numCases, 0.f);
4688 vector<float> outputs;
4690 spec.assembly = shader;
4691 spec.numWorkGroups = IVec3(numCases, 1, 1);
4693 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
4694 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
4695 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
4696 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
4697 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
4698 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
4700 outputs.push_back(0.f);
4701 outputs.push_back(-0.f);
4702 outputs.push_back(0.f);
4703 outputs.push_back(-0.f);
4704 outputs.push_back(0.f);
4705 outputs.push_back(-0.f);
4707 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4708 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4710 group->addChild(new SpvAsmComputeShaderCase(
4711 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
4715 ComputeShaderSpec spec;
4716 const deUint8 numCases = 6;
4717 vector<float> inputs (numCases, 0.f);
4718 vector<float> outputs;
4720 spec.assembly = shader;
4721 spec.numWorkGroups = IVec3(numCases, 1, 1);
4723 for (deUint8 idx = 0; idx < 6; ++idx)
4725 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
4726 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
4727 outputs.push_back(f);
4730 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4731 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4733 group->addChild(new SpvAsmComputeShaderCase(
4734 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
4738 ComputeShaderSpec spec;
4739 const deUint8 numCases = 4;
4740 vector<float> inputs (numCases, 0.f);
4741 vector<float> outputs;
4743 spec.assembly = shader;
4744 spec.numWorkGroups = IVec3(numCases, 1, 1);
4745 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
4747 outputs.push_back(constructNormalizedFloat(8, 0x300300));
4748 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
4749 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
4750 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
4752 for (deUint8 idx = 0; idx < numCases; ++idx)
4753 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
4755 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4756 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4758 group->addChild(new SpvAsmComputeShaderCase(
4759 testCtx, "rounded", "Check that are rounded when needed", spec));
4762 return group.release();
4765 // Checks that constant null/composite values can be used in computation.
4766 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
4768 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
4769 ComputeShaderSpec spec;
4770 de::Random rnd (deStringHash(group->getName()));
4771 const int numElements = 100;
4772 vector<float> positiveFloats (numElements, 0);
4773 vector<float> negativeFloats (numElements, 0);
4775 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4777 for (size_t ndx = 0; ndx < numElements; ++ndx)
4778 negativeFloats[ndx] = -positiveFloats[ndx];
4781 "OpCapability Shader\n"
4782 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
4783 "OpMemoryModel Logical GLSL450\n"
4784 "OpEntryPoint GLCompute %main \"main\" %id\n"
4785 "OpExecutionMode %main LocalSize 1 1 1\n"
4787 "OpSource GLSL 430\n"
4788 "OpName %main \"main\"\n"
4789 "OpName %id \"gl_GlobalInvocationID\"\n"
4791 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4793 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4795 "%fmat = OpTypeMatrix %fvec3 3\n"
4796 "%ten = OpConstant %u32 10\n"
4797 "%f32arr10 = OpTypeArray %f32 %ten\n"
4798 "%fst = OpTypeStruct %f32 %f32\n"
4800 + string(getComputeAsmInputOutputBuffer()) +
4802 "%id = OpVariable %uvec3ptr Input\n"
4803 "%zero = OpConstant %i32 0\n"
4805 // Create a bunch of null values
4806 "%unull = OpConstantNull %u32\n"
4807 "%fnull = OpConstantNull %f32\n"
4808 "%vnull = OpConstantNull %fvec3\n"
4809 "%mnull = OpConstantNull %fmat\n"
4810 "%anull = OpConstantNull %f32arr10\n"
4811 "%snull = OpConstantComposite %fst %fnull %fnull\n"
4813 "%main = OpFunction %void None %voidf\n"
4814 "%label = OpLabel\n"
4815 "%idval = OpLoad %uvec3 %id\n"
4816 "%x = OpCompositeExtract %u32 %idval 0\n"
4817 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4818 "%inval = OpLoad %f32 %inloc\n"
4819 "%neg = OpFNegate %f32 %inval\n"
4821 // Get the abs() of (a certain element of) those null values
4822 "%unull_cov = OpConvertUToF %f32 %unull\n"
4823 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
4824 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
4825 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
4826 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
4827 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
4828 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
4829 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
4830 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
4831 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
4832 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
4835 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
4836 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
4837 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
4838 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
4839 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
4840 "%final = OpFAdd %f32 %add5 %snull_abs\n"
4842 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4843 " OpStore %outloc %final\n" // write to output
4846 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4847 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4848 spec.numWorkGroups = IVec3(numElements, 1, 1);
4850 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
4852 return group.release();
4855 // Assembly code used for testing loop 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 // output_data.elements[x] = input_data.elements[x];
4868 // for (uint i = 0; i < 4; ++i)
4869 // output_data.elements[x] += 1.f;
4871 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
4873 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
4874 vector<CaseParameter> cases;
4875 de::Random rnd (deStringHash(group->getName()));
4876 const int numElements = 100;
4877 vector<float> inputFloats (numElements, 0);
4878 vector<float> outputFloats (numElements, 0);
4879 const StringTemplate shaderTemplate (
4880 string(getComputeAsmShaderPreamble()) +
4882 "OpSource GLSL 430\n"
4883 "OpName %main \"main\"\n"
4884 "OpName %id \"gl_GlobalInvocationID\"\n"
4886 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4888 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4890 "%u32ptr = OpTypePointer Function %u32\n"
4892 "%id = OpVariable %uvec3ptr Input\n"
4893 "%zero = OpConstant %i32 0\n"
4894 "%uzero = OpConstant %u32 0\n"
4895 "%one = OpConstant %i32 1\n"
4896 "%constf1 = OpConstant %f32 1.0\n"
4897 "%four = OpConstant %u32 4\n"
4899 "%main = OpFunction %void None %voidf\n"
4900 "%entry = OpLabel\n"
4901 "%i = OpVariable %u32ptr Function\n"
4902 " OpStore %i %uzero\n"
4904 "%idval = OpLoad %uvec3 %id\n"
4905 "%x = OpCompositeExtract %u32 %idval 0\n"
4906 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4907 "%inval = OpLoad %f32 %inloc\n"
4908 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4909 " OpStore %outloc %inval\n"
4910 " OpBranch %loop_entry\n"
4912 "%loop_entry = OpLabel\n"
4913 "%i_val = OpLoad %u32 %i\n"
4914 "%cmp_lt = OpULessThan %bool %i_val %four\n"
4915 " OpLoopMerge %loop_merge %loop_body ${CONTROL}\n"
4916 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
4917 "%loop_body = OpLabel\n"
4918 "%outval = OpLoad %f32 %outloc\n"
4919 "%addf1 = OpFAdd %f32 %outval %constf1\n"
4920 " OpStore %outloc %addf1\n"
4921 "%new_i = OpIAdd %u32 %i_val %one\n"
4922 " OpStore %i %new_i\n"
4923 " OpBranch %loop_entry\n"
4924 "%loop_merge = OpLabel\n"
4926 " OpFunctionEnd\n");
4928 cases.push_back(CaseParameter("none", "None"));
4929 cases.push_back(CaseParameter("unroll", "Unroll"));
4930 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
4931 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
4933 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4935 for (size_t ndx = 0; ndx < numElements; ++ndx)
4936 outputFloats[ndx] = inputFloats[ndx] + 4.f;
4938 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4940 map<string, string> specializations;
4941 ComputeShaderSpec spec;
4943 specializations["CONTROL"] = cases[caseNdx].param;
4944 spec.assembly = shaderTemplate.specialize(specializations);
4945 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4946 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4947 spec.numWorkGroups = IVec3(numElements, 1, 1);
4949 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4952 group->addChild(new SpvAsmLoopControlDependencyLengthCase(testCtx, "dependency_length", "dependency_length"));
4953 group->addChild(new SpvAsmLoopControlDependencyInfiniteCase(testCtx, "dependency_infinite", "dependency_infinite"));
4955 return group.release();
4958 // Assembly code used for testing selection control is based on GLSL source code:
4961 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4962 // float elements[];
4964 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4965 // float elements[];
4969 // uint x = gl_GlobalInvocationID.x;
4970 // float val = input_data.elements[x];
4972 // output_data.elements[x] = val + 1.f;
4974 // output_data.elements[x] = val - 1.f;
4976 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
4978 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
4979 vector<CaseParameter> cases;
4980 de::Random rnd (deStringHash(group->getName()));
4981 const int numElements = 100;
4982 vector<float> inputFloats (numElements, 0);
4983 vector<float> outputFloats (numElements, 0);
4984 const StringTemplate shaderTemplate (
4985 string(getComputeAsmShaderPreamble()) +
4987 "OpSource GLSL 430\n"
4988 "OpName %main \"main\"\n"
4989 "OpName %id \"gl_GlobalInvocationID\"\n"
4991 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4993 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4995 "%id = OpVariable %uvec3ptr Input\n"
4996 "%zero = OpConstant %i32 0\n"
4997 "%constf1 = OpConstant %f32 1.0\n"
4998 "%constf10 = OpConstant %f32 10.0\n"
5000 "%main = OpFunction %void None %voidf\n"
5001 "%entry = OpLabel\n"
5002 "%idval = OpLoad %uvec3 %id\n"
5003 "%x = OpCompositeExtract %u32 %idval 0\n"
5004 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5005 "%inval = OpLoad %f32 %inloc\n"
5006 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5007 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
5009 " OpSelectionMerge %if_end ${CONTROL}\n"
5010 " OpBranchConditional %cmp_gt %if_true %if_false\n"
5011 "%if_true = OpLabel\n"
5012 "%addf1 = OpFAdd %f32 %inval %constf1\n"
5013 " OpStore %outloc %addf1\n"
5014 " OpBranch %if_end\n"
5015 "%if_false = OpLabel\n"
5016 "%subf1 = OpFSub %f32 %inval %constf1\n"
5017 " OpStore %outloc %subf1\n"
5018 " OpBranch %if_end\n"
5019 "%if_end = OpLabel\n"
5021 " OpFunctionEnd\n");
5023 cases.push_back(CaseParameter("none", "None"));
5024 cases.push_back(CaseParameter("flatten", "Flatten"));
5025 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
5026 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
5028 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5030 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
5031 floorAll(inputFloats);
5033 for (size_t ndx = 0; ndx < numElements; ++ndx)
5034 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
5036 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5038 map<string, string> specializations;
5039 ComputeShaderSpec spec;
5041 specializations["CONTROL"] = cases[caseNdx].param;
5042 spec.assembly = shaderTemplate.specialize(specializations);
5043 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5044 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5045 spec.numWorkGroups = IVec3(numElements, 1, 1);
5047 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5050 return group.release();
5053 // Assembly code used for testing function control is based on GLSL source code:
5057 // layout(std140, set = 0, binding = 0) readonly buffer Input {
5058 // float elements[];
5060 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
5061 // float elements[];
5064 // float const10() { return 10.f; }
5067 // uint x = gl_GlobalInvocationID.x;
5068 // output_data.elements[x] = input_data.elements[x] + const10();
5070 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
5072 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
5073 vector<CaseParameter> cases;
5074 de::Random rnd (deStringHash(group->getName()));
5075 const int numElements = 100;
5076 vector<float> inputFloats (numElements, 0);
5077 vector<float> outputFloats (numElements, 0);
5078 const StringTemplate shaderTemplate (
5079 string(getComputeAsmShaderPreamble()) +
5081 "OpSource GLSL 430\n"
5082 "OpName %main \"main\"\n"
5083 "OpName %func_const10 \"const10(\"\n"
5084 "OpName %id \"gl_GlobalInvocationID\"\n"
5086 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5088 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5090 "%f32f = OpTypeFunction %f32\n"
5091 "%id = OpVariable %uvec3ptr Input\n"
5092 "%zero = OpConstant %i32 0\n"
5093 "%constf10 = OpConstant %f32 10.0\n"
5095 "%main = OpFunction %void None %voidf\n"
5096 "%entry = OpLabel\n"
5097 "%idval = OpLoad %uvec3 %id\n"
5098 "%x = OpCompositeExtract %u32 %idval 0\n"
5099 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5100 "%inval = OpLoad %f32 %inloc\n"
5101 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
5102 "%fadd = OpFAdd %f32 %inval %ret_10\n"
5103 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5104 " OpStore %outloc %fadd\n"
5108 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
5109 "%label = OpLabel\n"
5110 " OpReturnValue %constf10\n"
5111 " OpFunctionEnd\n");
5113 cases.push_back(CaseParameter("none", "None"));
5114 cases.push_back(CaseParameter("inline", "Inline"));
5115 cases.push_back(CaseParameter("dont_inline", "DontInline"));
5116 cases.push_back(CaseParameter("pure", "Pure"));
5117 cases.push_back(CaseParameter("const", "Const"));
5118 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
5119 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
5120 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
5121 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
5123 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5125 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
5126 floorAll(inputFloats);
5128 for (size_t ndx = 0; ndx < numElements; ++ndx)
5129 outputFloats[ndx] = inputFloats[ndx] + 10.f;
5131 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5133 map<string, string> specializations;
5134 ComputeShaderSpec spec;
5136 specializations["CONTROL"] = cases[caseNdx].param;
5137 spec.assembly = shaderTemplate.specialize(specializations);
5138 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5139 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5140 spec.numWorkGroups = IVec3(numElements, 1, 1);
5142 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5145 return group.release();
5148 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
5150 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
5151 vector<CaseParameter> cases;
5152 de::Random rnd (deStringHash(group->getName()));
5153 const int numElements = 100;
5154 vector<float> inputFloats (numElements, 0);
5155 vector<float> outputFloats (numElements, 0);
5156 const StringTemplate shaderTemplate (
5157 string(getComputeAsmShaderPreamble()) +
5159 "OpSource GLSL 430\n"
5160 "OpName %main \"main\"\n"
5161 "OpName %id \"gl_GlobalInvocationID\"\n"
5163 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5165 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5167 "%f32ptr_f = OpTypePointer Function %f32\n"
5169 "%id = OpVariable %uvec3ptr Input\n"
5170 "%zero = OpConstant %i32 0\n"
5171 "%four = OpConstant %i32 4\n"
5173 "%main = OpFunction %void None %voidf\n"
5174 "%label = OpLabel\n"
5175 "%copy = OpVariable %f32ptr_f Function\n"
5176 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
5177 "%x = OpCompositeExtract %u32 %idval 0\n"
5178 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5179 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5180 " OpCopyMemory %copy %inloc ${ACCESS}\n"
5181 "%val1 = OpLoad %f32 %copy\n"
5182 "%val2 = OpLoad %f32 %inloc\n"
5183 "%add = OpFAdd %f32 %val1 %val2\n"
5184 " OpStore %outloc %add ${ACCESS}\n"
5186 " OpFunctionEnd\n");
5188 cases.push_back(CaseParameter("null", ""));
5189 cases.push_back(CaseParameter("none", "None"));
5190 cases.push_back(CaseParameter("volatile", "Volatile"));
5191 cases.push_back(CaseParameter("aligned", "Aligned 4"));
5192 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
5193 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
5194 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
5196 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5198 for (size_t ndx = 0; ndx < numElements; ++ndx)
5199 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
5201 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5203 map<string, string> specializations;
5204 ComputeShaderSpec spec;
5206 specializations["ACCESS"] = cases[caseNdx].param;
5207 spec.assembly = shaderTemplate.specialize(specializations);
5208 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5209 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5210 spec.numWorkGroups = IVec3(numElements, 1, 1);
5212 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5215 return group.release();
5218 // Checks that we can get undefined values for various types, without exercising a computation with it.
5219 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
5221 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
5222 vector<CaseParameter> cases;
5223 de::Random rnd (deStringHash(group->getName()));
5224 const int numElements = 100;
5225 vector<float> positiveFloats (numElements, 0);
5226 vector<float> negativeFloats (numElements, 0);
5227 const StringTemplate shaderTemplate (
5228 string(getComputeAsmShaderPreamble()) +
5230 "OpSource GLSL 430\n"
5231 "OpName %main \"main\"\n"
5232 "OpName %id \"gl_GlobalInvocationID\"\n"
5234 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5236 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
5237 "%uvec2 = OpTypeVector %u32 2\n"
5238 "%fvec4 = OpTypeVector %f32 4\n"
5239 "%fmat33 = OpTypeMatrix %fvec3 3\n"
5240 "%image = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
5241 "%sampler = OpTypeSampler\n"
5242 "%simage = OpTypeSampledImage %image\n"
5243 "%const100 = OpConstant %u32 100\n"
5244 "%uarr100 = OpTypeArray %i32 %const100\n"
5245 "%struct = OpTypeStruct %f32 %i32 %u32\n"
5246 "%pointer = OpTypePointer Function %i32\n"
5247 + string(getComputeAsmInputOutputBuffer()) +
5249 "%id = OpVariable %uvec3ptr Input\n"
5250 "%zero = OpConstant %i32 0\n"
5252 "%main = OpFunction %void None %voidf\n"
5253 "%label = OpLabel\n"
5255 "%undef = OpUndef ${TYPE}\n"
5257 "%idval = OpLoad %uvec3 %id\n"
5258 "%x = OpCompositeExtract %u32 %idval 0\n"
5260 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5261 "%inval = OpLoad %f32 %inloc\n"
5262 "%neg = OpFNegate %f32 %inval\n"
5263 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5264 " OpStore %outloc %neg\n"
5266 " OpFunctionEnd\n");
5268 cases.push_back(CaseParameter("bool", "%bool"));
5269 cases.push_back(CaseParameter("sint32", "%i32"));
5270 cases.push_back(CaseParameter("uint32", "%u32"));
5271 cases.push_back(CaseParameter("float32", "%f32"));
5272 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
5273 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
5274 cases.push_back(CaseParameter("matrix", "%fmat33"));
5275 cases.push_back(CaseParameter("image", "%image"));
5276 cases.push_back(CaseParameter("sampler", "%sampler"));
5277 cases.push_back(CaseParameter("sampledimage", "%simage"));
5278 cases.push_back(CaseParameter("array", "%uarr100"));
5279 cases.push_back(CaseParameter("runtimearray", "%f32arr"));
5280 cases.push_back(CaseParameter("struct", "%struct"));
5281 cases.push_back(CaseParameter("pointer", "%pointer"));
5283 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
5285 for (size_t ndx = 0; ndx < numElements; ++ndx)
5286 negativeFloats[ndx] = -positiveFloats[ndx];
5288 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5290 map<string, string> specializations;
5291 ComputeShaderSpec spec;
5293 specializations["TYPE"] = cases[caseNdx].param;
5294 spec.assembly = shaderTemplate.specialize(specializations);
5295 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
5296 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
5297 spec.numWorkGroups = IVec3(numElements, 1, 1);
5299 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5302 return group.release();
5306 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
5308 struct NameCodePair { string name, code; };
5309 RGBA defaultColors[4];
5310 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
5311 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
5312 map<string, string> fragments = passthruFragments();
5313 const NameCodePair tests[] =
5315 {"unknown", "OpSource Unknown 321"},
5316 {"essl", "OpSource ESSL 310"},
5317 {"glsl", "OpSource GLSL 450"},
5318 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
5319 {"opencl_c", "OpSource OpenCL_C 120"},
5320 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
5321 {"file", opsourceGLSLWithFile},
5322 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
5323 // Longest possible source string: SPIR-V limits instructions to 65535
5324 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
5325 // contain 65530 UTF8 characters (one word each) plus one last word
5326 // containing 3 ASCII characters and \0.
5327 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
5330 getDefaultColors(defaultColors);
5331 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5333 fragments["debug"] = tests[testNdx].code;
5334 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5337 return opSourceTests.release();
5340 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
5342 struct NameCodePair { string name, code; };
5343 RGBA defaultColors[4];
5344 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
5345 map<string, string> fragments = passthruFragments();
5346 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
5347 const NameCodePair tests[] =
5349 {"empty", opsource + "OpSourceContinued \"\""},
5350 {"short", opsource + "OpSourceContinued \"abcde\""},
5351 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
5352 // Longest possible source string: SPIR-V limits instructions to 65535
5353 // words, of which the first one is OpSourceContinued/length; the rest
5354 // will contain 65533 UTF8 characters (one word each) plus one last word
5355 // containing 3 ASCII characters and \0.
5356 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
5359 getDefaultColors(defaultColors);
5360 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5362 fragments["debug"] = tests[testNdx].code;
5363 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5366 return opSourceTests.release();
5368 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
5370 RGBA defaultColors[4];
5371 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
5372 map<string, string> fragments;
5373 getDefaultColors(defaultColors);
5374 fragments["debug"] =
5375 "%name = OpString \"name\"\n";
5377 fragments["pre_main"] =
5380 "OpLine %name 1 1\n"
5382 "OpLine %name 1 1\n"
5383 "OpLine %name 1 1\n"
5384 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5386 "OpLine %name 1 1\n"
5388 "OpLine %name 1 1\n"
5389 "OpLine %name 1 1\n"
5390 "%second_param1 = OpFunctionParameter %v4f32\n"
5393 "%label_secondfunction = OpLabel\n"
5395 "OpReturnValue %second_param1\n"
5400 fragments["testfun"] =
5401 // A %test_code function that returns its argument unchanged.
5404 "OpLine %name 1 1\n"
5405 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5407 "%param1 = OpFunctionParameter %v4f32\n"
5410 "%label_testfun = OpLabel\n"
5412 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5413 "OpReturnValue %val1\n"
5415 "OpLine %name 1 1\n"
5418 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
5420 return opLineTests.release();
5423 tcu::TestCaseGroup* createOpModuleProcessedTests(tcu::TestContext& testCtx)
5425 RGBA defaultColors[4];
5426 de::MovePtr<tcu::TestCaseGroup> opModuleProcessedTests (new tcu::TestCaseGroup(testCtx, "opmoduleprocessed", "OpModuleProcessed instruction"));
5427 map<string, string> fragments;
5428 std::vector<std::string> noExtensions;
5429 GraphicsResources resources;
5431 getDefaultColors(defaultColors);
5432 resources.verifyBinary = veryfiBinaryShader;
5433 resources.spirvVersion = SPIRV_VERSION_1_3;
5435 fragments["moduleprocessed"] =
5436 "OpModuleProcessed \"VULKAN CTS\"\n"
5437 "OpModuleProcessed \"Negative values\"\n"
5438 "OpModuleProcessed \"Date: 2017/09/21\"\n";
5440 fragments["pre_main"] =
5441 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5442 "%second_param1 = OpFunctionParameter %v4f32\n"
5443 "%label_secondfunction = OpLabel\n"
5444 "OpReturnValue %second_param1\n"
5447 fragments["testfun"] =
5448 // A %test_code function that returns its argument unchanged.
5449 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5450 "%param1 = OpFunctionParameter %v4f32\n"
5451 "%label_testfun = OpLabel\n"
5452 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5453 "OpReturnValue %val1\n"
5456 createTestsForAllStages ("opmoduleprocessed", defaultColors, defaultColors, fragments, resources, noExtensions, opModuleProcessedTests.get());
5458 return opModuleProcessedTests.release();
5462 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
5464 RGBA defaultColors[4];
5465 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
5466 map<string, string> fragments;
5467 std::vector<std::pair<std::string, std::string> > problemStrings;
5469 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
5470 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
5471 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
5472 getDefaultColors(defaultColors);
5474 fragments["debug"] =
5475 "%other_name = OpString \"other_name\"\n";
5477 fragments["pre_main"] =
5478 "OpLine %file_name 32 0\n"
5479 "OpLine %file_name 32 32\n"
5480 "OpLine %file_name 32 40\n"
5481 "OpLine %other_name 32 40\n"
5482 "OpLine %other_name 0 100\n"
5483 "OpLine %other_name 0 4294967295\n"
5484 "OpLine %other_name 4294967295 0\n"
5485 "OpLine %other_name 32 40\n"
5486 "OpLine %file_name 0 0\n"
5487 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5488 "OpLine %file_name 1 0\n"
5489 "%second_param1 = OpFunctionParameter %v4f32\n"
5490 "OpLine %file_name 1 3\n"
5491 "OpLine %file_name 1 2\n"
5492 "%label_secondfunction = OpLabel\n"
5493 "OpLine %file_name 0 2\n"
5494 "OpReturnValue %second_param1\n"
5496 "OpLine %file_name 0 2\n"
5497 "OpLine %file_name 0 2\n";
5499 fragments["testfun"] =
5500 // A %test_code function that returns its argument unchanged.
5501 "OpLine %file_name 1 0\n"
5502 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5503 "OpLine %file_name 16 330\n"
5504 "%param1 = OpFunctionParameter %v4f32\n"
5505 "OpLine %file_name 14 442\n"
5506 "%label_testfun = OpLabel\n"
5507 "OpLine %file_name 11 1024\n"
5508 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5509 "OpLine %file_name 2 97\n"
5510 "OpReturnValue %val1\n"
5512 "OpLine %file_name 5 32\n";
5514 for (size_t i = 0; i < problemStrings.size(); ++i)
5516 map<string, string> testFragments = fragments;
5517 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
5518 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
5521 return opLineTests.release();
5524 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
5526 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
5530 const char functionStart[] =
5531 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5532 "%param1 = OpFunctionParameter %v4f32\n"
5535 const char functionEnd[] =
5536 "OpReturnValue %transformed_param\n"
5539 struct NameConstantsCode
5546 NameConstantsCode tests[] =
5550 "%cnull = OpConstantNull %v4f32\n",
5551 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
5555 "%cnull = OpConstantNull %f32\n",
5556 "%vp = OpVariable %fp_v4f32 Function\n"
5557 "%v = OpLoad %v4f32 %vp\n"
5558 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
5559 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
5560 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
5561 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
5562 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
5566 "%cnull = OpConstantNull %bool\n",
5567 "%v = OpVariable %fp_v4f32 Function\n"
5568 " OpStore %v %param1\n"
5569 " OpSelectionMerge %false_label None\n"
5570 " OpBranchConditional %cnull %true_label %false_label\n"
5571 "%true_label = OpLabel\n"
5572 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
5573 " OpBranch %false_label\n"
5574 "%false_label = OpLabel\n"
5575 "%transformed_param = OpLoad %v4f32 %v\n"
5579 "%cnull = OpConstantNull %i32\n",
5580 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
5581 "%b = OpIEqual %bool %cnull %c_i32_0\n"
5582 " OpSelectionMerge %false_label None\n"
5583 " OpBranchConditional %b %true_label %false_label\n"
5584 "%true_label = OpLabel\n"
5585 " OpStore %v %param1\n"
5586 " OpBranch %false_label\n"
5587 "%false_label = OpLabel\n"
5588 "%transformed_param = OpLoad %v4f32 %v\n"
5592 "%stype = OpTypeStruct %f32 %v4f32\n"
5593 "%fp_stype = OpTypePointer Function %stype\n"
5594 "%cnull = OpConstantNull %stype\n",
5595 "%v = OpVariable %fp_stype Function %cnull\n"
5596 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
5597 "%f_val = OpLoad %v4f32 %f\n"
5598 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
5602 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
5603 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
5604 "%cnull = OpConstantNull %a4_v4f32\n",
5605 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
5606 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5607 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
5608 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
5609 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
5610 "%f_val = OpLoad %v4f32 %f\n"
5611 "%f1_val = OpLoad %v4f32 %f1\n"
5612 "%f2_val = OpLoad %v4f32 %f2\n"
5613 "%f3_val = OpLoad %v4f32 %f3\n"
5614 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
5615 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
5616 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
5617 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
5621 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5622 "%cnull = OpConstantNull %mat4x4_f32\n",
5623 // Our null matrix * any vector should result in a zero vector.
5624 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
5625 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
5629 getHalfColorsFullAlpha(colors);
5631 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5633 map<string, string> fragments;
5634 fragments["pre_main"] = tests[testNdx].constants;
5635 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5636 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
5638 return opConstantNullTests.release();
5640 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
5642 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
5643 RGBA inputColors[4];
5644 RGBA outputColors[4];
5647 const char functionStart[] =
5648 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5649 "%param1 = OpFunctionParameter %v4f32\n"
5652 const char functionEnd[] =
5653 "OpReturnValue %transformed_param\n"
5656 struct NameConstantsCode
5663 NameConstantsCode tests[] =
5668 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
5669 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
5674 "%stype = OpTypeStruct %v4f32 %f32\n"
5675 "%fp_stype = OpTypePointer Function %stype\n"
5676 "%f32_n_1 = OpConstant %f32 -1.0\n"
5677 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5678 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
5679 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
5681 "%v = OpVariable %fp_stype Function %cval\n"
5682 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5683 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
5684 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
5685 "%f32_val = OpLoad %f32 %f32_ptr\n"
5686 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
5687 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
5688 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
5691 // [1|0|0|0.5] [x] = x + 0.5
5692 // [0|1|0|0.5] [y] = y + 0.5
5693 // [0|0|1|0.5] [z] = z + 0.5
5694 // [0|0|0|1 ] [1] = 1
5697 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5698 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
5699 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
5700 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
5701 "%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"
5702 "%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",
5704 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
5709 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5710 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5711 "%f32_n_1 = OpConstant %f32 -1.0\n"
5712 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5713 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
5715 "%v = OpVariable %fp_a4f32 Function %carr\n"
5716 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
5717 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
5718 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5719 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
5720 "%f_val = OpLoad %f32 %f\n"
5721 "%f1_val = OpLoad %f32 %f1\n"
5722 "%f2_val = OpLoad %f32 %f2\n"
5723 "%f3_val = OpLoad %f32 %f3\n"
5724 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
5725 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
5726 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
5727 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
5728 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5735 // [ 1.0, 1.0, 1.0, 1.0]
5739 // [ 0.0, 0.5, 0.0, 0.0]
5743 // [ 1.0, 1.0, 1.0, 1.0]
5746 "array_of_struct_of_array",
5748 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5749 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5750 "%stype = OpTypeStruct %f32 %a4f32\n"
5751 "%a3stype = OpTypeArray %stype %c_u32_3\n"
5752 "%fp_a3stype = OpTypePointer Function %a3stype\n"
5753 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
5754 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5755 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
5756 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
5757 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
5759 "%v = OpVariable %fp_a3stype Function %carr\n"
5760 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
5761 "%f_l = OpLoad %f32 %f\n"
5762 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
5763 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5767 getHalfColorsFullAlpha(inputColors);
5768 outputColors[0] = RGBA(255, 255, 255, 255);
5769 outputColors[1] = RGBA(255, 127, 127, 255);
5770 outputColors[2] = RGBA(127, 255, 127, 255);
5771 outputColors[3] = RGBA(127, 127, 255, 255);
5773 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5775 map<string, string> fragments;
5776 fragments["pre_main"] = tests[testNdx].constants;
5777 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5778 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
5780 return opConstantCompositeTests.release();
5783 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
5785 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
5786 RGBA inputColors[4];
5787 RGBA outputColors[4];
5788 map<string, string> fragments;
5790 // vec4 test_code(vec4 param) {
5791 // vec4 result = param;
5792 // for (int i = 0; i < 4; ++i) {
5793 // if (i == 0) result[i] = 0.;
5794 // else result[i] = 1. - result[i];
5798 const char function[] =
5799 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5800 "%param1 = OpFunctionParameter %v4f32\n"
5802 "%iptr = OpVariable %fp_i32 Function\n"
5803 "%result = OpVariable %fp_v4f32 Function\n"
5804 " OpStore %iptr %c_i32_0\n"
5805 " OpStore %result %param1\n"
5808 // Loop entry block.
5810 "%ival = OpLoad %i32 %iptr\n"
5811 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5812 " OpLoopMerge %exit %if_entry None\n"
5813 " OpBranchConditional %lt_4 %if_entry %exit\n"
5815 // Merge block for loop.
5817 "%ret = OpLoad %v4f32 %result\n"
5818 " OpReturnValue %ret\n"
5820 // If-statement entry block.
5821 "%if_entry = OpLabel\n"
5822 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5823 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
5824 " OpSelectionMerge %if_exit None\n"
5825 " OpBranchConditional %eq_0 %if_true %if_false\n"
5827 // False branch for if-statement.
5828 "%if_false = OpLabel\n"
5829 "%val = OpLoad %f32 %loc\n"
5830 "%sub = OpFSub %f32 %c_f32_1 %val\n"
5831 " OpStore %loc %sub\n"
5832 " OpBranch %if_exit\n"
5834 // Merge block for if-statement.
5835 "%if_exit = OpLabel\n"
5836 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5837 " OpStore %iptr %ival_next\n"
5840 // True branch for if-statement.
5841 "%if_true = OpLabel\n"
5842 " OpStore %loc %c_f32_0\n"
5843 " OpBranch %if_exit\n"
5847 fragments["testfun"] = function;
5849 inputColors[0] = RGBA(127, 127, 127, 0);
5850 inputColors[1] = RGBA(127, 0, 0, 0);
5851 inputColors[2] = RGBA(0, 127, 0, 0);
5852 inputColors[3] = RGBA(0, 0, 127, 0);
5854 outputColors[0] = RGBA(0, 128, 128, 255);
5855 outputColors[1] = RGBA(0, 255, 255, 255);
5856 outputColors[2] = RGBA(0, 128, 255, 255);
5857 outputColors[3] = RGBA(0, 255, 128, 255);
5859 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5861 return group.release();
5864 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
5866 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
5867 RGBA inputColors[4];
5868 RGBA outputColors[4];
5869 map<string, string> fragments;
5871 const char typesAndConstants[] =
5872 "%c_f32_p2 = OpConstant %f32 0.2\n"
5873 "%c_f32_p4 = OpConstant %f32 0.4\n"
5874 "%c_f32_p6 = OpConstant %f32 0.6\n"
5875 "%c_f32_p8 = OpConstant %f32 0.8\n";
5877 // vec4 test_code(vec4 param) {
5878 // vec4 result = param;
5879 // for (int i = 0; i < 4; ++i) {
5881 // case 0: result[i] += .2; break;
5882 // case 1: result[i] += .6; break;
5883 // case 2: result[i] += .4; break;
5884 // case 3: result[i] += .8; break;
5885 // default: break; // unreachable
5890 const char function[] =
5891 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5892 "%param1 = OpFunctionParameter %v4f32\n"
5894 "%iptr = OpVariable %fp_i32 Function\n"
5895 "%result = OpVariable %fp_v4f32 Function\n"
5896 " OpStore %iptr %c_i32_0\n"
5897 " OpStore %result %param1\n"
5900 // Loop entry block.
5902 "%ival = OpLoad %i32 %iptr\n"
5903 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5904 " OpLoopMerge %exit %switch_exit None\n"
5905 " OpBranchConditional %lt_4 %switch_entry %exit\n"
5907 // Merge block for loop.
5909 "%ret = OpLoad %v4f32 %result\n"
5910 " OpReturnValue %ret\n"
5912 // Switch-statement entry block.
5913 "%switch_entry = OpLabel\n"
5914 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5915 "%val = OpLoad %f32 %loc\n"
5916 " OpSelectionMerge %switch_exit None\n"
5917 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5919 "%case2 = OpLabel\n"
5920 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
5921 " OpStore %loc %addp4\n"
5922 " OpBranch %switch_exit\n"
5924 "%switch_default = OpLabel\n"
5927 "%case3 = OpLabel\n"
5928 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
5929 " OpStore %loc %addp8\n"
5930 " OpBranch %switch_exit\n"
5932 "%case0 = OpLabel\n"
5933 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
5934 " OpStore %loc %addp2\n"
5935 " OpBranch %switch_exit\n"
5937 // Merge block for switch-statement.
5938 "%switch_exit = OpLabel\n"
5939 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5940 " OpStore %iptr %ival_next\n"
5943 "%case1 = OpLabel\n"
5944 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
5945 " OpStore %loc %addp6\n"
5946 " OpBranch %switch_exit\n"
5950 fragments["pre_main"] = typesAndConstants;
5951 fragments["testfun"] = function;
5953 inputColors[0] = RGBA(127, 27, 127, 51);
5954 inputColors[1] = RGBA(127, 0, 0, 51);
5955 inputColors[2] = RGBA(0, 27, 0, 51);
5956 inputColors[3] = RGBA(0, 0, 127, 51);
5958 outputColors[0] = RGBA(178, 180, 229, 255);
5959 outputColors[1] = RGBA(178, 153, 102, 255);
5960 outputColors[2] = RGBA(51, 180, 102, 255);
5961 outputColors[3] = RGBA(51, 153, 229, 255);
5963 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5965 return group.release();
5968 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
5970 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
5971 RGBA inputColors[4];
5972 RGBA outputColors[4];
5973 map<string, string> fragments;
5975 const char decorations[] =
5976 "OpDecorate %array_group ArrayStride 4\n"
5977 "OpDecorate %struct_member_group Offset 0\n"
5978 "%array_group = OpDecorationGroup\n"
5979 "%struct_member_group = OpDecorationGroup\n"
5981 "OpDecorate %group1 RelaxedPrecision\n"
5982 "OpDecorate %group3 RelaxedPrecision\n"
5983 "OpDecorate %group3 Invariant\n"
5984 "OpDecorate %group3 Restrict\n"
5985 "%group0 = OpDecorationGroup\n"
5986 "%group1 = OpDecorationGroup\n"
5987 "%group3 = OpDecorationGroup\n";
5989 const char typesAndConstants[] =
5990 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
5991 "%struct1 = OpTypeStruct %a3f32\n"
5992 "%struct2 = OpTypeStruct %a3f32\n"
5993 "%fp_struct1 = OpTypePointer Function %struct1\n"
5994 "%fp_struct2 = OpTypePointer Function %struct2\n"
5995 "%c_f32_2 = OpConstant %f32 2.\n"
5996 "%c_f32_n2 = OpConstant %f32 -2.\n"
5998 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
5999 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
6000 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
6001 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
6003 const char function[] =
6004 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6005 "%param = OpFunctionParameter %v4f32\n"
6006 "%entry = OpLabel\n"
6007 "%result = OpVariable %fp_v4f32 Function\n"
6008 "%v_struct1 = OpVariable %fp_struct1 Function\n"
6009 "%v_struct2 = OpVariable %fp_struct2 Function\n"
6010 " OpStore %result %param\n"
6011 " OpStore %v_struct1 %c_struct1\n"
6012 " OpStore %v_struct2 %c_struct2\n"
6013 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
6014 "%val1 = OpLoad %f32 %ptr1\n"
6015 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
6016 "%val2 = OpLoad %f32 %ptr2\n"
6017 "%addvalues = OpFAdd %f32 %val1 %val2\n"
6018 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
6019 "%val = OpLoad %f32 %ptr\n"
6020 "%addresult = OpFAdd %f32 %addvalues %val\n"
6021 " OpStore %ptr %addresult\n"
6022 "%ret = OpLoad %v4f32 %result\n"
6023 " OpReturnValue %ret\n"
6026 struct CaseNameDecoration
6032 CaseNameDecoration tests[] =
6035 "same_decoration_group_on_multiple_types",
6036 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
6039 "empty_decoration_group",
6040 "OpGroupDecorate %group0 %a3f32\n"
6041 "OpGroupDecorate %group0 %result\n"
6044 "one_element_decoration_group",
6045 "OpGroupDecorate %array_group %a3f32\n"
6048 "multiple_elements_decoration_group",
6049 "OpGroupDecorate %group3 %v_struct1\n"
6052 "multiple_decoration_groups_on_same_variable",
6053 "OpGroupDecorate %group0 %v_struct2\n"
6054 "OpGroupDecorate %group1 %v_struct2\n"
6055 "OpGroupDecorate %group3 %v_struct2\n"
6058 "same_decoration_group_multiple_times",
6059 "OpGroupDecorate %group1 %addvalues\n"
6060 "OpGroupDecorate %group1 %addvalues\n"
6061 "OpGroupDecorate %group1 %addvalues\n"
6066 getHalfColorsFullAlpha(inputColors);
6067 getHalfColorsFullAlpha(outputColors);
6069 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
6071 fragments["decoration"] = decorations + tests[idx].decoration;
6072 fragments["pre_main"] = typesAndConstants;
6073 fragments["testfun"] = function;
6075 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
6078 return group.release();
6081 struct SpecConstantTwoIntGraphicsCase
6083 const char* caseName;
6084 const char* scDefinition0;
6085 const char* scDefinition1;
6086 const char* scResultType;
6087 const char* scOperation;
6088 deInt32 scActualValue0;
6089 deInt32 scActualValue1;
6090 const char* resultOperation;
6091 RGBA expectedColors[4];
6093 SpecConstantTwoIntGraphicsCase (const char* name,
6094 const char* definition0,
6095 const char* definition1,
6096 const char* resultType,
6097 const char* operation,
6100 const char* resultOp,
6101 const RGBA (&output)[4])
6103 , scDefinition0 (definition0)
6104 , scDefinition1 (definition1)
6105 , scResultType (resultType)
6106 , scOperation (operation)
6107 , scActualValue0 (value0)
6108 , scActualValue1 (value1)
6109 , resultOperation (resultOp)
6111 expectedColors[0] = output[0];
6112 expectedColors[1] = output[1];
6113 expectedColors[2] = output[2];
6114 expectedColors[3] = output[3];
6118 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
6120 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
6121 vector<SpecConstantTwoIntGraphicsCase> cases;
6122 RGBA inputColors[4];
6123 RGBA outputColors0[4];
6124 RGBA outputColors1[4];
6125 RGBA outputColors2[4];
6127 const char decorations1[] =
6128 "OpDecorate %sc_0 SpecId 0\n"
6129 "OpDecorate %sc_1 SpecId 1\n";
6131 const char typesAndConstants1[] =
6132 "${OPTYPE_DEFINITIONS:opt}"
6133 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
6134 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
6135 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
6137 const char function1[] =
6138 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6139 "%param = OpFunctionParameter %v4f32\n"
6140 "%label = OpLabel\n"
6141 "${TYPE_CONVERT:opt}"
6142 "%result = OpVariable %fp_v4f32 Function\n"
6143 " OpStore %result %param\n"
6144 "%gen = ${GEN_RESULT}\n"
6145 "%index = OpIAdd %i32 %gen %c_i32_1\n"
6146 "%loc = OpAccessChain %fp_f32 %result %index\n"
6147 "%val = OpLoad %f32 %loc\n"
6148 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6149 " OpStore %loc %add\n"
6150 "%ret = OpLoad %v4f32 %result\n"
6151 " OpReturnValue %ret\n"
6154 inputColors[0] = RGBA(127, 127, 127, 255);
6155 inputColors[1] = RGBA(127, 0, 0, 255);
6156 inputColors[2] = RGBA(0, 127, 0, 255);
6157 inputColors[3] = RGBA(0, 0, 127, 255);
6159 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
6160 outputColors0[0] = RGBA(255, 127, 127, 255);
6161 outputColors0[1] = RGBA(255, 0, 0, 255);
6162 outputColors0[2] = RGBA(128, 127, 0, 255);
6163 outputColors0[3] = RGBA(128, 0, 127, 255);
6165 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
6166 outputColors1[0] = RGBA(127, 255, 127, 255);
6167 outputColors1[1] = RGBA(127, 128, 0, 255);
6168 outputColors1[2] = RGBA(0, 255, 0, 255);
6169 outputColors1[3] = RGBA(0, 128, 127, 255);
6171 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
6172 outputColors2[0] = RGBA(127, 127, 255, 255);
6173 outputColors2[1] = RGBA(127, 0, 128, 255);
6174 outputColors2[2] = RGBA(0, 127, 128, 255);
6175 outputColors2[3] = RGBA(0, 0, 255, 255);
6177 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
6178 const char addZeroToSc32[] = "OpIAdd %i32 %c_i32_0 %sc_op32";
6179 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
6180 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
6182 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
6183 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
6184 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
6185 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
6186 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
6187 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6188 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6189 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
6190 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
6191 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
6192 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
6193 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
6194 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
6195 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
6196 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
6197 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
6198 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6199 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
6200 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
6201 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
6202 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6203 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
6204 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
6205 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6206 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6207 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6208 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6209 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
6210 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
6211 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
6212 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
6213 cases.push_back(SpecConstantTwoIntGraphicsCase("sconvert", " %i32 0", " %i32 0", "%i16", "SConvert %sc_0", -1, 0, addZeroToSc32, outputColors0));
6214 // -1082130432 stored as 32-bit two's complement is the binary representation of -1 as IEEE-754 Float
6215 cases.push_back(SpecConstantTwoIntGraphicsCase("fconvert", " %f32 0", " %f32 0", "%f64", "FConvert %sc_0", -1082130432, 0, addZeroToSc32, outputColors0));
6216 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
6218 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
6220 map<string, string> specializations;
6221 map<string, string> fragments;
6222 vector<deInt32> specConstants;
6223 vector<string> features;
6224 PushConstants noPushConstants;
6225 GraphicsResources noResources;
6226 GraphicsInterfaces noInterfaces;
6227 std::vector<std::string> noExtensions;
6229 // Special SPIR-V code for SConvert-case
6230 if (strcmp(cases[caseNdx].caseName, "sconvert") == 0)
6232 features.push_back("shaderInt16");
6233 fragments["capability"] = "OpCapability Int16\n"; // Adds 16-bit integer capability
6234 specializations["OPTYPE_DEFINITIONS"] = "%i16 = OpTypeInt 16 1\n"; // Adds 16-bit integer type
6235 specializations["TYPE_CONVERT"] = "%sc_op32 = OpSConvert %i32 %sc_op\n"; // Converts 16-bit integer to 32-bit integer
6238 // Special SPIR-V code for FConvert-case
6239 if (strcmp(cases[caseNdx].caseName, "fconvert") == 0)
6241 features.push_back("shaderFloat64");
6242 fragments["capability"] = "OpCapability Float64\n"; // Adds 64-bit float capability
6243 specializations["OPTYPE_DEFINITIONS"] = "%f64 = OpTypeFloat 64\n"; // Adds 64-bit float type
6244 specializations["TYPE_CONVERT"] = "%sc_op32 = OpConvertFToS %i32 %sc_op\n"; // Converts 64-bit float to 32-bit integer
6247 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
6248 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
6249 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
6250 specializations["SC_OP"] = cases[caseNdx].scOperation;
6251 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
6253 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
6254 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
6255 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
6257 specConstants.push_back(cases[caseNdx].scActualValue0);
6258 specConstants.push_back(cases[caseNdx].scActualValue1);
6260 createTestsForAllStages(
6261 cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants,
6262 noPushConstants, noResources, noInterfaces, noExtensions, features, VulkanFeatures(), group.get());
6265 const char decorations2[] =
6266 "OpDecorate %sc_0 SpecId 0\n"
6267 "OpDecorate %sc_1 SpecId 1\n"
6268 "OpDecorate %sc_2 SpecId 2\n";
6270 const char typesAndConstants2[] =
6271 "%v3i32 = OpTypeVector %i32 3\n"
6272 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
6273 "%vec3_undef = OpUndef %v3i32\n"
6275 "%sc_0 = OpSpecConstant %i32 0\n"
6276 "%sc_1 = OpSpecConstant %i32 0\n"
6277 "%sc_2 = OpSpecConstant %i32 0\n"
6278 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
6279 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
6280 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
6281 "%sc_vec3_0_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
6282 "%sc_vec3_1_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
6283 "%sc_vec3_2_s = OpSpecConstantOp %v3i32 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
6284 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
6285 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
6286 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
6287 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
6288 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
6289 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
6290 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
6292 const char function2[] =
6293 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6294 "%param = OpFunctionParameter %v4f32\n"
6295 "%label = OpLabel\n"
6296 "%result = OpVariable %fp_v4f32 Function\n"
6297 " OpStore %result %param\n"
6298 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
6299 "%val = OpLoad %f32 %loc\n"
6300 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6301 " OpStore %loc %add\n"
6302 "%ret = OpLoad %v4f32 %result\n"
6303 " OpReturnValue %ret\n"
6306 map<string, string> fragments;
6307 vector<deInt32> specConstants;
6309 fragments["decoration"] = decorations2;
6310 fragments["pre_main"] = typesAndConstants2;
6311 fragments["testfun"] = function2;
6313 specConstants.push_back(56789);
6314 specConstants.push_back(-2);
6315 specConstants.push_back(56788);
6317 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
6319 return group.release();
6322 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
6324 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
6325 RGBA inputColors[4];
6326 RGBA outputColors1[4];
6327 RGBA outputColors2[4];
6328 RGBA outputColors3[4];
6329 map<string, string> fragments1;
6330 map<string, string> fragments2;
6331 map<string, string> fragments3;
6333 const char typesAndConstants1[] =
6334 "%c_f32_p2 = OpConstant %f32 0.2\n"
6335 "%c_f32_p4 = OpConstant %f32 0.4\n"
6336 "%c_f32_p5 = OpConstant %f32 0.5\n"
6337 "%c_f32_p8 = OpConstant %f32 0.8\n";
6339 // vec4 test_code(vec4 param) {
6340 // vec4 result = param;
6341 // for (int i = 0; i < 4; ++i) {
6344 // case 0: operand = .2; break;
6345 // case 1: operand = .5; break;
6346 // case 2: operand = .4; break;
6347 // case 3: operand = .0; break;
6348 // default: break; // unreachable
6350 // result[i] += operand;
6354 const char function1[] =
6355 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6356 "%param1 = OpFunctionParameter %v4f32\n"
6358 "%iptr = OpVariable %fp_i32 Function\n"
6359 "%result = OpVariable %fp_v4f32 Function\n"
6360 " OpStore %iptr %c_i32_0\n"
6361 " OpStore %result %param1\n"
6365 "%ival = OpLoad %i32 %iptr\n"
6366 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6367 " OpLoopMerge %exit %phi None\n"
6368 " OpBranchConditional %lt_4 %entry %exit\n"
6370 "%entry = OpLabel\n"
6371 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6372 "%val = OpLoad %f32 %loc\n"
6373 " OpSelectionMerge %phi None\n"
6374 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6376 "%case0 = OpLabel\n"
6378 "%case1 = OpLabel\n"
6380 "%case2 = OpLabel\n"
6382 "%case3 = OpLabel\n"
6385 "%default = OpLabel\n"
6389 "%operand = OpPhi %f32 %c_f32_p4 %case2 %c_f32_p5 %case1 %c_f32_p2 %case0 %c_f32_0 %case3\n" // not in the order of blocks
6390 "%add = OpFAdd %f32 %val %operand\n"
6391 " OpStore %loc %add\n"
6392 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6393 " OpStore %iptr %ival_next\n"
6397 "%ret = OpLoad %v4f32 %result\n"
6398 " OpReturnValue %ret\n"
6402 fragments1["pre_main"] = typesAndConstants1;
6403 fragments1["testfun"] = function1;
6405 getHalfColorsFullAlpha(inputColors);
6407 outputColors1[0] = RGBA(178, 255, 229, 255);
6408 outputColors1[1] = RGBA(178, 127, 102, 255);
6409 outputColors1[2] = RGBA(51, 255, 102, 255);
6410 outputColors1[3] = RGBA(51, 127, 229, 255);
6412 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
6414 const char typesAndConstants2[] =
6415 "%c_f32_p2 = OpConstant %f32 0.2\n";
6417 // Add .4 to the second element of the given parameter.
6418 const char function2[] =
6419 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6420 "%param = OpFunctionParameter %v4f32\n"
6421 "%entry = OpLabel\n"
6422 "%result = OpVariable %fp_v4f32 Function\n"
6423 " OpStore %result %param\n"
6424 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6425 "%val = OpLoad %f32 %loc\n"
6429 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
6430 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
6431 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
6432 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
6433 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
6434 " OpLoopMerge %exit %phi None\n"
6435 " OpBranchConditional %still_loop %phi %exit\n"
6438 " OpStore %loc %accum\n"
6439 "%ret = OpLoad %v4f32 %result\n"
6440 " OpReturnValue %ret\n"
6444 fragments2["pre_main"] = typesAndConstants2;
6445 fragments2["testfun"] = function2;
6447 outputColors2[0] = RGBA(127, 229, 127, 255);
6448 outputColors2[1] = RGBA(127, 102, 0, 255);
6449 outputColors2[2] = RGBA(0, 229, 0, 255);
6450 outputColors2[3] = RGBA(0, 102, 127, 255);
6452 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
6454 const char typesAndConstants3[] =
6455 "%true = OpConstantTrue %bool\n"
6456 "%false = OpConstantFalse %bool\n"
6457 "%c_f32_p2 = OpConstant %f32 0.2\n";
6459 // Swap the second and the third element of the given parameter.
6460 const char function3[] =
6461 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6462 "%param = OpFunctionParameter %v4f32\n"
6463 "%entry = OpLabel\n"
6464 "%result = OpVariable %fp_v4f32 Function\n"
6465 " OpStore %result %param\n"
6466 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6467 "%a_init = OpLoad %f32 %a_loc\n"
6468 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
6469 "%b_init = OpLoad %f32 %b_loc\n"
6473 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
6474 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
6475 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
6476 " OpLoopMerge %exit %phi None\n"
6477 " OpBranchConditional %still_loop %phi %exit\n"
6480 " OpStore %a_loc %a_next\n"
6481 " OpStore %b_loc %b_next\n"
6482 "%ret = OpLoad %v4f32 %result\n"
6483 " OpReturnValue %ret\n"
6487 fragments3["pre_main"] = typesAndConstants3;
6488 fragments3["testfun"] = function3;
6490 outputColors3[0] = RGBA(127, 127, 127, 255);
6491 outputColors3[1] = RGBA(127, 0, 0, 255);
6492 outputColors3[2] = RGBA(0, 0, 127, 255);
6493 outputColors3[3] = RGBA(0, 127, 0, 255);
6495 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
6497 return group.release();
6500 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
6502 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
6503 RGBA inputColors[4];
6504 RGBA outputColors[4];
6506 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
6507 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
6508 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
6509 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
6510 const char constantsAndTypes[] =
6511 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
6512 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
6513 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
6514 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
6515 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n";
6517 const char function[] =
6518 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6519 "%param = OpFunctionParameter %v4f32\n"
6520 "%label = OpLabel\n"
6521 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
6522 "%var2 = OpVariable %fp_f32 Function\n"
6523 "%red = OpCompositeExtract %f32 %param 0\n"
6524 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
6525 " OpStore %var2 %plus_red\n"
6526 "%val1 = OpLoad %f32 %var1\n"
6527 "%val2 = OpLoad %f32 %var2\n"
6528 "%mul = OpFMul %f32 %val1 %val2\n"
6529 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
6530 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
6531 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
6532 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
6533 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
6534 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
6535 " OpReturnValue %ret\n"
6538 struct CaseNameDecoration
6545 CaseNameDecoration tests[] = {
6546 {"multiplication", "OpDecorate %mul NoContraction"},
6547 {"addition", "OpDecorate %add NoContraction"},
6548 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
6551 getHalfColorsFullAlpha(inputColors);
6553 for (deUint8 idx = 0; idx < 4; ++idx)
6555 inputColors[idx].setRed(0);
6556 outputColors[idx] = RGBA(0, 0, 0, 255);
6559 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
6561 map<string, string> fragments;
6563 fragments["decoration"] = tests[testNdx].decoration;
6564 fragments["pre_main"] = constantsAndTypes;
6565 fragments["testfun"] = function;
6567 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
6570 return group.release();
6573 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
6575 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
6578 const char constantsAndTypes[] =
6579 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
6580 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
6581 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
6582 "%fp_stype = OpTypePointer Function %stype\n";
6584 const char function[] =
6585 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6586 "%param1 = OpFunctionParameter %v4f32\n"
6588 "%v1 = OpVariable %fp_v4f32 Function\n"
6589 "%v2 = OpVariable %fp_a2f32 Function\n"
6590 "%v3 = OpVariable %fp_f32 Function\n"
6591 "%v = OpVariable %fp_stype Function\n"
6592 "%vv = OpVariable %fp_stype Function\n"
6593 "%vvv = OpVariable %fp_f32 Function\n"
6595 " OpStore %v1 %c_v4f32_1_1_1_1\n"
6596 " OpStore %v2 %c_a2f32_1\n"
6597 " OpStore %v3 %c_f32_1\n"
6599 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
6600 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
6601 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
6602 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
6603 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
6604 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
6606 " OpStore %p_v4f32 %v1_v ${access_type}\n"
6607 " OpStore %p_a2f32 %v2_v ${access_type}\n"
6608 " OpStore %p_f32 %v3_v ${access_type}\n"
6610 " OpCopyMemory %vv %v ${access_type}\n"
6611 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
6613 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
6614 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
6615 "%v_f32_3 = OpLoad %f32 %vvv\n"
6617 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
6618 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
6619 " OpReturnValue %ret2\n"
6622 struct NameMemoryAccess
6629 NameMemoryAccess tests[] =
6632 { "volatile", "Volatile" },
6633 { "aligned", "Aligned 1" },
6634 { "volatile_aligned", "Volatile|Aligned 1" },
6635 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
6636 { "volatile_nontemporal", "Volatile|Nontemporal" },
6637 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
6640 getHalfColorsFullAlpha(colors);
6642 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
6644 map<string, string> fragments;
6645 map<string, string> memoryAccess;
6646 memoryAccess["access_type"] = tests[testNdx].accessType;
6648 fragments["pre_main"] = constantsAndTypes;
6649 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
6650 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
6652 return memoryAccessTests.release();
6654 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
6656 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
6657 RGBA defaultColors[4];
6658 map<string, string> fragments;
6659 getDefaultColors(defaultColors);
6661 // First, simple cases that don't do anything with the OpUndef result.
6662 struct NameCodePair { string name, decl, type; };
6663 const NameCodePair tests[] =
6665 {"bool", "", "%bool"},
6666 {"vec2uint32", "", "%v2u32"},
6667 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown", "%type"},
6668 {"sampler", "%type = OpTypeSampler", "%type"},
6669 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img", "%type"},
6670 {"pointer", "", "%fp_i32"},
6671 {"runtimearray", "%type = OpTypeRuntimeArray %f32", "%type"},
6672 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100", "%type"},
6673 {"struct", "%type = OpTypeStruct %f32 %i32 %u32", "%type"}};
6674 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
6676 fragments["undef_type"] = tests[testNdx].type;
6677 fragments["testfun"] = StringTemplate(
6678 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6679 "%param1 = OpFunctionParameter %v4f32\n"
6680 "%label_testfun = OpLabel\n"
6681 "%undef = OpUndef ${undef_type}\n"
6682 "OpReturnValue %param1\n"
6683 "OpFunctionEnd\n").specialize(fragments);
6684 fragments["pre_main"] = tests[testNdx].decl;
6685 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
6689 fragments["testfun"] =
6690 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6691 "%param1 = OpFunctionParameter %v4f32\n"
6692 "%label_testfun = OpLabel\n"
6693 "%undef = OpUndef %f32\n"
6694 "%zero = OpFMul %f32 %undef %c_f32_0\n"
6695 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
6696 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
6697 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6698 "%b = OpFAdd %f32 %a %actually_zero\n"
6699 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
6700 "OpReturnValue %ret\n"
6703 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6705 fragments["testfun"] =
6706 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6707 "%param1 = OpFunctionParameter %v4f32\n"
6708 "%label_testfun = OpLabel\n"
6709 "%undef = OpUndef %i32\n"
6710 "%zero = OpIMul %i32 %undef %c_i32_0\n"
6711 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6712 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6713 "OpReturnValue %ret\n"
6716 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6718 fragments["testfun"] =
6719 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6720 "%param1 = OpFunctionParameter %v4f32\n"
6721 "%label_testfun = OpLabel\n"
6722 "%undef = OpUndef %u32\n"
6723 "%zero = OpIMul %u32 %undef %c_i32_0\n"
6724 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6725 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6726 "OpReturnValue %ret\n"
6729 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6731 fragments["testfun"] =
6732 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6733 "%param1 = OpFunctionParameter %v4f32\n"
6734 "%label_testfun = OpLabel\n"
6735 "%undef = OpUndef %v4f32\n"
6736 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
6737 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
6738 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
6739 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
6740 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
6741 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6742 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6743 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6744 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6745 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6746 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
6747 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
6748 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
6749 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6750 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6751 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6752 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6753 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6754 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6755 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6756 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6757 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6758 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6759 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6760 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6761 "OpReturnValue %ret\n"
6764 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6766 fragments["pre_main"] =
6767 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
6768 fragments["testfun"] =
6769 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6770 "%param1 = OpFunctionParameter %v4f32\n"
6771 "%label_testfun = OpLabel\n"
6772 "%undef = OpUndef %m2x2f32\n"
6773 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
6774 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
6775 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
6776 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
6777 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
6778 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6779 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6780 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6781 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6782 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6783 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
6784 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
6785 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
6786 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6787 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6788 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6789 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6790 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6791 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6792 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6793 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6794 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6795 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6796 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6797 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6798 "OpReturnValue %ret\n"
6801 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
6803 return opUndefTests.release();
6806 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
6808 const RGBA inputColors[4] =
6811 RGBA(0, 0, 255, 255),
6812 RGBA(0, 255, 0, 255),
6813 RGBA(0, 255, 255, 255)
6816 const RGBA expectedColors[4] =
6818 RGBA(255, 0, 0, 255),
6819 RGBA(255, 0, 0, 255),
6820 RGBA(255, 0, 0, 255),
6821 RGBA(255, 0, 0, 255)
6824 const struct SingleFP16Possibility
6827 const char* constant; // Value to assign to %test_constant.
6829 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
6835 -constructNormalizedFloat(1, 0x300000),
6836 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6841 constructNormalizedFloat(7, 0x000000),
6842 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6844 // SPIR-V requires that OpQuantizeToF16 flushes
6845 // any numbers that would end up denormalized in F16 to zero.
6849 std::ldexp(1.5f, -140),
6850 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6855 -std::ldexp(1.5f, -140),
6856 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6861 std::ldexp(1.0f, -16),
6862 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6863 }, // too small positive
6865 "negative_too_small",
6867 -std::ldexp(1.0f, -32),
6868 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6869 }, // too small negative
6873 -std::ldexp(1.0f, 128),
6875 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6876 "%inf = OpIsInf %bool %c\n"
6877 "%cond = OpLogicalAnd %bool %gz %inf\n"
6882 std::ldexp(1.0f, 128),
6884 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6885 "%inf = OpIsInf %bool %c\n"
6886 "%cond = OpLogicalAnd %bool %gz %inf\n"
6889 "round_to_negative_inf",
6891 -std::ldexp(1.0f, 32),
6893 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6894 "%inf = OpIsInf %bool %c\n"
6895 "%cond = OpLogicalAnd %bool %gz %inf\n"
6900 std::ldexp(1.0f, 16),
6902 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6903 "%inf = OpIsInf %bool %c\n"
6904 "%cond = OpLogicalAnd %bool %gz %inf\n"
6909 std::numeric_limits<float>::quiet_NaN(),
6911 // Test for any NaN value, as NaNs are not preserved
6912 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6913 "%cond = OpIsNan %bool %direct_quant\n"
6918 std::numeric_limits<float>::quiet_NaN(),
6920 // Test for any NaN value, as NaNs are not preserved
6921 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6922 "%cond = OpIsNan %bool %direct_quant\n"
6925 const char* constants =
6926 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
6928 StringTemplate function (
6929 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6930 "%param1 = OpFunctionParameter %v4f32\n"
6931 "%label_testfun = OpLabel\n"
6932 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6933 "%b = OpFAdd %f32 %test_constant %a\n"
6934 "%c = OpQuantizeToF16 %f32 %b\n"
6936 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6937 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
6938 " OpReturnValue %retval\n"
6942 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
6943 const char* specConstants =
6944 "%test_constant = OpSpecConstant %f32 0.\n"
6945 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
6947 StringTemplate specConstantFunction(
6948 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6949 "%param1 = OpFunctionParameter %v4f32\n"
6950 "%label_testfun = OpLabel\n"
6952 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6953 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
6954 " OpReturnValue %retval\n"
6958 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6960 map<string, string> codeSpecialization;
6961 map<string, string> fragments;
6962 codeSpecialization["condition"] = tests[idx].condition;
6963 fragments["testfun"] = function.specialize(codeSpecialization);
6964 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
6965 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6968 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6970 map<string, string> codeSpecialization;
6971 map<string, string> fragments;
6972 vector<deInt32> passConstants;
6973 deInt32 specConstant;
6975 codeSpecialization["condition"] = tests[idx].condition;
6976 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
6977 fragments["decoration"] = specDecorations;
6978 fragments["pre_main"] = specConstants;
6980 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
6981 passConstants.push_back(specConstant);
6983 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6987 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
6989 RGBA inputColors[4] = {
6991 RGBA(0, 0, 255, 255),
6992 RGBA(0, 255, 0, 255),
6993 RGBA(0, 255, 255, 255)
6996 RGBA expectedColors[4] =
6998 RGBA(255, 0, 0, 255),
6999 RGBA(255, 0, 0, 255),
7000 RGBA(255, 0, 0, 255),
7001 RGBA(255, 0, 0, 255)
7004 struct DualFP16Possibility
7009 const char* possibleOutput1;
7010 const char* possibleOutput2;
7013 "positive_round_up_or_round_down",
7015 constructNormalizedFloat(8, 0x300300),
7020 "negative_round_up_or_round_down",
7022 -constructNormalizedFloat(-7, 0x600800),
7029 constructNormalizedFloat(2, 0x01e000),
7034 "carry_to_exponent",
7036 constructNormalizedFloat(1, 0xffe000),
7041 StringTemplate constants (
7042 "%input_const = OpConstant %f32 ${input}\n"
7043 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7044 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7047 StringTemplate specConstants (
7048 "%input_const = OpSpecConstant %f32 0.\n"
7049 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7050 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7053 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
7055 const char* function =
7056 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7057 "%param1 = OpFunctionParameter %v4f32\n"
7058 "%label_testfun = OpLabel\n"
7059 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7060 // For the purposes of this test we assume that 0.f will always get
7061 // faithfully passed through the pipeline stages.
7062 "%b = OpFAdd %f32 %input_const %a\n"
7063 "%c = OpQuantizeToF16 %f32 %b\n"
7064 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
7065 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
7066 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
7067 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7068 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
7069 " OpReturnValue %retval\n"
7072 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7073 map<string, string> fragments;
7074 map<string, string> constantSpecialization;
7076 constantSpecialization["input"] = tests[idx].input;
7077 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7078 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7079 fragments["testfun"] = function;
7080 fragments["pre_main"] = constants.specialize(constantSpecialization);
7081 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7084 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7085 map<string, string> fragments;
7086 map<string, string> constantSpecialization;
7087 vector<deInt32> passConstants;
7088 deInt32 specConstant;
7090 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7091 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7092 fragments["testfun"] = function;
7093 fragments["decoration"] = specDecorations;
7094 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
7096 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
7097 passConstants.push_back(specConstant);
7099 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7103 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
7105 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
7106 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
7107 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
7108 return opQuantizeTests.release();
7111 struct ShaderPermutation
7113 deUint8 vertexPermutation;
7114 deUint8 geometryPermutation;
7115 deUint8 tesscPermutation;
7116 deUint8 tessePermutation;
7117 deUint8 fragmentPermutation;
7120 ShaderPermutation getShaderPermutation(deUint8 inputValue)
7122 ShaderPermutation permutation =
7124 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
7125 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
7126 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
7127 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
7128 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
7133 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
7135 RGBA defaultColors[4];
7136 RGBA invertedColors[4];
7137 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
7139 const ShaderElement combinedPipeline[] =
7141 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
7142 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
7143 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7144 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7145 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
7148 getDefaultColors(defaultColors);
7149 getInvertedDefaultColors(invertedColors);
7150 addFunctionCaseWithPrograms<InstanceContext>(
7151 moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline,
7152 createInstanceContext(combinedPipeline, map<string, string>()));
7154 const char* numbers[] =
7159 for (deInt8 idx = 0; idx < 32; ++idx)
7161 ShaderPermutation permutation = getShaderPermutation(idx);
7162 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
7163 const ShaderElement pipeline[] =
7165 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
7166 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
7167 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7168 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7169 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
7172 // If there are an even number of swaps, then it should be no-op.
7173 // If there are an odd number, the color should be flipped.
7174 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
7176 addFunctionCaseWithPrograms<InstanceContext>(
7177 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
7178 createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
7182 addFunctionCaseWithPrograms<InstanceContext>(
7183 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
7184 createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
7187 return moduleTests.release();
7190 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
7192 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
7193 RGBA defaultColors[4];
7194 getDefaultColors(defaultColors);
7195 map<string, string> fragments;
7196 fragments["pre_main"] =
7197 "%c_f32_5 = OpConstant %f32 5.\n";
7199 // A loop with a single block. The Continue Target is the loop block
7200 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
7201 // -- the "continue construct" forms the entire loop.
7202 fragments["testfun"] =
7203 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7204 "%param1 = OpFunctionParameter %v4f32\n"
7206 "%entry = OpLabel\n"
7207 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7210 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7212 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7213 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
7214 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7215 "%val = OpFAdd %f32 %val1 %delta\n"
7216 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
7217 "%count__ = OpISub %i32 %count %c_i32_1\n"
7218 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7219 "OpLoopMerge %exit %loop None\n"
7220 "OpBranchConditional %again %loop %exit\n"
7223 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7224 "OpReturnValue %result\n"
7228 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
7230 // Body comprised of multiple basic blocks.
7231 const StringTemplate multiBlock(
7232 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7233 "%param1 = OpFunctionParameter %v4f32\n"
7235 "%entry = OpLabel\n"
7236 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7239 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7241 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
7242 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
7243 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
7244 // There are several possibilities for the Continue Target below. Each
7245 // will be specialized into a separate test case.
7246 "OpLoopMerge %exit ${continue_target} None\n"
7250 ";delta_next = (delta > 0) ? -1 : 1;\n"
7251 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
7252 "OpSelectionMerge %gather DontFlatten\n"
7253 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
7256 "OpBranch %gather\n"
7259 "OpBranch %gather\n"
7261 "%gather = OpLabel\n"
7262 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
7263 "%val = OpFAdd %f32 %val1 %delta\n"
7264 "%count__ = OpISub %i32 %count %c_i32_1\n"
7265 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7266 "OpBranchConditional %again %loop %exit\n"
7269 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7270 "OpReturnValue %result\n"
7274 map<string, string> continue_target;
7276 // The Continue Target is the loop block itself.
7277 continue_target["continue_target"] = "%loop";
7278 fragments["testfun"] = multiBlock.specialize(continue_target);
7279 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
7281 // The Continue Target is at the end of the loop.
7282 continue_target["continue_target"] = "%gather";
7283 fragments["testfun"] = multiBlock.specialize(continue_target);
7284 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
7286 // A loop with continue statement.
7287 fragments["testfun"] =
7288 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7289 "%param1 = OpFunctionParameter %v4f32\n"
7291 "%entry = OpLabel\n"
7292 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7295 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
7297 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7298 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
7299 "OpLoopMerge %exit %continue None\n"
7303 ";skip if %count==2\n"
7304 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
7305 "OpSelectionMerge %continue DontFlatten\n"
7306 "OpBranchConditional %eq2 %continue %body\n"
7309 "%fcount = OpConvertSToF %f32 %count\n"
7310 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7311 "OpBranch %continue\n"
7313 "%continue = OpLabel\n"
7314 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
7315 "%count__ = OpISub %i32 %count %c_i32_1\n"
7316 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7317 "OpBranchConditional %again %loop %exit\n"
7320 "%same = OpFSub %f32 %val %c_f32_8\n"
7321 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7322 "OpReturnValue %result\n"
7324 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
7326 // A loop with break.
7327 fragments["testfun"] =
7328 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7329 "%param1 = OpFunctionParameter %v4f32\n"
7331 "%entry = OpLabel\n"
7332 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7333 "%dot = OpDot %f32 %param1 %param1\n"
7334 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7335 "%zero = OpConvertFToU %u32 %div\n"
7336 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7337 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7340 ";adds 4 and 3 to %val0 (exits early)\n"
7342 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7343 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7344 "OpLoopMerge %exit %continue None\n"
7348 ";end loop if %count==%two\n"
7349 "%above2 = OpSGreaterThan %bool %count %two\n"
7350 "OpSelectionMerge %continue DontFlatten\n"
7351 "OpBranchConditional %above2 %body %exit\n"
7354 "%fcount = OpConvertSToF %f32 %count\n"
7355 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7356 "OpBranch %continue\n"
7358 "%continue = OpLabel\n"
7359 "%count__ = OpISub %i32 %count %c_i32_1\n"
7360 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7361 "OpBranchConditional %again %loop %exit\n"
7364 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
7365 "%same = OpFSub %f32 %val_post %c_f32_7\n"
7366 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7367 "OpReturnValue %result\n"
7369 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
7371 // A loop with return.
7372 fragments["testfun"] =
7373 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7374 "%param1 = OpFunctionParameter %v4f32\n"
7376 "%entry = OpLabel\n"
7377 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7378 "%dot = OpDot %f32 %param1 %param1\n"
7379 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7380 "%zero = OpConvertFToU %u32 %div\n"
7381 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7382 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7385 ";returns early without modifying %param1\n"
7387 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7388 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7389 "OpLoopMerge %exit %continue None\n"
7393 ";return if %count==%two\n"
7394 "%above2 = OpSGreaterThan %bool %count %two\n"
7395 "OpSelectionMerge %continue DontFlatten\n"
7396 "OpBranchConditional %above2 %body %early_exit\n"
7398 "%early_exit = OpLabel\n"
7399 "OpReturnValue %param1\n"
7402 "%fcount = OpConvertSToF %f32 %count\n"
7403 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7404 "OpBranch %continue\n"
7406 "%continue = OpLabel\n"
7407 "%count__ = OpISub %i32 %count %c_i32_1\n"
7408 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7409 "OpBranchConditional %again %loop %exit\n"
7412 ";should never get here, so return an incorrect result\n"
7413 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
7414 "OpReturnValue %result\n"
7416 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
7418 // Continue inside a switch block to break to enclosing loop's merge block.
7419 // Matches roughly the following GLSL code:
7420 // for (; keep_going; keep_going = false)
7422 // switch (int(param1.x))
7424 // case 0: continue;
7425 // case 1: continue;
7426 // default: continue;
7428 // dead code: modify return value to invalid result.
7430 fragments["pre_main"] =
7431 "%fp_bool = OpTypePointer Function %bool\n"
7432 "%true = OpConstantTrue %bool\n"
7433 "%false = OpConstantFalse %bool\n";
7435 fragments["testfun"] =
7436 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7437 "%param1 = OpFunctionParameter %v4f32\n"
7439 "%entry = OpLabel\n"
7440 "%keep_going = OpVariable %fp_bool Function\n"
7441 "%val_ptr = OpVariable %fp_f32 Function\n"
7442 "%param1_x = OpCompositeExtract %f32 %param1 0\n"
7443 "OpStore %keep_going %true\n"
7444 "OpBranch %forloop_begin\n"
7446 "%forloop_begin = OpLabel\n"
7447 "OpLoopMerge %forloop_merge %forloop_continue None\n"
7448 "OpBranch %forloop\n"
7450 "%forloop = OpLabel\n"
7451 "%for_condition = OpLoad %bool %keep_going\n"
7452 "OpBranchConditional %for_condition %forloop_body %forloop_merge\n"
7454 "%forloop_body = OpLabel\n"
7455 "OpStore %val_ptr %param1_x\n"
7456 "%param1_x_int = OpConvertFToS %i32 %param1_x\n"
7458 "OpSelectionMerge %switch_merge None\n"
7459 "OpSwitch %param1_x_int %default 0 %case_0 1 %case_1\n"
7460 "%case_0 = OpLabel\n"
7461 "OpBranch %forloop_continue\n"
7462 "%case_1 = OpLabel\n"
7463 "OpBranch %forloop_continue\n"
7464 "%default = OpLabel\n"
7465 "OpBranch %forloop_continue\n"
7466 "%switch_merge = OpLabel\n"
7467 ";should never get here, so change the return value to invalid result\n"
7468 "OpStore %val_ptr %c_f32_1\n"
7469 "OpBranch %forloop_continue\n"
7471 "%forloop_continue = OpLabel\n"
7472 "OpStore %keep_going %false\n"
7473 "OpBranch %forloop_begin\n"
7474 "%forloop_merge = OpLabel\n"
7476 "%val = OpLoad %f32 %val_ptr\n"
7477 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7478 "OpReturnValue %result\n"
7480 createTestsForAllStages("switch_continue", defaultColors, defaultColors, fragments, testGroup.get());
7482 return testGroup.release();
7485 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
7486 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
7488 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
7489 map<string, string> fragments;
7491 // A barrier inside a function body.
7492 fragments["pre_main"] =
7493 "%Workgroup = OpConstant %i32 2\n"
7494 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
7495 fragments["testfun"] =
7496 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7497 "%param1 = OpFunctionParameter %v4f32\n"
7498 "%label_testfun = OpLabel\n"
7499 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7500 "OpReturnValue %param1\n"
7502 addTessCtrlTest(testGroup.get(), "in_function", fragments);
7504 // Common setup code for the following tests.
7505 fragments["pre_main"] =
7506 "%Workgroup = OpConstant %i32 2\n"
7507 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7508 "%c_f32_5 = OpConstant %f32 5.\n";
7509 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
7510 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7511 "%param1 = OpFunctionParameter %v4f32\n"
7512 "%entry = OpLabel\n"
7513 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7514 "%dot = OpDot %f32 %param1 %param1\n"
7515 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7516 "%zero = OpConvertFToU %u32 %div\n";
7518 // Barriers inside OpSwitch branches.
7519 fragments["testfun"] =
7521 "OpSelectionMerge %switch_exit None\n"
7522 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
7524 "%case1 = OpLabel\n"
7525 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7526 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7527 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7528 "OpBranch %switch_exit\n"
7530 "%switch_default = OpLabel\n"
7531 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7532 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7533 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7534 "OpBranch %switch_exit\n"
7536 "%case0 = OpLabel\n"
7537 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7538 "OpBranch %switch_exit\n"
7540 "%switch_exit = OpLabel\n"
7541 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
7542 "OpReturnValue %ret\n"
7544 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
7546 // Barriers inside if-then-else.
7547 fragments["testfun"] =
7549 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
7550 "OpSelectionMerge %exit DontFlatten\n"
7551 "OpBranchConditional %eq0 %then %else\n"
7554 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7555 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7556 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7560 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7564 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
7565 "OpReturnValue %ret\n"
7567 addTessCtrlTest(testGroup.get(), "in_if", fragments);
7569 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
7570 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
7571 fragments["testfun"] =
7573 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
7574 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
7575 "OpSelectionMerge %exit DontFlatten\n"
7576 "OpBranchConditional %thread0 %then %else\n"
7579 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7583 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
7587 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
7588 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7589 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
7590 "OpReturnValue %ret\n"
7592 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
7594 // A barrier inside a loop.
7595 fragments["pre_main"] =
7596 "%Workgroup = OpConstant %i32 2\n"
7597 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7598 "%c_f32_10 = OpConstant %f32 10.\n";
7599 fragments["testfun"] =
7600 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7601 "%param1 = OpFunctionParameter %v4f32\n"
7602 "%entry = OpLabel\n"
7603 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7606 ";adds 4, 3, 2, and 1 to %val0\n"
7608 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7609 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7610 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7611 "%fcount = OpConvertSToF %f32 %count\n"
7612 "%val = OpFAdd %f32 %val1 %fcount\n"
7613 "%count__ = OpISub %i32 %count %c_i32_1\n"
7614 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7615 "OpLoopMerge %exit %loop None\n"
7616 "OpBranchConditional %again %loop %exit\n"
7619 "%same = OpFSub %f32 %val %c_f32_10\n"
7620 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7621 "OpReturnValue %ret\n"
7623 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
7625 return testGroup.release();
7628 // Test for the OpFRem instruction.
7629 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
7631 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
7632 map<string, string> fragments;
7633 RGBA inputColors[4];
7634 RGBA outputColors[4];
7636 fragments["pre_main"] =
7637 "%c_f32_3 = OpConstant %f32 3.0\n"
7638 "%c_f32_n3 = OpConstant %f32 -3.0\n"
7639 "%c_f32_4 = OpConstant %f32 4.0\n"
7640 "%c_f32_p75 = OpConstant %f32 0.75\n"
7641 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
7642 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
7643 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
7645 // The test does the following.
7646 // vec4 result = (param1 * 8.0) - 4.0;
7647 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
7648 fragments["testfun"] =
7649 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7650 "%param1 = OpFunctionParameter %v4f32\n"
7651 "%label_testfun = OpLabel\n"
7652 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
7653 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
7654 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
7655 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
7656 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
7657 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
7658 "OpReturnValue %xy_0_1\n"
7662 inputColors[0] = RGBA(16, 16, 0, 255);
7663 inputColors[1] = RGBA(232, 232, 0, 255);
7664 inputColors[2] = RGBA(232, 16, 0, 255);
7665 inputColors[3] = RGBA(16, 232, 0, 255);
7667 outputColors[0] = RGBA(64, 64, 0, 255);
7668 outputColors[1] = RGBA(255, 255, 0, 255);
7669 outputColors[2] = RGBA(255, 64, 0, 255);
7670 outputColors[3] = RGBA(64, 255, 0, 255);
7672 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
7673 return testGroup.release();
7676 // Test for the OpSRem instruction.
7677 tcu::TestCaseGroup* createOpSRemGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
7679 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "srem", "OpSRem"));
7680 map<string, string> fragments;
7682 fragments["pre_main"] =
7683 "%c_f32_255 = OpConstant %f32 255.0\n"
7684 "%c_i32_128 = OpConstant %i32 128\n"
7685 "%c_i32_255 = OpConstant %i32 255\n"
7686 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
7687 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
7688 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
7690 // The test does the following.
7691 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
7692 // ivec4 result = ivec4(srem(ints.x, ints.y), srem(ints.y, ints.z), srem(ints.z, ints.x), 255);
7693 // return float(result + 128) / 255.0;
7694 fragments["testfun"] =
7695 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7696 "%param1 = OpFunctionParameter %v4f32\n"
7697 "%label_testfun = OpLabel\n"
7698 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
7699 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
7700 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
7701 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
7702 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
7703 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
7704 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
7705 "%x_out = OpSRem %i32 %x_in %y_in\n"
7706 "%y_out = OpSRem %i32 %y_in %z_in\n"
7707 "%z_out = OpSRem %i32 %z_in %x_in\n"
7708 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
7709 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
7710 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
7711 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
7712 "OpReturnValue %float_out\n"
7715 const struct CaseParams
7718 const char* failMessageTemplate; // customized status message
7719 qpTestResult failResult; // override status on failure
7720 int operands[4][3]; // four (x, y, z) vectors of operands
7721 int results[4][3]; // four (x, y, z) vectors of results
7727 QP_TEST_RESULT_FAIL,
7728 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
7729 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
7733 "Inconsistent results, but within specification: ${reason}",
7734 negFailResult, // negative operands, not required by the spec
7735 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
7736 { { 5, 12, -2 }, { 0, -5, 2 }, { 3, 8, -6 }, { 25, -60, 0 } }, // results
7739 // If either operand is negative the result is undefined. Some implementations may still return correct values.
7741 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
7743 const CaseParams& params = cases[caseNdx];
7744 RGBA inputColors[4];
7745 RGBA outputColors[4];
7747 for (int i = 0; i < 4; ++i)
7749 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
7750 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
7753 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
7756 return testGroup.release();
7759 // Test for the OpSMod instruction.
7760 tcu::TestCaseGroup* createOpSModGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
7762 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "smod", "OpSMod"));
7763 map<string, string> fragments;
7765 fragments["pre_main"] =
7766 "%c_f32_255 = OpConstant %f32 255.0\n"
7767 "%c_i32_128 = OpConstant %i32 128\n"
7768 "%c_i32_255 = OpConstant %i32 255\n"
7769 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
7770 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
7771 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
7773 // The test does the following.
7774 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
7775 // ivec4 result = ivec4(smod(ints.x, ints.y), smod(ints.y, ints.z), smod(ints.z, ints.x), 255);
7776 // return float(result + 128) / 255.0;
7777 fragments["testfun"] =
7778 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7779 "%param1 = OpFunctionParameter %v4f32\n"
7780 "%label_testfun = OpLabel\n"
7781 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
7782 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
7783 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
7784 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
7785 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
7786 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
7787 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
7788 "%x_out = OpSMod %i32 %x_in %y_in\n"
7789 "%y_out = OpSMod %i32 %y_in %z_in\n"
7790 "%z_out = OpSMod %i32 %z_in %x_in\n"
7791 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
7792 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
7793 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
7794 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
7795 "OpReturnValue %float_out\n"
7798 const struct CaseParams
7801 const char* failMessageTemplate; // customized status message
7802 qpTestResult failResult; // override status on failure
7803 int operands[4][3]; // four (x, y, z) vectors of operands
7804 int results[4][3]; // four (x, y, z) vectors of results
7810 QP_TEST_RESULT_FAIL,
7811 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
7812 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
7816 "Inconsistent results, but within specification: ${reason}",
7817 negFailResult, // negative operands, not required by the spec
7818 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
7819 { { 5, -5, 3 }, { 0, 2, -3 }, { 3, -73, 69 }, { -35, 40, 0 } }, // results
7822 // If either operand is negative the result is undefined. Some implementations may still return correct values.
7824 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
7826 const CaseParams& params = cases[caseNdx];
7827 RGBA inputColors[4];
7828 RGBA outputColors[4];
7830 for (int i = 0; i < 4; ++i)
7832 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
7833 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
7836 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
7838 return testGroup.release();
7844 INTEGER_TYPE_SIGNED_16,
7845 INTEGER_TYPE_SIGNED_32,
7846 INTEGER_TYPE_SIGNED_64,
7848 INTEGER_TYPE_UNSIGNED_16,
7849 INTEGER_TYPE_UNSIGNED_32,
7850 INTEGER_TYPE_UNSIGNED_64,
7853 const string getBitWidthStr (IntegerType type)
7857 case INTEGER_TYPE_SIGNED_16:
7858 case INTEGER_TYPE_UNSIGNED_16: return "16";
7860 case INTEGER_TYPE_SIGNED_32:
7861 case INTEGER_TYPE_UNSIGNED_32: return "32";
7863 case INTEGER_TYPE_SIGNED_64:
7864 case INTEGER_TYPE_UNSIGNED_64: return "64";
7866 default: DE_ASSERT(false);
7871 const string getByteWidthStr (IntegerType type)
7875 case INTEGER_TYPE_SIGNED_16:
7876 case INTEGER_TYPE_UNSIGNED_16: return "2";
7878 case INTEGER_TYPE_SIGNED_32:
7879 case INTEGER_TYPE_UNSIGNED_32: return "4";
7881 case INTEGER_TYPE_SIGNED_64:
7882 case INTEGER_TYPE_UNSIGNED_64: return "8";
7884 default: DE_ASSERT(false);
7889 bool isSigned (IntegerType type)
7891 return (type <= INTEGER_TYPE_SIGNED_64);
7894 const string getTypeName (IntegerType type)
7896 string prefix = isSigned(type) ? "" : "u";
7897 return prefix + "int" + getBitWidthStr(type);
7900 const string getTestName (IntegerType from, IntegerType to)
7902 return getTypeName(from) + "_to_" + getTypeName(to);
7905 const string getAsmTypeDeclaration (IntegerType type)
7907 string sign = isSigned(type) ? " 1" : " 0";
7908 return "OpTypeInt " + getBitWidthStr(type) + sign;
7911 const string getAsmTypeName (IntegerType type)
7913 const string prefix = isSigned(type) ? "%i" : "%u";
7914 return prefix + getBitWidthStr(type);
7917 template<typename T>
7918 BufferSp getSpecializedBuffer (deInt64 number)
7920 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
7923 BufferSp getBuffer (IntegerType type, deInt64 number)
7927 case INTEGER_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
7928 case INTEGER_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
7929 case INTEGER_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
7931 case INTEGER_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
7932 case INTEGER_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
7933 case INTEGER_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
7935 default: DE_ASSERT(false);
7936 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
7940 bool usesInt16 (IntegerType from, IntegerType to)
7942 return (from == INTEGER_TYPE_SIGNED_16 || from == INTEGER_TYPE_UNSIGNED_16
7943 || to == INTEGER_TYPE_SIGNED_16 || to == INTEGER_TYPE_UNSIGNED_16);
7946 bool usesInt64 (IntegerType from, IntegerType to)
7948 return (from == INTEGER_TYPE_SIGNED_64 || from == INTEGER_TYPE_UNSIGNED_64
7949 || to == INTEGER_TYPE_SIGNED_64 || to == INTEGER_TYPE_UNSIGNED_64);
7952 ComputeTestFeatures getConversionUsedFeatures (IntegerType from, IntegerType to)
7954 if (usesInt16(from, to))
7956 if (usesInt64(from, to))
7958 return COMPUTE_TEST_USES_INT16_INT64;
7962 return COMPUTE_TEST_USES_INT16;
7967 return COMPUTE_TEST_USES_INT64;
7973 ConvertCase (IntegerType from, IntegerType to, deInt64 number)
7976 , m_features (getConversionUsedFeatures(from, to))
7977 , m_name (getTestName(from, to))
7978 , m_inputBuffer (getBuffer(from, number))
7979 , m_outputBuffer (getBuffer(to, number))
7981 m_asmTypes["inputType"] = getAsmTypeName(from);
7982 m_asmTypes["outputType"] = getAsmTypeName(to);
7984 if (m_features == COMPUTE_TEST_USES_INT16)
7986 m_asmTypes["int_capabilities"] = "OpCapability Int16\n"
7987 "OpCapability StorageUniformBufferBlock16\n";
7988 m_asmTypes["int_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
7989 "%u16 = OpTypeInt 16 0\n";
7990 m_asmTypes["int_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
7992 else if (m_features == COMPUTE_TEST_USES_INT64)
7994 m_asmTypes["int_capabilities"] = "OpCapability Int64\n";
7995 m_asmTypes["int_additional_decl"] = "%i64 = OpTypeInt 64 1\n"
7996 "%u64 = OpTypeInt 64 0\n";
7997 m_asmTypes["int_extensions"] = "";
7999 else if (m_features == COMPUTE_TEST_USES_INT16_INT64)
8001 m_asmTypes["int_capabilities"] = "OpCapability Int16\n"
8002 "OpCapability StorageUniformBufferBlock16\n"
8003 "OpCapability Int64\n";
8004 m_asmTypes["int_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
8005 "%u16 = OpTypeInt 16 0\n"
8006 "%i64 = OpTypeInt 64 1\n"
8007 "%u64 = OpTypeInt 64 0\n";
8008 m_asmTypes["int_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
8016 IntegerType m_fromType;
8017 IntegerType m_toType;
8018 ComputeTestFeatures m_features;
8020 map<string, string> m_asmTypes;
8021 BufferSp m_inputBuffer;
8022 BufferSp m_outputBuffer;
8025 const string getConvertCaseShaderStr (const string& instruction, const ConvertCase& convertCase)
8027 map<string, string> params = convertCase.m_asmTypes;
8029 params["instruction"] = instruction;
8031 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
8032 params["outDecorator"] = getByteWidthStr(convertCase.m_toType);
8034 const StringTemplate shader (
8035 "OpCapability Shader\n"
8036 "${int_capabilities}"
8038 "OpMemoryModel Logical GLSL450\n"
8039 "OpEntryPoint GLCompute %main \"main\" %id\n"
8040 "OpExecutionMode %main LocalSize 1 1 1\n"
8041 "OpSource GLSL 430\n"
8042 "OpName %main \"main\"\n"
8043 "OpName %id \"gl_GlobalInvocationID\"\n"
8045 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8046 "OpDecorate %indata DescriptorSet 0\n"
8047 "OpDecorate %indata Binding 0\n"
8048 "OpDecorate %outdata DescriptorSet 0\n"
8049 "OpDecorate %outdata Binding 1\n"
8050 "OpDecorate %in_arr ArrayStride ${inDecorator}\n"
8051 "OpDecorate %out_arr ArrayStride ${outDecorator}\n"
8052 "OpDecorate %in_buf BufferBlock\n"
8053 "OpDecorate %out_buf BufferBlock\n"
8054 "OpMemberDecorate %in_buf 0 Offset 0\n"
8055 "OpMemberDecorate %out_buf 0 Offset 0\n"
8057 "%void = OpTypeVoid\n"
8058 "%voidf = OpTypeFunction %void\n"
8059 "%u32 = OpTypeInt 32 0\n"
8060 "%i32 = OpTypeInt 32 1\n"
8061 "${int_additional_decl}"
8062 "%uvec3 = OpTypeVector %u32 3\n"
8063 "%uvec3ptr = OpTypePointer Input %uvec3\n"
8065 "%in_ptr = OpTypePointer Uniform ${inputType}\n"
8066 "%out_ptr = OpTypePointer Uniform ${outputType}\n"
8067 "%in_arr = OpTypeRuntimeArray ${inputType}\n"
8068 "%out_arr = OpTypeRuntimeArray ${outputType}\n"
8069 "%in_buf = OpTypeStruct %in_arr\n"
8070 "%out_buf = OpTypeStruct %out_arr\n"
8071 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
8072 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
8073 "%indata = OpVariable %in_bufptr Uniform\n"
8074 "%outdata = OpVariable %out_bufptr Uniform\n"
8075 "%inputptr = OpTypePointer Input ${inputType}\n"
8076 "%id = OpVariable %uvec3ptr Input\n"
8078 "%zero = OpConstant %i32 0\n"
8080 "%main = OpFunction %void None %voidf\n"
8081 "%label = OpLabel\n"
8082 "%idval = OpLoad %uvec3 %id\n"
8083 "%x = OpCompositeExtract %u32 %idval 0\n"
8084 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
8085 "%outloc = OpAccessChain %out_ptr %outdata %zero %x\n"
8086 "%inval = OpLoad ${inputType} %inloc\n"
8087 "%conv = ${instruction} ${outputType} %inval\n"
8088 " OpStore %outloc %conv\n"
8093 return shader.specialize(params);
8096 void createSConvertCases (vector<ConvertCase>& testCases)
8098 // Convert int to int
8099 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_32, 14669));
8100 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_64, 3341));
8102 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_SIGNED_64, 973610259));
8104 // Convert int to unsigned int
8105 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_32, 9288));
8106 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_64, 15460));
8108 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_UNSIGNED_64, 346213461));
8111 // Test for the OpSConvert instruction.
8112 tcu::TestCaseGroup* createSConvertTests (tcu::TestContext& testCtx)
8114 const string instruction ("OpSConvert");
8115 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "sconvert", "OpSConvert"));
8116 vector<ConvertCase> testCases;
8117 createSConvertCases(testCases);
8119 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8121 ComputeShaderSpec spec;
8123 spec.assembly = getConvertCaseShaderStr(instruction, *test);
8124 spec.inputs.push_back(test->m_inputBuffer);
8125 spec.outputs.push_back(test->m_outputBuffer);
8126 spec.numWorkGroups = IVec3(1, 1, 1);
8128 if (test->m_features == COMPUTE_TEST_USES_INT16 || test->m_features == COMPUTE_TEST_USES_INT16_INT64)
8130 spec.extensions.push_back("VK_KHR_16bit_storage");
8131 spec.requestedVulkanFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_UNIFORM_BUFFER_BLOCK;
8134 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpSConvert.", spec, test->m_features));
8137 return group.release();
8140 void createUConvertCases (vector<ConvertCase>& testCases)
8142 // Convert unsigned int to unsigned int
8143 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_32, 60653));
8144 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_64, 17991));
8146 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_UNSIGNED_64, 904256275));
8149 // Test for the OpUConvert instruction.
8150 tcu::TestCaseGroup* createUConvertTests (tcu::TestContext& testCtx)
8152 const string instruction ("OpUConvert");
8153 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "uconvert", "OpUConvert"));
8154 vector<ConvertCase> testCases;
8155 createUConvertCases(testCases);
8157 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8159 ComputeShaderSpec spec;
8161 spec.assembly = getConvertCaseShaderStr(instruction, *test);
8162 spec.inputs.push_back(test->m_inputBuffer);
8163 spec.outputs.push_back(test->m_outputBuffer);
8164 spec.numWorkGroups = IVec3(1, 1, 1);
8166 if (test->m_features == COMPUTE_TEST_USES_INT16 || test->m_features == COMPUTE_TEST_USES_INT16_INT64)
8168 spec.extensions.push_back("VK_KHR_16bit_storage");
8169 spec.requestedVulkanFeatures.ext16BitStorage = EXT16BITSTORAGEFEATURES_UNIFORM_BUFFER_BLOCK;
8172 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpUConvert.", spec, test->m_features));
8174 return group.release();
8177 const string getNumberTypeName (const NumberType type)
8179 if (type == NUMBERTYPE_INT32)
8183 else if (type == NUMBERTYPE_UINT32)
8187 else if (type == NUMBERTYPE_FLOAT32)
8198 deInt32 getInt(de::Random& rnd)
8200 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
8203 const string repeatString (const string& str, int times)
8206 for (int i = 0; i < times; ++i)
8213 const string getRandomConstantString (const NumberType type, de::Random& rnd)
8215 if (type == NUMBERTYPE_INT32)
8217 return numberToString<deInt32>(getInt(rnd));
8219 else if (type == NUMBERTYPE_UINT32)
8221 return numberToString<deUint32>(rnd.getUint32());
8223 else if (type == NUMBERTYPE_FLOAT32)
8225 return numberToString<float>(rnd.getFloat());
8234 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8236 map<string, string> params;
8239 for (int width = 2; width <= 4; ++width)
8241 const string randomConst = numberToString(getInt(rnd));
8242 const string widthStr = numberToString(width);
8243 const string composite_type = "${customType}vec" + widthStr;
8244 const int index = rnd.getInt(0, width-1);
8246 params["type"] = "vec";
8247 params["name"] = params["type"] + "_" + widthStr;
8248 params["compositeDecl"] = composite_type + " = OpTypeVector ${customType} " + widthStr +"\n";
8249 params["compositeType"] = composite_type;
8250 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8251 params["compositeConstruct"] = "%instance = OpCompositeConstruct " + composite_type + repeatString(" %filler", width) + "\n";
8252 params["indexes"] = numberToString(index);
8253 testCases.push_back(params);
8257 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8259 const int limit = 10;
8260 map<string, string> params;
8262 for (int width = 2; width <= limit; ++width)
8264 string randomConst = numberToString(getInt(rnd));
8265 string widthStr = numberToString(width);
8266 int index = rnd.getInt(0, width-1);
8268 params["type"] = "array";
8269 params["name"] = params["type"] + "_" + widthStr;
8270 params["compositeDecl"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
8271 + "%composite = OpTypeArray ${customType} %arraywidth\n";
8272 params["compositeType"] = "%composite";
8273 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8274 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8275 params["indexes"] = numberToString(index);
8276 testCases.push_back(params);
8280 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8282 const int limit = 10;
8283 map<string, string> params;
8285 for (int width = 2; width <= limit; ++width)
8287 string randomConst = numberToString(getInt(rnd));
8288 int index = rnd.getInt(0, width-1);
8290 params["type"] = "struct";
8291 params["name"] = params["type"] + "_" + numberToString(width);
8292 params["compositeDecl"] = "%composite = OpTypeStruct" + repeatString(" ${customType}", width) + "\n";
8293 params["compositeType"] = "%composite";
8294 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8295 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8296 params["indexes"] = numberToString(index);
8297 testCases.push_back(params);
8301 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8303 map<string, string> params;
8306 for (int width = 2; width <= 4; ++width)
8308 string widthStr = numberToString(width);
8310 for (int column = 2 ; column <= 4; ++column)
8312 int index_0 = rnd.getInt(0, column-1);
8313 int index_1 = rnd.getInt(0, width-1);
8314 string columnStr = numberToString(column);
8316 params["type"] = "matrix";
8317 params["name"] = params["type"] + "_" + widthStr + "x" + columnStr;
8318 params["compositeDecl"] = string("%vectype = OpTypeVector ${customType} " + widthStr + "\n")
8319 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
8320 params["compositeType"] = "%composite";
8322 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n"
8323 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
8325 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
8326 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
8327 testCases.push_back(params);
8332 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8334 createVectorCompositeCases(testCases, rnd, type);
8335 createArrayCompositeCases(testCases, rnd, type);
8336 createStructCompositeCases(testCases, rnd, type);
8337 // Matrix only supports float types
8338 if (type == NUMBERTYPE_FLOAT32)
8340 createMatrixCompositeCases(testCases, rnd, type);
8344 const string getAssemblyTypeDeclaration (const NumberType type)
8348 case NUMBERTYPE_INT32: return "OpTypeInt 32 1";
8349 case NUMBERTYPE_UINT32: return "OpTypeInt 32 0";
8350 case NUMBERTYPE_FLOAT32: return "OpTypeFloat 32";
8351 default: DE_ASSERT(false); return "";
8355 const string getAssemblyTypeName (const NumberType type)
8359 case NUMBERTYPE_INT32: return "%i32";
8360 case NUMBERTYPE_UINT32: return "%u32";
8361 case NUMBERTYPE_FLOAT32: return "%f32";
8362 default: DE_ASSERT(false); return "";
8366 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
8368 map<string, string> parameters(params);
8370 const string customType = getAssemblyTypeName(type);
8371 map<string, string> substCustomType;
8372 substCustomType["customType"] = customType;
8373 parameters["compositeDecl"] = StringTemplate(parameters.at("compositeDecl")).specialize(substCustomType);
8374 parameters["compositeType"] = StringTemplate(parameters.at("compositeType")).specialize(substCustomType);
8375 parameters["compositeConstruct"] = StringTemplate(parameters.at("compositeConstruct")).specialize(substCustomType);
8376 parameters["filler"] = StringTemplate(parameters.at("filler")).specialize(substCustomType);
8377 parameters["customType"] = customType;
8378 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
8380 if (parameters.at("compositeType") != "%u32vec3")
8382 parameters["u32vec3Decl"] = "%u32vec3 = OpTypeVector %u32 3\n";
8385 return StringTemplate(
8386 "OpCapability Shader\n"
8387 "OpCapability Matrix\n"
8388 "OpMemoryModel Logical GLSL450\n"
8389 "OpEntryPoint GLCompute %main \"main\" %id\n"
8390 "OpExecutionMode %main LocalSize 1 1 1\n"
8392 "OpSource GLSL 430\n"
8393 "OpName %main \"main\"\n"
8394 "OpName %id \"gl_GlobalInvocationID\"\n"
8397 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8398 "OpDecorate %buf BufferBlock\n"
8399 "OpDecorate %indata DescriptorSet 0\n"
8400 "OpDecorate %indata Binding 0\n"
8401 "OpDecorate %outdata DescriptorSet 0\n"
8402 "OpDecorate %outdata Binding 1\n"
8403 "OpDecorate %customarr ArrayStride 4\n"
8404 "${compositeDecorator}"
8405 "OpMemberDecorate %buf 0 Offset 0\n"
8408 "%void = OpTypeVoid\n"
8409 "%voidf = OpTypeFunction %void\n"
8410 "%u32 = OpTypeInt 32 0\n"
8411 "%i32 = OpTypeInt 32 1\n"
8412 "%f32 = OpTypeFloat 32\n"
8414 // Composite declaration
8420 "${u32vec3Decl:opt}"
8421 "%uvec3ptr = OpTypePointer Input %u32vec3\n"
8423 // Inherited from custom
8424 "%customptr = OpTypePointer Uniform ${customType}\n"
8425 "%customarr = OpTypeRuntimeArray ${customType}\n"
8426 "%buf = OpTypeStruct %customarr\n"
8427 "%bufptr = OpTypePointer Uniform %buf\n"
8429 "%indata = OpVariable %bufptr Uniform\n"
8430 "%outdata = OpVariable %bufptr Uniform\n"
8432 "%id = OpVariable %uvec3ptr Input\n"
8433 "%zero = OpConstant %i32 0\n"
8435 "%main = OpFunction %void None %voidf\n"
8436 "%label = OpLabel\n"
8437 "%idval = OpLoad %u32vec3 %id\n"
8438 "%x = OpCompositeExtract %u32 %idval 0\n"
8440 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
8441 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
8442 // Read the input value
8443 "%inval = OpLoad ${customType} %inloc\n"
8444 // Create the composite and fill it
8445 "${compositeConstruct}"
8446 // Insert the input value to a place
8447 "%instance2 = OpCompositeInsert ${compositeType} %inval %instance ${indexes}\n"
8448 // Read back the value from the position
8449 "%out_val = OpCompositeExtract ${customType} %instance2 ${indexes}\n"
8450 // Store it in the output position
8451 " OpStore %outloc %out_val\n"
8454 ).specialize(parameters);
8457 template<typename T>
8458 BufferSp createCompositeBuffer(T number)
8460 return BufferSp(new Buffer<T>(vector<T>(1, number)));
8463 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
8465 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
8466 de::Random rnd (deStringHash(group->getName()));
8468 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8470 NumberType numberType = NumberType(type);
8471 const string typeName = getNumberTypeName(numberType);
8472 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
8473 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8474 vector<map<string, string> > testCases;
8476 createCompositeCases(testCases, rnd, numberType);
8478 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8480 ComputeShaderSpec spec;
8482 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
8486 case NUMBERTYPE_INT32:
8488 deInt32 number = getInt(rnd);
8489 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8490 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8493 case NUMBERTYPE_UINT32:
8495 deUint32 number = rnd.getUint32();
8496 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8497 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8500 case NUMBERTYPE_FLOAT32:
8502 float number = rnd.getFloat();
8503 spec.inputs.push_back(createCompositeBuffer<float>(number));
8504 spec.outputs.push_back(createCompositeBuffer<float>(number));
8511 spec.numWorkGroups = IVec3(1, 1, 1);
8512 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
8514 group->addChild(subGroup.release());
8516 return group.release();
8519 struct AssemblyStructInfo
8521 AssemblyStructInfo (const deUint32 comp, const deUint32 idx)
8526 deUint32 components;
8530 const string specializeInBoundsShaderTemplate (const NumberType type, const AssemblyStructInfo& structInfo, const map<string, string>& params)
8532 // Create the full index string
8533 string fullIndex = numberToString(structInfo.index) + " " + params.at("indexes");
8534 // Convert it to list of indexes
8535 vector<string> indexes = de::splitString(fullIndex, ' ');
8537 map<string, string> parameters (params);
8538 parameters["structType"] = repeatString(" ${compositeType}", structInfo.components);
8539 parameters["structConstruct"] = repeatString(" %instance", structInfo.components);
8540 parameters["insertIndexes"] = fullIndex;
8542 // In matrix cases the last two index is the CompositeExtract indexes
8543 const deUint32 extractIndexes = (parameters["type"] == "matrix") ? 2 : 1;
8545 // Construct the extractIndex
8546 for (vector<string>::const_iterator index = indexes.end() - extractIndexes; index != indexes.end(); ++index)
8548 parameters["extractIndexes"] += " " + *index;
8551 // Remove the last 1 or 2 element depends on matrix case or not
8552 indexes.erase(indexes.end() - extractIndexes, indexes.end());
8555 // Generate AccessChain index expressions (except for the last one, because we use ptr to the composite)
8556 for (vector<string>::const_iterator index = indexes.begin(); index != indexes.end(); ++index)
8558 string indexId = "%index_" + numberToString(id++);
8559 parameters["accessChainConstDeclaration"] += indexId + " = OpConstant %u32 " + *index + "\n";
8560 parameters["accessChainIndexes"] += " " + indexId;
8563 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
8565 const string customType = getAssemblyTypeName(type);
8566 map<string, string> substCustomType;
8567 substCustomType["customType"] = customType;
8568 parameters["compositeDecl"] = StringTemplate(parameters.at("compositeDecl")).specialize(substCustomType);
8569 parameters["compositeType"] = StringTemplate(parameters.at("compositeType")).specialize(substCustomType);
8570 parameters["compositeConstruct"] = StringTemplate(parameters.at("compositeConstruct")).specialize(substCustomType);
8571 parameters["filler"] = StringTemplate(parameters.at("filler")).specialize(substCustomType);
8572 parameters["customType"] = customType;
8574 const string compositeType = parameters.at("compositeType");
8575 map<string, string> substCompositeType;
8576 substCompositeType["compositeType"] = compositeType;
8577 parameters["structType"] = StringTemplate(parameters.at("structType")).specialize(substCompositeType);
8578 if (compositeType != "%u32vec3")
8580 parameters["u32vec3Decl"] = "%u32vec3 = OpTypeVector %u32 3\n";
8583 return StringTemplate(
8584 "OpCapability Shader\n"
8585 "OpCapability Matrix\n"
8586 "OpMemoryModel Logical GLSL450\n"
8587 "OpEntryPoint GLCompute %main \"main\" %id\n"
8588 "OpExecutionMode %main LocalSize 1 1 1\n"
8590 "OpSource GLSL 430\n"
8591 "OpName %main \"main\"\n"
8592 "OpName %id \"gl_GlobalInvocationID\"\n"
8594 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8595 "OpDecorate %buf BufferBlock\n"
8596 "OpDecorate %indata DescriptorSet 0\n"
8597 "OpDecorate %indata Binding 0\n"
8598 "OpDecorate %outdata DescriptorSet 0\n"
8599 "OpDecorate %outdata Binding 1\n"
8600 "OpDecorate %customarr ArrayStride 4\n"
8601 "${compositeDecorator}"
8602 "OpMemberDecorate %buf 0 Offset 0\n"
8604 "%void = OpTypeVoid\n"
8605 "%voidf = OpTypeFunction %void\n"
8606 "%i32 = OpTypeInt 32 1\n"
8607 "%u32 = OpTypeInt 32 0\n"
8608 "%f32 = OpTypeFloat 32\n"
8611 // %u32vec3 if not already declared in ${compositeDecl}
8612 "${u32vec3Decl:opt}"
8613 "%uvec3ptr = OpTypePointer Input %u32vec3\n"
8614 // Inherited from composite
8615 "%composite_p = OpTypePointer Function ${compositeType}\n"
8616 "%struct_t = OpTypeStruct${structType}\n"
8617 "%struct_p = OpTypePointer Function %struct_t\n"
8620 "${accessChainConstDeclaration}"
8621 // Inherited from custom
8622 "%customptr = OpTypePointer Uniform ${customType}\n"
8623 "%customarr = OpTypeRuntimeArray ${customType}\n"
8624 "%buf = OpTypeStruct %customarr\n"
8625 "%bufptr = OpTypePointer Uniform %buf\n"
8626 "%indata = OpVariable %bufptr Uniform\n"
8627 "%outdata = OpVariable %bufptr Uniform\n"
8629 "%id = OpVariable %uvec3ptr Input\n"
8630 "%zero = OpConstant %u32 0\n"
8631 "%main = OpFunction %void None %voidf\n"
8632 "%label = OpLabel\n"
8633 "%struct_v = OpVariable %struct_p Function\n"
8634 "%idval = OpLoad %u32vec3 %id\n"
8635 "%x = OpCompositeExtract %u32 %idval 0\n"
8636 // Create the input/output type
8637 "%inloc = OpInBoundsAccessChain %customptr %indata %zero %x\n"
8638 "%outloc = OpInBoundsAccessChain %customptr %outdata %zero %x\n"
8639 // Read the input value
8640 "%inval = OpLoad ${customType} %inloc\n"
8641 // Create the composite and fill it
8642 "${compositeConstruct}"
8643 // Create the struct and fill it with the composite
8644 "%struct = OpCompositeConstruct %struct_t${structConstruct}\n"
8646 "%comp_obj = OpCompositeInsert %struct_t %inval %struct ${insertIndexes}\n"
8648 " OpStore %struct_v %comp_obj\n"
8649 // Get deepest possible composite pointer
8650 "%inner_ptr = OpInBoundsAccessChain %composite_p %struct_v${accessChainIndexes}\n"
8651 "%read_obj = OpLoad ${compositeType} %inner_ptr\n"
8652 // Read back the stored value
8653 "%read_val = OpCompositeExtract ${customType} %read_obj${extractIndexes}\n"
8654 " OpStore %outloc %read_val\n"
8657 ).specialize(parameters);
8660 tcu::TestCaseGroup* createOpInBoundsAccessChainGroup (tcu::TestContext& testCtx)
8662 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opinboundsaccesschain", "Test the OpInBoundsAccessChain instruction"));
8663 de::Random rnd (deStringHash(group->getName()));
8665 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8667 NumberType numberType = NumberType(type);
8668 const string typeName = getNumberTypeName(numberType);
8669 const string description = "Test the OpInBoundsAccessChain instruction with " + typeName + "s";
8670 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8672 vector<map<string, string> > testCases;
8673 createCompositeCases(testCases, rnd, numberType);
8675 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8677 ComputeShaderSpec spec;
8679 // Number of components inside of a struct
8680 deUint32 structComponents = rnd.getInt(2, 8);
8681 // Component index value
8682 deUint32 structIndex = rnd.getInt(0, structComponents - 1);
8683 AssemblyStructInfo structInfo(structComponents, structIndex);
8685 spec.assembly = specializeInBoundsShaderTemplate(numberType, structInfo, *test);
8689 case NUMBERTYPE_INT32:
8691 deInt32 number = getInt(rnd);
8692 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8693 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8696 case NUMBERTYPE_UINT32:
8698 deUint32 number = rnd.getUint32();
8699 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8700 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8703 case NUMBERTYPE_FLOAT32:
8705 float number = rnd.getFloat();
8706 spec.inputs.push_back(createCompositeBuffer<float>(number));
8707 spec.outputs.push_back(createCompositeBuffer<float>(number));
8713 spec.numWorkGroups = IVec3(1, 1, 1);
8714 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpInBoundsAccessChain test", spec));
8716 group->addChild(subGroup.release());
8718 return group.release();
8721 // If the params missing, uninitialized case
8722 const string specializeDefaultOutputShaderTemplate (const NumberType type, const map<string, string>& params = map<string, string>())
8724 map<string, string> parameters(params);
8726 parameters["customType"] = getAssemblyTypeName(type);
8728 // Declare the const value, and use it in the initializer
8729 if (params.find("constValue") != params.end())
8731 parameters["variableInitializer"] = " %const";
8733 // Uninitialized case
8736 parameters["commentDecl"] = ";";
8739 return StringTemplate(
8740 "OpCapability Shader\n"
8741 "OpMemoryModel Logical GLSL450\n"
8742 "OpEntryPoint GLCompute %main \"main\" %id\n"
8743 "OpExecutionMode %main LocalSize 1 1 1\n"
8744 "OpSource GLSL 430\n"
8745 "OpName %main \"main\"\n"
8746 "OpName %id \"gl_GlobalInvocationID\"\n"
8748 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8749 "OpDecorate %indata DescriptorSet 0\n"
8750 "OpDecorate %indata Binding 0\n"
8751 "OpDecorate %outdata DescriptorSet 0\n"
8752 "OpDecorate %outdata Binding 1\n"
8753 "OpDecorate %in_arr ArrayStride 4\n"
8754 "OpDecorate %in_buf BufferBlock\n"
8755 "OpMemberDecorate %in_buf 0 Offset 0\n"
8757 "%void = OpTypeVoid\n"
8758 "%voidf = OpTypeFunction %void\n"
8759 "%u32 = OpTypeInt 32 0\n"
8760 "%i32 = OpTypeInt 32 1\n"
8761 "%f32 = OpTypeFloat 32\n"
8762 "%uvec3 = OpTypeVector %u32 3\n"
8763 "%uvec3ptr = OpTypePointer Input %uvec3\n"
8764 "${commentDecl:opt}%const = OpConstant ${customType} ${constValue:opt}\n"
8766 "%in_ptr = OpTypePointer Uniform ${customType}\n"
8767 "%in_arr = OpTypeRuntimeArray ${customType}\n"
8768 "%in_buf = OpTypeStruct %in_arr\n"
8769 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
8770 "%indata = OpVariable %in_bufptr Uniform\n"
8771 "%outdata = OpVariable %in_bufptr Uniform\n"
8772 "%id = OpVariable %uvec3ptr Input\n"
8773 "%var_ptr = OpTypePointer Function ${customType}\n"
8775 "%zero = OpConstant %i32 0\n"
8777 "%main = OpFunction %void None %voidf\n"
8778 "%label = OpLabel\n"
8779 "%out_var = OpVariable %var_ptr Function${variableInitializer:opt}\n"
8780 "%idval = OpLoad %uvec3 %id\n"
8781 "%x = OpCompositeExtract %u32 %idval 0\n"
8782 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
8783 "%outloc = OpAccessChain %in_ptr %outdata %zero %x\n"
8785 "%outval = OpLoad ${customType} %out_var\n"
8786 " OpStore %outloc %outval\n"
8789 ).specialize(parameters);
8792 bool compareFloats (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
8794 DE_ASSERT(outputAllocs.size() != 0);
8795 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
8797 // Use custom epsilon because of the float->string conversion
8798 const float epsilon = 0.00001f;
8800 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
8802 vector<deUint8> expectedBytes;
8806 expectedOutputs[outputNdx]->getBytes(expectedBytes);
8807 memcpy(&expected, &expectedBytes.front(), expectedBytes.size());
8808 memcpy(&actual, outputAllocs[outputNdx]->getHostPtr(), expectedBytes.size());
8810 // Test with epsilon
8811 if (fabs(expected - actual) > epsilon)
8813 log << TestLog::Message << "Error: The actual and expected values not matching."
8814 << " Expected: " << expected << " Actual: " << actual << " Epsilon: " << epsilon << TestLog::EndMessage;
8821 // Checks if the driver crash with uninitialized cases
8822 bool passthruVerify (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
8824 DE_ASSERT(outputAllocs.size() != 0);
8825 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
8827 // Copy and discard the result.
8828 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
8830 vector<deUint8> expectedBytes;
8831 expectedOutputs[outputNdx]->getBytes(expectedBytes);
8833 const size_t width = expectedBytes.size();
8834 vector<char> data (width);
8836 memcpy(&data[0], outputAllocs[outputNdx]->getHostPtr(), width);
8841 tcu::TestCaseGroup* createShaderDefaultOutputGroup (tcu::TestContext& testCtx)
8843 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "shader_default_output", "Test shader default output."));
8844 de::Random rnd (deStringHash(group->getName()));
8846 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8848 NumberType numberType = NumberType(type);
8849 const string typeName = getNumberTypeName(numberType);
8850 const string description = "Test the OpVariable initializer with " + typeName + ".";
8851 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8853 // 2 similar subcases (initialized and uninitialized)
8854 for (int subCase = 0; subCase < 2; ++subCase)
8856 ComputeShaderSpec spec;
8857 spec.numWorkGroups = IVec3(1, 1, 1);
8859 map<string, string> params;
8863 case NUMBERTYPE_INT32:
8865 deInt32 number = getInt(rnd);
8866 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8867 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8868 params["constValue"] = numberToString(number);
8871 case NUMBERTYPE_UINT32:
8873 deUint32 number = rnd.getUint32();
8874 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8875 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8876 params["constValue"] = numberToString(number);
8879 case NUMBERTYPE_FLOAT32:
8881 float number = rnd.getFloat();
8882 spec.inputs.push_back(createCompositeBuffer<float>(number));
8883 spec.outputs.push_back(createCompositeBuffer<float>(number));
8884 spec.verifyIO = &compareFloats;
8885 params["constValue"] = numberToString(number);
8892 // Initialized subcase
8895 spec.assembly = specializeDefaultOutputShaderTemplate(numberType, params);
8896 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "initialized", "OpVariable initializer tests.", spec));
8898 // Uninitialized subcase
8901 spec.assembly = specializeDefaultOutputShaderTemplate(numberType);
8902 spec.verifyIO = &passthruVerify;
8903 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "uninitialized", "OpVariable initializer tests.", spec));
8906 group->addChild(subGroup.release());
8908 return group.release();
8911 tcu::TestCaseGroup* createOpNopTests (tcu::TestContext& testCtx)
8913 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
8914 RGBA defaultColors[4];
8915 map<string, string> opNopFragments;
8917 getDefaultColors(defaultColors);
8919 opNopFragments["testfun"] =
8920 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
8921 "%param1 = OpFunctionParameter %v4f32\n"
8922 "%label_testfun = OpLabel\n"
8931 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
8932 "%b = OpFAdd %f32 %a %a\n"
8934 "%c = OpFSub %f32 %b %a\n"
8935 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
8938 "OpReturnValue %ret\n"
8941 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, testGroup.get());
8943 return testGroup.release();
8946 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
8948 const bool testComputePipeline = true;
8950 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
8951 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
8952 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
8954 computeTests->addChild(createSpivVersionCheckTests(testCtx, testComputePipeline));
8955 computeTests->addChild(createLocalSizeGroup(testCtx));
8956 computeTests->addChild(createOpNopGroup(testCtx));
8957 computeTests->addChild(createOpFUnordGroup(testCtx));
8958 computeTests->addChild(createOpAtomicGroup(testCtx, false));
8959 computeTests->addChild(createOpAtomicGroup(testCtx, true)); // Using new StorageBuffer decoration
8960 computeTests->addChild(createOpLineGroup(testCtx));
8961 computeTests->addChild(createOpModuleProcessedGroup(testCtx));
8962 computeTests->addChild(createOpNoLineGroup(testCtx));
8963 computeTests->addChild(createOpConstantNullGroup(testCtx));
8964 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
8965 computeTests->addChild(createOpConstantUsageGroup(testCtx));
8966 computeTests->addChild(createSpecConstantGroup(testCtx));
8967 computeTests->addChild(createOpSourceGroup(testCtx));
8968 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
8969 computeTests->addChild(createDecorationGroupGroup(testCtx));
8970 computeTests->addChild(createOpPhiGroup(testCtx));
8971 computeTests->addChild(createLoopControlGroup(testCtx));
8972 computeTests->addChild(createFunctionControlGroup(testCtx));
8973 computeTests->addChild(createSelectionControlGroup(testCtx));
8974 computeTests->addChild(createBlockOrderGroup(testCtx));
8975 computeTests->addChild(createMultipleShaderGroup(testCtx));
8976 computeTests->addChild(createMemoryAccessGroup(testCtx));
8977 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
8978 computeTests->addChild(createOpCopyObjectGroup(testCtx));
8979 computeTests->addChild(createNoContractionGroup(testCtx));
8980 computeTests->addChild(createOpUndefGroup(testCtx));
8981 computeTests->addChild(createOpUnreachableGroup(testCtx));
8982 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
8983 computeTests ->addChild(createOpFRemGroup(testCtx));
8984 computeTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_PASS));
8985 computeTests->addChild(createOpSRemComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
8986 computeTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_PASS));
8987 computeTests->addChild(createOpSModComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
8988 computeTests->addChild(createSConvertTests(testCtx));
8989 computeTests->addChild(createUConvertTests(testCtx));
8990 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
8991 computeTests->addChild(createOpInBoundsAccessChainGroup(testCtx));
8992 computeTests->addChild(createShaderDefaultOutputGroup(testCtx));
8993 computeTests->addChild(createOpNMinGroup(testCtx));
8994 computeTests->addChild(createOpNMaxGroup(testCtx));
8995 computeTests->addChild(createOpNClampGroup(testCtx));
8997 de::MovePtr<tcu::TestCaseGroup> computeAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
8999 computeAndroidTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9000 computeAndroidTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9002 computeTests->addChild(computeAndroidTests.release());
9005 computeTests->addChild(create16BitStorageComputeGroup(testCtx));
9006 computeTests->addChild(createUboMatrixPaddingComputeGroup(testCtx));
9007 computeTests->addChild(createConditionalBranchComputeGroup(testCtx));
9008 computeTests->addChild(createIndexingComputeGroup(testCtx));
9009 computeTests->addChild(createVariablePointersComputeGroup(testCtx));
9010 computeTests->addChild(createImageSamplerComputeGroup(testCtx));
9011 graphicsTests->addChild(createSpivVersionCheckTests(testCtx, !testComputePipeline));
9012 graphicsTests->addChild(createOpNopTests(testCtx));
9013 graphicsTests->addChild(createOpSourceTests(testCtx));
9014 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
9015 graphicsTests->addChild(createOpModuleProcessedTests(testCtx));
9016 graphicsTests->addChild(createOpLineTests(testCtx));
9017 graphicsTests->addChild(createOpNoLineTests(testCtx));
9018 graphicsTests->addChild(createOpConstantNullTests(testCtx));
9019 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
9020 graphicsTests->addChild(createMemoryAccessTests(testCtx));
9021 graphicsTests->addChild(createOpUndefTests(testCtx));
9022 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
9023 graphicsTests->addChild(createModuleTests(testCtx));
9024 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
9025 graphicsTests->addChild(createOpPhiTests(testCtx));
9026 graphicsTests->addChild(createNoContractionTests(testCtx));
9027 graphicsTests->addChild(createOpQuantizeTests(testCtx));
9028 graphicsTests->addChild(createLoopTests(testCtx));
9029 graphicsTests->addChild(createSpecConstantTests(testCtx));
9030 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
9031 graphicsTests->addChild(createBarrierTests(testCtx));
9032 graphicsTests->addChild(createDecorationGroupTests(testCtx));
9033 graphicsTests->addChild(createFRemTests(testCtx));
9034 graphicsTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
9035 graphicsTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
9038 de::MovePtr<tcu::TestCaseGroup> graphicsAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
9040 graphicsAndroidTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9041 graphicsAndroidTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9043 graphicsTests->addChild(graphicsAndroidTests.release());
9046 graphicsTests->addChild(create16BitStorageGraphicsGroup(testCtx));
9047 graphicsTests->addChild(createUboMatrixPaddingGraphicsGroup(testCtx));
9048 graphicsTests->addChild(createConditionalBranchGraphicsGroup(testCtx));
9049 graphicsTests->addChild(createIndexingGraphicsGroup(testCtx));
9050 graphicsTests->addChild(createVariablePointersGraphicsGroup(testCtx));
9051 graphicsTests->addChild(createImageSamplerGraphicsGroup(testCtx));
9053 instructionTests->addChild(computeTests.release());
9054 instructionTests->addChild(graphicsTests.release());
9056 return instructionTests.release();