1 /*-------------------------------------------------------------------------
2 * Vulkan Conformance Tests
3 * ------------------------
5 * Copyright (c) 2015 Google Inc.
6 * Copyright (c) 2016 The Khronos Group Inc.
8 * Licensed under the Apache License, Version 2.0 (the "License");
9 * you may not use this file except in compliance with the License.
10 * You may obtain a copy of the License at
12 * http://www.apache.org/licenses/LICENSE-2.0
14 * Unless required by applicable law or agreed to in writing, software
15 * distributed under the License is distributed on an "AS IS" BASIS,
16 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17 * See the License for the specific language governing permissions and
18 * limitations under the License.
22 * \brief SPIR-V Assembly Tests for Instructions (special opcode/operand)
23 *//*--------------------------------------------------------------------*/
25 #include "vktSpvAsmInstructionTests.hpp"
27 #include "tcuCommandLine.hpp"
28 #include "tcuFormatUtil.hpp"
29 #include "tcuFloat.hpp"
30 #include "tcuRGBA.hpp"
31 #include "tcuStringTemplate.hpp"
32 #include "tcuTestLog.hpp"
33 #include "tcuVectorUtil.hpp"
34 #include "tcuInterval.hpp"
37 #include "vkDeviceUtil.hpp"
38 #include "vkMemUtil.hpp"
39 #include "vkPlatform.hpp"
40 #include "vkPrograms.hpp"
41 #include "vkQueryUtil.hpp"
43 #include "vkRefUtil.hpp"
44 #include "vkStrUtil.hpp"
45 #include "vkTypeUtil.hpp"
47 #include "deStringUtil.hpp"
48 #include "deUniquePtr.hpp"
50 #include "tcuStringTemplate.hpp"
52 #include "vktSpvAsmCrossStageInterfaceTests.hpp"
53 #include "vktSpvAsm16bitStorageTests.hpp"
54 #include "vktSpvAsmUboMatrixPaddingTests.hpp"
55 #include "vktSpvAsmConditionalBranchTests.hpp"
56 #include "vktSpvAsmIndexingTests.hpp"
57 #include "vktSpvAsmImageSamplerTests.hpp"
58 #include "vktSpvAsmComputeShaderCase.hpp"
59 #include "vktSpvAsmComputeShaderTestUtil.hpp"
60 #include "vktSpvAsmGraphicsShaderTestUtil.hpp"
61 #include "vktSpvAsmVariablePointersTests.hpp"
62 #include "vktSpvAsmVariableInitTests.hpp"
63 #include "vktSpvAsmSpirvVersionTests.hpp"
64 #include "vktTestCaseUtil.hpp"
65 #include "vktSpvAsmLoopDepLenTests.hpp"
66 #include "vktSpvAsmLoopDepInfTests.hpp"
78 namespace SpirVAssembly
92 using tcu::TestStatus;
95 using tcu::StringTemplate;
99 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0)
101 T* const typedPtr = (T*)dst;
102 for (int ndx = 0; ndx < numValues; ndx++)
103 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue);
106 // Filter is a function that returns true if a value should pass, false otherwise.
107 template<typename T, typename FilterT>
108 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, FilterT filter, int offset = 0)
110 T* const typedPtr = (T*)dst;
112 for (int ndx = 0; ndx < numValues; ndx++)
115 value = randomScalar<T>(rnd, minValue, maxValue);
116 while (!filter(value));
118 typedPtr[offset + ndx] = value;
122 // Gets a 64-bit integer with a more logarithmic distribution
123 deInt64 randomInt64LogDistributed (de::Random& rnd)
125 deInt64 val = rnd.getUint64();
126 val &= (1ull << rnd.getInt(1, 63)) - 1;
132 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues)
134 for (int ndx = 0; ndx < numValues; ndx++)
135 dst[ndx] = randomInt64LogDistributed(rnd);
138 template<typename FilterT>
139 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues, FilterT filter)
141 for (int ndx = 0; ndx < numValues; ndx++)
145 value = randomInt64LogDistributed(rnd);
146 } while (!filter(value));
151 inline bool filterNonNegative (const deInt64 value)
156 inline bool filterPositive (const deInt64 value)
161 inline bool filterNotZero (const deInt64 value)
166 static void floorAll (vector<float>& values)
168 for (size_t i = 0; i < values.size(); i++)
169 values[i] = deFloatFloor(values[i]);
172 static void floorAll (vector<Vec4>& values)
174 for (size_t i = 0; i < values.size(); i++)
175 values[i] = floor(values[i]);
183 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {}
186 // Assembly code used for testing LocalSize, OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code:
190 // layout(std140, set = 0, binding = 0) readonly buffer Input {
193 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
197 // layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
200 // uint x = gl_GlobalInvocationID.x;
201 // output_data.elements[x] = -input_data.elements[x];
204 static string getAsmForLocalSizeTest(bool useLiteralLocalSize, bool useSpecConstantWorkgroupSize, IVec3 workGroupSize, deUint32 ndx)
206 std::ostringstream out;
207 out << getComputeAsmShaderPreambleWithoutLocalSize();
209 if (useLiteralLocalSize)
211 out << "OpExecutionMode %main LocalSize "
212 << workGroupSize.x() << " " << workGroupSize.y() << " " << workGroupSize.z() << "\n";
215 out << "OpSource GLSL 430\n"
216 "OpName %main \"main\"\n"
217 "OpName %id \"gl_GlobalInvocationID\"\n"
218 "OpDecorate %id BuiltIn GlobalInvocationId\n";
220 if (useSpecConstantWorkgroupSize)
222 out << "OpDecorate %spec_0 SpecId 100\n"
223 << "OpDecorate %spec_1 SpecId 101\n"
224 << "OpDecorate %spec_2 SpecId 102\n"
225 << "OpDecorate %gl_WorkGroupSize BuiltIn WorkgroupSize\n";
228 out << getComputeAsmInputOutputBufferTraits()
229 << getComputeAsmCommonTypes()
230 << getComputeAsmInputOutputBuffer()
231 << "%id = OpVariable %uvec3ptr Input\n"
232 << "%zero = OpConstant %i32 0 \n";
234 if (useSpecConstantWorkgroupSize)
236 out << "%spec_0 = OpSpecConstant %u32 "<< workGroupSize.x() << "\n"
237 << "%spec_1 = OpSpecConstant %u32 "<< workGroupSize.y() << "\n"
238 << "%spec_2 = OpSpecConstant %u32 "<< workGroupSize.z() << "\n"
239 << "%gl_WorkGroupSize = OpSpecConstantComposite %uvec3 %spec_0 %spec_1 %spec_2\n";
242 out << "%main = OpFunction %void None %voidf\n"
243 << "%label = OpLabel\n"
244 << "%idval = OpLoad %uvec3 %id\n"
245 << "%ndx = OpCompositeExtract %u32 %idval " << ndx << "\n"
247 "%inloc = OpAccessChain %f32ptr %indata %zero %ndx\n"
248 "%inval = OpLoad %f32 %inloc\n"
249 "%neg = OpFNegate %f32 %inval\n"
250 "%outloc = OpAccessChain %f32ptr %outdata %zero %ndx\n"
251 " OpStore %outloc %neg\n"
257 tcu::TestCaseGroup* createLocalSizeGroup (tcu::TestContext& testCtx)
259 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "localsize", ""));
260 ComputeShaderSpec spec;
261 de::Random rnd (deStringHash(group->getName()));
262 const deUint32 numElements = 64u;
263 vector<float> positiveFloats (numElements, 0);
264 vector<float> negativeFloats (numElements, 0);
266 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
268 for (size_t ndx = 0; ndx < numElements; ++ndx)
269 negativeFloats[ndx] = -positiveFloats[ndx];
271 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
272 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
274 spec.numWorkGroups = IVec3(numElements, 1, 1);
276 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, 1, 1), 0u);
277 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize", "", spec));
279 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, 1, 1), 0u);
280 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize", "", spec));
282 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, 1, 1), 0u);
283 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize", "", spec));
285 spec.numWorkGroups = IVec3(1, 1, 1);
287 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(numElements, 1, 1), 0u);
288 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_x", "", spec));
290 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(numElements, 1, 1), 0u);
291 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_x", "", spec));
293 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(numElements, 1, 1), 0u);
294 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_x", "", spec));
296 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, numElements, 1), 1u);
297 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_y", "", spec));
299 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, numElements, 1), 1u);
300 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_y", "", spec));
302 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, numElements, 1), 1u);
303 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_y", "", spec));
305 spec.assembly = getAsmForLocalSizeTest(true, false, IVec3(1, 1, numElements), 2u);
306 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_localsize_z", "", spec));
308 spec.assembly = getAsmForLocalSizeTest(true, true, IVec3(1, 1, numElements), 2u);
309 group->addChild(new SpvAsmComputeShaderCase(testCtx, "literal_and_specid_localsize_z", "", spec));
311 spec.assembly = getAsmForLocalSizeTest(false, true, IVec3(1, 1, numElements), 2u);
312 group->addChild(new SpvAsmComputeShaderCase(testCtx, "specid_localsize_z", "", spec));
314 return group.release();
317 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
319 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
320 ComputeShaderSpec spec;
321 de::Random rnd (deStringHash(group->getName()));
322 const int numElements = 100;
323 vector<float> positiveFloats (numElements, 0);
324 vector<float> negativeFloats (numElements, 0);
326 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
328 for (size_t ndx = 0; ndx < numElements; ++ndx)
329 negativeFloats[ndx] = -positiveFloats[ndx];
332 string(getComputeAsmShaderPreamble()) +
334 "OpSource GLSL 430\n"
335 "OpName %main \"main\"\n"
336 "OpName %id \"gl_GlobalInvocationID\"\n"
338 "OpDecorate %id BuiltIn GlobalInvocationId\n"
340 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes())
342 + string(getComputeAsmInputOutputBuffer()) +
344 "%id = OpVariable %uvec3ptr Input\n"
345 "%zero = OpConstant %i32 0\n"
347 "%main = OpFunction %void None %voidf\n"
349 "%idval = OpLoad %uvec3 %id\n"
350 "%x = OpCompositeExtract %u32 %idval 0\n"
352 " OpNop\n" // Inside a function body
354 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
355 "%inval = OpLoad %f32 %inloc\n"
356 "%neg = OpFNegate %f32 %inval\n"
357 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
358 " OpStore %outloc %neg\n"
361 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
362 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
363 spec.numWorkGroups = IVec3(numElements, 1, 1);
365 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
367 return group.release();
370 bool compareFUnord (const std::vector<BufferSp>& inputs, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
372 if (outputAllocs.size() != 1)
375 vector<deUint8> input1Bytes;
376 vector<deUint8> input2Bytes;
377 vector<deUint8> expectedBytes;
379 inputs[0]->getBytes(input1Bytes);
380 inputs[1]->getBytes(input2Bytes);
381 expectedOutputs[0]->getBytes(expectedBytes);
383 const deInt32* const expectedOutputAsInt = reinterpret_cast<const deInt32* const>(&expectedBytes.front());
384 const deInt32* const outputAsInt = static_cast<const deInt32* const>(outputAllocs[0]->getHostPtr());
385 const float* const input1AsFloat = reinterpret_cast<const float* const>(&input1Bytes.front());
386 const float* const input2AsFloat = reinterpret_cast<const float* const>(&input2Bytes.front());
387 bool returnValue = true;
389 for (size_t idx = 0; idx < expectedBytes.size() / sizeof(deInt32); ++idx)
391 if (outputAsInt[idx] != expectedOutputAsInt[idx])
393 log << TestLog::Message << "ERROR: Sub-case failed. inputs: " << input1AsFloat[idx] << "," << input2AsFloat[idx] << " output: " << outputAsInt[idx]<< " expected output: " << expectedOutputAsInt[idx] << TestLog::EndMessage;
400 typedef VkBool32 (*compareFuncType) (float, float);
406 compareFuncType compareFunc;
408 OpFUnordCase (const char* _name, const char* _opCode, compareFuncType _compareFunc)
411 , compareFunc (_compareFunc) {}
414 #define ADD_OPFUNORD_CASE(NAME, OPCODE, OPERATOR) \
416 struct compare_##NAME { static VkBool32 compare(float x, float y) { return (x OPERATOR y) ? VK_TRUE : VK_FALSE; } }; \
417 cases.push_back(OpFUnordCase(#NAME, OPCODE, compare_##NAME::compare)); \
418 } while (deGetFalse())
420 tcu::TestCaseGroup* createOpFUnordGroup (tcu::TestContext& testCtx)
422 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfunord", "Test the OpFUnord* opcodes"));
423 de::Random rnd (deStringHash(group->getName()));
424 const int numElements = 100;
425 vector<OpFUnordCase> cases;
427 const StringTemplate shaderTemplate (
429 string(getComputeAsmShaderPreamble()) +
431 "OpSource GLSL 430\n"
432 "OpName %main \"main\"\n"
433 "OpName %id \"gl_GlobalInvocationID\"\n"
435 "OpDecorate %id BuiltIn GlobalInvocationId\n"
437 "OpDecorate %buf BufferBlock\n"
438 "OpDecorate %buf2 BufferBlock\n"
439 "OpDecorate %indata1 DescriptorSet 0\n"
440 "OpDecorate %indata1 Binding 0\n"
441 "OpDecorate %indata2 DescriptorSet 0\n"
442 "OpDecorate %indata2 Binding 1\n"
443 "OpDecorate %outdata DescriptorSet 0\n"
444 "OpDecorate %outdata Binding 2\n"
445 "OpDecorate %f32arr ArrayStride 4\n"
446 "OpDecorate %i32arr ArrayStride 4\n"
447 "OpMemberDecorate %buf 0 Offset 0\n"
448 "OpMemberDecorate %buf2 0 Offset 0\n"
450 + string(getComputeAsmCommonTypes()) +
452 "%buf = OpTypeStruct %f32arr\n"
453 "%bufptr = OpTypePointer Uniform %buf\n"
454 "%indata1 = OpVariable %bufptr Uniform\n"
455 "%indata2 = OpVariable %bufptr Uniform\n"
457 "%buf2 = OpTypeStruct %i32arr\n"
458 "%buf2ptr = OpTypePointer Uniform %buf2\n"
459 "%outdata = OpVariable %buf2ptr Uniform\n"
461 "%id = OpVariable %uvec3ptr Input\n"
462 "%zero = OpConstant %i32 0\n"
463 "%consti1 = OpConstant %i32 1\n"
464 "%constf1 = OpConstant %f32 1.0\n"
466 "%main = OpFunction %void None %voidf\n"
468 "%idval = OpLoad %uvec3 %id\n"
469 "%x = OpCompositeExtract %u32 %idval 0\n"
471 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
472 "%inval1 = OpLoad %f32 %inloc1\n"
473 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
474 "%inval2 = OpLoad %f32 %inloc2\n"
475 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
477 "%result = ${OPCODE} %bool %inval1 %inval2\n"
478 "%int_res = OpSelect %i32 %result %consti1 %zero\n"
479 " OpStore %outloc %int_res\n"
484 ADD_OPFUNORD_CASE(equal, "OpFUnordEqual", ==);
485 ADD_OPFUNORD_CASE(less, "OpFUnordLessThan", <);
486 ADD_OPFUNORD_CASE(lessequal, "OpFUnordLessThanEqual", <=);
487 ADD_OPFUNORD_CASE(greater, "OpFUnordGreaterThan", >);
488 ADD_OPFUNORD_CASE(greaterequal, "OpFUnordGreaterThanEqual", >=);
489 ADD_OPFUNORD_CASE(notequal, "OpFUnordNotEqual", !=);
491 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
493 map<string, string> specializations;
494 ComputeShaderSpec spec;
495 const float NaN = std::numeric_limits<float>::quiet_NaN();
496 vector<float> inputFloats1 (numElements, 0);
497 vector<float> inputFloats2 (numElements, 0);
498 vector<deInt32> expectedInts (numElements, 0);
500 specializations["OPCODE"] = cases[caseNdx].opCode;
501 spec.assembly = shaderTemplate.specialize(specializations);
503 fillRandomScalars(rnd, 1.f, 100.f, &inputFloats1[0], numElements);
504 for (size_t ndx = 0; ndx < numElements; ++ndx)
508 case 0: inputFloats2[ndx] = inputFloats1[ndx] + 1.0f; break;
509 case 1: inputFloats2[ndx] = inputFloats1[ndx] - 1.0f; break;
510 case 2: inputFloats2[ndx] = inputFloats1[ndx]; break;
511 case 3: inputFloats2[ndx] = NaN; break;
512 case 4: inputFloats2[ndx] = inputFloats1[ndx]; inputFloats1[ndx] = NaN; break;
513 case 5: inputFloats2[ndx] = NaN; inputFloats1[ndx] = NaN; break;
515 expectedInts[ndx] = tcu::Float32(inputFloats1[ndx]).isNaN() || tcu::Float32(inputFloats2[ndx]).isNaN() || cases[caseNdx].compareFunc(inputFloats1[ndx], inputFloats2[ndx]);
518 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
519 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
520 spec.outputs.push_back(BufferSp(new Int32Buffer(expectedInts)));
521 spec.numWorkGroups = IVec3(numElements, 1, 1);
522 spec.verifyIO = &compareFUnord;
523 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
526 return group.release();
532 const char* assembly;
533 OpAtomicType opAtomic;
534 deInt32 numOutputElements;
536 OpAtomicCase (const char* _name, const char* _assembly, OpAtomicType _opAtomic, deInt32 _numOutputElements)
538 , assembly (_assembly)
539 , opAtomic (_opAtomic)
540 , numOutputElements (_numOutputElements) {}
543 tcu::TestCaseGroup* createOpAtomicGroup (tcu::TestContext& testCtx, bool useStorageBuffer)
545 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx,
546 useStorageBuffer ? "opatomic_storage_buffer" : "opatomic",
547 "Test the OpAtomic* opcodes"));
548 const int numElements = 65535;
549 vector<OpAtomicCase> cases;
551 const StringTemplate shaderTemplate (
553 string("OpCapability Shader\n") +
554 (useStorageBuffer ? "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" : "") +
555 "OpMemoryModel Logical GLSL450\n"
556 "OpEntryPoint GLCompute %main \"main\" %id\n"
557 "OpExecutionMode %main LocalSize 1 1 1\n" +
559 "OpSource GLSL 430\n"
560 "OpName %main \"main\"\n"
561 "OpName %id \"gl_GlobalInvocationID\"\n"
563 "OpDecorate %id BuiltIn GlobalInvocationId\n"
565 "OpDecorate %buf ${BLOCK_DECORATION}\n"
566 "OpDecorate %indata DescriptorSet 0\n"
567 "OpDecorate %indata Binding 0\n"
568 "OpDecorate %i32arr ArrayStride 4\n"
569 "OpMemberDecorate %buf 0 Offset 0\n"
571 "OpDecorate %sumbuf ${BLOCK_DECORATION}\n"
572 "OpDecorate %sum DescriptorSet 0\n"
573 "OpDecorate %sum Binding 1\n"
574 "OpMemberDecorate %sumbuf 0 Coherent\n"
575 "OpMemberDecorate %sumbuf 0 Offset 0\n"
577 + getComputeAsmCommonTypes("${BLOCK_POINTER_TYPE}") +
579 "%buf = OpTypeStruct %i32arr\n"
580 "%bufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %buf\n"
581 "%indata = OpVariable %bufptr ${BLOCK_POINTER_TYPE}\n"
583 "%sumbuf = OpTypeStruct %i32arr\n"
584 "%sumbufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %sumbuf\n"
585 "%sum = OpVariable %sumbufptr ${BLOCK_POINTER_TYPE}\n"
587 "%id = OpVariable %uvec3ptr Input\n"
588 "%minusone = OpConstant %i32 -1\n"
589 "%zero = OpConstant %i32 0\n"
590 "%one = OpConstant %u32 1\n"
591 "%two = OpConstant %i32 2\n"
593 "%main = OpFunction %void None %voidf\n"
595 "%idval = OpLoad %uvec3 %id\n"
596 "%x = OpCompositeExtract %u32 %idval 0\n"
598 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
599 "%inval = OpLoad %i32 %inloc\n"
601 "%outloc = OpAccessChain %i32ptr %sum %zero ${INDEX}\n"
607 #define ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, NUM_OUTPUT_ELEMENTS) \
609 DE_STATIC_ASSERT((NUM_OUTPUT_ELEMENTS) == 1 || (NUM_OUTPUT_ELEMENTS) == numElements); \
610 cases.push_back(OpAtomicCase(#NAME, ASSEMBLY, OPATOMIC, NUM_OUTPUT_ELEMENTS)); \
611 } while (deGetFalse())
612 #define ADD_OPATOMIC_CASE_1(NAME, ASSEMBLY, OPATOMIC) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, 1)
613 #define ADD_OPATOMIC_CASE_N(NAME, ASSEMBLY, OPATOMIC) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, numElements)
615 ADD_OPATOMIC_CASE_1(iadd, "%unused = OpAtomicIAdd %i32 %outloc %one %zero %inval\n", OPATOMIC_IADD );
616 ADD_OPATOMIC_CASE_1(isub, "%unused = OpAtomicISub %i32 %outloc %one %zero %inval\n", OPATOMIC_ISUB );
617 ADD_OPATOMIC_CASE_1(iinc, "%unused = OpAtomicIIncrement %i32 %outloc %one %zero\n", OPATOMIC_IINC );
618 ADD_OPATOMIC_CASE_1(idec, "%unused = OpAtomicIDecrement %i32 %outloc %one %zero\n", OPATOMIC_IDEC );
619 ADD_OPATOMIC_CASE_N(load, "%inval2 = OpAtomicLoad %i32 %inloc %one %zero\n"
620 " OpStore %outloc %inval2\n", OPATOMIC_LOAD );
621 ADD_OPATOMIC_CASE_N(store, " OpAtomicStore %outloc %one %zero %inval\n", OPATOMIC_STORE );
622 ADD_OPATOMIC_CASE_N(compex, "%even = OpSMod %i32 %inval %two\n"
623 " OpStore %outloc %even\n"
624 "%unused = OpAtomicCompareExchange %i32 %outloc %one %zero %zero %minusone %zero\n", OPATOMIC_COMPEX );
626 #undef ADD_OPATOMIC_CASE
627 #undef ADD_OPATOMIC_CASE_1
628 #undef ADD_OPATOMIC_CASE_N
630 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
632 map<string, string> specializations;
633 ComputeShaderSpec spec;
634 vector<deInt32> inputInts (numElements, 0);
635 vector<deInt32> expected (cases[caseNdx].numOutputElements, -1);
637 specializations["INDEX"] = (cases[caseNdx].numOutputElements == 1) ? "%zero" : "%x";
638 specializations["INSTRUCTION"] = cases[caseNdx].assembly;
639 specializations["BLOCK_DECORATION"] = useStorageBuffer ? "Block" : "BufferBlock";
640 specializations["BLOCK_POINTER_TYPE"] = useStorageBuffer ? "StorageBuffer" : "Uniform";
641 spec.assembly = shaderTemplate.specialize(specializations);
643 if (useStorageBuffer)
644 spec.extensions.push_back("VK_KHR_storage_buffer_storage_class");
646 spec.inputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_INPUT)));
647 spec.outputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_EXPECTED)));
648 spec.numWorkGroups = IVec3(numElements, 1, 1);
649 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
652 return group.release();
655 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
657 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
658 ComputeShaderSpec spec;
659 de::Random rnd (deStringHash(group->getName()));
660 const int numElements = 100;
661 vector<float> positiveFloats (numElements, 0);
662 vector<float> negativeFloats (numElements, 0);
664 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
666 for (size_t ndx = 0; ndx < numElements; ++ndx)
667 negativeFloats[ndx] = -positiveFloats[ndx];
670 string(getComputeAsmShaderPreamble()) +
672 "%fname1 = OpString \"negateInputs.comp\"\n"
673 "%fname2 = OpString \"negateInputs\"\n"
675 "OpSource GLSL 430\n"
676 "OpName %main \"main\"\n"
677 "OpName %id \"gl_GlobalInvocationID\"\n"
679 "OpDecorate %id BuiltIn GlobalInvocationId\n"
681 + string(getComputeAsmInputOutputBufferTraits()) +
683 "OpLine %fname1 0 0\n" // At the earliest possible position
685 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
687 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
688 "OpLine %fname2 1 0\n" // Different filenames
689 "OpLine %fname1 1000 100000\n"
691 "%id = OpVariable %uvec3ptr Input\n"
692 "%zero = OpConstant %i32 0\n"
694 "OpLine %fname1 1 1\n" // Before a function
696 "%main = OpFunction %void None %voidf\n"
699 "OpLine %fname1 1 1\n" // In a function
701 "%idval = OpLoad %uvec3 %id\n"
702 "%x = OpCompositeExtract %u32 %idval 0\n"
703 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
704 "%inval = OpLoad %f32 %inloc\n"
705 "%neg = OpFNegate %f32 %inval\n"
706 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
707 " OpStore %outloc %neg\n"
710 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
711 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
712 spec.numWorkGroups = IVec3(numElements, 1, 1);
714 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
716 return group.release();
719 bool veryfiBinaryShader (const ProgramBinary& binary)
721 const size_t paternCount = 3u;
722 bool paternsCheck[paternCount] =
726 const string patersns[paternCount] =
732 size_t paternNdx = 0u;
734 for (size_t ndx = 0u; ndx < binary.getSize(); ++ndx)
736 if (false == paternsCheck[paternNdx] &&
737 patersns[paternNdx][0] == static_cast<char>(binary.getBinary()[ndx]) &&
738 deMemoryEqual((const char*)&binary.getBinary()[ndx], &patersns[paternNdx][0], patersns[paternNdx].length()))
740 paternsCheck[paternNdx]= true;
742 if (paternNdx == paternCount)
747 for (size_t ndx = 0u; ndx < paternCount; ++ndx)
749 if (!paternsCheck[ndx])
756 tcu::TestCaseGroup* createOpModuleProcessedGroup (tcu::TestContext& testCtx)
758 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opmoduleprocessed", "Test the OpModuleProcessed instruction"));
759 ComputeShaderSpec spec;
760 de::Random rnd (deStringHash(group->getName()));
761 const int numElements = 10;
762 vector<float> positiveFloats (numElements, 0);
763 vector<float> negativeFloats (numElements, 0);
765 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
767 for (size_t ndx = 0; ndx < numElements; ++ndx)
768 negativeFloats[ndx] = -positiveFloats[ndx];
771 string(getComputeAsmShaderPreamble()) +
772 "%fname = OpString \"negateInputs.comp\"\n"
774 "OpSource GLSL 430\n"
775 "OpName %main \"main\"\n"
776 "OpName %id \"gl_GlobalInvocationID\"\n"
777 "OpModuleProcessed \"VULKAN CTS\"\n" //OpModuleProcessed;
778 "OpModuleProcessed \"Negative values\"\n"
779 "OpModuleProcessed \"Date: 2017/09/21\"\n"
780 "OpDecorate %id BuiltIn GlobalInvocationId\n"
782 + string(getComputeAsmInputOutputBufferTraits())
784 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
786 "OpLine %fname 0 1\n"
788 "OpLine %fname 1000 1\n"
790 "%id = OpVariable %uvec3ptr Input\n"
791 "%zero = OpConstant %i32 0\n"
792 "%main = OpFunction %void None %voidf\n"
795 "%idval = OpLoad %uvec3 %id\n"
796 "%x = OpCompositeExtract %u32 %idval 0\n"
798 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
799 "%inval = OpLoad %f32 %inloc\n"
800 "%neg = OpFNegate %f32 %inval\n"
801 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
802 " OpStore %outloc %neg\n"
805 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
806 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
807 spec.numWorkGroups = IVec3(numElements, 1, 1);
808 spec.verifyBinary = veryfiBinaryShader;
809 spec.spirvVersion = SPIRV_VERSION_1_3;
811 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpModuleProcessed Tests", spec));
813 return group.release();
816 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
818 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
819 ComputeShaderSpec spec;
820 de::Random rnd (deStringHash(group->getName()));
821 const int numElements = 100;
822 vector<float> positiveFloats (numElements, 0);
823 vector<float> negativeFloats (numElements, 0);
825 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
827 for (size_t ndx = 0; ndx < numElements; ++ndx)
828 negativeFloats[ndx] = -positiveFloats[ndx];
831 string(getComputeAsmShaderPreamble()) +
833 "%fname = OpString \"negateInputs.comp\"\n"
835 "OpSource GLSL 430\n"
836 "OpName %main \"main\"\n"
837 "OpName %id \"gl_GlobalInvocationID\"\n"
839 "OpDecorate %id BuiltIn GlobalInvocationId\n"
841 + string(getComputeAsmInputOutputBufferTraits()) +
843 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
845 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
847 "OpLine %fname 0 1\n"
848 "OpNoLine\n" // Immediately following a preceding OpLine
850 "OpLine %fname 1000 1\n"
852 "%id = OpVariable %uvec3ptr Input\n"
853 "%zero = OpConstant %i32 0\n"
855 "OpNoLine\n" // Contents after the previous OpLine
857 "%main = OpFunction %void None %voidf\n"
859 "%idval = OpLoad %uvec3 %id\n"
860 "%x = OpCompositeExtract %u32 %idval 0\n"
862 "OpNoLine\n" // Multiple OpNoLine
866 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
867 "%inval = OpLoad %f32 %inloc\n"
868 "%neg = OpFNegate %f32 %inval\n"
869 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
870 " OpStore %outloc %neg\n"
873 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
874 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
875 spec.numWorkGroups = IVec3(numElements, 1, 1);
877 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
879 return group.release();
882 // Compare instruction for the contraction compute case.
883 // Returns true if the output is what is expected from the test case.
884 bool compareNoContractCase(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
886 if (outputAllocs.size() != 1)
889 // Only size is needed because we are not comparing the exact values.
890 size_t byteSize = expectedOutputs[0]->getByteSize();
892 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
894 for(size_t i = 0; i < byteSize / sizeof(float); ++i) {
895 if (outputAsFloat[i] != 0.f &&
896 outputAsFloat[i] != -ldexp(1, -24)) {
904 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
906 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
907 vector<CaseParameter> cases;
908 const int numElements = 100;
909 vector<float> inputFloats1 (numElements, 0);
910 vector<float> inputFloats2 (numElements, 0);
911 vector<float> outputFloats (numElements, 0);
912 const StringTemplate shaderTemplate (
913 string(getComputeAsmShaderPreamble()) +
915 "OpName %main \"main\"\n"
916 "OpName %id \"gl_GlobalInvocationID\"\n"
918 "OpDecorate %id BuiltIn GlobalInvocationId\n"
922 "OpDecorate %buf BufferBlock\n"
923 "OpDecorate %indata1 DescriptorSet 0\n"
924 "OpDecorate %indata1 Binding 0\n"
925 "OpDecorate %indata2 DescriptorSet 0\n"
926 "OpDecorate %indata2 Binding 1\n"
927 "OpDecorate %outdata DescriptorSet 0\n"
928 "OpDecorate %outdata Binding 2\n"
929 "OpDecorate %f32arr ArrayStride 4\n"
930 "OpMemberDecorate %buf 0 Offset 0\n"
932 + string(getComputeAsmCommonTypes()) +
934 "%buf = OpTypeStruct %f32arr\n"
935 "%bufptr = OpTypePointer Uniform %buf\n"
936 "%indata1 = OpVariable %bufptr Uniform\n"
937 "%indata2 = OpVariable %bufptr Uniform\n"
938 "%outdata = OpVariable %bufptr Uniform\n"
940 "%id = OpVariable %uvec3ptr Input\n"
941 "%zero = OpConstant %i32 0\n"
942 "%c_f_m1 = OpConstant %f32 -1.\n"
944 "%main = OpFunction %void None %voidf\n"
946 "%idval = OpLoad %uvec3 %id\n"
947 "%x = OpCompositeExtract %u32 %idval 0\n"
948 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
949 "%inval1 = OpLoad %f32 %inloc1\n"
950 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
951 "%inval2 = OpLoad %f32 %inloc2\n"
952 "%mul = OpFMul %f32 %inval1 %inval2\n"
953 "%add = OpFAdd %f32 %mul %c_f_m1\n"
954 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
955 " OpStore %outloc %add\n"
959 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
960 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
961 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
963 for (size_t ndx = 0; ndx < numElements; ++ndx)
965 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
966 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
967 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
968 // conducted separately and the result is rounded to 1, or 0x1.fffffcp-1
969 // So the final result will be 0.f or 0x1p-24.
970 // If the operation is combined into a precise fused multiply-add, then the result would be
971 // 2^-46 (0xa8800000).
972 outputFloats[ndx] = 0.f;
975 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
977 map<string, string> specializations;
978 ComputeShaderSpec spec;
980 specializations["DECORATION"] = cases[caseNdx].param;
981 spec.assembly = shaderTemplate.specialize(specializations);
982 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
983 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
984 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
985 spec.numWorkGroups = IVec3(numElements, 1, 1);
986 // Check against the two possible answers based on rounding mode.
987 spec.verifyIO = &compareNoContractCase;
989 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
991 return group.release();
994 bool compareFRem(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
996 if (outputAllocs.size() != 1)
999 vector<deUint8> expectedBytes;
1000 expectedOutputs[0]->getBytes(expectedBytes);
1002 const float* expectedOutputAsFloat = reinterpret_cast<const float*>(&expectedBytes.front());
1003 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1005 for (size_t idx = 0; idx < expectedBytes.size() / sizeof(float); ++idx)
1007 const float f0 = expectedOutputAsFloat[idx];
1008 const float f1 = outputAsFloat[idx];
1009 // \todo relative error needs to be fairly high because FRem may be implemented as
1010 // (roughly) frac(a/b)*b, so LSB errors can be magnified. But this should be fine for now.
1011 if (deFloatAbs((f1 - f0) / f0) > 0.02)
1018 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
1020 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
1021 ComputeShaderSpec spec;
1022 de::Random rnd (deStringHash(group->getName()));
1023 const int numElements = 200;
1024 vector<float> inputFloats1 (numElements, 0);
1025 vector<float> inputFloats2 (numElements, 0);
1026 vector<float> outputFloats (numElements, 0);
1028 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1029 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
1031 for (size_t ndx = 0; ndx < numElements; ++ndx)
1033 // Guard against divisors near zero.
1034 if (std::fabs(inputFloats2[ndx]) < 1e-3)
1035 inputFloats2[ndx] = 8.f;
1037 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1038 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
1042 string(getComputeAsmShaderPreamble()) +
1044 "OpName %main \"main\"\n"
1045 "OpName %id \"gl_GlobalInvocationID\"\n"
1047 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1049 "OpDecorate %buf BufferBlock\n"
1050 "OpDecorate %indata1 DescriptorSet 0\n"
1051 "OpDecorate %indata1 Binding 0\n"
1052 "OpDecorate %indata2 DescriptorSet 0\n"
1053 "OpDecorate %indata2 Binding 1\n"
1054 "OpDecorate %outdata DescriptorSet 0\n"
1055 "OpDecorate %outdata Binding 2\n"
1056 "OpDecorate %f32arr ArrayStride 4\n"
1057 "OpMemberDecorate %buf 0 Offset 0\n"
1059 + string(getComputeAsmCommonTypes()) +
1061 "%buf = OpTypeStruct %f32arr\n"
1062 "%bufptr = OpTypePointer Uniform %buf\n"
1063 "%indata1 = OpVariable %bufptr Uniform\n"
1064 "%indata2 = OpVariable %bufptr Uniform\n"
1065 "%outdata = OpVariable %bufptr Uniform\n"
1067 "%id = OpVariable %uvec3ptr Input\n"
1068 "%zero = OpConstant %i32 0\n"
1070 "%main = OpFunction %void None %voidf\n"
1071 "%label = OpLabel\n"
1072 "%idval = OpLoad %uvec3 %id\n"
1073 "%x = OpCompositeExtract %u32 %idval 0\n"
1074 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1075 "%inval1 = OpLoad %f32 %inloc1\n"
1076 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1077 "%inval2 = OpLoad %f32 %inloc2\n"
1078 "%rem = OpFRem %f32 %inval1 %inval2\n"
1079 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1080 " OpStore %outloc %rem\n"
1084 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1085 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1086 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1087 spec.numWorkGroups = IVec3(numElements, 1, 1);
1088 spec.verifyIO = &compareFRem;
1090 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1092 return group.release();
1095 bool compareNMin (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1097 if (outputAllocs.size() != 1)
1100 const BufferSp& expectedOutput (expectedOutputs[0]);
1101 std::vector<deUint8> data;
1102 expectedOutput->getBytes(data);
1104 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1105 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1107 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float); ++idx)
1109 const float f0 = expectedOutputAsFloat[idx];
1110 const float f1 = outputAsFloat[idx];
1112 // For NMin, we accept NaN as output if both inputs were NaN.
1113 // Otherwise the NaN is the wrong choise, as on architectures that
1114 // do not handle NaN, those are huge values.
1115 if (!(tcu::Float32(f1).isNaN() && tcu::Float32(f0).isNaN()) && deFloatAbs(f1 - f0) > 0.00001f)
1122 tcu::TestCaseGroup* createOpNMinGroup (tcu::TestContext& testCtx)
1124 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnmin", "Test the OpNMin instruction"));
1125 ComputeShaderSpec spec;
1126 de::Random rnd (deStringHash(group->getName()));
1127 const int numElements = 200;
1128 vector<float> inputFloats1 (numElements, 0);
1129 vector<float> inputFloats2 (numElements, 0);
1130 vector<float> outputFloats (numElements, 0);
1132 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1133 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1135 // Make the first case a full-NAN case.
1136 inputFloats1[0] = TCU_NAN;
1137 inputFloats2[0] = TCU_NAN;
1139 for (size_t ndx = 0; ndx < numElements; ++ndx)
1141 // By default, pick the smallest
1142 outputFloats[ndx] = std::min(inputFloats1[ndx], inputFloats2[ndx]);
1144 // Make half of the cases NaN cases
1147 // Alternate between the NaN operand
1150 outputFloats[ndx] = inputFloats2[ndx];
1151 inputFloats1[ndx] = TCU_NAN;
1155 outputFloats[ndx] = inputFloats1[ndx];
1156 inputFloats2[ndx] = TCU_NAN;
1162 "OpCapability Shader\n"
1163 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1164 "OpMemoryModel Logical GLSL450\n"
1165 "OpEntryPoint GLCompute %main \"main\" %id\n"
1166 "OpExecutionMode %main LocalSize 1 1 1\n"
1168 "OpName %main \"main\"\n"
1169 "OpName %id \"gl_GlobalInvocationID\"\n"
1171 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1173 "OpDecorate %buf BufferBlock\n"
1174 "OpDecorate %indata1 DescriptorSet 0\n"
1175 "OpDecorate %indata1 Binding 0\n"
1176 "OpDecorate %indata2 DescriptorSet 0\n"
1177 "OpDecorate %indata2 Binding 1\n"
1178 "OpDecorate %outdata DescriptorSet 0\n"
1179 "OpDecorate %outdata Binding 2\n"
1180 "OpDecorate %f32arr ArrayStride 4\n"
1181 "OpMemberDecorate %buf 0 Offset 0\n"
1183 + string(getComputeAsmCommonTypes()) +
1185 "%buf = OpTypeStruct %f32arr\n"
1186 "%bufptr = OpTypePointer Uniform %buf\n"
1187 "%indata1 = OpVariable %bufptr Uniform\n"
1188 "%indata2 = OpVariable %bufptr Uniform\n"
1189 "%outdata = OpVariable %bufptr Uniform\n"
1191 "%id = OpVariable %uvec3ptr Input\n"
1192 "%zero = OpConstant %i32 0\n"
1194 "%main = OpFunction %void None %voidf\n"
1195 "%label = OpLabel\n"
1196 "%idval = OpLoad %uvec3 %id\n"
1197 "%x = OpCompositeExtract %u32 %idval 0\n"
1198 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1199 "%inval1 = OpLoad %f32 %inloc1\n"
1200 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1201 "%inval2 = OpLoad %f32 %inloc2\n"
1202 "%rem = OpExtInst %f32 %std450 NMin %inval1 %inval2\n"
1203 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1204 " OpStore %outloc %rem\n"
1208 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1209 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1210 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1211 spec.numWorkGroups = IVec3(numElements, 1, 1);
1212 spec.verifyIO = &compareNMin;
1214 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1216 return group.release();
1219 bool compareNMax (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1221 if (outputAllocs.size() != 1)
1224 const BufferSp& expectedOutput = expectedOutputs[0];
1225 std::vector<deUint8> data;
1226 expectedOutput->getBytes(data);
1228 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1229 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1231 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float); ++idx)
1233 const float f0 = expectedOutputAsFloat[idx];
1234 const float f1 = outputAsFloat[idx];
1236 // For NMax, NaN is considered acceptable result, since in
1237 // architectures that do not handle NaNs, those are huge values.
1238 if (!tcu::Float32(f1).isNaN() && deFloatAbs(f1 - f0) > 0.00001f)
1245 tcu::TestCaseGroup* createOpNMaxGroup (tcu::TestContext& testCtx)
1247 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, "opnmax", "Test the OpNMax instruction"));
1248 ComputeShaderSpec spec;
1249 de::Random rnd (deStringHash(group->getName()));
1250 const int numElements = 200;
1251 vector<float> inputFloats1 (numElements, 0);
1252 vector<float> inputFloats2 (numElements, 0);
1253 vector<float> outputFloats (numElements, 0);
1255 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1256 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1258 // Make the first case a full-NAN case.
1259 inputFloats1[0] = TCU_NAN;
1260 inputFloats2[0] = TCU_NAN;
1262 for (size_t ndx = 0; ndx < numElements; ++ndx)
1264 // By default, pick the biggest
1265 outputFloats[ndx] = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1267 // Make half of the cases NaN cases
1270 // Alternate between the NaN operand
1273 outputFloats[ndx] = inputFloats2[ndx];
1274 inputFloats1[ndx] = TCU_NAN;
1278 outputFloats[ndx] = inputFloats1[ndx];
1279 inputFloats2[ndx] = TCU_NAN;
1285 "OpCapability Shader\n"
1286 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1287 "OpMemoryModel Logical GLSL450\n"
1288 "OpEntryPoint GLCompute %main \"main\" %id\n"
1289 "OpExecutionMode %main LocalSize 1 1 1\n"
1291 "OpName %main \"main\"\n"
1292 "OpName %id \"gl_GlobalInvocationID\"\n"
1294 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1296 "OpDecorate %buf BufferBlock\n"
1297 "OpDecorate %indata1 DescriptorSet 0\n"
1298 "OpDecorate %indata1 Binding 0\n"
1299 "OpDecorate %indata2 DescriptorSet 0\n"
1300 "OpDecorate %indata2 Binding 1\n"
1301 "OpDecorate %outdata DescriptorSet 0\n"
1302 "OpDecorate %outdata Binding 2\n"
1303 "OpDecorate %f32arr ArrayStride 4\n"
1304 "OpMemberDecorate %buf 0 Offset 0\n"
1306 + string(getComputeAsmCommonTypes()) +
1308 "%buf = OpTypeStruct %f32arr\n"
1309 "%bufptr = OpTypePointer Uniform %buf\n"
1310 "%indata1 = OpVariable %bufptr Uniform\n"
1311 "%indata2 = OpVariable %bufptr Uniform\n"
1312 "%outdata = OpVariable %bufptr Uniform\n"
1314 "%id = OpVariable %uvec3ptr Input\n"
1315 "%zero = OpConstant %i32 0\n"
1317 "%main = OpFunction %void None %voidf\n"
1318 "%label = OpLabel\n"
1319 "%idval = OpLoad %uvec3 %id\n"
1320 "%x = OpCompositeExtract %u32 %idval 0\n"
1321 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1322 "%inval1 = OpLoad %f32 %inloc1\n"
1323 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1324 "%inval2 = OpLoad %f32 %inloc2\n"
1325 "%rem = OpExtInst %f32 %std450 NMax %inval1 %inval2\n"
1326 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1327 " OpStore %outloc %rem\n"
1331 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1332 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1333 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1334 spec.numWorkGroups = IVec3(numElements, 1, 1);
1335 spec.verifyIO = &compareNMax;
1337 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1339 return group.release();
1342 bool compareNClamp (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1344 if (outputAllocs.size() != 1)
1347 const BufferSp& expectedOutput = expectedOutputs[0];
1348 std::vector<deUint8> data;
1349 expectedOutput->getBytes(data);
1351 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1352 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1354 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float) / 2; ++idx)
1356 const float e0 = expectedOutputAsFloat[idx * 2];
1357 const float e1 = expectedOutputAsFloat[idx * 2 + 1];
1358 const float res = outputAsFloat[idx];
1360 // For NClamp, we have two possible outcomes based on
1361 // whether NaNs are handled or not.
1362 // If either min or max value is NaN, the result is undefined,
1363 // so this test doesn't stress those. If the clamped value is
1364 // NaN, and NaNs are handled, the result is min; if NaNs are not
1365 // handled, they are big values that result in max.
1366 // If all three parameters are NaN, the result should be NaN.
1367 if (!((tcu::Float32(e0).isNaN() && tcu::Float32(res).isNaN()) ||
1368 (deFloatAbs(e0 - res) < 0.00001f) ||
1369 (deFloatAbs(e1 - res) < 0.00001f)))
1376 tcu::TestCaseGroup* createOpNClampGroup (tcu::TestContext& testCtx)
1378 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnclamp", "Test the OpNClamp instruction"));
1379 ComputeShaderSpec spec;
1380 de::Random rnd (deStringHash(group->getName()));
1381 const int numElements = 200;
1382 vector<float> inputFloats1 (numElements, 0);
1383 vector<float> inputFloats2 (numElements, 0);
1384 vector<float> inputFloats3 (numElements, 0);
1385 vector<float> outputFloats (numElements * 2, 0);
1387 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1388 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1389 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats3[0], numElements);
1391 for (size_t ndx = 0; ndx < numElements; ++ndx)
1393 // Results are only defined if max value is bigger than min value.
1394 if (inputFloats2[ndx] > inputFloats3[ndx])
1396 float t = inputFloats2[ndx];
1397 inputFloats2[ndx] = inputFloats3[ndx];
1398 inputFloats3[ndx] = t;
1401 // By default, do the clamp, setting both possible answers
1402 float defaultRes = std::min(std::max(inputFloats1[ndx], inputFloats2[ndx]), inputFloats3[ndx]);
1404 float maxResA = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1405 float maxResB = maxResA;
1407 // Alternate between the NaN cases
1410 inputFloats1[ndx] = TCU_NAN;
1411 // If NaN is handled, the result should be same as the clamp minimum.
1412 // If NaN is not handled, the result should clamp to the clamp maximum.
1413 maxResA = inputFloats2[ndx];
1414 maxResB = inputFloats3[ndx];
1418 // Not a NaN case - only one legal result.
1419 maxResA = defaultRes;
1420 maxResB = defaultRes;
1423 outputFloats[ndx * 2] = maxResA;
1424 outputFloats[ndx * 2 + 1] = maxResB;
1427 // Make the first case a full-NAN case.
1428 inputFloats1[0] = TCU_NAN;
1429 inputFloats2[0] = TCU_NAN;
1430 inputFloats3[0] = TCU_NAN;
1431 outputFloats[0] = TCU_NAN;
1432 outputFloats[1] = TCU_NAN;
1435 "OpCapability Shader\n"
1436 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1437 "OpMemoryModel Logical GLSL450\n"
1438 "OpEntryPoint GLCompute %main \"main\" %id\n"
1439 "OpExecutionMode %main LocalSize 1 1 1\n"
1441 "OpName %main \"main\"\n"
1442 "OpName %id \"gl_GlobalInvocationID\"\n"
1444 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1446 "OpDecorate %buf BufferBlock\n"
1447 "OpDecorate %indata1 DescriptorSet 0\n"
1448 "OpDecorate %indata1 Binding 0\n"
1449 "OpDecorate %indata2 DescriptorSet 0\n"
1450 "OpDecorate %indata2 Binding 1\n"
1451 "OpDecorate %indata3 DescriptorSet 0\n"
1452 "OpDecorate %indata3 Binding 2\n"
1453 "OpDecorate %outdata DescriptorSet 0\n"
1454 "OpDecorate %outdata Binding 3\n"
1455 "OpDecorate %f32arr ArrayStride 4\n"
1456 "OpMemberDecorate %buf 0 Offset 0\n"
1458 + string(getComputeAsmCommonTypes()) +
1460 "%buf = OpTypeStruct %f32arr\n"
1461 "%bufptr = OpTypePointer Uniform %buf\n"
1462 "%indata1 = OpVariable %bufptr Uniform\n"
1463 "%indata2 = OpVariable %bufptr Uniform\n"
1464 "%indata3 = OpVariable %bufptr Uniform\n"
1465 "%outdata = OpVariable %bufptr Uniform\n"
1467 "%id = OpVariable %uvec3ptr Input\n"
1468 "%zero = OpConstant %i32 0\n"
1470 "%main = OpFunction %void None %voidf\n"
1471 "%label = OpLabel\n"
1472 "%idval = OpLoad %uvec3 %id\n"
1473 "%x = OpCompositeExtract %u32 %idval 0\n"
1474 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1475 "%inval1 = OpLoad %f32 %inloc1\n"
1476 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1477 "%inval2 = OpLoad %f32 %inloc2\n"
1478 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1479 "%inval3 = OpLoad %f32 %inloc3\n"
1480 "%rem = OpExtInst %f32 %std450 NClamp %inval1 %inval2 %inval3\n"
1481 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1482 " OpStore %outloc %rem\n"
1486 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1487 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1488 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1489 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1490 spec.numWorkGroups = IVec3(numElements, 1, 1);
1491 spec.verifyIO = &compareNClamp;
1493 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1495 return group.release();
1498 tcu::TestCaseGroup* createOpSRemComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1500 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem", "Test the OpSRem instruction"));
1501 de::Random rnd (deStringHash(group->getName()));
1502 const int numElements = 200;
1504 const struct CaseParams
1507 const char* failMessage; // customized status message
1508 qpTestResult failResult; // override status on failure
1509 int op1Min, op1Max; // operand ranges
1513 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1514 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1516 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1518 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1520 const CaseParams& params = cases[caseNdx];
1521 ComputeShaderSpec spec;
1522 vector<deInt32> inputInts1 (numElements, 0);
1523 vector<deInt32> inputInts2 (numElements, 0);
1524 vector<deInt32> outputInts (numElements, 0);
1526 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1527 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1529 for (int ndx = 0; ndx < numElements; ++ndx)
1531 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1532 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1536 string(getComputeAsmShaderPreamble()) +
1538 "OpName %main \"main\"\n"
1539 "OpName %id \"gl_GlobalInvocationID\"\n"
1541 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1543 "OpDecorate %buf BufferBlock\n"
1544 "OpDecorate %indata1 DescriptorSet 0\n"
1545 "OpDecorate %indata1 Binding 0\n"
1546 "OpDecorate %indata2 DescriptorSet 0\n"
1547 "OpDecorate %indata2 Binding 1\n"
1548 "OpDecorate %outdata DescriptorSet 0\n"
1549 "OpDecorate %outdata Binding 2\n"
1550 "OpDecorate %i32arr ArrayStride 4\n"
1551 "OpMemberDecorate %buf 0 Offset 0\n"
1553 + string(getComputeAsmCommonTypes()) +
1555 "%buf = OpTypeStruct %i32arr\n"
1556 "%bufptr = OpTypePointer Uniform %buf\n"
1557 "%indata1 = OpVariable %bufptr Uniform\n"
1558 "%indata2 = OpVariable %bufptr Uniform\n"
1559 "%outdata = OpVariable %bufptr Uniform\n"
1561 "%id = OpVariable %uvec3ptr Input\n"
1562 "%zero = OpConstant %i32 0\n"
1564 "%main = OpFunction %void None %voidf\n"
1565 "%label = OpLabel\n"
1566 "%idval = OpLoad %uvec3 %id\n"
1567 "%x = OpCompositeExtract %u32 %idval 0\n"
1568 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1569 "%inval1 = OpLoad %i32 %inloc1\n"
1570 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1571 "%inval2 = OpLoad %i32 %inloc2\n"
1572 "%rem = OpSRem %i32 %inval1 %inval2\n"
1573 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1574 " OpStore %outloc %rem\n"
1578 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1579 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1580 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1581 spec.numWorkGroups = IVec3(numElements, 1, 1);
1582 spec.failResult = params.failResult;
1583 spec.failMessage = params.failMessage;
1585 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1588 return group.release();
1591 tcu::TestCaseGroup* createOpSRemComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1593 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem64", "Test the 64-bit OpSRem instruction"));
1594 de::Random rnd (deStringHash(group->getName()));
1595 const int numElements = 200;
1597 const struct CaseParams
1600 const char* failMessage; // customized status message
1601 qpTestResult failResult; // override status on failure
1605 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1606 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1608 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1610 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1612 const CaseParams& params = cases[caseNdx];
1613 ComputeShaderSpec spec;
1614 vector<deInt64> inputInts1 (numElements, 0);
1615 vector<deInt64> inputInts2 (numElements, 0);
1616 vector<deInt64> outputInts (numElements, 0);
1618 if (params.positive)
1620 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1621 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1625 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1626 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1629 for (int ndx = 0; ndx < numElements; ++ndx)
1631 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1632 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1636 "OpCapability Int64\n"
1638 + string(getComputeAsmShaderPreamble()) +
1640 "OpName %main \"main\"\n"
1641 "OpName %id \"gl_GlobalInvocationID\"\n"
1643 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1645 "OpDecorate %buf BufferBlock\n"
1646 "OpDecorate %indata1 DescriptorSet 0\n"
1647 "OpDecorate %indata1 Binding 0\n"
1648 "OpDecorate %indata2 DescriptorSet 0\n"
1649 "OpDecorate %indata2 Binding 1\n"
1650 "OpDecorate %outdata DescriptorSet 0\n"
1651 "OpDecorate %outdata Binding 2\n"
1652 "OpDecorate %i64arr ArrayStride 8\n"
1653 "OpMemberDecorate %buf 0 Offset 0\n"
1655 + string(getComputeAsmCommonTypes())
1656 + string(getComputeAsmCommonInt64Types()) +
1658 "%buf = OpTypeStruct %i64arr\n"
1659 "%bufptr = OpTypePointer Uniform %buf\n"
1660 "%indata1 = OpVariable %bufptr Uniform\n"
1661 "%indata2 = OpVariable %bufptr Uniform\n"
1662 "%outdata = OpVariable %bufptr Uniform\n"
1664 "%id = OpVariable %uvec3ptr Input\n"
1665 "%zero = OpConstant %i64 0\n"
1667 "%main = OpFunction %void None %voidf\n"
1668 "%label = OpLabel\n"
1669 "%idval = OpLoad %uvec3 %id\n"
1670 "%x = OpCompositeExtract %u32 %idval 0\n"
1671 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1672 "%inval1 = OpLoad %i64 %inloc1\n"
1673 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1674 "%inval2 = OpLoad %i64 %inloc2\n"
1675 "%rem = OpSRem %i64 %inval1 %inval2\n"
1676 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1677 " OpStore %outloc %rem\n"
1681 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1682 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1683 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1684 spec.numWorkGroups = IVec3(numElements, 1, 1);
1685 spec.failResult = params.failResult;
1686 spec.failMessage = params.failMessage;
1688 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1691 return group.release();
1694 tcu::TestCaseGroup* createOpSModComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1696 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod", "Test the OpSMod instruction"));
1697 de::Random rnd (deStringHash(group->getName()));
1698 const int numElements = 200;
1700 const struct CaseParams
1703 const char* failMessage; // customized status message
1704 qpTestResult failResult; // override status on failure
1705 int op1Min, op1Max; // operand ranges
1709 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1710 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1712 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1714 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1716 const CaseParams& params = cases[caseNdx];
1718 ComputeShaderSpec spec;
1719 vector<deInt32> inputInts1 (numElements, 0);
1720 vector<deInt32> inputInts2 (numElements, 0);
1721 vector<deInt32> outputInts (numElements, 0);
1723 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1724 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1726 for (int ndx = 0; ndx < numElements; ++ndx)
1728 deInt32 rem = inputInts1[ndx] % inputInts2[ndx];
1731 outputInts[ndx] = 0;
1733 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1735 // They have the same sign
1736 outputInts[ndx] = rem;
1740 // They have opposite sign. The remainder operation takes the
1741 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1742 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1743 // the result has the correct sign and that it is still
1744 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1746 // See also http://mathforum.org/library/drmath/view/52343.html
1747 outputInts[ndx] = rem + inputInts2[ndx];
1752 string(getComputeAsmShaderPreamble()) +
1754 "OpName %main \"main\"\n"
1755 "OpName %id \"gl_GlobalInvocationID\"\n"
1757 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1759 "OpDecorate %buf BufferBlock\n"
1760 "OpDecorate %indata1 DescriptorSet 0\n"
1761 "OpDecorate %indata1 Binding 0\n"
1762 "OpDecorate %indata2 DescriptorSet 0\n"
1763 "OpDecorate %indata2 Binding 1\n"
1764 "OpDecorate %outdata DescriptorSet 0\n"
1765 "OpDecorate %outdata Binding 2\n"
1766 "OpDecorate %i32arr ArrayStride 4\n"
1767 "OpMemberDecorate %buf 0 Offset 0\n"
1769 + string(getComputeAsmCommonTypes()) +
1771 "%buf = OpTypeStruct %i32arr\n"
1772 "%bufptr = OpTypePointer Uniform %buf\n"
1773 "%indata1 = OpVariable %bufptr Uniform\n"
1774 "%indata2 = OpVariable %bufptr Uniform\n"
1775 "%outdata = OpVariable %bufptr Uniform\n"
1777 "%id = OpVariable %uvec3ptr Input\n"
1778 "%zero = OpConstant %i32 0\n"
1780 "%main = OpFunction %void None %voidf\n"
1781 "%label = OpLabel\n"
1782 "%idval = OpLoad %uvec3 %id\n"
1783 "%x = OpCompositeExtract %u32 %idval 0\n"
1784 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1785 "%inval1 = OpLoad %i32 %inloc1\n"
1786 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1787 "%inval2 = OpLoad %i32 %inloc2\n"
1788 "%rem = OpSMod %i32 %inval1 %inval2\n"
1789 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1790 " OpStore %outloc %rem\n"
1794 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1795 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1796 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1797 spec.numWorkGroups = IVec3(numElements, 1, 1);
1798 spec.failResult = params.failResult;
1799 spec.failMessage = params.failMessage;
1801 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1804 return group.release();
1807 tcu::TestCaseGroup* createOpSModComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1809 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod64", "Test the OpSMod instruction"));
1810 de::Random rnd (deStringHash(group->getName()));
1811 const int numElements = 200;
1813 const struct CaseParams
1816 const char* failMessage; // customized status message
1817 qpTestResult failResult; // override status on failure
1821 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1822 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1824 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1826 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1828 const CaseParams& params = cases[caseNdx];
1830 ComputeShaderSpec spec;
1831 vector<deInt64> inputInts1 (numElements, 0);
1832 vector<deInt64> inputInts2 (numElements, 0);
1833 vector<deInt64> outputInts (numElements, 0);
1836 if (params.positive)
1838 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1839 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1843 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1844 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1847 for (int ndx = 0; ndx < numElements; ++ndx)
1849 deInt64 rem = inputInts1[ndx] % inputInts2[ndx];
1852 outputInts[ndx] = 0;
1854 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1856 // They have the same sign
1857 outputInts[ndx] = rem;
1861 // They have opposite sign. The remainder operation takes the
1862 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1863 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1864 // the result has the correct sign and that it is still
1865 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1867 // See also http://mathforum.org/library/drmath/view/52343.html
1868 outputInts[ndx] = rem + inputInts2[ndx];
1873 "OpCapability Int64\n"
1875 + string(getComputeAsmShaderPreamble()) +
1877 "OpName %main \"main\"\n"
1878 "OpName %id \"gl_GlobalInvocationID\"\n"
1880 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1882 "OpDecorate %buf BufferBlock\n"
1883 "OpDecorate %indata1 DescriptorSet 0\n"
1884 "OpDecorate %indata1 Binding 0\n"
1885 "OpDecorate %indata2 DescriptorSet 0\n"
1886 "OpDecorate %indata2 Binding 1\n"
1887 "OpDecorate %outdata DescriptorSet 0\n"
1888 "OpDecorate %outdata Binding 2\n"
1889 "OpDecorate %i64arr ArrayStride 8\n"
1890 "OpMemberDecorate %buf 0 Offset 0\n"
1892 + string(getComputeAsmCommonTypes())
1893 + string(getComputeAsmCommonInt64Types()) +
1895 "%buf = OpTypeStruct %i64arr\n"
1896 "%bufptr = OpTypePointer Uniform %buf\n"
1897 "%indata1 = OpVariable %bufptr Uniform\n"
1898 "%indata2 = OpVariable %bufptr Uniform\n"
1899 "%outdata = OpVariable %bufptr Uniform\n"
1901 "%id = OpVariable %uvec3ptr Input\n"
1902 "%zero = OpConstant %i64 0\n"
1904 "%main = OpFunction %void None %voidf\n"
1905 "%label = OpLabel\n"
1906 "%idval = OpLoad %uvec3 %id\n"
1907 "%x = OpCompositeExtract %u32 %idval 0\n"
1908 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1909 "%inval1 = OpLoad %i64 %inloc1\n"
1910 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1911 "%inval2 = OpLoad %i64 %inloc2\n"
1912 "%rem = OpSMod %i64 %inval1 %inval2\n"
1913 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1914 " OpStore %outloc %rem\n"
1918 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1919 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1920 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1921 spec.numWorkGroups = IVec3(numElements, 1, 1);
1922 spec.failResult = params.failResult;
1923 spec.failMessage = params.failMessage;
1925 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1928 return group.release();
1931 // Copy contents in the input buffer to the output buffer.
1932 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
1934 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
1935 de::Random rnd (deStringHash(group->getName()));
1936 const int numElements = 100;
1938 // 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.
1939 ComputeShaderSpec spec1;
1940 vector<Vec4> inputFloats1 (numElements);
1941 vector<Vec4> outputFloats1 (numElements);
1943 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
1945 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1946 floorAll(inputFloats1);
1948 for (size_t ndx = 0; ndx < numElements; ++ndx)
1949 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
1952 string(getComputeAsmShaderPreamble()) +
1954 "OpName %main \"main\"\n"
1955 "OpName %id \"gl_GlobalInvocationID\"\n"
1957 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1958 "OpDecorate %vec4arr ArrayStride 16\n"
1960 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1962 "%vec4 = OpTypeVector %f32 4\n"
1963 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
1964 "%vec4ptr_f = OpTypePointer Function %vec4\n"
1965 "%vec4arr = OpTypeRuntimeArray %vec4\n"
1966 "%buf = OpTypeStruct %vec4arr\n"
1967 "%bufptr = OpTypePointer Uniform %buf\n"
1968 "%indata = OpVariable %bufptr Uniform\n"
1969 "%outdata = OpVariable %bufptr Uniform\n"
1971 "%id = OpVariable %uvec3ptr Input\n"
1972 "%zero = OpConstant %i32 0\n"
1973 "%c_f_0 = OpConstant %f32 0.\n"
1974 "%c_f_0_5 = OpConstant %f32 0.5\n"
1975 "%c_f_1_5 = OpConstant %f32 1.5\n"
1976 "%c_f_2_5 = OpConstant %f32 2.5\n"
1977 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
1979 "%main = OpFunction %void None %voidf\n"
1980 "%label = OpLabel\n"
1981 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
1982 "%idval = OpLoad %uvec3 %id\n"
1983 "%x = OpCompositeExtract %u32 %idval 0\n"
1984 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
1985 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
1986 " OpCopyMemory %v_vec4 %inloc\n"
1987 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
1988 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
1989 " OpStore %outloc %add\n"
1993 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
1994 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
1995 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1997 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
1999 // The following case copies a float[100] variable from the input buffer to the output buffer.
2000 ComputeShaderSpec spec2;
2001 vector<float> inputFloats2 (numElements);
2002 vector<float> outputFloats2 (numElements);
2004 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
2006 for (size_t ndx = 0; ndx < numElements; ++ndx)
2007 outputFloats2[ndx] = inputFloats2[ndx];
2010 string(getComputeAsmShaderPreamble()) +
2012 "OpName %main \"main\"\n"
2013 "OpName %id \"gl_GlobalInvocationID\"\n"
2015 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2016 "OpDecorate %f32arr100 ArrayStride 4\n"
2018 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2020 "%hundred = OpConstant %u32 100\n"
2021 "%f32arr100 = OpTypeArray %f32 %hundred\n"
2022 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
2023 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
2024 "%buf = OpTypeStruct %f32arr100\n"
2025 "%bufptr = OpTypePointer Uniform %buf\n"
2026 "%indata = OpVariable %bufptr Uniform\n"
2027 "%outdata = OpVariable %bufptr Uniform\n"
2029 "%id = OpVariable %uvec3ptr Input\n"
2030 "%zero = OpConstant %i32 0\n"
2032 "%main = OpFunction %void None %voidf\n"
2033 "%label = OpLabel\n"
2034 "%var = OpVariable %f32arr100ptr_f Function\n"
2035 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
2036 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
2037 " OpCopyMemory %var %inarr\n"
2038 " OpCopyMemory %outarr %var\n"
2042 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2043 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2044 spec2.numWorkGroups = IVec3(1, 1, 1);
2046 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
2048 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
2049 ComputeShaderSpec spec3;
2050 vector<float> inputFloats3 (16);
2051 vector<float> outputFloats3 (16);
2053 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
2055 for (size_t ndx = 0; ndx < 16; ++ndx)
2056 outputFloats3[ndx] = inputFloats3[ndx];
2059 string(getComputeAsmShaderPreamble()) +
2061 "OpName %main \"main\"\n"
2062 "OpName %id \"gl_GlobalInvocationID\"\n"
2064 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2065 "OpMemberDecorate %buf 0 Offset 0\n"
2066 "OpMemberDecorate %buf 1 Offset 16\n"
2067 "OpMemberDecorate %buf 2 Offset 32\n"
2068 "OpMemberDecorate %buf 3 Offset 48\n"
2070 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2072 "%vec4 = OpTypeVector %f32 4\n"
2073 "%buf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
2074 "%bufptr = OpTypePointer Uniform %buf\n"
2075 "%indata = OpVariable %bufptr Uniform\n"
2076 "%outdata = OpVariable %bufptr Uniform\n"
2077 "%vec4stptr = OpTypePointer Function %buf\n"
2079 "%id = OpVariable %uvec3ptr Input\n"
2080 "%zero = OpConstant %i32 0\n"
2082 "%main = OpFunction %void None %voidf\n"
2083 "%label = OpLabel\n"
2084 "%var = OpVariable %vec4stptr Function\n"
2085 " OpCopyMemory %var %indata\n"
2086 " OpCopyMemory %outdata %var\n"
2090 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2091 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
2092 spec3.numWorkGroups = IVec3(1, 1, 1);
2094 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
2096 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
2097 ComputeShaderSpec spec4;
2098 vector<float> inputFloats4 (numElements);
2099 vector<float> outputFloats4 (numElements);
2101 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
2103 for (size_t ndx = 0; ndx < numElements; ++ndx)
2104 outputFloats4[ndx] = -inputFloats4[ndx];
2107 string(getComputeAsmShaderPreamble()) +
2109 "OpName %main \"main\"\n"
2110 "OpName %id \"gl_GlobalInvocationID\"\n"
2112 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2114 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2116 "%f32ptr_f = OpTypePointer Function %f32\n"
2117 "%id = OpVariable %uvec3ptr Input\n"
2118 "%zero = OpConstant %i32 0\n"
2120 "%main = OpFunction %void None %voidf\n"
2121 "%label = OpLabel\n"
2122 "%var = OpVariable %f32ptr_f Function\n"
2123 "%idval = OpLoad %uvec3 %id\n"
2124 "%x = OpCompositeExtract %u32 %idval 0\n"
2125 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2126 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2127 " OpCopyMemory %var %inloc\n"
2128 "%val = OpLoad %f32 %var\n"
2129 "%neg = OpFNegate %f32 %val\n"
2130 " OpStore %outloc %neg\n"
2134 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2135 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
2136 spec4.numWorkGroups = IVec3(numElements, 1, 1);
2138 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
2140 return group.release();
2143 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
2145 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
2146 ComputeShaderSpec spec;
2147 de::Random rnd (deStringHash(group->getName()));
2148 const int numElements = 100;
2149 vector<float> inputFloats (numElements, 0);
2150 vector<float> outputFloats (numElements, 0);
2152 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
2154 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2155 floorAll(inputFloats);
2157 for (size_t ndx = 0; ndx < numElements; ++ndx)
2158 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
2161 string(getComputeAsmShaderPreamble()) +
2163 "OpName %main \"main\"\n"
2164 "OpName %id \"gl_GlobalInvocationID\"\n"
2166 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2168 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2170 "%fmat = OpTypeMatrix %fvec3 3\n"
2171 "%three = OpConstant %u32 3\n"
2172 "%farr = OpTypeArray %f32 %three\n"
2173 "%fst = OpTypeStruct %f32 %f32\n"
2175 + string(getComputeAsmInputOutputBuffer()) +
2177 "%id = OpVariable %uvec3ptr Input\n"
2178 "%zero = OpConstant %i32 0\n"
2179 "%c_f = OpConstant %f32 1.5\n"
2180 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
2181 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
2182 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
2183 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
2185 "%main = OpFunction %void None %voidf\n"
2186 "%label = OpLabel\n"
2187 "%c_f_copy = OpCopyObject %f32 %c_f\n"
2188 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
2189 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
2190 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
2191 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
2192 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
2193 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
2194 "%farr_elem = OpCompositeExtract %f32 %c_farr_copy 2\n"
2195 "%fst_elem = OpCompositeExtract %f32 %c_fst_copy 1\n"
2196 // Add up. 1.5 * 5 = 7.5.
2197 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
2198 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
2199 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
2200 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
2202 "%idval = OpLoad %uvec3 %id\n"
2203 "%x = OpCompositeExtract %u32 %idval 0\n"
2204 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2205 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2206 "%inval = OpLoad %f32 %inloc\n"
2207 "%add = OpFAdd %f32 %add4 %inval\n"
2208 " OpStore %outloc %add\n"
2211 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2212 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2213 spec.numWorkGroups = IVec3(numElements, 1, 1);
2215 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
2217 return group.release();
2219 // Assembly code used for testing OpUnreachable is based on GLSL source code:
2223 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2224 // float elements[];
2226 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2227 // float elements[];
2230 // void not_called_func() {
2231 // // place OpUnreachable here
2234 // uint modulo4(uint val) {
2235 // switch (val % uint(4)) {
2236 // case 0: return 3;
2237 // case 1: return 2;
2238 // case 2: return 1;
2239 // case 3: return 0;
2240 // default: return 100; // place OpUnreachable here
2246 // // place OpUnreachable here
2250 // uint x = gl_GlobalInvocationID.x;
2251 // if (const5() > modulo4(1000)) {
2252 // output_data.elements[x] = -input_data.elements[x];
2254 // // place OpUnreachable here
2255 // output_data.elements[x] = input_data.elements[x];
2259 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
2261 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
2262 ComputeShaderSpec spec;
2263 de::Random rnd (deStringHash(group->getName()));
2264 const int numElements = 100;
2265 vector<float> positiveFloats (numElements, 0);
2266 vector<float> negativeFloats (numElements, 0);
2268 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2270 for (size_t ndx = 0; ndx < numElements; ++ndx)
2271 negativeFloats[ndx] = -positiveFloats[ndx];
2274 string(getComputeAsmShaderPreamble()) +
2276 "OpSource GLSL 430\n"
2277 "OpName %main \"main\"\n"
2278 "OpName %func_not_called_func \"not_called_func(\"\n"
2279 "OpName %func_modulo4 \"modulo4(u1;\"\n"
2280 "OpName %func_const5 \"const5(\"\n"
2281 "OpName %id \"gl_GlobalInvocationID\"\n"
2283 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2285 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2287 "%u32ptr = OpTypePointer Function %u32\n"
2288 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
2289 "%unitf = OpTypeFunction %u32\n"
2291 "%id = OpVariable %uvec3ptr Input\n"
2292 "%zero = OpConstant %u32 0\n"
2293 "%one = OpConstant %u32 1\n"
2294 "%two = OpConstant %u32 2\n"
2295 "%three = OpConstant %u32 3\n"
2296 "%four = OpConstant %u32 4\n"
2297 "%five = OpConstant %u32 5\n"
2298 "%hundred = OpConstant %u32 100\n"
2299 "%thousand = OpConstant %u32 1000\n"
2301 + string(getComputeAsmInputOutputBuffer()) +
2304 "%main = OpFunction %void None %voidf\n"
2305 "%main_entry = OpLabel\n"
2306 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
2307 "%idval = OpLoad %uvec3 %id\n"
2308 "%x = OpCompositeExtract %u32 %idval 0\n"
2309 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2310 "%inval = OpLoad %f32 %inloc\n"
2311 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2312 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
2313 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
2314 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
2315 " OpSelectionMerge %if_end None\n"
2316 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2317 "%if_true = OpLabel\n"
2318 "%negate = OpFNegate %f32 %inval\n"
2319 " OpStore %outloc %negate\n"
2320 " OpBranch %if_end\n"
2321 "%if_false = OpLabel\n"
2322 " OpUnreachable\n" // Unreachable else branch for if statement
2323 "%if_end = OpLabel\n"
2327 // not_called_function()
2328 "%func_not_called_func = OpFunction %void None %voidf\n"
2329 "%not_called_func_entry = OpLabel\n"
2330 " OpUnreachable\n" // Unreachable entry block in not called static function
2334 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
2335 "%valptr = OpFunctionParameter %u32ptr\n"
2336 "%modulo4_entry = OpLabel\n"
2337 "%val = OpLoad %u32 %valptr\n"
2338 "%modulo = OpUMod %u32 %val %four\n"
2339 " OpSelectionMerge %switch_merge None\n"
2340 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
2341 "%case0 = OpLabel\n"
2342 " OpReturnValue %three\n"
2343 "%case1 = OpLabel\n"
2344 " OpReturnValue %two\n"
2345 "%case2 = OpLabel\n"
2346 " OpReturnValue %one\n"
2347 "%case3 = OpLabel\n"
2348 " OpReturnValue %zero\n"
2349 "%default = OpLabel\n"
2350 " OpUnreachable\n" // Unreachable default case for switch statement
2351 "%switch_merge = OpLabel\n"
2352 " OpUnreachable\n" // Unreachable merge block for switch statement
2356 "%func_const5 = OpFunction %u32 None %unitf\n"
2357 "%const5_entry = OpLabel\n"
2358 " OpReturnValue %five\n"
2359 "%unreachable = OpLabel\n"
2360 " OpUnreachable\n" // Unreachable block in function
2362 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2363 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2364 spec.numWorkGroups = IVec3(numElements, 1, 1);
2366 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
2368 return group.release();
2371 // Assembly code used for testing decoration group is based on GLSL source code:
2375 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
2376 // float elements[];
2378 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
2379 // float elements[];
2381 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
2382 // float elements[];
2384 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
2385 // float elements[];
2387 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
2388 // float elements[];
2390 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
2391 // float elements[];
2395 // uint x = gl_GlobalInvocationID.x;
2396 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
2398 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
2400 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
2401 ComputeShaderSpec spec;
2402 de::Random rnd (deStringHash(group->getName()));
2403 const int numElements = 100;
2404 vector<float> inputFloats0 (numElements, 0);
2405 vector<float> inputFloats1 (numElements, 0);
2406 vector<float> inputFloats2 (numElements, 0);
2407 vector<float> inputFloats3 (numElements, 0);
2408 vector<float> inputFloats4 (numElements, 0);
2409 vector<float> outputFloats (numElements, 0);
2411 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
2412 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
2413 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
2414 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
2415 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
2417 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2418 floorAll(inputFloats0);
2419 floorAll(inputFloats1);
2420 floorAll(inputFloats2);
2421 floorAll(inputFloats3);
2422 floorAll(inputFloats4);
2424 for (size_t ndx = 0; ndx < numElements; ++ndx)
2425 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
2428 string(getComputeAsmShaderPreamble()) +
2430 "OpSource GLSL 430\n"
2431 "OpName %main \"main\"\n"
2432 "OpName %id \"gl_GlobalInvocationID\"\n"
2434 // Not using group decoration on variable.
2435 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2436 // Not using group decoration on type.
2437 "OpDecorate %f32arr ArrayStride 4\n"
2439 "OpDecorate %groups BufferBlock\n"
2440 "OpDecorate %groupm Offset 0\n"
2441 "%groups = OpDecorationGroup\n"
2442 "%groupm = OpDecorationGroup\n"
2444 // Group decoration on multiple structs.
2445 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
2446 // Group decoration on multiple struct members.
2447 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
2449 "OpDecorate %group1 DescriptorSet 0\n"
2450 "OpDecorate %group3 DescriptorSet 0\n"
2451 "OpDecorate %group3 NonWritable\n"
2452 "OpDecorate %group3 Restrict\n"
2453 "%group0 = OpDecorationGroup\n"
2454 "%group1 = OpDecorationGroup\n"
2455 "%group3 = OpDecorationGroup\n"
2457 // Applying the same decoration group multiple times.
2458 "OpGroupDecorate %group1 %outdata\n"
2459 "OpGroupDecorate %group1 %outdata\n"
2460 "OpGroupDecorate %group1 %outdata\n"
2461 "OpDecorate %outdata DescriptorSet 0\n"
2462 "OpDecorate %outdata Binding 5\n"
2463 // Applying decoration group containing nothing.
2464 "OpGroupDecorate %group0 %indata0\n"
2465 "OpDecorate %indata0 DescriptorSet 0\n"
2466 "OpDecorate %indata0 Binding 0\n"
2467 // Applying decoration group containing one decoration.
2468 "OpGroupDecorate %group1 %indata1\n"
2469 "OpDecorate %indata1 Binding 1\n"
2470 // Applying decoration group containing multiple decorations.
2471 "OpGroupDecorate %group3 %indata2 %indata3\n"
2472 "OpDecorate %indata2 Binding 2\n"
2473 "OpDecorate %indata3 Binding 3\n"
2474 // Applying multiple decoration groups (with overlapping).
2475 "OpGroupDecorate %group0 %indata4\n"
2476 "OpGroupDecorate %group1 %indata4\n"
2477 "OpGroupDecorate %group3 %indata4\n"
2478 "OpDecorate %indata4 Binding 4\n"
2480 + string(getComputeAsmCommonTypes()) +
2482 "%id = OpVariable %uvec3ptr Input\n"
2483 "%zero = OpConstant %i32 0\n"
2485 "%outbuf = OpTypeStruct %f32arr\n"
2486 "%outbufptr = OpTypePointer Uniform %outbuf\n"
2487 "%outdata = OpVariable %outbufptr Uniform\n"
2488 "%inbuf0 = OpTypeStruct %f32arr\n"
2489 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
2490 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
2491 "%inbuf1 = OpTypeStruct %f32arr\n"
2492 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
2493 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
2494 "%inbuf2 = OpTypeStruct %f32arr\n"
2495 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
2496 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
2497 "%inbuf3 = OpTypeStruct %f32arr\n"
2498 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
2499 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
2500 "%inbuf4 = OpTypeStruct %f32arr\n"
2501 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
2502 "%indata4 = OpVariable %inbufptr Uniform\n"
2504 "%main = OpFunction %void None %voidf\n"
2505 "%label = OpLabel\n"
2506 "%idval = OpLoad %uvec3 %id\n"
2507 "%x = OpCompositeExtract %u32 %idval 0\n"
2508 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
2509 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
2510 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
2511 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
2512 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
2513 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2514 "%inval0 = OpLoad %f32 %inloc0\n"
2515 "%inval1 = OpLoad %f32 %inloc1\n"
2516 "%inval2 = OpLoad %f32 %inloc2\n"
2517 "%inval3 = OpLoad %f32 %inloc3\n"
2518 "%inval4 = OpLoad %f32 %inloc4\n"
2519 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
2520 "%add1 = OpFAdd %f32 %add0 %inval2\n"
2521 "%add2 = OpFAdd %f32 %add1 %inval3\n"
2522 "%add = OpFAdd %f32 %add2 %inval4\n"
2523 " OpStore %outloc %add\n"
2526 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
2527 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
2528 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2529 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2530 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2531 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2532 spec.numWorkGroups = IVec3(numElements, 1, 1);
2534 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
2536 return group.release();
2539 struct SpecConstantTwoIntCase
2541 const char* caseName;
2542 const char* scDefinition0;
2543 const char* scDefinition1;
2544 const char* scResultType;
2545 const char* scOperation;
2546 deInt32 scActualValue0;
2547 deInt32 scActualValue1;
2548 const char* resultOperation;
2549 vector<deInt32> expectedOutput;
2551 SpecConstantTwoIntCase (const char* name,
2552 const char* definition0,
2553 const char* definition1,
2554 const char* resultType,
2555 const char* operation,
2558 const char* resultOp,
2559 const vector<deInt32>& output)
2561 , scDefinition0 (definition0)
2562 , scDefinition1 (definition1)
2563 , scResultType (resultType)
2564 , scOperation (operation)
2565 , scActualValue0 (value0)
2566 , scActualValue1 (value1)
2567 , resultOperation (resultOp)
2568 , expectedOutput (output) {}
2571 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
2573 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
2574 vector<SpecConstantTwoIntCase> cases;
2575 de::Random rnd (deStringHash(group->getName()));
2576 const int numElements = 100;
2577 vector<deInt32> inputInts (numElements, 0);
2578 vector<deInt32> outputInts1 (numElements, 0);
2579 vector<deInt32> outputInts2 (numElements, 0);
2580 vector<deInt32> outputInts3 (numElements, 0);
2581 vector<deInt32> outputInts4 (numElements, 0);
2582 const StringTemplate shaderTemplate (
2583 "${CAPABILITIES:opt}"
2584 + string(getComputeAsmShaderPreamble()) +
2586 "OpName %main \"main\"\n"
2587 "OpName %id \"gl_GlobalInvocationID\"\n"
2589 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2590 "OpDecorate %sc_0 SpecId 0\n"
2591 "OpDecorate %sc_1 SpecId 1\n"
2592 "OpDecorate %i32arr ArrayStride 4\n"
2594 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2596 "${OPTYPE_DEFINITIONS:opt}"
2597 "%buf = OpTypeStruct %i32arr\n"
2598 "%bufptr = OpTypePointer Uniform %buf\n"
2599 "%indata = OpVariable %bufptr Uniform\n"
2600 "%outdata = OpVariable %bufptr Uniform\n"
2602 "%id = OpVariable %uvec3ptr Input\n"
2603 "%zero = OpConstant %i32 0\n"
2605 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
2606 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
2607 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
2609 "%main = OpFunction %void None %voidf\n"
2610 "%label = OpLabel\n"
2611 "${TYPE_CONVERT:opt}"
2612 "%idval = OpLoad %uvec3 %id\n"
2613 "%x = OpCompositeExtract %u32 %idval 0\n"
2614 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2615 "%inval = OpLoad %i32 %inloc\n"
2616 "%final = ${GEN_RESULT}\n"
2617 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2618 " OpStore %outloc %final\n"
2620 " OpFunctionEnd\n");
2622 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
2624 for (size_t ndx = 0; ndx < numElements; ++ndx)
2626 outputInts1[ndx] = inputInts[ndx] + 42;
2627 outputInts2[ndx] = inputInts[ndx];
2628 outputInts3[ndx] = inputInts[ndx] - 11200;
2629 outputInts4[ndx] = inputInts[ndx] + 1;
2632 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
2633 const char addSc32ToInput[] = "OpIAdd %i32 %inval %sc_final32";
2634 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
2635 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
2637 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
2638 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
2639 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
2640 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
2641 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
2642 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2643 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2644 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
2645 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
2646 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
2647 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
2648 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2649 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2650 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
2651 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
2652 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
2653 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2654 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
2655 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
2656 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
2657 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2658 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
2659 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
2660 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2661 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2662 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2663 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2664 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
2665 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
2666 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
2667 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
2668 cases.push_back(SpecConstantTwoIntCase("sconvert", " %i32 0", " %i32 0", "%i16", "SConvert %sc_0", -11200, 0, addSc32ToInput, outputInts3));
2669 // -969998336 stored as 32-bit two's complement is the binary representation of -11200 as IEEE-754 Float
2670 cases.push_back(SpecConstantTwoIntCase("fconvert", " %f32 0", " %f32 0", "%f64", "FConvert %sc_0", -969998336, 0, addSc32ToInput, outputInts3));
2672 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2674 map<string, string> specializations;
2675 ComputeShaderSpec spec;
2676 ComputeTestFeatures features = COMPUTE_TEST_USES_NONE;
2678 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
2679 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
2680 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
2681 specializations["SC_OP"] = cases[caseNdx].scOperation;
2682 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
2684 // Special SPIR-V code for SConvert-case
2685 if (strcmp(cases[caseNdx].caseName, "sconvert") == 0)
2687 features = COMPUTE_TEST_USES_INT16;
2688 specializations["CAPABILITIES"] = "OpCapability Int16\n"; // Adds 16-bit integer capability
2689 specializations["OPTYPE_DEFINITIONS"] = "%i16 = OpTypeInt 16 1\n"; // Adds 16-bit integer type
2690 specializations["TYPE_CONVERT"] = "%sc_final32 = OpSConvert %i32 %sc_final\n"; // Converts 16-bit integer to 32-bit integer
2693 // Special SPIR-V code for FConvert-case
2694 if (strcmp(cases[caseNdx].caseName, "fconvert") == 0)
2696 features = COMPUTE_TEST_USES_FLOAT64;
2697 specializations["CAPABILITIES"] = "OpCapability Float64\n"; // Adds 64-bit float capability
2698 specializations["OPTYPE_DEFINITIONS"] = "%f64 = OpTypeFloat 64\n"; // Adds 64-bit float type
2699 specializations["TYPE_CONVERT"] = "%sc_final32 = OpConvertFToS %i32 %sc_final\n"; // Converts 64-bit float to 32-bit integer
2702 spec.assembly = shaderTemplate.specialize(specializations);
2703 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2704 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
2705 spec.numWorkGroups = IVec3(numElements, 1, 1);
2706 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
2707 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
2709 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec, features));
2712 ComputeShaderSpec spec;
2715 string(getComputeAsmShaderPreamble()) +
2717 "OpName %main \"main\"\n"
2718 "OpName %id \"gl_GlobalInvocationID\"\n"
2720 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2721 "OpDecorate %sc_0 SpecId 0\n"
2722 "OpDecorate %sc_1 SpecId 1\n"
2723 "OpDecorate %sc_2 SpecId 2\n"
2724 "OpDecorate %i32arr ArrayStride 4\n"
2726 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2728 "%ivec3 = OpTypeVector %i32 3\n"
2729 "%buf = OpTypeStruct %i32arr\n"
2730 "%bufptr = OpTypePointer Uniform %buf\n"
2731 "%indata = OpVariable %bufptr Uniform\n"
2732 "%outdata = OpVariable %bufptr Uniform\n"
2734 "%id = OpVariable %uvec3ptr Input\n"
2735 "%zero = OpConstant %i32 0\n"
2736 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
2737 "%vec3_undef = OpUndef %ivec3\n"
2739 "%sc_0 = OpSpecConstant %i32 0\n"
2740 "%sc_1 = OpSpecConstant %i32 0\n"
2741 "%sc_2 = OpSpecConstant %i32 0\n"
2742 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
2743 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
2744 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
2745 "%sc_vec3_0_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
2746 "%sc_vec3_1_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
2747 "%sc_vec3_2_s = OpSpecConstantOp %ivec3 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
2748 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
2749 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
2750 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
2751 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
2752 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
2753 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
2754 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
2756 "%main = OpFunction %void None %voidf\n"
2757 "%label = OpLabel\n"
2758 "%idval = OpLoad %uvec3 %id\n"
2759 "%x = OpCompositeExtract %u32 %idval 0\n"
2760 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2761 "%inval = OpLoad %i32 %inloc\n"
2762 "%final = OpIAdd %i32 %inval %sc_final\n"
2763 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2764 " OpStore %outloc %final\n"
2767 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2768 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
2769 spec.numWorkGroups = IVec3(numElements, 1, 1);
2770 spec.specConstants.push_back(123);
2771 spec.specConstants.push_back(56);
2772 spec.specConstants.push_back(-77);
2774 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
2776 return group.release();
2779 void createOpPhiVartypeTests (de::MovePtr<tcu::TestCaseGroup>& group, tcu::TestContext& testCtx)
2781 ComputeShaderSpec specInt;
2782 ComputeShaderSpec specFloat;
2783 ComputeShaderSpec specVec3;
2784 ComputeShaderSpec specMat4;
2785 ComputeShaderSpec specArray;
2786 ComputeShaderSpec specStruct;
2787 de::Random rnd (deStringHash(group->getName()));
2788 const int numElements = 100;
2789 vector<float> inputFloats (numElements, 0);
2790 vector<float> outputFloats (numElements, 0);
2792 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
2794 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2795 floorAll(inputFloats);
2797 for (size_t ndx = 0; ndx < numElements; ++ndx)
2799 // Just check if the value is positive or not
2800 outputFloats[ndx] = (inputFloats[ndx] > 0) ? 1.0f : -1.0f;
2803 // All of the tests are of the form:
2807 // if (inputdata > 0)
2814 specFloat.assembly =
2815 string(getComputeAsmShaderPreamble()) +
2817 "OpSource GLSL 430\n"
2818 "OpName %main \"main\"\n"
2819 "OpName %id \"gl_GlobalInvocationID\"\n"
2821 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2823 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2825 "%id = OpVariable %uvec3ptr Input\n"
2826 "%zero = OpConstant %i32 0\n"
2827 "%float_0 = OpConstant %f32 0.0\n"
2828 "%float_1 = OpConstant %f32 1.0\n"
2829 "%float_n1 = OpConstant %f32 -1.0\n"
2831 "%main = OpFunction %void None %voidf\n"
2832 "%entry = OpLabel\n"
2833 "%idval = OpLoad %uvec3 %id\n"
2834 "%x = OpCompositeExtract %u32 %idval 0\n"
2835 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2836 "%inval = OpLoad %f32 %inloc\n"
2838 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2839 " OpSelectionMerge %cm None\n"
2840 " OpBranchConditional %comp %tb %fb\n"
2846 "%res = OpPhi %f32 %float_1 %tb %float_n1 %fb\n"
2848 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2849 " OpStore %outloc %res\n"
2853 specFloat.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2854 specFloat.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2855 specFloat.numWorkGroups = IVec3(numElements, 1, 1);
2858 string(getComputeAsmShaderPreamble()) +
2860 "OpSource GLSL 430\n"
2861 "OpName %main \"main\"\n"
2862 "OpName %id \"gl_GlobalInvocationID\"\n"
2864 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2866 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2868 "%id = OpVariable %uvec3ptr Input\n"
2869 "%v4f32 = OpTypeVector %f32 4\n"
2870 "%mat4v4f32 = OpTypeMatrix %v4f32 4\n"
2871 "%zero = OpConstant %i32 0\n"
2872 "%float_0 = OpConstant %f32 0.0\n"
2873 "%float_1 = OpConstant %f32 1.0\n"
2874 "%float_n1 = OpConstant %f32 -1.0\n"
2875 "%m11 = OpConstantComposite %v4f32 %float_1 %float_0 %float_0 %float_0\n"
2876 "%m12 = OpConstantComposite %v4f32 %float_0 %float_1 %float_0 %float_0\n"
2877 "%m13 = OpConstantComposite %v4f32 %float_0 %float_0 %float_1 %float_0\n"
2878 "%m14 = OpConstantComposite %v4f32 %float_0 %float_0 %float_0 %float_1\n"
2879 "%m1 = OpConstantComposite %mat4v4f32 %m11 %m12 %m13 %m14\n"
2880 "%m21 = OpConstantComposite %v4f32 %float_n1 %float_0 %float_0 %float_0\n"
2881 "%m22 = OpConstantComposite %v4f32 %float_0 %float_n1 %float_0 %float_0\n"
2882 "%m23 = OpConstantComposite %v4f32 %float_0 %float_0 %float_n1 %float_0\n"
2883 "%m24 = OpConstantComposite %v4f32 %float_0 %float_0 %float_0 %float_n1\n"
2884 "%m2 = OpConstantComposite %mat4v4f32 %m21 %m22 %m23 %m24\n"
2886 "%main = OpFunction %void None %voidf\n"
2887 "%entry = OpLabel\n"
2888 "%idval = OpLoad %uvec3 %id\n"
2889 "%x = OpCompositeExtract %u32 %idval 0\n"
2890 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2891 "%inval = OpLoad %f32 %inloc\n"
2893 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2894 " OpSelectionMerge %cm None\n"
2895 " OpBranchConditional %comp %tb %fb\n"
2901 "%mres = OpPhi %mat4v4f32 %m1 %tb %m2 %fb\n"
2902 "%res = OpCompositeExtract %f32 %mres 2 2\n"
2904 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2905 " OpStore %outloc %res\n"
2909 specMat4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2910 specMat4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2911 specMat4.numWorkGroups = IVec3(numElements, 1, 1);
2914 string(getComputeAsmShaderPreamble()) +
2916 "OpSource GLSL 430\n"
2917 "OpName %main \"main\"\n"
2918 "OpName %id \"gl_GlobalInvocationID\"\n"
2920 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2922 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2924 "%id = OpVariable %uvec3ptr Input\n"
2925 "%zero = OpConstant %i32 0\n"
2926 "%float_0 = OpConstant %f32 0.0\n"
2927 "%float_1 = OpConstant %f32 1.0\n"
2928 "%float_n1 = OpConstant %f32 -1.0\n"
2929 "%v1 = OpConstantComposite %fvec3 %float_1 %float_1 %float_1\n"
2930 "%v2 = OpConstantComposite %fvec3 %float_n1 %float_n1 %float_n1\n"
2932 "%main = OpFunction %void None %voidf\n"
2933 "%entry = OpLabel\n"
2934 "%idval = OpLoad %uvec3 %id\n"
2935 "%x = OpCompositeExtract %u32 %idval 0\n"
2936 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2937 "%inval = OpLoad %f32 %inloc\n"
2939 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2940 " OpSelectionMerge %cm None\n"
2941 " OpBranchConditional %comp %tb %fb\n"
2947 "%vres = OpPhi %fvec3 %v1 %tb %v2 %fb\n"
2948 "%res = OpCompositeExtract %f32 %vres 2\n"
2950 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2951 " OpStore %outloc %res\n"
2955 specVec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2956 specVec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2957 specVec3.numWorkGroups = IVec3(numElements, 1, 1);
2960 string(getComputeAsmShaderPreamble()) +
2962 "OpSource GLSL 430\n"
2963 "OpName %main \"main\"\n"
2964 "OpName %id \"gl_GlobalInvocationID\"\n"
2966 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2968 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2970 "%id = OpVariable %uvec3ptr Input\n"
2971 "%zero = OpConstant %i32 0\n"
2972 "%float_0 = OpConstant %f32 0.0\n"
2973 "%i1 = OpConstant %i32 1\n"
2974 "%i2 = OpConstant %i32 -1\n"
2976 "%main = OpFunction %void None %voidf\n"
2977 "%entry = OpLabel\n"
2978 "%idval = OpLoad %uvec3 %id\n"
2979 "%x = OpCompositeExtract %u32 %idval 0\n"
2980 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2981 "%inval = OpLoad %f32 %inloc\n"
2983 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
2984 " OpSelectionMerge %cm None\n"
2985 " OpBranchConditional %comp %tb %fb\n"
2991 "%ires = OpPhi %i32 %i1 %tb %i2 %fb\n"
2992 "%res = OpConvertSToF %f32 %ires\n"
2994 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2995 " OpStore %outloc %res\n"
2999 specInt.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3000 specInt.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3001 specInt.numWorkGroups = IVec3(numElements, 1, 1);
3003 specArray.assembly =
3004 string(getComputeAsmShaderPreamble()) +
3006 "OpSource GLSL 430\n"
3007 "OpName %main \"main\"\n"
3008 "OpName %id \"gl_GlobalInvocationID\"\n"
3010 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3012 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3014 "%id = OpVariable %uvec3ptr Input\n"
3015 "%zero = OpConstant %i32 0\n"
3016 "%u7 = OpConstant %u32 7\n"
3017 "%float_0 = OpConstant %f32 0.0\n"
3018 "%float_1 = OpConstant %f32 1.0\n"
3019 "%float_n1 = OpConstant %f32 -1.0\n"
3020 "%f32a7 = OpTypeArray %f32 %u7\n"
3021 "%a1 = OpConstantComposite %f32a7 %float_1 %float_1 %float_1 %float_1 %float_1 %float_1 %float_1\n"
3022 "%a2 = OpConstantComposite %f32a7 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1 %float_n1\n"
3023 "%main = OpFunction %void None %voidf\n"
3024 "%entry = OpLabel\n"
3025 "%idval = OpLoad %uvec3 %id\n"
3026 "%x = OpCompositeExtract %u32 %idval 0\n"
3027 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3028 "%inval = OpLoad %f32 %inloc\n"
3030 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3031 " OpSelectionMerge %cm None\n"
3032 " OpBranchConditional %comp %tb %fb\n"
3038 "%ares = OpPhi %f32a7 %a1 %tb %a2 %fb\n"
3039 "%res = OpCompositeExtract %f32 %ares 5\n"
3041 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3042 " OpStore %outloc %res\n"
3046 specArray.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3047 specArray.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3048 specArray.numWorkGroups = IVec3(numElements, 1, 1);
3050 specStruct.assembly =
3051 string(getComputeAsmShaderPreamble()) +
3053 "OpSource GLSL 430\n"
3054 "OpName %main \"main\"\n"
3055 "OpName %id \"gl_GlobalInvocationID\"\n"
3057 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3059 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3061 "%id = OpVariable %uvec3ptr Input\n"
3062 "%zero = OpConstant %i32 0\n"
3063 "%float_0 = OpConstant %f32 0.0\n"
3064 "%float_1 = OpConstant %f32 1.0\n"
3065 "%float_n1 = OpConstant %f32 -1.0\n"
3067 "%v2f32 = OpTypeVector %f32 2\n"
3068 "%Data2 = OpTypeStruct %f32 %v2f32\n"
3069 "%Data = OpTypeStruct %Data2 %f32\n"
3071 "%in1a = OpConstantComposite %v2f32 %float_1 %float_1\n"
3072 "%in1b = OpConstantComposite %Data2 %float_1 %in1a\n"
3073 "%s1 = OpConstantComposite %Data %in1b %float_1\n"
3074 "%in2a = OpConstantComposite %v2f32 %float_n1 %float_n1\n"
3075 "%in2b = OpConstantComposite %Data2 %float_n1 %in2a\n"
3076 "%s2 = OpConstantComposite %Data %in2b %float_n1\n"
3078 "%main = OpFunction %void None %voidf\n"
3079 "%entry = OpLabel\n"
3080 "%idval = OpLoad %uvec3 %id\n"
3081 "%x = OpCompositeExtract %u32 %idval 0\n"
3082 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3083 "%inval = OpLoad %f32 %inloc\n"
3085 "%comp = OpFOrdGreaterThan %bool %inval %float_0\n"
3086 " OpSelectionMerge %cm None\n"
3087 " OpBranchConditional %comp %tb %fb\n"
3093 "%sres = OpPhi %Data %s1 %tb %s2 %fb\n"
3094 "%res = OpCompositeExtract %f32 %sres 0 0\n"
3096 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3097 " OpStore %outloc %res\n"
3101 specStruct.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3102 specStruct.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3103 specStruct.numWorkGroups = IVec3(numElements, 1, 1);
3105 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_int", "OpPhi with int variables", specInt));
3106 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_float", "OpPhi with float variables", specFloat));
3107 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_vec3", "OpPhi with vec3 variables", specVec3));
3108 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_mat4", "OpPhi with mat4 variables", specMat4));
3109 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_array", "OpPhi with array variables", specArray));
3110 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vartype_struct", "OpPhi with struct variables", specStruct));
3113 string generateConstantDefinitions (int count)
3115 std::ostringstream r;
3116 for (int i = 0; i < count; i++)
3117 r << "%cf" << (i * 10 + 5) << " = OpConstant %f32 " <<(i * 10 + 5) << ".0\n";
3122 string generateSwitchCases (int count)
3124 std::ostringstream r;
3125 for (int i = 0; i < count; i++)
3126 r << " " << i << " %case" << i;
3131 string generateSwitchTargets (int count)
3133 std::ostringstream r;
3134 for (int i = 0; i < count; i++)
3135 r << "%case" << i << " = OpLabel\n OpBranch %phi\n";
3140 string generateOpPhiParams (int count)
3142 std::ostringstream r;
3143 for (int i = 0; i < count; i++)
3144 r << " %cf" << (i * 10 + 5) << " %case" << i;
3149 string generateIntWidth (int value)
3151 std::ostringstream r;
3156 // Expand input string by injecting "ABC" between the input
3157 // string characters. The acc/add/treshold parameters are used
3158 // to skip some of the injections to make the result less
3159 // uniform (and a lot shorter).
3160 string expandOpPhiCase5 (const string& s, int &acc, int add, int treshold)
3162 std::ostringstream res;
3163 const char* p = s.c_str();
3179 // Calculate expected result based on the code string
3180 float calcOpPhiCase5 (float val, const string& s)
3182 const char* p = s.c_str();
3185 const float tv[8] = { 0.5f, 1.5f, 3.5f, 7.5f, 15.5f, 31.5f, 63.5f, 127.5f };
3186 const float v = deFloatAbs(val);
3191 for (int i = 7; i >= 0; --i)
3192 x[i] = std::fmod((float)v, (float)(2 << i));
3193 for (int i = 7; i >= 0; --i)
3194 b[i] = x[i] > tv[i];
3201 if (skip == 0 && b[depth])
3212 if (b[depth] || skip)
3226 // In the code string, the letters represent the following:
3229 // if (certain bit is set)
3240 // AABCBC leads to if(){r++;if(){r++;}else{}}else{}
3241 // ABABCC leads to if(){r++;}else{if(){r++;}else{}}
3242 // ABCABC leads to if(){r++;}else{}if(){r++;}else{}
3244 // Code generation gets a bit complicated due to the else-branches,
3245 // which do not generate new values. Thus, the generator needs to
3246 // keep track of the previous variable change seen by the else
3248 string generateOpPhiCase5 (const string& s)
3250 std::stack<int> idStack;
3251 std::stack<std::string> value;
3252 std::stack<std::string> valueLabel;
3253 std::stack<std::string> mergeLeft;
3254 std::stack<std::string> mergeRight;
3255 std::ostringstream res;
3256 const char* p = s.c_str();
3262 value.push("%f32_0");
3263 valueLabel.push("%f32_0 %entry");
3271 idStack.push(currId);
3272 res << "\tOpSelectionMerge %m" << currId << " None\n";
3273 res << "\tOpBranchConditional %b" << depth << " %t" << currId << " %f" << currId << "\n";
3274 res << "%t" << currId << " = OpLabel\n";
3275 res << "%rt" << currId << " = OpFAdd %f32 " << value.top() << " %f32_1\n";
3276 std::ostringstream tag;
3277 tag << "%rt" << currId;
3278 value.push(tag.str());
3279 tag << " %t" << currId;
3280 valueLabel.push(tag.str());
3285 mergeLeft.push(valueLabel.top());
3288 res << "\tOpBranch %m" << currId << "\n";
3289 res << "%f" << currId << " = OpLabel\n";
3290 std::ostringstream tag;
3291 tag << value.top() << " %f" << currId;
3293 valueLabel.push(tag.str());
3298 mergeRight.push(valueLabel.top());
3299 res << "\tOpBranch %m" << currId << "\n";
3300 res << "%m" << currId << " = OpLabel\n";
3302 res << "%res"; // last result goes to %res
3304 res << "%rm" << currId;
3305 res << " = OpPhi %f32 " << mergeLeft.top() << " " << mergeRight.top() << "\n";
3306 std::ostringstream tag;
3307 tag << "%rm" << currId;
3309 value.push(tag.str());
3310 tag << " %m" << currId;
3312 valueLabel.push(tag.str());
3317 currId = idStack.top();
3325 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
3327 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
3328 ComputeShaderSpec spec1;
3329 ComputeShaderSpec spec2;
3330 ComputeShaderSpec spec3;
3331 ComputeShaderSpec spec4;
3332 ComputeShaderSpec spec5;
3333 de::Random rnd (deStringHash(group->getName()));
3334 const int numElements = 100;
3335 vector<float> inputFloats (numElements, 0);
3336 vector<float> outputFloats1 (numElements, 0);
3337 vector<float> outputFloats2 (numElements, 0);
3338 vector<float> outputFloats3 (numElements, 0);
3339 vector<float> outputFloats4 (numElements, 0);
3340 vector<float> outputFloats5 (numElements, 0);
3341 std::string codestring = "ABC";
3342 const int test4Width = 1024;
3344 // Build case 5 code string. Each iteration makes the hierarchy more complicated.
3345 // 9 iterations with (7, 24) parameters makes the hierarchy 8 deep with about 1500 lines of
3347 for (int i = 0, acc = 0; i < 9; i++)
3348 codestring = expandOpPhiCase5(codestring, acc, 7, 24);
3350 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
3352 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3353 floorAll(inputFloats);
3355 for (size_t ndx = 0; ndx < numElements; ++ndx)
3359 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
3360 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
3361 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
3364 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
3365 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
3367 int index4 = (int)deFloor(deAbs((float)ndx * inputFloats[ndx]));
3368 outputFloats4[ndx] = (float)(index4 % test4Width) * 10.0f + 5.0f;
3370 outputFloats5[ndx] = calcOpPhiCase5(inputFloats[ndx], codestring);
3374 string(getComputeAsmShaderPreamble()) +
3376 "OpSource GLSL 430\n"
3377 "OpName %main \"main\"\n"
3378 "OpName %id \"gl_GlobalInvocationID\"\n"
3380 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3382 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3384 "%id = OpVariable %uvec3ptr Input\n"
3385 "%zero = OpConstant %i32 0\n"
3386 "%three = OpConstant %u32 3\n"
3387 "%constf5p5 = OpConstant %f32 5.5\n"
3388 "%constf20p5 = OpConstant %f32 20.5\n"
3389 "%constf1p75 = OpConstant %f32 1.75\n"
3390 "%constf8p5 = OpConstant %f32 8.5\n"
3391 "%constf6p5 = OpConstant %f32 6.5\n"
3393 "%main = OpFunction %void None %voidf\n"
3394 "%entry = OpLabel\n"
3395 "%idval = OpLoad %uvec3 %id\n"
3396 "%x = OpCompositeExtract %u32 %idval 0\n"
3397 "%selector = OpUMod %u32 %x %three\n"
3398 " OpSelectionMerge %phi None\n"
3399 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
3401 // Case 1 before OpPhi.
3402 "%case1 = OpLabel\n"
3405 "%default = OpLabel\n"
3409 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
3410 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3411 "%inval = OpLoad %f32 %inloc\n"
3412 "%add = OpFAdd %f32 %inval %operand\n"
3413 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3414 " OpStore %outloc %add\n"
3417 // Case 0 after OpPhi.
3418 "%case0 = OpLabel\n"
3422 // Case 2 after OpPhi.
3423 "%case2 = OpLabel\n"
3427 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3428 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
3429 spec1.numWorkGroups = IVec3(numElements, 1, 1);
3431 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
3434 string(getComputeAsmShaderPreamble()) +
3436 "OpName %main \"main\"\n"
3437 "OpName %id \"gl_GlobalInvocationID\"\n"
3439 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3441 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3443 "%id = OpVariable %uvec3ptr Input\n"
3444 "%zero = OpConstant %i32 0\n"
3445 "%one = OpConstant %i32 1\n"
3446 "%three = OpConstant %i32 3\n"
3447 "%constf6p5 = OpConstant %f32 6.5\n"
3449 "%main = OpFunction %void None %voidf\n"
3450 "%entry = OpLabel\n"
3451 "%idval = OpLoad %uvec3 %id\n"
3452 "%x = OpCompositeExtract %u32 %idval 0\n"
3453 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3454 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3455 "%inval = OpLoad %f32 %inloc\n"
3459 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
3460 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
3461 "%step_next = OpIAdd %i32 %step %one\n"
3462 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
3463 "%still_loop = OpSLessThan %bool %step %three\n"
3464 " OpLoopMerge %exit %phi None\n"
3465 " OpBranchConditional %still_loop %phi %exit\n"
3468 " OpStore %outloc %accum\n"
3471 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3472 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
3473 spec2.numWorkGroups = IVec3(numElements, 1, 1);
3475 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
3478 string(getComputeAsmShaderPreamble()) +
3480 "OpName %main \"main\"\n"
3481 "OpName %id \"gl_GlobalInvocationID\"\n"
3483 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3485 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3487 "%f32ptr_f = OpTypePointer Function %f32\n"
3488 "%id = OpVariable %uvec3ptr Input\n"
3489 "%true = OpConstantTrue %bool\n"
3490 "%false = OpConstantFalse %bool\n"
3491 "%zero = OpConstant %i32 0\n"
3492 "%constf8p5 = OpConstant %f32 8.5\n"
3494 "%main = OpFunction %void None %voidf\n"
3495 "%entry = OpLabel\n"
3496 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
3497 "%idval = OpLoad %uvec3 %id\n"
3498 "%x = OpCompositeExtract %u32 %idval 0\n"
3499 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3500 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3501 "%a_init = OpLoad %f32 %inloc\n"
3502 "%b_init = OpLoad %f32 %b\n"
3506 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
3507 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
3508 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
3509 " OpLoopMerge %exit %phi None\n"
3510 " OpBranchConditional %still_loop %phi %exit\n"
3513 "%sub = OpFSub %f32 %a_next %b_next\n"
3514 " OpStore %outloc %sub\n"
3517 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3518 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
3519 spec3.numWorkGroups = IVec3(numElements, 1, 1);
3521 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
3524 "OpCapability Shader\n"
3525 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
3526 "OpMemoryModel Logical GLSL450\n"
3527 "OpEntryPoint GLCompute %main \"main\" %id\n"
3528 "OpExecutionMode %main LocalSize 1 1 1\n"
3530 "OpSource GLSL 430\n"
3531 "OpName %main \"main\"\n"
3532 "OpName %id \"gl_GlobalInvocationID\"\n"
3534 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3536 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3538 "%id = OpVariable %uvec3ptr Input\n"
3539 "%zero = OpConstant %i32 0\n"
3540 "%cimod = OpConstant %u32 " + generateIntWidth(test4Width) + "\n"
3542 + generateConstantDefinitions(test4Width) +
3544 "%main = OpFunction %void None %voidf\n"
3545 "%entry = OpLabel\n"
3546 "%idval = OpLoad %uvec3 %id\n"
3547 "%x = OpCompositeExtract %u32 %idval 0\n"
3548 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3549 "%inval = OpLoad %f32 %inloc\n"
3550 "%xf = OpConvertUToF %f32 %x\n"
3551 "%xm = OpFMul %f32 %xf %inval\n"
3552 "%xa = OpExtInst %f32 %ext FAbs %xm\n"
3553 "%xi = OpConvertFToU %u32 %xa\n"
3554 "%selector = OpUMod %u32 %xi %cimod\n"
3555 " OpSelectionMerge %phi None\n"
3556 " OpSwitch %selector %default "
3558 + generateSwitchCases(test4Width) +
3560 "%default = OpLabel\n"
3563 + generateSwitchTargets(test4Width) +
3566 "%result = OpPhi %f32"
3568 + generateOpPhiParams(test4Width) +
3570 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3571 " OpStore %outloc %result\n"
3575 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3576 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
3577 spec4.numWorkGroups = IVec3(numElements, 1, 1);
3579 group->addChild(new SpvAsmComputeShaderCase(testCtx, "wide", "OpPhi with a lot of parameters", spec4));
3582 "OpCapability Shader\n"
3583 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
3584 "OpMemoryModel Logical GLSL450\n"
3585 "OpEntryPoint GLCompute %main \"main\" %id\n"
3586 "OpExecutionMode %main LocalSize 1 1 1\n"
3587 "%code = OpString \"" + codestring + "\"\n"
3589 "OpSource GLSL 430\n"
3590 "OpName %main \"main\"\n"
3591 "OpName %id \"gl_GlobalInvocationID\"\n"
3593 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3595 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3597 "%id = OpVariable %uvec3ptr Input\n"
3598 "%zero = OpConstant %i32 0\n"
3599 "%f32_0 = OpConstant %f32 0.0\n"
3600 "%f32_0_5 = OpConstant %f32 0.5\n"
3601 "%f32_1 = OpConstant %f32 1.0\n"
3602 "%f32_1_5 = OpConstant %f32 1.5\n"
3603 "%f32_2 = OpConstant %f32 2.0\n"
3604 "%f32_3_5 = OpConstant %f32 3.5\n"
3605 "%f32_4 = OpConstant %f32 4.0\n"
3606 "%f32_7_5 = OpConstant %f32 7.5\n"
3607 "%f32_8 = OpConstant %f32 8.0\n"
3608 "%f32_15_5 = OpConstant %f32 15.5\n"
3609 "%f32_16 = OpConstant %f32 16.0\n"
3610 "%f32_31_5 = OpConstant %f32 31.5\n"
3611 "%f32_32 = OpConstant %f32 32.0\n"
3612 "%f32_63_5 = OpConstant %f32 63.5\n"
3613 "%f32_64 = OpConstant %f32 64.0\n"
3614 "%f32_127_5 = OpConstant %f32 127.5\n"
3615 "%f32_128 = OpConstant %f32 128.0\n"
3616 "%f32_256 = OpConstant %f32 256.0\n"
3618 "%main = OpFunction %void None %voidf\n"
3619 "%entry = OpLabel\n"
3620 "%idval = OpLoad %uvec3 %id\n"
3621 "%x = OpCompositeExtract %u32 %idval 0\n"
3622 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3623 "%inval = OpLoad %f32 %inloc\n"
3625 "%xabs = OpExtInst %f32 %ext FAbs %inval\n"
3626 "%x8 = OpFMod %f32 %xabs %f32_256\n"
3627 "%x7 = OpFMod %f32 %xabs %f32_128\n"
3628 "%x6 = OpFMod %f32 %xabs %f32_64\n"
3629 "%x5 = OpFMod %f32 %xabs %f32_32\n"
3630 "%x4 = OpFMod %f32 %xabs %f32_16\n"
3631 "%x3 = OpFMod %f32 %xabs %f32_8\n"
3632 "%x2 = OpFMod %f32 %xabs %f32_4\n"
3633 "%x1 = OpFMod %f32 %xabs %f32_2\n"
3635 "%b7 = OpFOrdGreaterThanEqual %bool %x8 %f32_127_5\n"
3636 "%b6 = OpFOrdGreaterThanEqual %bool %x7 %f32_63_5\n"
3637 "%b5 = OpFOrdGreaterThanEqual %bool %x6 %f32_31_5\n"
3638 "%b4 = OpFOrdGreaterThanEqual %bool %x5 %f32_15_5\n"
3639 "%b3 = OpFOrdGreaterThanEqual %bool %x4 %f32_7_5\n"
3640 "%b2 = OpFOrdGreaterThanEqual %bool %x3 %f32_3_5\n"
3641 "%b1 = OpFOrdGreaterThanEqual %bool %x2 %f32_1_5\n"
3642 "%b0 = OpFOrdGreaterThanEqual %bool %x1 %f32_0_5\n"
3644 + generateOpPhiCase5(codestring) +
3646 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3647 " OpStore %outloc %res\n"
3651 spec5.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3652 spec5.outputs.push_back(BufferSp(new Float32Buffer(outputFloats5)));
3653 spec5.numWorkGroups = IVec3(numElements, 1, 1);
3655 group->addChild(new SpvAsmComputeShaderCase(testCtx, "nested", "Stress OpPhi with a lot of nesting", spec5));
3657 createOpPhiVartypeTests(group, testCtx);
3659 return group.release();
3662 // Assembly code used for testing block order is based on GLSL source code:
3666 // layout(std140, set = 0, binding = 0) readonly buffer Input {
3667 // float elements[];
3669 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
3670 // float elements[];
3674 // uint x = gl_GlobalInvocationID.x;
3675 // output_data.elements[x] = input_data.elements[x];
3676 // if (x > uint(50)) {
3677 // switch (x % uint(3)) {
3678 // case 0: output_data.elements[x] += 1.5f; break;
3679 // case 1: output_data.elements[x] += 42.f; break;
3680 // case 2: output_data.elements[x] -= 27.f; break;
3684 // output_data.elements[x] = -input_data.elements[x];
3687 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
3689 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
3690 ComputeShaderSpec spec;
3691 de::Random rnd (deStringHash(group->getName()));
3692 const int numElements = 100;
3693 vector<float> inputFloats (numElements, 0);
3694 vector<float> outputFloats (numElements, 0);
3696 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3698 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3699 floorAll(inputFloats);
3701 for (size_t ndx = 0; ndx <= 50; ++ndx)
3702 outputFloats[ndx] = -inputFloats[ndx];
3704 for (size_t ndx = 51; ndx < numElements; ++ndx)
3708 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
3709 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
3710 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
3716 string(getComputeAsmShaderPreamble()) +
3718 "OpSource GLSL 430\n"
3719 "OpName %main \"main\"\n"
3720 "OpName %id \"gl_GlobalInvocationID\"\n"
3722 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3724 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3726 "%u32ptr = OpTypePointer Function %u32\n"
3727 "%u32ptr_input = OpTypePointer Input %u32\n"
3729 + string(getComputeAsmInputOutputBuffer()) +
3731 "%id = OpVariable %uvec3ptr Input\n"
3732 "%zero = OpConstant %i32 0\n"
3733 "%const3 = OpConstant %u32 3\n"
3734 "%const50 = OpConstant %u32 50\n"
3735 "%constf1p5 = OpConstant %f32 1.5\n"
3736 "%constf27 = OpConstant %f32 27.0\n"
3737 "%constf42 = OpConstant %f32 42.0\n"
3739 "%main = OpFunction %void None %voidf\n"
3742 "%entry = OpLabel\n"
3744 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
3745 "%xvar = OpVariable %u32ptr Function\n"
3746 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
3747 "%x = OpLoad %u32 %xptr\n"
3748 " OpStore %xvar %x\n"
3750 "%cmp = OpUGreaterThan %bool %x %const50\n"
3751 " OpSelectionMerge %if_merge None\n"
3752 " OpBranchConditional %cmp %if_true %if_false\n"
3754 // False branch for if-statement: placed in the middle of switch cases and before true branch.
3755 "%if_false = OpLabel\n"
3756 "%x_f = OpLoad %u32 %xvar\n"
3757 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
3758 "%inval_f = OpLoad %f32 %inloc_f\n"
3759 "%negate = OpFNegate %f32 %inval_f\n"
3760 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
3761 " OpStore %outloc_f %negate\n"
3762 " OpBranch %if_merge\n"
3764 // Merge block for if-statement: placed in the middle of true and false branch.
3765 "%if_merge = OpLabel\n"
3768 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
3769 "%if_true = OpLabel\n"
3770 "%xval_t = OpLoad %u32 %xvar\n"
3771 "%mod = OpUMod %u32 %xval_t %const3\n"
3772 " OpSelectionMerge %switch_merge None\n"
3773 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
3775 // Merge block for switch-statement: placed before the case
3776 // bodies. But it must follow OpSwitch which dominates it.
3777 "%switch_merge = OpLabel\n"
3778 " OpBranch %if_merge\n"
3780 // Case 1 for switch-statement: placed before case 0.
3781 // It must follow the OpSwitch that dominates it.
3782 "%case1 = OpLabel\n"
3783 "%x_1 = OpLoad %u32 %xvar\n"
3784 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
3785 "%inval_1 = OpLoad %f32 %inloc_1\n"
3786 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
3787 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
3788 " OpStore %outloc_1 %addf42\n"
3789 " OpBranch %switch_merge\n"
3791 // Case 2 for switch-statement.
3792 "%case2 = OpLabel\n"
3793 "%x_2 = OpLoad %u32 %xvar\n"
3794 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
3795 "%inval_2 = OpLoad %f32 %inloc_2\n"
3796 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
3797 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
3798 " OpStore %outloc_2 %subf27\n"
3799 " OpBranch %switch_merge\n"
3801 // Default case for switch-statement: placed in the middle of normal cases.
3802 "%default = OpLabel\n"
3803 " OpBranch %switch_merge\n"
3805 // Case 0 for switch-statement: out of order.
3806 "%case0 = OpLabel\n"
3807 "%x_0 = OpLoad %u32 %xvar\n"
3808 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
3809 "%inval_0 = OpLoad %f32 %inloc_0\n"
3810 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
3811 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
3812 " OpStore %outloc_0 %addf1p5\n"
3813 " OpBranch %switch_merge\n"
3816 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3817 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3818 spec.numWorkGroups = IVec3(numElements, 1, 1);
3820 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
3822 return group.release();
3825 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
3827 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
3828 ComputeShaderSpec spec1;
3829 ComputeShaderSpec spec2;
3830 de::Random rnd (deStringHash(group->getName()));
3831 const int numElements = 100;
3832 vector<float> inputFloats (numElements, 0);
3833 vector<float> outputFloats1 (numElements, 0);
3834 vector<float> outputFloats2 (numElements, 0);
3835 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
3837 for (size_t ndx = 0; ndx < numElements; ++ndx)
3839 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
3840 outputFloats2[ndx] = -inputFloats[ndx];
3843 const string assembly(
3844 "OpCapability Shader\n"
3845 "OpCapability ClipDistance\n"
3846 "OpMemoryModel Logical GLSL450\n"
3847 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
3848 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
3849 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
3850 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
3851 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
3852 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
3854 "OpName %comp_main1 \"entrypoint1\"\n"
3855 "OpName %comp_main2 \"entrypoint2\"\n"
3856 "OpName %vert_main \"entrypoint2\"\n"
3857 "OpName %id \"gl_GlobalInvocationID\"\n"
3858 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
3859 "OpName %vertexIndex \"gl_VertexIndex\"\n"
3860 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
3861 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
3862 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
3863 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
3865 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3866 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
3867 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
3868 "OpDecorate %vert_builtin_st Block\n"
3869 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
3870 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
3871 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
3873 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3875 "%zero = OpConstant %i32 0\n"
3876 "%one = OpConstant %u32 1\n"
3877 "%c_f32_1 = OpConstant %f32 1\n"
3879 "%i32inputptr = OpTypePointer Input %i32\n"
3880 "%vec4 = OpTypeVector %f32 4\n"
3881 "%vec4ptr = OpTypePointer Output %vec4\n"
3882 "%f32arr1 = OpTypeArray %f32 %one\n"
3883 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
3884 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
3885 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
3887 "%id = OpVariable %uvec3ptr Input\n"
3888 "%vertexIndex = OpVariable %i32inputptr Input\n"
3889 "%instanceIndex = OpVariable %i32inputptr Input\n"
3890 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
3892 // gl_Position = vec4(1.);
3893 "%vert_main = OpFunction %void None %voidf\n"
3894 "%vert_entry = OpLabel\n"
3895 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
3896 " OpStore %position %c_vec4_1\n"
3901 "%comp_main1 = OpFunction %void None %voidf\n"
3902 "%comp1_entry = OpLabel\n"
3903 "%idval1 = OpLoad %uvec3 %id\n"
3904 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
3905 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
3906 "%inval1 = OpLoad %f32 %inloc1\n"
3907 "%add = OpFAdd %f32 %inval1 %inval1\n"
3908 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
3909 " OpStore %outloc1 %add\n"
3914 "%comp_main2 = OpFunction %void None %voidf\n"
3915 "%comp2_entry = OpLabel\n"
3916 "%idval2 = OpLoad %uvec3 %id\n"
3917 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
3918 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
3919 "%inval2 = OpLoad %f32 %inloc2\n"
3920 "%neg = OpFNegate %f32 %inval2\n"
3921 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
3922 " OpStore %outloc2 %neg\n"
3924 " OpFunctionEnd\n");
3926 spec1.assembly = assembly;
3927 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3928 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
3929 spec1.numWorkGroups = IVec3(numElements, 1, 1);
3930 spec1.entryPoint = "entrypoint1";
3932 spec2.assembly = assembly;
3933 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3934 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
3935 spec2.numWorkGroups = IVec3(numElements, 1, 1);
3936 spec2.entryPoint = "entrypoint2";
3938 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
3939 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
3941 return group.release();
3944 inline std::string makeLongUTF8String (size_t num4ByteChars)
3946 // An example of a longest valid UTF-8 character. Be explicit about the
3947 // character type because Microsoft compilers can otherwise interpret the
3948 // character string as being over wide (16-bit) characters. Ideally, we
3949 // would just use a C++11 UTF-8 string literal, but we want to support older
3950 // Microsoft compilers.
3951 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
3952 std::string longString;
3953 longString.reserve(num4ByteChars * 4);
3954 for (size_t count = 0; count < num4ByteChars; count++)
3956 longString += earthAfrica;
3961 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
3963 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
3964 vector<CaseParameter> cases;
3965 de::Random rnd (deStringHash(group->getName()));
3966 const int numElements = 100;
3967 vector<float> positiveFloats (numElements, 0);
3968 vector<float> negativeFloats (numElements, 0);
3969 const StringTemplate shaderTemplate (
3970 "OpCapability Shader\n"
3971 "OpMemoryModel Logical GLSL450\n"
3973 "OpEntryPoint GLCompute %main \"main\" %id\n"
3974 "OpExecutionMode %main LocalSize 1 1 1\n"
3978 "OpName %main \"main\"\n"
3979 "OpName %id \"gl_GlobalInvocationID\"\n"
3981 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3983 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3985 "%id = OpVariable %uvec3ptr Input\n"
3986 "%zero = OpConstant %i32 0\n"
3988 "%main = OpFunction %void None %voidf\n"
3989 "%label = OpLabel\n"
3990 "%idval = OpLoad %uvec3 %id\n"
3991 "%x = OpCompositeExtract %u32 %idval 0\n"
3992 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3993 "%inval = OpLoad %f32 %inloc\n"
3994 "%neg = OpFNegate %f32 %inval\n"
3995 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3996 " OpStore %outloc %neg\n"
3998 " OpFunctionEnd\n");
4000 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
4001 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
4002 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
4003 "OpSource GLSL 430 %fname"));
4004 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
4005 "OpSource GLSL 430 %fname"));
4006 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
4007 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
4008 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
4009 "OpSource GLSL 430 %fname \"\""));
4010 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
4011 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
4012 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
4013 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
4014 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
4015 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
4016 "OpSourceContinued \"id main() {}\""));
4017 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
4018 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4019 "OpSourceContinued \"\""));
4020 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
4021 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4022 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
4023 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
4024 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
4025 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
4026 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
4027 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
4028 "OpSourceContinued \"void\"\n"
4029 "OpSourceContinued \"main()\"\n"
4030 "OpSourceContinued \"{}\""));
4031 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
4032 "OpSource GLSL 430 %fname \"\"\n"
4033 "OpSourceContinued \"#version 430\nvoid main() {}\""));
4035 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4037 for (size_t ndx = 0; ndx < numElements; ++ndx)
4038 negativeFloats[ndx] = -positiveFloats[ndx];
4040 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4042 map<string, string> specializations;
4043 ComputeShaderSpec spec;
4045 specializations["SOURCE"] = cases[caseNdx].param;
4046 spec.assembly = shaderTemplate.specialize(specializations);
4047 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4048 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4049 spec.numWorkGroups = IVec3(numElements, 1, 1);
4051 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4054 return group.release();
4057 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
4059 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
4060 vector<CaseParameter> cases;
4061 de::Random rnd (deStringHash(group->getName()));
4062 const int numElements = 100;
4063 vector<float> inputFloats (numElements, 0);
4064 vector<float> outputFloats (numElements, 0);
4065 const StringTemplate shaderTemplate (
4066 string(getComputeAsmShaderPreamble()) +
4068 "OpSourceExtension \"${EXTENSION}\"\n"
4070 "OpName %main \"main\"\n"
4071 "OpName %id \"gl_GlobalInvocationID\"\n"
4073 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4075 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4077 "%id = OpVariable %uvec3ptr Input\n"
4078 "%zero = OpConstant %i32 0\n"
4080 "%main = OpFunction %void None %voidf\n"
4081 "%label = OpLabel\n"
4082 "%idval = OpLoad %uvec3 %id\n"
4083 "%x = OpCompositeExtract %u32 %idval 0\n"
4084 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4085 "%inval = OpLoad %f32 %inloc\n"
4086 "%neg = OpFNegate %f32 %inval\n"
4087 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4088 " OpStore %outloc %neg\n"
4090 " OpFunctionEnd\n");
4092 cases.push_back(CaseParameter("empty_extension", ""));
4093 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
4094 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
4095 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
4096 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
4098 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
4100 for (size_t ndx = 0; ndx < numElements; ++ndx)
4101 outputFloats[ndx] = -inputFloats[ndx];
4103 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4105 map<string, string> specializations;
4106 ComputeShaderSpec spec;
4108 specializations["EXTENSION"] = cases[caseNdx].param;
4109 spec.assembly = shaderTemplate.specialize(specializations);
4110 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4111 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4112 spec.numWorkGroups = IVec3(numElements, 1, 1);
4114 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4117 return group.release();
4120 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
4121 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
4123 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
4124 vector<CaseParameter> cases;
4125 de::Random rnd (deStringHash(group->getName()));
4126 const int numElements = 100;
4127 vector<float> positiveFloats (numElements, 0);
4128 vector<float> negativeFloats (numElements, 0);
4129 const StringTemplate shaderTemplate (
4130 string(getComputeAsmShaderPreamble()) +
4132 "OpSource GLSL 430\n"
4133 "OpName %main \"main\"\n"
4134 "OpName %id \"gl_GlobalInvocationID\"\n"
4136 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4138 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4139 "%uvec2 = OpTypeVector %u32 2\n"
4140 "%bvec3 = OpTypeVector %bool 3\n"
4141 "%fvec4 = OpTypeVector %f32 4\n"
4142 "%fmat33 = OpTypeMatrix %fvec3 3\n"
4143 "%const100 = OpConstant %u32 100\n"
4144 "%uarr100 = OpTypeArray %i32 %const100\n"
4145 "%struct = OpTypeStruct %f32 %i32 %u32\n"
4146 "%pointer = OpTypePointer Function %i32\n"
4147 + string(getComputeAsmInputOutputBuffer()) +
4149 "%null = OpConstantNull ${TYPE}\n"
4151 "%id = OpVariable %uvec3ptr Input\n"
4152 "%zero = OpConstant %i32 0\n"
4154 "%main = OpFunction %void None %voidf\n"
4155 "%label = OpLabel\n"
4156 "%idval = OpLoad %uvec3 %id\n"
4157 "%x = OpCompositeExtract %u32 %idval 0\n"
4158 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4159 "%inval = OpLoad %f32 %inloc\n"
4160 "%neg = OpFNegate %f32 %inval\n"
4161 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4162 " OpStore %outloc %neg\n"
4164 " OpFunctionEnd\n");
4166 cases.push_back(CaseParameter("bool", "%bool"));
4167 cases.push_back(CaseParameter("sint32", "%i32"));
4168 cases.push_back(CaseParameter("uint32", "%u32"));
4169 cases.push_back(CaseParameter("float32", "%f32"));
4170 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
4171 cases.push_back(CaseParameter("vec3bool", "%bvec3"));
4172 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
4173 cases.push_back(CaseParameter("matrix", "%fmat33"));
4174 cases.push_back(CaseParameter("array", "%uarr100"));
4175 cases.push_back(CaseParameter("struct", "%struct"));
4176 cases.push_back(CaseParameter("pointer", "%pointer"));
4178 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4180 for (size_t ndx = 0; ndx < numElements; ++ndx)
4181 negativeFloats[ndx] = -positiveFloats[ndx];
4183 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4185 map<string, string> specializations;
4186 ComputeShaderSpec spec;
4188 specializations["TYPE"] = cases[caseNdx].param;
4189 spec.assembly = shaderTemplate.specialize(specializations);
4190 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4191 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4192 spec.numWorkGroups = IVec3(numElements, 1, 1);
4194 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4197 return group.release();
4200 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
4201 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
4203 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
4204 vector<CaseParameter> cases;
4205 de::Random rnd (deStringHash(group->getName()));
4206 const int numElements = 100;
4207 vector<float> positiveFloats (numElements, 0);
4208 vector<float> negativeFloats (numElements, 0);
4209 const StringTemplate shaderTemplate (
4210 string(getComputeAsmShaderPreamble()) +
4212 "OpSource GLSL 430\n"
4213 "OpName %main \"main\"\n"
4214 "OpName %id \"gl_GlobalInvocationID\"\n"
4216 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4218 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4220 "%id = OpVariable %uvec3ptr Input\n"
4221 "%zero = OpConstant %i32 0\n"
4225 "%main = OpFunction %void None %voidf\n"
4226 "%label = OpLabel\n"
4227 "%idval = OpLoad %uvec3 %id\n"
4228 "%x = OpCompositeExtract %u32 %idval 0\n"
4229 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4230 "%inval = OpLoad %f32 %inloc\n"
4231 "%neg = OpFNegate %f32 %inval\n"
4232 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4233 " OpStore %outloc %neg\n"
4235 " OpFunctionEnd\n");
4237 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
4238 "%const = OpConstantComposite %uvec3 %five %zero %five"));
4239 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
4240 "%ten = OpConstant %f32 10.\n"
4241 "%fzero = OpConstant %f32 0.\n"
4242 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
4243 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
4244 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
4245 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
4246 "%fzero = OpConstant %f32 0.\n"
4247 "%one = OpConstant %f32 1.\n"
4248 "%point5 = OpConstant %f32 0.5\n"
4249 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
4250 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
4251 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
4252 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
4253 "%st2 = OpTypeStruct %i32 %i32\n"
4254 "%struct = OpTypeStruct %st1 %st2\n"
4255 "%point5 = OpConstant %f32 0.5\n"
4256 "%one = OpConstant %u32 1\n"
4257 "%ten = OpConstant %i32 10\n"
4258 "%st1val = OpConstantComposite %st1 %one %point5\n"
4259 "%st2val = OpConstantComposite %st2 %ten %ten\n"
4260 "%const = OpConstantComposite %struct %st1val %st2val"));
4262 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4264 for (size_t ndx = 0; ndx < numElements; ++ndx)
4265 negativeFloats[ndx] = -positiveFloats[ndx];
4267 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4269 map<string, string> specializations;
4270 ComputeShaderSpec spec;
4272 specializations["CONSTANT"] = cases[caseNdx].param;
4273 spec.assembly = shaderTemplate.specialize(specializations);
4274 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4275 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4276 spec.numWorkGroups = IVec3(numElements, 1, 1);
4278 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4281 return group.release();
4284 // Creates a floating point number with the given exponent, and significand
4285 // bits set. It can only create normalized numbers. Only the least significant
4286 // 24 bits of the significand will be examined. The final bit of the
4287 // significand will also be ignored. This allows alignment to be written
4288 // similarly to C99 hex-floats.
4289 // For example if you wanted to write 0x1.7f34p-12 you would call
4290 // constructNormalizedFloat(-12, 0x7f3400)
4291 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
4295 for (deInt32 idx = 0; idx < 23; ++idx)
4297 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
4301 return std::ldexp(f, exponent);
4304 // Compare instruction for the OpQuantizeF16 compute exact case.
4305 // Returns true if the output is what is expected from the test case.
4306 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
4308 if (outputAllocs.size() != 1)
4311 // Only size is needed because we cannot compare Nans.
4312 size_t byteSize = expectedOutputs[0]->getByteSize();
4314 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
4316 if (byteSize != 4*sizeof(float)) {
4320 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
4321 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
4326 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
4327 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
4332 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
4333 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
4338 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
4339 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
4346 // Checks that every output from a test-case is a float NaN.
4347 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
4349 if (outputAllocs.size() != 1)
4352 // Only size is needed because we cannot compare Nans.
4353 size_t byteSize = expectedOutputs[0]->getByteSize();
4355 const float* const output_as_float = static_cast<const float* const>(outputAllocs[0]->getHostPtr());
4357 for (size_t idx = 0; idx < byteSize / sizeof(float); ++idx)
4359 if (!deFloatIsNaN(output_as_float[idx]))
4368 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
4369 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
4371 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
4373 const std::string shader (
4374 string(getComputeAsmShaderPreamble()) +
4376 "OpSource GLSL 430\n"
4377 "OpName %main \"main\"\n"
4378 "OpName %id \"gl_GlobalInvocationID\"\n"
4380 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4382 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4384 "%id = OpVariable %uvec3ptr Input\n"
4385 "%zero = OpConstant %i32 0\n"
4387 "%main = OpFunction %void None %voidf\n"
4388 "%label = OpLabel\n"
4389 "%idval = OpLoad %uvec3 %id\n"
4390 "%x = OpCompositeExtract %u32 %idval 0\n"
4391 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4392 "%inval = OpLoad %f32 %inloc\n"
4393 "%quant = OpQuantizeToF16 %f32 %inval\n"
4394 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4395 " OpStore %outloc %quant\n"
4397 " OpFunctionEnd\n");
4400 ComputeShaderSpec spec;
4401 const deUint32 numElements = 100;
4402 vector<float> infinities;
4403 vector<float> results;
4405 infinities.reserve(numElements);
4406 results.reserve(numElements);
4408 for (size_t idx = 0; idx < numElements; ++idx)
4413 infinities.push_back(std::numeric_limits<float>::infinity());
4414 results.push_back(std::numeric_limits<float>::infinity());
4417 infinities.push_back(-std::numeric_limits<float>::infinity());
4418 results.push_back(-std::numeric_limits<float>::infinity());
4421 infinities.push_back(std::ldexp(1.0f, 16));
4422 results.push_back(std::numeric_limits<float>::infinity());
4425 infinities.push_back(std::ldexp(-1.0f, 32));
4426 results.push_back(-std::numeric_limits<float>::infinity());
4431 spec.assembly = shader;
4432 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
4433 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
4434 spec.numWorkGroups = IVec3(numElements, 1, 1);
4436 group->addChild(new SpvAsmComputeShaderCase(
4437 testCtx, "infinities", "Check that infinities propagated and created", spec));
4441 ComputeShaderSpec spec;
4443 const deUint32 numElements = 100;
4445 nans.reserve(numElements);
4447 for (size_t idx = 0; idx < numElements; ++idx)
4451 nans.push_back(std::numeric_limits<float>::quiet_NaN());
4455 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
4459 spec.assembly = shader;
4460 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
4461 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
4462 spec.numWorkGroups = IVec3(numElements, 1, 1);
4463 spec.verifyIO = &compareNan;
4465 group->addChild(new SpvAsmComputeShaderCase(
4466 testCtx, "propagated_nans", "Check that nans are propagated", spec));
4470 ComputeShaderSpec spec;
4471 vector<float> small;
4472 vector<float> zeros;
4473 const deUint32 numElements = 100;
4475 small.reserve(numElements);
4476 zeros.reserve(numElements);
4478 for (size_t idx = 0; idx < numElements; ++idx)
4483 small.push_back(0.f);
4484 zeros.push_back(0.f);
4487 small.push_back(-0.f);
4488 zeros.push_back(-0.f);
4491 small.push_back(std::ldexp(1.0f, -16));
4492 zeros.push_back(0.f);
4495 small.push_back(std::ldexp(-1.0f, -32));
4496 zeros.push_back(-0.f);
4499 small.push_back(std::ldexp(1.0f, -127));
4500 zeros.push_back(0.f);
4503 small.push_back(-std::ldexp(1.0f, -128));
4504 zeros.push_back(-0.f);
4509 spec.assembly = shader;
4510 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
4511 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
4512 spec.numWorkGroups = IVec3(numElements, 1, 1);
4514 group->addChild(new SpvAsmComputeShaderCase(
4515 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
4519 ComputeShaderSpec spec;
4520 vector<float> exact;
4521 const deUint32 numElements = 200;
4523 exact.reserve(numElements);
4525 for (size_t idx = 0; idx < numElements; ++idx)
4526 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
4528 spec.assembly = shader;
4529 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
4530 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
4531 spec.numWorkGroups = IVec3(numElements, 1, 1);
4533 group->addChild(new SpvAsmComputeShaderCase(
4534 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
4538 ComputeShaderSpec spec;
4539 vector<float> inputs;
4540 const deUint32 numElements = 4;
4542 inputs.push_back(constructNormalizedFloat(8, 0x300300));
4543 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
4544 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
4545 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
4547 spec.assembly = shader;
4548 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
4549 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4550 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
4551 spec.numWorkGroups = IVec3(numElements, 1, 1);
4553 group->addChild(new SpvAsmComputeShaderCase(
4554 testCtx, "rounded", "Check that are rounded when needed", spec));
4557 return group.release();
4560 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
4562 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
4564 const std::string shader (
4565 string(getComputeAsmShaderPreamble()) +
4567 "OpName %main \"main\"\n"
4568 "OpName %id \"gl_GlobalInvocationID\"\n"
4570 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4572 "OpDecorate %sc_0 SpecId 0\n"
4573 "OpDecorate %sc_1 SpecId 1\n"
4574 "OpDecorate %sc_2 SpecId 2\n"
4575 "OpDecorate %sc_3 SpecId 3\n"
4576 "OpDecorate %sc_4 SpecId 4\n"
4577 "OpDecorate %sc_5 SpecId 5\n"
4579 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4581 "%id = OpVariable %uvec3ptr Input\n"
4582 "%zero = OpConstant %i32 0\n"
4583 "%c_u32_6 = OpConstant %u32 6\n"
4585 "%sc_0 = OpSpecConstant %f32 0.\n"
4586 "%sc_1 = OpSpecConstant %f32 0.\n"
4587 "%sc_2 = OpSpecConstant %f32 0.\n"
4588 "%sc_3 = OpSpecConstant %f32 0.\n"
4589 "%sc_4 = OpSpecConstant %f32 0.\n"
4590 "%sc_5 = OpSpecConstant %f32 0.\n"
4592 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
4593 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
4594 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
4595 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
4596 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
4597 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
4599 "%main = OpFunction %void None %voidf\n"
4600 "%label = OpLabel\n"
4601 "%idval = OpLoad %uvec3 %id\n"
4602 "%x = OpCompositeExtract %u32 %idval 0\n"
4603 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4604 "%selector = OpUMod %u32 %x %c_u32_6\n"
4605 " OpSelectionMerge %exit None\n"
4606 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
4608 "%case0 = OpLabel\n"
4609 " OpStore %outloc %sc_0_quant\n"
4612 "%case1 = OpLabel\n"
4613 " OpStore %outloc %sc_1_quant\n"
4616 "%case2 = OpLabel\n"
4617 " OpStore %outloc %sc_2_quant\n"
4620 "%case3 = OpLabel\n"
4621 " OpStore %outloc %sc_3_quant\n"
4624 "%case4 = OpLabel\n"
4625 " OpStore %outloc %sc_4_quant\n"
4628 "%case5 = OpLabel\n"
4629 " OpStore %outloc %sc_5_quant\n"
4635 " OpFunctionEnd\n");
4638 ComputeShaderSpec spec;
4639 const deUint8 numCases = 4;
4640 vector<float> inputs (numCases, 0.f);
4641 vector<float> outputs;
4643 spec.assembly = shader;
4644 spec.numWorkGroups = IVec3(numCases, 1, 1);
4646 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
4647 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
4648 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
4649 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
4651 outputs.push_back(std::numeric_limits<float>::infinity());
4652 outputs.push_back(-std::numeric_limits<float>::infinity());
4653 outputs.push_back(std::numeric_limits<float>::infinity());
4654 outputs.push_back(-std::numeric_limits<float>::infinity());
4656 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4657 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4659 group->addChild(new SpvAsmComputeShaderCase(
4660 testCtx, "infinities", "Check that infinities propagated and created", spec));
4664 ComputeShaderSpec spec;
4665 const deUint8 numCases = 2;
4666 vector<float> inputs (numCases, 0.f);
4667 vector<float> outputs;
4669 spec.assembly = shader;
4670 spec.numWorkGroups = IVec3(numCases, 1, 1);
4671 spec.verifyIO = &compareNan;
4673 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
4674 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
4676 for (deUint8 idx = 0; idx < numCases; ++idx)
4677 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
4679 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4680 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4682 group->addChild(new SpvAsmComputeShaderCase(
4683 testCtx, "propagated_nans", "Check that nans are propagated", spec));
4687 ComputeShaderSpec spec;
4688 const deUint8 numCases = 6;
4689 vector<float> inputs (numCases, 0.f);
4690 vector<float> outputs;
4692 spec.assembly = shader;
4693 spec.numWorkGroups = IVec3(numCases, 1, 1);
4695 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
4696 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
4697 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
4698 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
4699 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
4700 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
4702 outputs.push_back(0.f);
4703 outputs.push_back(-0.f);
4704 outputs.push_back(0.f);
4705 outputs.push_back(-0.f);
4706 outputs.push_back(0.f);
4707 outputs.push_back(-0.f);
4709 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4710 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4712 group->addChild(new SpvAsmComputeShaderCase(
4713 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
4717 ComputeShaderSpec spec;
4718 const deUint8 numCases = 6;
4719 vector<float> inputs (numCases, 0.f);
4720 vector<float> outputs;
4722 spec.assembly = shader;
4723 spec.numWorkGroups = IVec3(numCases, 1, 1);
4725 for (deUint8 idx = 0; idx < 6; ++idx)
4727 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
4728 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
4729 outputs.push_back(f);
4732 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4733 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4735 group->addChild(new SpvAsmComputeShaderCase(
4736 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
4740 ComputeShaderSpec spec;
4741 const deUint8 numCases = 4;
4742 vector<float> inputs (numCases, 0.f);
4743 vector<float> outputs;
4745 spec.assembly = shader;
4746 spec.numWorkGroups = IVec3(numCases, 1, 1);
4747 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
4749 outputs.push_back(constructNormalizedFloat(8, 0x300300));
4750 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
4751 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
4752 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
4754 for (deUint8 idx = 0; idx < numCases; ++idx)
4755 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
4757 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
4758 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
4760 group->addChild(new SpvAsmComputeShaderCase(
4761 testCtx, "rounded", "Check that are rounded when needed", spec));
4764 return group.release();
4767 // Checks that constant null/composite values can be used in computation.
4768 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
4770 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
4771 ComputeShaderSpec spec;
4772 de::Random rnd (deStringHash(group->getName()));
4773 const int numElements = 100;
4774 vector<float> positiveFloats (numElements, 0);
4775 vector<float> negativeFloats (numElements, 0);
4777 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4779 for (size_t ndx = 0; ndx < numElements; ++ndx)
4780 negativeFloats[ndx] = -positiveFloats[ndx];
4783 "OpCapability Shader\n"
4784 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
4785 "OpMemoryModel Logical GLSL450\n"
4786 "OpEntryPoint GLCompute %main \"main\" %id\n"
4787 "OpExecutionMode %main LocalSize 1 1 1\n"
4789 "OpSource GLSL 430\n"
4790 "OpName %main \"main\"\n"
4791 "OpName %id \"gl_GlobalInvocationID\"\n"
4793 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4795 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4797 "%fmat = OpTypeMatrix %fvec3 3\n"
4798 "%ten = OpConstant %u32 10\n"
4799 "%f32arr10 = OpTypeArray %f32 %ten\n"
4800 "%fst = OpTypeStruct %f32 %f32\n"
4802 + string(getComputeAsmInputOutputBuffer()) +
4804 "%id = OpVariable %uvec3ptr Input\n"
4805 "%zero = OpConstant %i32 0\n"
4807 // Create a bunch of null values
4808 "%unull = OpConstantNull %u32\n"
4809 "%fnull = OpConstantNull %f32\n"
4810 "%vnull = OpConstantNull %fvec3\n"
4811 "%mnull = OpConstantNull %fmat\n"
4812 "%anull = OpConstantNull %f32arr10\n"
4813 "%snull = OpConstantComposite %fst %fnull %fnull\n"
4815 "%main = OpFunction %void None %voidf\n"
4816 "%label = OpLabel\n"
4817 "%idval = OpLoad %uvec3 %id\n"
4818 "%x = OpCompositeExtract %u32 %idval 0\n"
4819 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4820 "%inval = OpLoad %f32 %inloc\n"
4821 "%neg = OpFNegate %f32 %inval\n"
4823 // Get the abs() of (a certain element of) those null values
4824 "%unull_cov = OpConvertUToF %f32 %unull\n"
4825 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
4826 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
4827 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
4828 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
4829 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
4830 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
4831 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
4832 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
4833 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
4834 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
4837 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
4838 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
4839 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
4840 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
4841 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
4842 "%final = OpFAdd %f32 %add5 %snull_abs\n"
4844 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4845 " OpStore %outloc %final\n" // write to output
4848 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4849 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4850 spec.numWorkGroups = IVec3(numElements, 1, 1);
4852 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
4854 return group.release();
4857 // Assembly code used for testing loop control is based on GLSL source code:
4860 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4861 // float elements[];
4863 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4864 // float elements[];
4868 // uint x = gl_GlobalInvocationID.x;
4869 // output_data.elements[x] = input_data.elements[x];
4870 // for (uint i = 0; i < 4; ++i)
4871 // output_data.elements[x] += 1.f;
4873 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
4875 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
4876 vector<CaseParameter> cases;
4877 de::Random rnd (deStringHash(group->getName()));
4878 const int numElements = 100;
4879 vector<float> inputFloats (numElements, 0);
4880 vector<float> outputFloats (numElements, 0);
4881 const StringTemplate shaderTemplate (
4882 string(getComputeAsmShaderPreamble()) +
4884 "OpSource GLSL 430\n"
4885 "OpName %main \"main\"\n"
4886 "OpName %id \"gl_GlobalInvocationID\"\n"
4888 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4890 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4892 "%u32ptr = OpTypePointer Function %u32\n"
4894 "%id = OpVariable %uvec3ptr Input\n"
4895 "%zero = OpConstant %i32 0\n"
4896 "%uzero = OpConstant %u32 0\n"
4897 "%one = OpConstant %i32 1\n"
4898 "%constf1 = OpConstant %f32 1.0\n"
4899 "%four = OpConstant %u32 4\n"
4901 "%main = OpFunction %void None %voidf\n"
4902 "%entry = OpLabel\n"
4903 "%i = OpVariable %u32ptr Function\n"
4904 " OpStore %i %uzero\n"
4906 "%idval = OpLoad %uvec3 %id\n"
4907 "%x = OpCompositeExtract %u32 %idval 0\n"
4908 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4909 "%inval = OpLoad %f32 %inloc\n"
4910 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4911 " OpStore %outloc %inval\n"
4912 " OpBranch %loop_entry\n"
4914 "%loop_entry = OpLabel\n"
4915 "%i_val = OpLoad %u32 %i\n"
4916 "%cmp_lt = OpULessThan %bool %i_val %four\n"
4917 " OpLoopMerge %loop_merge %loop_body ${CONTROL}\n"
4918 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
4919 "%loop_body = OpLabel\n"
4920 "%outval = OpLoad %f32 %outloc\n"
4921 "%addf1 = OpFAdd %f32 %outval %constf1\n"
4922 " OpStore %outloc %addf1\n"
4923 "%new_i = OpIAdd %u32 %i_val %one\n"
4924 " OpStore %i %new_i\n"
4925 " OpBranch %loop_entry\n"
4926 "%loop_merge = OpLabel\n"
4928 " OpFunctionEnd\n");
4930 cases.push_back(CaseParameter("none", "None"));
4931 cases.push_back(CaseParameter("unroll", "Unroll"));
4932 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
4933 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
4935 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4937 for (size_t ndx = 0; ndx < numElements; ++ndx)
4938 outputFloats[ndx] = inputFloats[ndx] + 4.f;
4940 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4942 map<string, string> specializations;
4943 ComputeShaderSpec spec;
4945 specializations["CONTROL"] = cases[caseNdx].param;
4946 spec.assembly = shaderTemplate.specialize(specializations);
4947 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4948 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4949 spec.numWorkGroups = IVec3(numElements, 1, 1);
4951 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4954 group->addChild(new SpvAsmLoopControlDependencyLengthCase(testCtx, "dependency_length", "dependency_length"));
4955 group->addChild(new SpvAsmLoopControlDependencyInfiniteCase(testCtx, "dependency_infinite", "dependency_infinite"));
4957 return group.release();
4960 // Assembly code used for testing selection control is based on GLSL source code:
4963 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4964 // float elements[];
4966 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4967 // float elements[];
4971 // uint x = gl_GlobalInvocationID.x;
4972 // float val = input_data.elements[x];
4974 // output_data.elements[x] = val + 1.f;
4976 // output_data.elements[x] = val - 1.f;
4978 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
4980 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
4981 vector<CaseParameter> cases;
4982 de::Random rnd (deStringHash(group->getName()));
4983 const int numElements = 100;
4984 vector<float> inputFloats (numElements, 0);
4985 vector<float> outputFloats (numElements, 0);
4986 const StringTemplate shaderTemplate (
4987 string(getComputeAsmShaderPreamble()) +
4989 "OpSource GLSL 430\n"
4990 "OpName %main \"main\"\n"
4991 "OpName %id \"gl_GlobalInvocationID\"\n"
4993 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4995 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4997 "%id = OpVariable %uvec3ptr Input\n"
4998 "%zero = OpConstant %i32 0\n"
4999 "%constf1 = OpConstant %f32 1.0\n"
5000 "%constf10 = OpConstant %f32 10.0\n"
5002 "%main = OpFunction %void None %voidf\n"
5003 "%entry = OpLabel\n"
5004 "%idval = OpLoad %uvec3 %id\n"
5005 "%x = OpCompositeExtract %u32 %idval 0\n"
5006 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5007 "%inval = OpLoad %f32 %inloc\n"
5008 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5009 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
5011 " OpSelectionMerge %if_end ${CONTROL}\n"
5012 " OpBranchConditional %cmp_gt %if_true %if_false\n"
5013 "%if_true = OpLabel\n"
5014 "%addf1 = OpFAdd %f32 %inval %constf1\n"
5015 " OpStore %outloc %addf1\n"
5016 " OpBranch %if_end\n"
5017 "%if_false = OpLabel\n"
5018 "%subf1 = OpFSub %f32 %inval %constf1\n"
5019 " OpStore %outloc %subf1\n"
5020 " OpBranch %if_end\n"
5021 "%if_end = OpLabel\n"
5023 " OpFunctionEnd\n");
5025 cases.push_back(CaseParameter("none", "None"));
5026 cases.push_back(CaseParameter("flatten", "Flatten"));
5027 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
5028 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
5030 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5032 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
5033 floorAll(inputFloats);
5035 for (size_t ndx = 0; ndx < numElements; ++ndx)
5036 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
5038 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5040 map<string, string> specializations;
5041 ComputeShaderSpec spec;
5043 specializations["CONTROL"] = cases[caseNdx].param;
5044 spec.assembly = shaderTemplate.specialize(specializations);
5045 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5046 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5047 spec.numWorkGroups = IVec3(numElements, 1, 1);
5049 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5052 return group.release();
5055 tcu::TestCaseGroup* createOpNameGroup(tcu::TestContext& testCtx)
5057 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opname", "Tests OpName cases"));
5058 de::MovePtr<tcu::TestCaseGroup> entryMainGroup (new tcu::TestCaseGroup(testCtx, "entry_main", "OpName tests with entry main"));
5059 de::MovePtr<tcu::TestCaseGroup> entryNotGroup (new tcu::TestCaseGroup(testCtx, "entry_rdc", "OpName tests with entry rdc"));
5060 vector<CaseParameter> cases;
5061 vector<string> testFunc;
5062 de::Random rnd (deStringHash(group->getName()));
5063 const int numElements = 100;
5064 vector<float> inputFloats (numElements, 0);
5065 vector<float> outputFloats (numElements, 0);
5067 fillRandomScalars(rnd, -100.0f, 100.0f, &inputFloats[0], numElements);
5069 for(size_t ndx = 0; ndx < numElements; ++ndx)
5070 outputFloats[ndx] = -inputFloats[ndx];
5072 const StringTemplate shaderTemplate (
5073 "OpCapability Shader\n"
5074 "OpMemoryModel Logical GLSL450\n"
5075 "OpEntryPoint GLCompute %main \"${ENTRY}\" %id\n"
5076 "OpExecutionMode %main LocalSize 1 1 1\n"
5078 "OpName %${FUNC_ID} \"${NAME}\"\n"
5080 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5082 + string(getComputeAsmInputOutputBufferTraits())
5084 + string(getComputeAsmCommonTypes())
5086 + string(getComputeAsmInputOutputBuffer()) +
5088 "%id = OpVariable %uvec3ptr Input\n"
5089 "%zero = OpConstant %i32 0\n"
5091 "%func = OpFunction %void None %voidf\n"
5096 "%main = OpFunction %void None %voidf\n"
5097 "%entry = OpLabel\n"
5098 "%7 = OpFunctionCall %void %func\n"
5100 "%idval = OpLoad %uvec3 %id\n"
5101 "%x = OpCompositeExtract %u32 %idval 0\n"
5103 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5104 "%inval = OpLoad %f32 %inloc\n"
5105 "%neg = OpFNegate %f32 %inval\n"
5106 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5107 " OpStore %outloc %neg\n"
5111 " OpFunctionEnd\n");
5113 cases.push_back(CaseParameter("_is_main", "main"));
5114 cases.push_back(CaseParameter("_is_not_main", "not_main"));
5116 testFunc.push_back("main");
5117 testFunc.push_back("func");
5119 for(size_t fNdx = 0; fNdx < testFunc.size(); ++fNdx)
5121 for(size_t ndx = 0; ndx < cases.size(); ++ndx)
5123 map<string, string> specializations;
5124 ComputeShaderSpec spec;
5126 specializations["ENTRY"] = "main";
5127 specializations["FUNC_ID"] = testFunc[fNdx];
5128 specializations["NAME"] = cases[ndx].param;
5129 spec.assembly = shaderTemplate.specialize(specializations);
5130 spec.numWorkGroups = IVec3(numElements, 1, 1);
5131 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5132 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5134 entryMainGroup->addChild(new SpvAsmComputeShaderCase(testCtx, (testFunc[fNdx] + cases[ndx].name).c_str(), cases[ndx].name, spec));
5138 cases.push_back(CaseParameter("_is_entry", "rdc"));
5140 for(size_t fNdx = 0; fNdx < testFunc.size(); ++fNdx)
5142 for(size_t ndx = 0; ndx < cases.size(); ++ndx)
5144 map<string, string> specializations;
5145 ComputeShaderSpec spec;
5147 specializations["ENTRY"] = "rdc";
5148 specializations["FUNC_ID"] = testFunc[fNdx];
5149 specializations["NAME"] = cases[ndx].param;
5150 spec.assembly = shaderTemplate.specialize(specializations);
5151 spec.numWorkGroups = IVec3(numElements, 1, 1);
5152 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5153 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5154 spec.entryPoint = "rdc";
5156 entryNotGroup->addChild(new SpvAsmComputeShaderCase(testCtx, (testFunc[fNdx] + cases[ndx].name).c_str(), cases[ndx].name, spec));
5160 group->addChild(entryMainGroup.release());
5161 group->addChild(entryNotGroup.release());
5163 return group.release();
5166 // Assembly code used for testing function control is based on GLSL source code:
5170 // layout(std140, set = 0, binding = 0) readonly buffer Input {
5171 // float elements[];
5173 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
5174 // float elements[];
5177 // float const10() { return 10.f; }
5180 // uint x = gl_GlobalInvocationID.x;
5181 // output_data.elements[x] = input_data.elements[x] + const10();
5183 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
5185 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
5186 vector<CaseParameter> cases;
5187 de::Random rnd (deStringHash(group->getName()));
5188 const int numElements = 100;
5189 vector<float> inputFloats (numElements, 0);
5190 vector<float> outputFloats (numElements, 0);
5191 const StringTemplate shaderTemplate (
5192 string(getComputeAsmShaderPreamble()) +
5194 "OpSource GLSL 430\n"
5195 "OpName %main \"main\"\n"
5196 "OpName %func_const10 \"const10(\"\n"
5197 "OpName %id \"gl_GlobalInvocationID\"\n"
5199 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5201 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5203 "%f32f = OpTypeFunction %f32\n"
5204 "%id = OpVariable %uvec3ptr Input\n"
5205 "%zero = OpConstant %i32 0\n"
5206 "%constf10 = OpConstant %f32 10.0\n"
5208 "%main = OpFunction %void None %voidf\n"
5209 "%entry = OpLabel\n"
5210 "%idval = OpLoad %uvec3 %id\n"
5211 "%x = OpCompositeExtract %u32 %idval 0\n"
5212 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5213 "%inval = OpLoad %f32 %inloc\n"
5214 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
5215 "%fadd = OpFAdd %f32 %inval %ret_10\n"
5216 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5217 " OpStore %outloc %fadd\n"
5221 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
5222 "%label = OpLabel\n"
5223 " OpReturnValue %constf10\n"
5224 " OpFunctionEnd\n");
5226 cases.push_back(CaseParameter("none", "None"));
5227 cases.push_back(CaseParameter("inline", "Inline"));
5228 cases.push_back(CaseParameter("dont_inline", "DontInline"));
5229 cases.push_back(CaseParameter("pure", "Pure"));
5230 cases.push_back(CaseParameter("const", "Const"));
5231 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
5232 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
5233 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
5234 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
5236 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5238 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
5239 floorAll(inputFloats);
5241 for (size_t ndx = 0; ndx < numElements; ++ndx)
5242 outputFloats[ndx] = inputFloats[ndx] + 10.f;
5244 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5246 map<string, string> specializations;
5247 ComputeShaderSpec spec;
5249 specializations["CONTROL"] = cases[caseNdx].param;
5250 spec.assembly = shaderTemplate.specialize(specializations);
5251 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5252 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5253 spec.numWorkGroups = IVec3(numElements, 1, 1);
5255 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5258 return group.release();
5261 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
5263 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
5264 vector<CaseParameter> cases;
5265 de::Random rnd (deStringHash(group->getName()));
5266 const int numElements = 100;
5267 vector<float> inputFloats (numElements, 0);
5268 vector<float> outputFloats (numElements, 0);
5269 const StringTemplate shaderTemplate (
5270 string(getComputeAsmShaderPreamble()) +
5272 "OpSource GLSL 430\n"
5273 "OpName %main \"main\"\n"
5274 "OpName %id \"gl_GlobalInvocationID\"\n"
5276 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5278 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
5280 "%f32ptr_f = OpTypePointer Function %f32\n"
5282 "%id = OpVariable %uvec3ptr Input\n"
5283 "%zero = OpConstant %i32 0\n"
5284 "%four = OpConstant %i32 4\n"
5286 "%main = OpFunction %void None %voidf\n"
5287 "%label = OpLabel\n"
5288 "%copy = OpVariable %f32ptr_f Function\n"
5289 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
5290 "%x = OpCompositeExtract %u32 %idval 0\n"
5291 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5292 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5293 " OpCopyMemory %copy %inloc ${ACCESS}\n"
5294 "%val1 = OpLoad %f32 %copy\n"
5295 "%val2 = OpLoad %f32 %inloc\n"
5296 "%add = OpFAdd %f32 %val1 %val2\n"
5297 " OpStore %outloc %add ${ACCESS}\n"
5299 " OpFunctionEnd\n");
5301 cases.push_back(CaseParameter("null", ""));
5302 cases.push_back(CaseParameter("none", "None"));
5303 cases.push_back(CaseParameter("volatile", "Volatile"));
5304 cases.push_back(CaseParameter("aligned", "Aligned 4"));
5305 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
5306 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
5307 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
5309 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
5311 for (size_t ndx = 0; ndx < numElements; ++ndx)
5312 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
5314 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5316 map<string, string> specializations;
5317 ComputeShaderSpec spec;
5319 specializations["ACCESS"] = cases[caseNdx].param;
5320 spec.assembly = shaderTemplate.specialize(specializations);
5321 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
5322 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
5323 spec.numWorkGroups = IVec3(numElements, 1, 1);
5325 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5328 return group.release();
5331 // Checks that we can get undefined values for various types, without exercising a computation with it.
5332 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
5334 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
5335 vector<CaseParameter> cases;
5336 de::Random rnd (deStringHash(group->getName()));
5337 const int numElements = 100;
5338 vector<float> positiveFloats (numElements, 0);
5339 vector<float> negativeFloats (numElements, 0);
5340 const StringTemplate shaderTemplate (
5341 string(getComputeAsmShaderPreamble()) +
5343 "OpSource GLSL 430\n"
5344 "OpName %main \"main\"\n"
5345 "OpName %id \"gl_GlobalInvocationID\"\n"
5347 "OpDecorate %id BuiltIn GlobalInvocationId\n"
5349 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
5350 "%uvec2 = OpTypeVector %u32 2\n"
5351 "%fvec4 = OpTypeVector %f32 4\n"
5352 "%fmat33 = OpTypeMatrix %fvec3 3\n"
5353 "%image = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
5354 "%sampler = OpTypeSampler\n"
5355 "%simage = OpTypeSampledImage %image\n"
5356 "%const100 = OpConstant %u32 100\n"
5357 "%uarr100 = OpTypeArray %i32 %const100\n"
5358 "%struct = OpTypeStruct %f32 %i32 %u32\n"
5359 "%pointer = OpTypePointer Function %i32\n"
5360 + string(getComputeAsmInputOutputBuffer()) +
5362 "%id = OpVariable %uvec3ptr Input\n"
5363 "%zero = OpConstant %i32 0\n"
5365 "%main = OpFunction %void None %voidf\n"
5366 "%label = OpLabel\n"
5368 "%undef = OpUndef ${TYPE}\n"
5370 "%idval = OpLoad %uvec3 %id\n"
5371 "%x = OpCompositeExtract %u32 %idval 0\n"
5373 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
5374 "%inval = OpLoad %f32 %inloc\n"
5375 "%neg = OpFNegate %f32 %inval\n"
5376 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
5377 " OpStore %outloc %neg\n"
5379 " OpFunctionEnd\n");
5381 cases.push_back(CaseParameter("bool", "%bool"));
5382 cases.push_back(CaseParameter("sint32", "%i32"));
5383 cases.push_back(CaseParameter("uint32", "%u32"));
5384 cases.push_back(CaseParameter("float32", "%f32"));
5385 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
5386 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
5387 cases.push_back(CaseParameter("matrix", "%fmat33"));
5388 cases.push_back(CaseParameter("image", "%image"));
5389 cases.push_back(CaseParameter("sampler", "%sampler"));
5390 cases.push_back(CaseParameter("sampledimage", "%simage"));
5391 cases.push_back(CaseParameter("array", "%uarr100"));
5392 cases.push_back(CaseParameter("runtimearray", "%f32arr"));
5393 cases.push_back(CaseParameter("struct", "%struct"));
5394 cases.push_back(CaseParameter("pointer", "%pointer"));
5396 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
5398 for (size_t ndx = 0; ndx < numElements; ++ndx)
5399 negativeFloats[ndx] = -positiveFloats[ndx];
5401 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5403 map<string, string> specializations;
5404 ComputeShaderSpec spec;
5406 specializations["TYPE"] = cases[caseNdx].param;
5407 spec.assembly = shaderTemplate.specialize(specializations);
5408 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
5409 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
5410 spec.numWorkGroups = IVec3(numElements, 1, 1);
5412 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
5415 return group.release();
5419 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
5421 struct NameCodePair { string name, code; };
5422 RGBA defaultColors[4];
5423 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
5424 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
5425 map<string, string> fragments = passthruFragments();
5426 const NameCodePair tests[] =
5428 {"unknown", "OpSource Unknown 321"},
5429 {"essl", "OpSource ESSL 310"},
5430 {"glsl", "OpSource GLSL 450"},
5431 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
5432 {"opencl_c", "OpSource OpenCL_C 120"},
5433 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
5434 {"file", opsourceGLSLWithFile},
5435 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
5436 // Longest possible source string: SPIR-V limits instructions to 65535
5437 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
5438 // contain 65530 UTF8 characters (one word each) plus one last word
5439 // containing 3 ASCII characters and \0.
5440 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
5443 getDefaultColors(defaultColors);
5444 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5446 fragments["debug"] = tests[testNdx].code;
5447 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5450 return opSourceTests.release();
5453 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
5455 struct NameCodePair { string name, code; };
5456 RGBA defaultColors[4];
5457 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
5458 map<string, string> fragments = passthruFragments();
5459 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
5460 const NameCodePair tests[] =
5462 {"empty", opsource + "OpSourceContinued \"\""},
5463 {"short", opsource + "OpSourceContinued \"abcde\""},
5464 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
5465 // Longest possible source string: SPIR-V limits instructions to 65535
5466 // words, of which the first one is OpSourceContinued/length; the rest
5467 // will contain 65533 UTF8 characters (one word each) plus one last word
5468 // containing 3 ASCII characters and \0.
5469 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
5472 getDefaultColors(defaultColors);
5473 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5475 fragments["debug"] = tests[testNdx].code;
5476 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5479 return opSourceTests.release();
5481 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
5483 RGBA defaultColors[4];
5484 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
5485 map<string, string> fragments;
5486 getDefaultColors(defaultColors);
5487 fragments["debug"] =
5488 "%name = OpString \"name\"\n";
5490 fragments["pre_main"] =
5493 "OpLine %name 1 1\n"
5495 "OpLine %name 1 1\n"
5496 "OpLine %name 1 1\n"
5497 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5499 "OpLine %name 1 1\n"
5501 "OpLine %name 1 1\n"
5502 "OpLine %name 1 1\n"
5503 "%second_param1 = OpFunctionParameter %v4f32\n"
5506 "%label_secondfunction = OpLabel\n"
5508 "OpReturnValue %second_param1\n"
5513 fragments["testfun"] =
5514 // A %test_code function that returns its argument unchanged.
5517 "OpLine %name 1 1\n"
5518 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5520 "%param1 = OpFunctionParameter %v4f32\n"
5523 "%label_testfun = OpLabel\n"
5525 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5526 "OpReturnValue %val1\n"
5528 "OpLine %name 1 1\n"
5531 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
5533 return opLineTests.release();
5536 tcu::TestCaseGroup* createOpModuleProcessedTests(tcu::TestContext& testCtx)
5538 RGBA defaultColors[4];
5539 de::MovePtr<tcu::TestCaseGroup> opModuleProcessedTests (new tcu::TestCaseGroup(testCtx, "opmoduleprocessed", "OpModuleProcessed instruction"));
5540 map<string, string> fragments;
5541 std::vector<std::string> noExtensions;
5542 GraphicsResources resources;
5544 getDefaultColors(defaultColors);
5545 resources.verifyBinary = veryfiBinaryShader;
5546 resources.spirvVersion = SPIRV_VERSION_1_3;
5548 fragments["moduleprocessed"] =
5549 "OpModuleProcessed \"VULKAN CTS\"\n"
5550 "OpModuleProcessed \"Negative values\"\n"
5551 "OpModuleProcessed \"Date: 2017/09/21\"\n";
5553 fragments["pre_main"] =
5554 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5555 "%second_param1 = OpFunctionParameter %v4f32\n"
5556 "%label_secondfunction = OpLabel\n"
5557 "OpReturnValue %second_param1\n"
5560 fragments["testfun"] =
5561 // A %test_code function that returns its argument unchanged.
5562 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5563 "%param1 = OpFunctionParameter %v4f32\n"
5564 "%label_testfun = OpLabel\n"
5565 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5566 "OpReturnValue %val1\n"
5569 createTestsForAllStages ("opmoduleprocessed", defaultColors, defaultColors, fragments, resources, noExtensions, opModuleProcessedTests.get());
5571 return opModuleProcessedTests.release();
5575 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
5577 RGBA defaultColors[4];
5578 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
5579 map<string, string> fragments;
5580 std::vector<std::pair<std::string, std::string> > problemStrings;
5582 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
5583 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
5584 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
5585 getDefaultColors(defaultColors);
5587 fragments["debug"] =
5588 "%other_name = OpString \"other_name\"\n";
5590 fragments["pre_main"] =
5591 "OpLine %file_name 32 0\n"
5592 "OpLine %file_name 32 32\n"
5593 "OpLine %file_name 32 40\n"
5594 "OpLine %other_name 32 40\n"
5595 "OpLine %other_name 0 100\n"
5596 "OpLine %other_name 0 4294967295\n"
5597 "OpLine %other_name 4294967295 0\n"
5598 "OpLine %other_name 32 40\n"
5599 "OpLine %file_name 0 0\n"
5600 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5601 "OpLine %file_name 1 0\n"
5602 "%second_param1 = OpFunctionParameter %v4f32\n"
5603 "OpLine %file_name 1 3\n"
5604 "OpLine %file_name 1 2\n"
5605 "%label_secondfunction = OpLabel\n"
5606 "OpLine %file_name 0 2\n"
5607 "OpReturnValue %second_param1\n"
5609 "OpLine %file_name 0 2\n"
5610 "OpLine %file_name 0 2\n";
5612 fragments["testfun"] =
5613 // A %test_code function that returns its argument unchanged.
5614 "OpLine %file_name 1 0\n"
5615 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5616 "OpLine %file_name 16 330\n"
5617 "%param1 = OpFunctionParameter %v4f32\n"
5618 "OpLine %file_name 14 442\n"
5619 "%label_testfun = OpLabel\n"
5620 "OpLine %file_name 11 1024\n"
5621 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5622 "OpLine %file_name 2 97\n"
5623 "OpReturnValue %val1\n"
5625 "OpLine %file_name 5 32\n";
5627 for (size_t i = 0; i < problemStrings.size(); ++i)
5629 map<string, string> testFragments = fragments;
5630 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
5631 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
5634 return opLineTests.release();
5637 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
5639 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
5643 const char functionStart[] =
5644 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5645 "%param1 = OpFunctionParameter %v4f32\n"
5648 const char functionEnd[] =
5649 "OpReturnValue %transformed_param\n"
5652 struct NameConstantsCode
5659 NameConstantsCode tests[] =
5663 "%cnull = OpConstantNull %v4f32\n",
5664 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
5668 "%cnull = OpConstantNull %f32\n",
5669 "%vp = OpVariable %fp_v4f32 Function\n"
5670 "%v = OpLoad %v4f32 %vp\n"
5671 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
5672 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
5673 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
5674 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
5675 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
5679 "%cnull = OpConstantNull %bool\n",
5680 "%v = OpVariable %fp_v4f32 Function\n"
5681 " OpStore %v %param1\n"
5682 " OpSelectionMerge %false_label None\n"
5683 " OpBranchConditional %cnull %true_label %false_label\n"
5684 "%true_label = OpLabel\n"
5685 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
5686 " OpBranch %false_label\n"
5687 "%false_label = OpLabel\n"
5688 "%transformed_param = OpLoad %v4f32 %v\n"
5692 "%cnull = OpConstantNull %i32\n",
5693 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
5694 "%b = OpIEqual %bool %cnull %c_i32_0\n"
5695 " OpSelectionMerge %false_label None\n"
5696 " OpBranchConditional %b %true_label %false_label\n"
5697 "%true_label = OpLabel\n"
5698 " OpStore %v %param1\n"
5699 " OpBranch %false_label\n"
5700 "%false_label = OpLabel\n"
5701 "%transformed_param = OpLoad %v4f32 %v\n"
5705 "%stype = OpTypeStruct %f32 %v4f32\n"
5706 "%fp_stype = OpTypePointer Function %stype\n"
5707 "%cnull = OpConstantNull %stype\n",
5708 "%v = OpVariable %fp_stype Function %cnull\n"
5709 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
5710 "%f_val = OpLoad %v4f32 %f\n"
5711 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
5715 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
5716 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
5717 "%cnull = OpConstantNull %a4_v4f32\n",
5718 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
5719 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5720 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
5721 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
5722 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
5723 "%f_val = OpLoad %v4f32 %f\n"
5724 "%f1_val = OpLoad %v4f32 %f1\n"
5725 "%f2_val = OpLoad %v4f32 %f2\n"
5726 "%f3_val = OpLoad %v4f32 %f3\n"
5727 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
5728 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
5729 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
5730 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
5734 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5735 "%cnull = OpConstantNull %mat4x4_f32\n",
5736 // Our null matrix * any vector should result in a zero vector.
5737 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
5738 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
5742 getHalfColorsFullAlpha(colors);
5744 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5746 map<string, string> fragments;
5747 fragments["pre_main"] = tests[testNdx].constants;
5748 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5749 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
5751 return opConstantNullTests.release();
5753 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
5755 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
5756 RGBA inputColors[4];
5757 RGBA outputColors[4];
5760 const char functionStart[] =
5761 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5762 "%param1 = OpFunctionParameter %v4f32\n"
5765 const char functionEnd[] =
5766 "OpReturnValue %transformed_param\n"
5769 struct NameConstantsCode
5776 NameConstantsCode tests[] =
5781 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
5782 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
5787 "%stype = OpTypeStruct %v4f32 %f32\n"
5788 "%fp_stype = OpTypePointer Function %stype\n"
5789 "%f32_n_1 = OpConstant %f32 -1.0\n"
5790 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5791 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
5792 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
5794 "%v = OpVariable %fp_stype Function %cval\n"
5795 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5796 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
5797 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
5798 "%f32_val = OpLoad %f32 %f32_ptr\n"
5799 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
5800 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
5801 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
5804 // [1|0|0|0.5] [x] = x + 0.5
5805 // [0|1|0|0.5] [y] = y + 0.5
5806 // [0|0|1|0.5] [z] = z + 0.5
5807 // [0|0|0|1 ] [1] = 1
5810 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5811 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
5812 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
5813 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
5814 "%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"
5815 "%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",
5817 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
5822 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5823 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5824 "%f32_n_1 = OpConstant %f32 -1.0\n"
5825 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5826 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
5828 "%v = OpVariable %fp_a4f32 Function %carr\n"
5829 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
5830 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
5831 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5832 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
5833 "%f_val = OpLoad %f32 %f\n"
5834 "%f1_val = OpLoad %f32 %f1\n"
5835 "%f2_val = OpLoad %f32 %f2\n"
5836 "%f3_val = OpLoad %f32 %f3\n"
5837 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
5838 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
5839 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
5840 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
5841 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5848 // [ 1.0, 1.0, 1.0, 1.0]
5852 // [ 0.0, 0.5, 0.0, 0.0]
5856 // [ 1.0, 1.0, 1.0, 1.0]
5859 "array_of_struct_of_array",
5861 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5862 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5863 "%stype = OpTypeStruct %f32 %a4f32\n"
5864 "%a3stype = OpTypeArray %stype %c_u32_3\n"
5865 "%fp_a3stype = OpTypePointer Function %a3stype\n"
5866 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
5867 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5868 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
5869 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
5870 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
5872 "%v = OpVariable %fp_a3stype Function %carr\n"
5873 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
5874 "%f_l = OpLoad %f32 %f\n"
5875 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
5876 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5880 getHalfColorsFullAlpha(inputColors);
5881 outputColors[0] = RGBA(255, 255, 255, 255);
5882 outputColors[1] = RGBA(255, 127, 127, 255);
5883 outputColors[2] = RGBA(127, 255, 127, 255);
5884 outputColors[3] = RGBA(127, 127, 255, 255);
5886 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5888 map<string, string> fragments;
5889 fragments["pre_main"] = tests[testNdx].constants;
5890 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5891 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
5893 return opConstantCompositeTests.release();
5896 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
5898 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
5899 RGBA inputColors[4];
5900 RGBA outputColors[4];
5901 map<string, string> fragments;
5903 // vec4 test_code(vec4 param) {
5904 // vec4 result = param;
5905 // for (int i = 0; i < 4; ++i) {
5906 // if (i == 0) result[i] = 0.;
5907 // else result[i] = 1. - result[i];
5911 const char function[] =
5912 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5913 "%param1 = OpFunctionParameter %v4f32\n"
5915 "%iptr = OpVariable %fp_i32 Function\n"
5916 "%result = OpVariable %fp_v4f32 Function\n"
5917 " OpStore %iptr %c_i32_0\n"
5918 " OpStore %result %param1\n"
5921 // Loop entry block.
5923 "%ival = OpLoad %i32 %iptr\n"
5924 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5925 " OpLoopMerge %exit %if_entry None\n"
5926 " OpBranchConditional %lt_4 %if_entry %exit\n"
5928 // Merge block for loop.
5930 "%ret = OpLoad %v4f32 %result\n"
5931 " OpReturnValue %ret\n"
5933 // If-statement entry block.
5934 "%if_entry = OpLabel\n"
5935 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5936 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
5937 " OpSelectionMerge %if_exit None\n"
5938 " OpBranchConditional %eq_0 %if_true %if_false\n"
5940 // False branch for if-statement.
5941 "%if_false = OpLabel\n"
5942 "%val = OpLoad %f32 %loc\n"
5943 "%sub = OpFSub %f32 %c_f32_1 %val\n"
5944 " OpStore %loc %sub\n"
5945 " OpBranch %if_exit\n"
5947 // Merge block for if-statement.
5948 "%if_exit = OpLabel\n"
5949 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5950 " OpStore %iptr %ival_next\n"
5953 // True branch for if-statement.
5954 "%if_true = OpLabel\n"
5955 " OpStore %loc %c_f32_0\n"
5956 " OpBranch %if_exit\n"
5960 fragments["testfun"] = function;
5962 inputColors[0] = RGBA(127, 127, 127, 0);
5963 inputColors[1] = RGBA(127, 0, 0, 0);
5964 inputColors[2] = RGBA(0, 127, 0, 0);
5965 inputColors[3] = RGBA(0, 0, 127, 0);
5967 outputColors[0] = RGBA(0, 128, 128, 255);
5968 outputColors[1] = RGBA(0, 255, 255, 255);
5969 outputColors[2] = RGBA(0, 128, 255, 255);
5970 outputColors[3] = RGBA(0, 255, 128, 255);
5972 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5974 return group.release();
5977 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
5979 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
5980 RGBA inputColors[4];
5981 RGBA outputColors[4];
5982 map<string, string> fragments;
5984 const char typesAndConstants[] =
5985 "%c_f32_p2 = OpConstant %f32 0.2\n"
5986 "%c_f32_p4 = OpConstant %f32 0.4\n"
5987 "%c_f32_p6 = OpConstant %f32 0.6\n"
5988 "%c_f32_p8 = OpConstant %f32 0.8\n";
5990 // vec4 test_code(vec4 param) {
5991 // vec4 result = param;
5992 // for (int i = 0; i < 4; ++i) {
5994 // case 0: result[i] += .2; break;
5995 // case 1: result[i] += .6; break;
5996 // case 2: result[i] += .4; break;
5997 // case 3: result[i] += .8; break;
5998 // default: break; // unreachable
6003 const char function[] =
6004 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6005 "%param1 = OpFunctionParameter %v4f32\n"
6007 "%iptr = OpVariable %fp_i32 Function\n"
6008 "%result = OpVariable %fp_v4f32 Function\n"
6009 " OpStore %iptr %c_i32_0\n"
6010 " OpStore %result %param1\n"
6013 // Loop entry block.
6015 "%ival = OpLoad %i32 %iptr\n"
6016 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6017 " OpLoopMerge %exit %switch_exit None\n"
6018 " OpBranchConditional %lt_4 %switch_entry %exit\n"
6020 // Merge block for loop.
6022 "%ret = OpLoad %v4f32 %result\n"
6023 " OpReturnValue %ret\n"
6025 // Switch-statement entry block.
6026 "%switch_entry = OpLabel\n"
6027 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6028 "%val = OpLoad %f32 %loc\n"
6029 " OpSelectionMerge %switch_exit None\n"
6030 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6032 "%case2 = OpLabel\n"
6033 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
6034 " OpStore %loc %addp4\n"
6035 " OpBranch %switch_exit\n"
6037 "%switch_default = OpLabel\n"
6040 "%case3 = OpLabel\n"
6041 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
6042 " OpStore %loc %addp8\n"
6043 " OpBranch %switch_exit\n"
6045 "%case0 = OpLabel\n"
6046 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
6047 " OpStore %loc %addp2\n"
6048 " OpBranch %switch_exit\n"
6050 // Merge block for switch-statement.
6051 "%switch_exit = OpLabel\n"
6052 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6053 " OpStore %iptr %ival_next\n"
6056 "%case1 = OpLabel\n"
6057 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
6058 " OpStore %loc %addp6\n"
6059 " OpBranch %switch_exit\n"
6063 fragments["pre_main"] = typesAndConstants;
6064 fragments["testfun"] = function;
6066 inputColors[0] = RGBA(127, 27, 127, 51);
6067 inputColors[1] = RGBA(127, 0, 0, 51);
6068 inputColors[2] = RGBA(0, 27, 0, 51);
6069 inputColors[3] = RGBA(0, 0, 127, 51);
6071 outputColors[0] = RGBA(178, 180, 229, 255);
6072 outputColors[1] = RGBA(178, 153, 102, 255);
6073 outputColors[2] = RGBA(51, 180, 102, 255);
6074 outputColors[3] = RGBA(51, 153, 229, 255);
6076 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
6078 return group.release();
6081 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
6083 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
6084 RGBA inputColors[4];
6085 RGBA outputColors[4];
6086 map<string, string> fragments;
6088 const char decorations[] =
6089 "OpDecorate %array_group ArrayStride 4\n"
6090 "OpDecorate %struct_member_group Offset 0\n"
6091 "%array_group = OpDecorationGroup\n"
6092 "%struct_member_group = OpDecorationGroup\n"
6094 "OpDecorate %group1 RelaxedPrecision\n"
6095 "OpDecorate %group3 RelaxedPrecision\n"
6096 "OpDecorate %group3 Invariant\n"
6097 "OpDecorate %group3 Restrict\n"
6098 "%group0 = OpDecorationGroup\n"
6099 "%group1 = OpDecorationGroup\n"
6100 "%group3 = OpDecorationGroup\n";
6102 const char typesAndConstants[] =
6103 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
6104 "%struct1 = OpTypeStruct %a3f32\n"
6105 "%struct2 = OpTypeStruct %a3f32\n"
6106 "%fp_struct1 = OpTypePointer Function %struct1\n"
6107 "%fp_struct2 = OpTypePointer Function %struct2\n"
6108 "%c_f32_2 = OpConstant %f32 2.\n"
6109 "%c_f32_n2 = OpConstant %f32 -2.\n"
6111 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
6112 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
6113 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
6114 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
6116 const char function[] =
6117 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6118 "%param = OpFunctionParameter %v4f32\n"
6119 "%entry = OpLabel\n"
6120 "%result = OpVariable %fp_v4f32 Function\n"
6121 "%v_struct1 = OpVariable %fp_struct1 Function\n"
6122 "%v_struct2 = OpVariable %fp_struct2 Function\n"
6123 " OpStore %result %param\n"
6124 " OpStore %v_struct1 %c_struct1\n"
6125 " OpStore %v_struct2 %c_struct2\n"
6126 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
6127 "%val1 = OpLoad %f32 %ptr1\n"
6128 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
6129 "%val2 = OpLoad %f32 %ptr2\n"
6130 "%addvalues = OpFAdd %f32 %val1 %val2\n"
6131 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
6132 "%val = OpLoad %f32 %ptr\n"
6133 "%addresult = OpFAdd %f32 %addvalues %val\n"
6134 " OpStore %ptr %addresult\n"
6135 "%ret = OpLoad %v4f32 %result\n"
6136 " OpReturnValue %ret\n"
6139 struct CaseNameDecoration
6145 CaseNameDecoration tests[] =
6148 "same_decoration_group_on_multiple_types",
6149 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
6152 "empty_decoration_group",
6153 "OpGroupDecorate %group0 %a3f32\n"
6154 "OpGroupDecorate %group0 %result\n"
6157 "one_element_decoration_group",
6158 "OpGroupDecorate %array_group %a3f32\n"
6161 "multiple_elements_decoration_group",
6162 "OpGroupDecorate %group3 %v_struct1\n"
6165 "multiple_decoration_groups_on_same_variable",
6166 "OpGroupDecorate %group0 %v_struct2\n"
6167 "OpGroupDecorate %group1 %v_struct2\n"
6168 "OpGroupDecorate %group3 %v_struct2\n"
6171 "same_decoration_group_multiple_times",
6172 "OpGroupDecorate %group1 %addvalues\n"
6173 "OpGroupDecorate %group1 %addvalues\n"
6174 "OpGroupDecorate %group1 %addvalues\n"
6179 getHalfColorsFullAlpha(inputColors);
6180 getHalfColorsFullAlpha(outputColors);
6182 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
6184 fragments["decoration"] = decorations + tests[idx].decoration;
6185 fragments["pre_main"] = typesAndConstants;
6186 fragments["testfun"] = function;
6188 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
6191 return group.release();
6194 struct SpecConstantTwoIntGraphicsCase
6196 const char* caseName;
6197 const char* scDefinition0;
6198 const char* scDefinition1;
6199 const char* scResultType;
6200 const char* scOperation;
6201 deInt32 scActualValue0;
6202 deInt32 scActualValue1;
6203 const char* resultOperation;
6204 RGBA expectedColors[4];
6206 SpecConstantTwoIntGraphicsCase (const char* name,
6207 const char* definition0,
6208 const char* definition1,
6209 const char* resultType,
6210 const char* operation,
6213 const char* resultOp,
6214 const RGBA (&output)[4])
6216 , scDefinition0 (definition0)
6217 , scDefinition1 (definition1)
6218 , scResultType (resultType)
6219 , scOperation (operation)
6220 , scActualValue0 (value0)
6221 , scActualValue1 (value1)
6222 , resultOperation (resultOp)
6224 expectedColors[0] = output[0];
6225 expectedColors[1] = output[1];
6226 expectedColors[2] = output[2];
6227 expectedColors[3] = output[3];
6231 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
6233 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
6234 vector<SpecConstantTwoIntGraphicsCase> cases;
6235 RGBA inputColors[4];
6236 RGBA outputColors0[4];
6237 RGBA outputColors1[4];
6238 RGBA outputColors2[4];
6240 const char decorations1[] =
6241 "OpDecorate %sc_0 SpecId 0\n"
6242 "OpDecorate %sc_1 SpecId 1\n";
6244 const char typesAndConstants1[] =
6245 "${OPTYPE_DEFINITIONS:opt}"
6246 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
6247 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
6248 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
6250 const char function1[] =
6251 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6252 "%param = OpFunctionParameter %v4f32\n"
6253 "%label = OpLabel\n"
6254 "${TYPE_CONVERT:opt}"
6255 "%result = OpVariable %fp_v4f32 Function\n"
6256 " OpStore %result %param\n"
6257 "%gen = ${GEN_RESULT}\n"
6258 "%index = OpIAdd %i32 %gen %c_i32_1\n"
6259 "%loc = OpAccessChain %fp_f32 %result %index\n"
6260 "%val = OpLoad %f32 %loc\n"
6261 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6262 " OpStore %loc %add\n"
6263 "%ret = OpLoad %v4f32 %result\n"
6264 " OpReturnValue %ret\n"
6267 inputColors[0] = RGBA(127, 127, 127, 255);
6268 inputColors[1] = RGBA(127, 0, 0, 255);
6269 inputColors[2] = RGBA(0, 127, 0, 255);
6270 inputColors[3] = RGBA(0, 0, 127, 255);
6272 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
6273 outputColors0[0] = RGBA(255, 127, 127, 255);
6274 outputColors0[1] = RGBA(255, 0, 0, 255);
6275 outputColors0[2] = RGBA(128, 127, 0, 255);
6276 outputColors0[3] = RGBA(128, 0, 127, 255);
6278 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
6279 outputColors1[0] = RGBA(127, 255, 127, 255);
6280 outputColors1[1] = RGBA(127, 128, 0, 255);
6281 outputColors1[2] = RGBA(0, 255, 0, 255);
6282 outputColors1[3] = RGBA(0, 128, 127, 255);
6284 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
6285 outputColors2[0] = RGBA(127, 127, 255, 255);
6286 outputColors2[1] = RGBA(127, 0, 128, 255);
6287 outputColors2[2] = RGBA(0, 127, 128, 255);
6288 outputColors2[3] = RGBA(0, 0, 255, 255);
6290 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
6291 const char addZeroToSc32[] = "OpIAdd %i32 %c_i32_0 %sc_op32";
6292 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
6293 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
6295 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
6296 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
6297 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
6298 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
6299 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
6300 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6301 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6302 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
6303 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
6304 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
6305 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
6306 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
6307 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
6308 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
6309 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
6310 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
6311 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6312 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
6313 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
6314 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
6315 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6316 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
6317 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
6318 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6319 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6320 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6321 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6322 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
6323 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
6324 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
6325 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
6326 cases.push_back(SpecConstantTwoIntGraphicsCase("sconvert", " %i32 0", " %i32 0", "%i16", "SConvert %sc_0", -1, 0, addZeroToSc32, outputColors0));
6327 // -1082130432 stored as 32-bit two's complement is the binary representation of -1 as IEEE-754 Float
6328 cases.push_back(SpecConstantTwoIntGraphicsCase("fconvert", " %f32 0", " %f32 0", "%f64", "FConvert %sc_0", -1082130432, 0, addZeroToSc32, outputColors0));
6329 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
6331 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
6333 map<string, string> specializations;
6334 map<string, string> fragments;
6335 vector<deInt32> specConstants;
6336 vector<string> features;
6337 PushConstants noPushConstants;
6338 GraphicsResources noResources;
6339 GraphicsInterfaces noInterfaces;
6340 std::vector<std::string> noExtensions;
6342 // Special SPIR-V code for SConvert-case
6343 if (strcmp(cases[caseNdx].caseName, "sconvert") == 0)
6345 features.push_back("shaderInt16");
6346 fragments["capability"] = "OpCapability Int16\n"; // Adds 16-bit integer capability
6347 specializations["OPTYPE_DEFINITIONS"] = "%i16 = OpTypeInt 16 1\n"; // Adds 16-bit integer type
6348 specializations["TYPE_CONVERT"] = "%sc_op32 = OpSConvert %i32 %sc_op\n"; // Converts 16-bit integer to 32-bit integer
6351 // Special SPIR-V code for FConvert-case
6352 if (strcmp(cases[caseNdx].caseName, "fconvert") == 0)
6354 features.push_back("shaderFloat64");
6355 fragments["capability"] = "OpCapability Float64\n"; // Adds 64-bit float capability
6356 specializations["OPTYPE_DEFINITIONS"] = "%f64 = OpTypeFloat 64\n"; // Adds 64-bit float type
6357 specializations["TYPE_CONVERT"] = "%sc_op32 = OpConvertFToS %i32 %sc_op\n"; // Converts 64-bit float to 32-bit integer
6360 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
6361 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
6362 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
6363 specializations["SC_OP"] = cases[caseNdx].scOperation;
6364 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
6366 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
6367 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
6368 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
6370 specConstants.push_back(cases[caseNdx].scActualValue0);
6371 specConstants.push_back(cases[caseNdx].scActualValue1);
6373 createTestsForAllStages(
6374 cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants,
6375 noPushConstants, noResources, noInterfaces, noExtensions, features, VulkanFeatures(), group.get());
6378 const char decorations2[] =
6379 "OpDecorate %sc_0 SpecId 0\n"
6380 "OpDecorate %sc_1 SpecId 1\n"
6381 "OpDecorate %sc_2 SpecId 2\n";
6383 const char typesAndConstants2[] =
6384 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
6385 "%vec3_undef = OpUndef %v3i32\n"
6387 "%sc_0 = OpSpecConstant %i32 0\n"
6388 "%sc_1 = OpSpecConstant %i32 0\n"
6389 "%sc_2 = OpSpecConstant %i32 0\n"
6390 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
6391 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
6392 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
6393 "%sc_vec3_0_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
6394 "%sc_vec3_1_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
6395 "%sc_vec3_2_s = OpSpecConstantOp %v3i32 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
6396 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
6397 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
6398 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
6399 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
6400 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
6401 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
6402 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
6404 const char function2[] =
6405 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6406 "%param = OpFunctionParameter %v4f32\n"
6407 "%label = OpLabel\n"
6408 "%result = OpVariable %fp_v4f32 Function\n"
6409 " OpStore %result %param\n"
6410 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
6411 "%val = OpLoad %f32 %loc\n"
6412 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6413 " OpStore %loc %add\n"
6414 "%ret = OpLoad %v4f32 %result\n"
6415 " OpReturnValue %ret\n"
6418 map<string, string> fragments;
6419 vector<deInt32> specConstants;
6421 fragments["decoration"] = decorations2;
6422 fragments["pre_main"] = typesAndConstants2;
6423 fragments["testfun"] = function2;
6425 specConstants.push_back(56789);
6426 specConstants.push_back(-2);
6427 specConstants.push_back(56788);
6429 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
6431 return group.release();
6434 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
6436 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
6437 RGBA inputColors[4];
6438 RGBA outputColors1[4];
6439 RGBA outputColors2[4];
6440 RGBA outputColors3[4];
6441 map<string, string> fragments1;
6442 map<string, string> fragments2;
6443 map<string, string> fragments3;
6445 const char typesAndConstants1[] =
6446 "%c_f32_p2 = OpConstant %f32 0.2\n"
6447 "%c_f32_p4 = OpConstant %f32 0.4\n"
6448 "%c_f32_p5 = OpConstant %f32 0.5\n"
6449 "%c_f32_p8 = OpConstant %f32 0.8\n";
6451 // vec4 test_code(vec4 param) {
6452 // vec4 result = param;
6453 // for (int i = 0; i < 4; ++i) {
6456 // case 0: operand = .2; break;
6457 // case 1: operand = .5; break;
6458 // case 2: operand = .4; break;
6459 // case 3: operand = .0; break;
6460 // default: break; // unreachable
6462 // result[i] += operand;
6466 const char function1[] =
6467 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6468 "%param1 = OpFunctionParameter %v4f32\n"
6470 "%iptr = OpVariable %fp_i32 Function\n"
6471 "%result = OpVariable %fp_v4f32 Function\n"
6472 " OpStore %iptr %c_i32_0\n"
6473 " OpStore %result %param1\n"
6477 "%ival = OpLoad %i32 %iptr\n"
6478 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6479 " OpLoopMerge %exit %phi None\n"
6480 " OpBranchConditional %lt_4 %entry %exit\n"
6482 "%entry = OpLabel\n"
6483 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6484 "%val = OpLoad %f32 %loc\n"
6485 " OpSelectionMerge %phi None\n"
6486 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6488 "%case0 = OpLabel\n"
6490 "%case1 = OpLabel\n"
6492 "%case2 = OpLabel\n"
6494 "%case3 = OpLabel\n"
6497 "%default = OpLabel\n"
6501 "%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
6502 "%add = OpFAdd %f32 %val %operand\n"
6503 " OpStore %loc %add\n"
6504 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6505 " OpStore %iptr %ival_next\n"
6509 "%ret = OpLoad %v4f32 %result\n"
6510 " OpReturnValue %ret\n"
6514 fragments1["pre_main"] = typesAndConstants1;
6515 fragments1["testfun"] = function1;
6517 getHalfColorsFullAlpha(inputColors);
6519 outputColors1[0] = RGBA(178, 255, 229, 255);
6520 outputColors1[1] = RGBA(178, 127, 102, 255);
6521 outputColors1[2] = RGBA(51, 255, 102, 255);
6522 outputColors1[3] = RGBA(51, 127, 229, 255);
6524 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
6526 const char typesAndConstants2[] =
6527 "%c_f32_p2 = OpConstant %f32 0.2\n";
6529 // Add .4 to the second element of the given parameter.
6530 const char function2[] =
6531 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6532 "%param = OpFunctionParameter %v4f32\n"
6533 "%entry = OpLabel\n"
6534 "%result = OpVariable %fp_v4f32 Function\n"
6535 " OpStore %result %param\n"
6536 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6537 "%val = OpLoad %f32 %loc\n"
6541 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
6542 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
6543 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
6544 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
6545 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
6546 " OpLoopMerge %exit %phi None\n"
6547 " OpBranchConditional %still_loop %phi %exit\n"
6550 " OpStore %loc %accum\n"
6551 "%ret = OpLoad %v4f32 %result\n"
6552 " OpReturnValue %ret\n"
6556 fragments2["pre_main"] = typesAndConstants2;
6557 fragments2["testfun"] = function2;
6559 outputColors2[0] = RGBA(127, 229, 127, 255);
6560 outputColors2[1] = RGBA(127, 102, 0, 255);
6561 outputColors2[2] = RGBA(0, 229, 0, 255);
6562 outputColors2[3] = RGBA(0, 102, 127, 255);
6564 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
6566 const char typesAndConstants3[] =
6567 "%true = OpConstantTrue %bool\n"
6568 "%false = OpConstantFalse %bool\n"
6569 "%c_f32_p2 = OpConstant %f32 0.2\n";
6571 // Swap the second and the third element of the given parameter.
6572 const char function3[] =
6573 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6574 "%param = OpFunctionParameter %v4f32\n"
6575 "%entry = OpLabel\n"
6576 "%result = OpVariable %fp_v4f32 Function\n"
6577 " OpStore %result %param\n"
6578 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6579 "%a_init = OpLoad %f32 %a_loc\n"
6580 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
6581 "%b_init = OpLoad %f32 %b_loc\n"
6585 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
6586 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
6587 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
6588 " OpLoopMerge %exit %phi None\n"
6589 " OpBranchConditional %still_loop %phi %exit\n"
6592 " OpStore %a_loc %a_next\n"
6593 " OpStore %b_loc %b_next\n"
6594 "%ret = OpLoad %v4f32 %result\n"
6595 " OpReturnValue %ret\n"
6599 fragments3["pre_main"] = typesAndConstants3;
6600 fragments3["testfun"] = function3;
6602 outputColors3[0] = RGBA(127, 127, 127, 255);
6603 outputColors3[1] = RGBA(127, 0, 0, 255);
6604 outputColors3[2] = RGBA(0, 0, 127, 255);
6605 outputColors3[3] = RGBA(0, 127, 0, 255);
6607 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
6609 return group.release();
6612 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
6614 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
6615 RGBA inputColors[4];
6616 RGBA outputColors[4];
6618 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
6619 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
6620 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
6621 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
6622 const char constantsAndTypes[] =
6623 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
6624 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
6625 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
6626 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
6627 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n";
6629 const char function[] =
6630 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6631 "%param = OpFunctionParameter %v4f32\n"
6632 "%label = OpLabel\n"
6633 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
6634 "%var2 = OpVariable %fp_f32 Function\n"
6635 "%red = OpCompositeExtract %f32 %param 0\n"
6636 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
6637 " OpStore %var2 %plus_red\n"
6638 "%val1 = OpLoad %f32 %var1\n"
6639 "%val2 = OpLoad %f32 %var2\n"
6640 "%mul = OpFMul %f32 %val1 %val2\n"
6641 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
6642 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
6643 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
6644 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
6645 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
6646 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
6647 " OpReturnValue %ret\n"
6650 struct CaseNameDecoration
6657 CaseNameDecoration tests[] = {
6658 {"multiplication", "OpDecorate %mul NoContraction"},
6659 {"addition", "OpDecorate %add NoContraction"},
6660 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
6663 getHalfColorsFullAlpha(inputColors);
6665 for (deUint8 idx = 0; idx < 4; ++idx)
6667 inputColors[idx].setRed(0);
6668 outputColors[idx] = RGBA(0, 0, 0, 255);
6671 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
6673 map<string, string> fragments;
6675 fragments["decoration"] = tests[testNdx].decoration;
6676 fragments["pre_main"] = constantsAndTypes;
6677 fragments["testfun"] = function;
6679 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
6682 return group.release();
6685 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
6687 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
6690 const char constantsAndTypes[] =
6691 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
6692 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
6693 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
6694 "%fp_stype = OpTypePointer Function %stype\n";
6696 const char function[] =
6697 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6698 "%param1 = OpFunctionParameter %v4f32\n"
6700 "%v1 = OpVariable %fp_v4f32 Function\n"
6701 "%v2 = OpVariable %fp_a2f32 Function\n"
6702 "%v3 = OpVariable %fp_f32 Function\n"
6703 "%v = OpVariable %fp_stype Function\n"
6704 "%vv = OpVariable %fp_stype Function\n"
6705 "%vvv = OpVariable %fp_f32 Function\n"
6707 " OpStore %v1 %c_v4f32_1_1_1_1\n"
6708 " OpStore %v2 %c_a2f32_1\n"
6709 " OpStore %v3 %c_f32_1\n"
6711 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
6712 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
6713 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
6714 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
6715 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
6716 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
6718 " OpStore %p_v4f32 %v1_v ${access_type}\n"
6719 " OpStore %p_a2f32 %v2_v ${access_type}\n"
6720 " OpStore %p_f32 %v3_v ${access_type}\n"
6722 " OpCopyMemory %vv %v ${access_type}\n"
6723 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
6725 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
6726 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
6727 "%v_f32_3 = OpLoad %f32 %vvv\n"
6729 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
6730 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
6731 " OpReturnValue %ret2\n"
6734 struct NameMemoryAccess
6741 NameMemoryAccess tests[] =
6744 { "volatile", "Volatile" },
6745 { "aligned", "Aligned 1" },
6746 { "volatile_aligned", "Volatile|Aligned 1" },
6747 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
6748 { "volatile_nontemporal", "Volatile|Nontemporal" },
6749 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
6752 getHalfColorsFullAlpha(colors);
6754 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
6756 map<string, string> fragments;
6757 map<string, string> memoryAccess;
6758 memoryAccess["access_type"] = tests[testNdx].accessType;
6760 fragments["pre_main"] = constantsAndTypes;
6761 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
6762 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
6764 return memoryAccessTests.release();
6766 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
6768 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
6769 RGBA defaultColors[4];
6770 map<string, string> fragments;
6771 getDefaultColors(defaultColors);
6773 // First, simple cases that don't do anything with the OpUndef result.
6774 struct NameCodePair { string name, decl, type; };
6775 const NameCodePair tests[] =
6777 {"bool", "", "%bool"},
6778 {"vec2uint32", "", "%v2u32"},
6779 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown", "%type"},
6780 {"sampler", "%type = OpTypeSampler", "%type"},
6781 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img", "%type"},
6782 {"pointer", "", "%fp_i32"},
6783 {"runtimearray", "%type = OpTypeRuntimeArray %f32", "%type"},
6784 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100", "%type"},
6785 {"struct", "%type = OpTypeStruct %f32 %i32 %u32", "%type"}};
6786 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
6788 fragments["undef_type"] = tests[testNdx].type;
6789 fragments["testfun"] = StringTemplate(
6790 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6791 "%param1 = OpFunctionParameter %v4f32\n"
6792 "%label_testfun = OpLabel\n"
6793 "%undef = OpUndef ${undef_type}\n"
6794 "OpReturnValue %param1\n"
6795 "OpFunctionEnd\n").specialize(fragments);
6796 fragments["pre_main"] = tests[testNdx].decl;
6797 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
6801 fragments["testfun"] =
6802 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6803 "%param1 = OpFunctionParameter %v4f32\n"
6804 "%label_testfun = OpLabel\n"
6805 "%undef = OpUndef %f32\n"
6806 "%zero = OpFMul %f32 %undef %c_f32_0\n"
6807 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
6808 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
6809 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6810 "%b = OpFAdd %f32 %a %actually_zero\n"
6811 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
6812 "OpReturnValue %ret\n"
6815 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6817 fragments["testfun"] =
6818 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6819 "%param1 = OpFunctionParameter %v4f32\n"
6820 "%label_testfun = OpLabel\n"
6821 "%undef = OpUndef %i32\n"
6822 "%zero = OpIMul %i32 %undef %c_i32_0\n"
6823 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6824 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6825 "OpReturnValue %ret\n"
6828 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6830 fragments["testfun"] =
6831 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6832 "%param1 = OpFunctionParameter %v4f32\n"
6833 "%label_testfun = OpLabel\n"
6834 "%undef = OpUndef %u32\n"
6835 "%zero = OpIMul %u32 %undef %c_i32_0\n"
6836 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6837 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6838 "OpReturnValue %ret\n"
6841 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6843 fragments["testfun"] =
6844 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6845 "%param1 = OpFunctionParameter %v4f32\n"
6846 "%label_testfun = OpLabel\n"
6847 "%undef = OpUndef %v4f32\n"
6848 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
6849 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
6850 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
6851 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
6852 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
6853 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6854 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6855 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6856 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6857 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6858 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
6859 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
6860 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
6861 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6862 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6863 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6864 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6865 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6866 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6867 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6868 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6869 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6870 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6871 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6872 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6873 "OpReturnValue %ret\n"
6876 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6878 fragments["pre_main"] =
6879 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
6880 fragments["testfun"] =
6881 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6882 "%param1 = OpFunctionParameter %v4f32\n"
6883 "%label_testfun = OpLabel\n"
6884 "%undef = OpUndef %m2x2f32\n"
6885 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
6886 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
6887 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
6888 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
6889 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
6890 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6891 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6892 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6893 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6894 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6895 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
6896 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
6897 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
6898 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6899 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6900 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6901 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6902 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6903 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6904 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6905 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6906 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6907 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6908 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6909 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6910 "OpReturnValue %ret\n"
6913 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
6915 return opUndefTests.release();
6918 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
6920 const RGBA inputColors[4] =
6923 RGBA(0, 0, 255, 255),
6924 RGBA(0, 255, 0, 255),
6925 RGBA(0, 255, 255, 255)
6928 const RGBA expectedColors[4] =
6930 RGBA(255, 0, 0, 255),
6931 RGBA(255, 0, 0, 255),
6932 RGBA(255, 0, 0, 255),
6933 RGBA(255, 0, 0, 255)
6936 const struct SingleFP16Possibility
6939 const char* constant; // Value to assign to %test_constant.
6941 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
6947 -constructNormalizedFloat(1, 0x300000),
6948 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6953 constructNormalizedFloat(7, 0x000000),
6954 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6956 // SPIR-V requires that OpQuantizeToF16 flushes
6957 // any numbers that would end up denormalized in F16 to zero.
6961 std::ldexp(1.5f, -140),
6962 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6967 -std::ldexp(1.5f, -140),
6968 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6973 std::ldexp(1.0f, -16),
6974 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6975 }, // too small positive
6977 "negative_too_small",
6979 -std::ldexp(1.0f, -32),
6980 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6981 }, // too small negative
6985 -std::ldexp(1.0f, 128),
6987 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6988 "%inf = OpIsInf %bool %c\n"
6989 "%cond = OpLogicalAnd %bool %gz %inf\n"
6994 std::ldexp(1.0f, 128),
6996 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6997 "%inf = OpIsInf %bool %c\n"
6998 "%cond = OpLogicalAnd %bool %gz %inf\n"
7001 "round_to_negative_inf",
7003 -std::ldexp(1.0f, 32),
7005 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
7006 "%inf = OpIsInf %bool %c\n"
7007 "%cond = OpLogicalAnd %bool %gz %inf\n"
7012 std::ldexp(1.0f, 16),
7014 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
7015 "%inf = OpIsInf %bool %c\n"
7016 "%cond = OpLogicalAnd %bool %gz %inf\n"
7021 std::numeric_limits<float>::quiet_NaN(),
7023 // Test for any NaN value, as NaNs are not preserved
7024 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
7025 "%cond = OpIsNan %bool %direct_quant\n"
7030 std::numeric_limits<float>::quiet_NaN(),
7032 // Test for any NaN value, as NaNs are not preserved
7033 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
7034 "%cond = OpIsNan %bool %direct_quant\n"
7037 const char* constants =
7038 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
7040 StringTemplate function (
7041 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7042 "%param1 = OpFunctionParameter %v4f32\n"
7043 "%label_testfun = OpLabel\n"
7044 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7045 "%b = OpFAdd %f32 %test_constant %a\n"
7046 "%c = OpQuantizeToF16 %f32 %b\n"
7048 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7049 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
7050 " OpReturnValue %retval\n"
7054 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
7055 const char* specConstants =
7056 "%test_constant = OpSpecConstant %f32 0.\n"
7057 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
7059 StringTemplate specConstantFunction(
7060 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7061 "%param1 = OpFunctionParameter %v4f32\n"
7062 "%label_testfun = OpLabel\n"
7064 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7065 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
7066 " OpReturnValue %retval\n"
7070 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
7072 map<string, string> codeSpecialization;
7073 map<string, string> fragments;
7074 codeSpecialization["condition"] = tests[idx].condition;
7075 fragments["testfun"] = function.specialize(codeSpecialization);
7076 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
7077 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7080 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
7082 map<string, string> codeSpecialization;
7083 map<string, string> fragments;
7084 vector<deInt32> passConstants;
7085 deInt32 specConstant;
7087 codeSpecialization["condition"] = tests[idx].condition;
7088 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
7089 fragments["decoration"] = specDecorations;
7090 fragments["pre_main"] = specConstants;
7092 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
7093 passConstants.push_back(specConstant);
7095 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7099 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
7101 RGBA inputColors[4] = {
7103 RGBA(0, 0, 255, 255),
7104 RGBA(0, 255, 0, 255),
7105 RGBA(0, 255, 255, 255)
7108 RGBA expectedColors[4] =
7110 RGBA(255, 0, 0, 255),
7111 RGBA(255, 0, 0, 255),
7112 RGBA(255, 0, 0, 255),
7113 RGBA(255, 0, 0, 255)
7116 struct DualFP16Possibility
7121 const char* possibleOutput1;
7122 const char* possibleOutput2;
7125 "positive_round_up_or_round_down",
7127 constructNormalizedFloat(8, 0x300300),
7132 "negative_round_up_or_round_down",
7134 -constructNormalizedFloat(-7, 0x600800),
7141 constructNormalizedFloat(2, 0x01e000),
7146 "carry_to_exponent",
7148 constructNormalizedFloat(1, 0xffe000),
7153 StringTemplate constants (
7154 "%input_const = OpConstant %f32 ${input}\n"
7155 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7156 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7159 StringTemplate specConstants (
7160 "%input_const = OpSpecConstant %f32 0.\n"
7161 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7162 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7165 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
7167 const char* function =
7168 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7169 "%param1 = OpFunctionParameter %v4f32\n"
7170 "%label_testfun = OpLabel\n"
7171 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7172 // For the purposes of this test we assume that 0.f will always get
7173 // faithfully passed through the pipeline stages.
7174 "%b = OpFAdd %f32 %input_const %a\n"
7175 "%c = OpQuantizeToF16 %f32 %b\n"
7176 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
7177 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
7178 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
7179 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7180 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
7181 " OpReturnValue %retval\n"
7184 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7185 map<string, string> fragments;
7186 map<string, string> constantSpecialization;
7188 constantSpecialization["input"] = tests[idx].input;
7189 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7190 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7191 fragments["testfun"] = function;
7192 fragments["pre_main"] = constants.specialize(constantSpecialization);
7193 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7196 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7197 map<string, string> fragments;
7198 map<string, string> constantSpecialization;
7199 vector<deInt32> passConstants;
7200 deInt32 specConstant;
7202 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7203 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7204 fragments["testfun"] = function;
7205 fragments["decoration"] = specDecorations;
7206 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
7208 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
7209 passConstants.push_back(specConstant);
7211 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7215 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
7217 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
7218 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
7219 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
7220 return opQuantizeTests.release();
7223 struct ShaderPermutation
7225 deUint8 vertexPermutation;
7226 deUint8 geometryPermutation;
7227 deUint8 tesscPermutation;
7228 deUint8 tessePermutation;
7229 deUint8 fragmentPermutation;
7232 ShaderPermutation getShaderPermutation(deUint8 inputValue)
7234 ShaderPermutation permutation =
7236 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
7237 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
7238 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
7239 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
7240 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
7245 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
7247 RGBA defaultColors[4];
7248 RGBA invertedColors[4];
7249 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
7251 const ShaderElement combinedPipeline[] =
7253 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
7254 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
7255 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7256 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7257 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
7260 getDefaultColors(defaultColors);
7261 getInvertedDefaultColors(invertedColors);
7262 addFunctionCaseWithPrograms<InstanceContext>(
7263 moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline,
7264 createInstanceContext(combinedPipeline, map<string, string>()));
7266 const char* numbers[] =
7271 for (deInt8 idx = 0; idx < 32; ++idx)
7273 ShaderPermutation permutation = getShaderPermutation(idx);
7274 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
7275 const ShaderElement pipeline[] =
7277 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
7278 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
7279 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7280 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7281 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
7284 // If there are an even number of swaps, then it should be no-op.
7285 // If there are an odd number, the color should be flipped.
7286 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
7288 addFunctionCaseWithPrograms<InstanceContext>(
7289 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
7290 createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
7294 addFunctionCaseWithPrograms<InstanceContext>(
7295 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
7296 createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
7299 return moduleTests.release();
7302 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
7304 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
7305 RGBA defaultColors[4];
7306 getDefaultColors(defaultColors);
7307 map<string, string> fragments;
7308 fragments["pre_main"] =
7309 "%c_f32_5 = OpConstant %f32 5.\n";
7311 // A loop with a single block. The Continue Target is the loop block
7312 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
7313 // -- the "continue construct" forms the entire loop.
7314 fragments["testfun"] =
7315 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7316 "%param1 = OpFunctionParameter %v4f32\n"
7318 "%entry = OpLabel\n"
7319 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7322 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7324 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7325 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
7326 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7327 "%val = OpFAdd %f32 %val1 %delta\n"
7328 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
7329 "%count__ = OpISub %i32 %count %c_i32_1\n"
7330 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7331 "OpLoopMerge %exit %loop None\n"
7332 "OpBranchConditional %again %loop %exit\n"
7335 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7336 "OpReturnValue %result\n"
7340 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
7342 // Body comprised of multiple basic blocks.
7343 const StringTemplate multiBlock(
7344 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7345 "%param1 = OpFunctionParameter %v4f32\n"
7347 "%entry = OpLabel\n"
7348 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7351 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7353 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
7354 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
7355 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
7356 // There are several possibilities for the Continue Target below. Each
7357 // will be specialized into a separate test case.
7358 "OpLoopMerge %exit ${continue_target} None\n"
7362 ";delta_next = (delta > 0) ? -1 : 1;\n"
7363 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
7364 "OpSelectionMerge %gather DontFlatten\n"
7365 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
7368 "OpBranch %gather\n"
7371 "OpBranch %gather\n"
7373 "%gather = OpLabel\n"
7374 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
7375 "%val = OpFAdd %f32 %val1 %delta\n"
7376 "%count__ = OpISub %i32 %count %c_i32_1\n"
7377 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7378 "OpBranchConditional %again %loop %exit\n"
7381 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7382 "OpReturnValue %result\n"
7386 map<string, string> continue_target;
7388 // The Continue Target is the loop block itself.
7389 continue_target["continue_target"] = "%loop";
7390 fragments["testfun"] = multiBlock.specialize(continue_target);
7391 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
7393 // The Continue Target is at the end of the loop.
7394 continue_target["continue_target"] = "%gather";
7395 fragments["testfun"] = multiBlock.specialize(continue_target);
7396 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
7398 // A loop with continue statement.
7399 fragments["testfun"] =
7400 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7401 "%param1 = OpFunctionParameter %v4f32\n"
7403 "%entry = OpLabel\n"
7404 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7407 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
7409 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7410 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
7411 "OpLoopMerge %exit %continue None\n"
7415 ";skip if %count==2\n"
7416 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
7417 "OpSelectionMerge %continue DontFlatten\n"
7418 "OpBranchConditional %eq2 %continue %body\n"
7421 "%fcount = OpConvertSToF %f32 %count\n"
7422 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7423 "OpBranch %continue\n"
7425 "%continue = OpLabel\n"
7426 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
7427 "%count__ = OpISub %i32 %count %c_i32_1\n"
7428 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7429 "OpBranchConditional %again %loop %exit\n"
7432 "%same = OpFSub %f32 %val %c_f32_8\n"
7433 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7434 "OpReturnValue %result\n"
7436 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
7438 // A loop with break.
7439 fragments["testfun"] =
7440 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7441 "%param1 = OpFunctionParameter %v4f32\n"
7443 "%entry = OpLabel\n"
7444 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7445 "%dot = OpDot %f32 %param1 %param1\n"
7446 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7447 "%zero = OpConvertFToU %u32 %div\n"
7448 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7449 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7452 ";adds 4 and 3 to %val0 (exits early)\n"
7454 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7455 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7456 "OpLoopMerge %exit %continue None\n"
7460 ";end loop if %count==%two\n"
7461 "%above2 = OpSGreaterThan %bool %count %two\n"
7462 "OpSelectionMerge %continue DontFlatten\n"
7463 "OpBranchConditional %above2 %body %exit\n"
7466 "%fcount = OpConvertSToF %f32 %count\n"
7467 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7468 "OpBranch %continue\n"
7470 "%continue = OpLabel\n"
7471 "%count__ = OpISub %i32 %count %c_i32_1\n"
7472 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7473 "OpBranchConditional %again %loop %exit\n"
7476 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
7477 "%same = OpFSub %f32 %val_post %c_f32_7\n"
7478 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7479 "OpReturnValue %result\n"
7481 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
7483 // A loop with return.
7484 fragments["testfun"] =
7485 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7486 "%param1 = OpFunctionParameter %v4f32\n"
7488 "%entry = OpLabel\n"
7489 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7490 "%dot = OpDot %f32 %param1 %param1\n"
7491 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7492 "%zero = OpConvertFToU %u32 %div\n"
7493 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7494 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7497 ";returns early without modifying %param1\n"
7499 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7500 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7501 "OpLoopMerge %exit %continue None\n"
7505 ";return if %count==%two\n"
7506 "%above2 = OpSGreaterThan %bool %count %two\n"
7507 "OpSelectionMerge %continue DontFlatten\n"
7508 "OpBranchConditional %above2 %body %early_exit\n"
7510 "%early_exit = OpLabel\n"
7511 "OpReturnValue %param1\n"
7514 "%fcount = OpConvertSToF %f32 %count\n"
7515 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7516 "OpBranch %continue\n"
7518 "%continue = OpLabel\n"
7519 "%count__ = OpISub %i32 %count %c_i32_1\n"
7520 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7521 "OpBranchConditional %again %loop %exit\n"
7524 ";should never get here, so return an incorrect result\n"
7525 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
7526 "OpReturnValue %result\n"
7528 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
7530 // Continue inside a switch block to break to enclosing loop's merge block.
7531 // Matches roughly the following GLSL code:
7532 // for (; keep_going; keep_going = false)
7534 // switch (int(param1.x))
7536 // case 0: continue;
7537 // case 1: continue;
7538 // default: continue;
7540 // dead code: modify return value to invalid result.
7542 fragments["pre_main"] =
7543 "%fp_bool = OpTypePointer Function %bool\n"
7544 "%true = OpConstantTrue %bool\n"
7545 "%false = OpConstantFalse %bool\n";
7547 fragments["testfun"] =
7548 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7549 "%param1 = OpFunctionParameter %v4f32\n"
7551 "%entry = OpLabel\n"
7552 "%keep_going = OpVariable %fp_bool Function\n"
7553 "%val_ptr = OpVariable %fp_f32 Function\n"
7554 "%param1_x = OpCompositeExtract %f32 %param1 0\n"
7555 "OpStore %keep_going %true\n"
7556 "OpBranch %forloop_begin\n"
7558 "%forloop_begin = OpLabel\n"
7559 "OpLoopMerge %forloop_merge %forloop_continue None\n"
7560 "OpBranch %forloop\n"
7562 "%forloop = OpLabel\n"
7563 "%for_condition = OpLoad %bool %keep_going\n"
7564 "OpBranchConditional %for_condition %forloop_body %forloop_merge\n"
7566 "%forloop_body = OpLabel\n"
7567 "OpStore %val_ptr %param1_x\n"
7568 "%param1_x_int = OpConvertFToS %i32 %param1_x\n"
7570 "OpSelectionMerge %switch_merge None\n"
7571 "OpSwitch %param1_x_int %default 0 %case_0 1 %case_1\n"
7572 "%case_0 = OpLabel\n"
7573 "OpBranch %forloop_continue\n"
7574 "%case_1 = OpLabel\n"
7575 "OpBranch %forloop_continue\n"
7576 "%default = OpLabel\n"
7577 "OpBranch %forloop_continue\n"
7578 "%switch_merge = OpLabel\n"
7579 ";should never get here, so change the return value to invalid result\n"
7580 "OpStore %val_ptr %c_f32_1\n"
7581 "OpBranch %forloop_continue\n"
7583 "%forloop_continue = OpLabel\n"
7584 "OpStore %keep_going %false\n"
7585 "OpBranch %forloop_begin\n"
7586 "%forloop_merge = OpLabel\n"
7588 "%val = OpLoad %f32 %val_ptr\n"
7589 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7590 "OpReturnValue %result\n"
7592 createTestsForAllStages("switch_continue", defaultColors, defaultColors, fragments, testGroup.get());
7594 return testGroup.release();
7597 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
7598 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
7600 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
7601 map<string, string> fragments;
7603 // A barrier inside a function body.
7604 fragments["pre_main"] =
7605 "%Workgroup = OpConstant %i32 2\n"
7606 "%WorkgroupAcquireRelease = OpConstant %i32 0x108\n";
7607 fragments["testfun"] =
7608 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7609 "%param1 = OpFunctionParameter %v4f32\n"
7610 "%label_testfun = OpLabel\n"
7611 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7612 "OpReturnValue %param1\n"
7614 addTessCtrlTest(testGroup.get(), "in_function", fragments);
7616 // Common setup code for the following tests.
7617 fragments["pre_main"] =
7618 "%Workgroup = OpConstant %i32 2\n"
7619 "%WorkgroupAcquireRelease = OpConstant %i32 0x108\n"
7620 "%c_f32_5 = OpConstant %f32 5.\n";
7621 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
7622 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7623 "%param1 = OpFunctionParameter %v4f32\n"
7624 "%entry = OpLabel\n"
7625 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7626 "%dot = OpDot %f32 %param1 %param1\n"
7627 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7628 "%zero = OpConvertFToU %u32 %div\n";
7630 // Barriers inside OpSwitch branches.
7631 fragments["testfun"] =
7633 "OpSelectionMerge %switch_exit None\n"
7634 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
7636 "%case1 = OpLabel\n"
7637 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7638 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7639 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7640 "OpBranch %switch_exit\n"
7642 "%switch_default = OpLabel\n"
7643 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7644 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7645 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7646 "OpBranch %switch_exit\n"
7648 "%case0 = OpLabel\n"
7649 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7650 "OpBranch %switch_exit\n"
7652 "%switch_exit = OpLabel\n"
7653 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
7654 "OpReturnValue %ret\n"
7656 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
7658 // Barriers inside if-then-else.
7659 fragments["testfun"] =
7661 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
7662 "OpSelectionMerge %exit DontFlatten\n"
7663 "OpBranchConditional %eq0 %then %else\n"
7666 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7667 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7668 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7672 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7676 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
7677 "OpReturnValue %ret\n"
7679 addTessCtrlTest(testGroup.get(), "in_if", fragments);
7681 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
7682 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
7683 fragments["testfun"] =
7685 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
7686 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
7687 "OpSelectionMerge %exit DontFlatten\n"
7688 "OpBranchConditional %thread0 %then %else\n"
7691 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7695 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
7699 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
7700 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7701 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
7702 "OpReturnValue %ret\n"
7704 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
7706 // A barrier inside a loop.
7707 fragments["pre_main"] =
7708 "%Workgroup = OpConstant %i32 2\n"
7709 "%WorkgroupAcquireRelease = OpConstant %i32 0x108\n"
7710 "%c_f32_10 = OpConstant %f32 10.\n";
7711 fragments["testfun"] =
7712 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7713 "%param1 = OpFunctionParameter %v4f32\n"
7714 "%entry = OpLabel\n"
7715 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7718 ";adds 4, 3, 2, and 1 to %val0\n"
7720 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7721 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7722 "OpControlBarrier %Workgroup %Workgroup %WorkgroupAcquireRelease\n"
7723 "%fcount = OpConvertSToF %f32 %count\n"
7724 "%val = OpFAdd %f32 %val1 %fcount\n"
7725 "%count__ = OpISub %i32 %count %c_i32_1\n"
7726 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7727 "OpLoopMerge %exit %loop None\n"
7728 "OpBranchConditional %again %loop %exit\n"
7731 "%same = OpFSub %f32 %val %c_f32_10\n"
7732 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7733 "OpReturnValue %ret\n"
7735 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
7737 return testGroup.release();
7740 // Test for the OpFRem instruction.
7741 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
7743 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
7744 map<string, string> fragments;
7745 RGBA inputColors[4];
7746 RGBA outputColors[4];
7748 fragments["pre_main"] =
7749 "%c_f32_3 = OpConstant %f32 3.0\n"
7750 "%c_f32_n3 = OpConstant %f32 -3.0\n"
7751 "%c_f32_4 = OpConstant %f32 4.0\n"
7752 "%c_f32_p75 = OpConstant %f32 0.75\n"
7753 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
7754 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
7755 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
7757 // The test does the following.
7758 // vec4 result = (param1 * 8.0) - 4.0;
7759 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
7760 fragments["testfun"] =
7761 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7762 "%param1 = OpFunctionParameter %v4f32\n"
7763 "%label_testfun = OpLabel\n"
7764 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
7765 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
7766 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
7767 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
7768 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
7769 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
7770 "OpReturnValue %xy_0_1\n"
7774 inputColors[0] = RGBA(16, 16, 0, 255);
7775 inputColors[1] = RGBA(232, 232, 0, 255);
7776 inputColors[2] = RGBA(232, 16, 0, 255);
7777 inputColors[3] = RGBA(16, 232, 0, 255);
7779 outputColors[0] = RGBA(64, 64, 0, 255);
7780 outputColors[1] = RGBA(255, 255, 0, 255);
7781 outputColors[2] = RGBA(255, 64, 0, 255);
7782 outputColors[3] = RGBA(64, 255, 0, 255);
7784 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
7785 return testGroup.release();
7788 // Test for the OpSRem instruction.
7789 tcu::TestCaseGroup* createOpSRemGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
7791 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "srem", "OpSRem"));
7792 map<string, string> fragments;
7794 fragments["pre_main"] =
7795 "%c_f32_255 = OpConstant %f32 255.0\n"
7796 "%c_i32_128 = OpConstant %i32 128\n"
7797 "%c_i32_255 = OpConstant %i32 255\n"
7798 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
7799 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
7800 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
7802 // The test does the following.
7803 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
7804 // ivec4 result = ivec4(srem(ints.x, ints.y), srem(ints.y, ints.z), srem(ints.z, ints.x), 255);
7805 // return float(result + 128) / 255.0;
7806 fragments["testfun"] =
7807 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7808 "%param1 = OpFunctionParameter %v4f32\n"
7809 "%label_testfun = OpLabel\n"
7810 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
7811 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
7812 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
7813 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
7814 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
7815 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
7816 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
7817 "%x_out = OpSRem %i32 %x_in %y_in\n"
7818 "%y_out = OpSRem %i32 %y_in %z_in\n"
7819 "%z_out = OpSRem %i32 %z_in %x_in\n"
7820 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
7821 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
7822 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
7823 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
7824 "OpReturnValue %float_out\n"
7827 const struct CaseParams
7830 const char* failMessageTemplate; // customized status message
7831 qpTestResult failResult; // override status on failure
7832 int operands[4][3]; // four (x, y, z) vectors of operands
7833 int results[4][3]; // four (x, y, z) vectors of results
7839 QP_TEST_RESULT_FAIL,
7840 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
7841 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
7845 "Inconsistent results, but within specification: ${reason}",
7846 negFailResult, // negative operands, not required by the spec
7847 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
7848 { { 5, 12, -2 }, { 0, -5, 2 }, { 3, 8, -6 }, { 25, -60, 0 } }, // results
7851 // If either operand is negative the result is undefined. Some implementations may still return correct values.
7853 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
7855 const CaseParams& params = cases[caseNdx];
7856 RGBA inputColors[4];
7857 RGBA outputColors[4];
7859 for (int i = 0; i < 4; ++i)
7861 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
7862 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
7865 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
7868 return testGroup.release();
7871 // Test for the OpSMod instruction.
7872 tcu::TestCaseGroup* createOpSModGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
7874 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "smod", "OpSMod"));
7875 map<string, string> fragments;
7877 fragments["pre_main"] =
7878 "%c_f32_255 = OpConstant %f32 255.0\n"
7879 "%c_i32_128 = OpConstant %i32 128\n"
7880 "%c_i32_255 = OpConstant %i32 255\n"
7881 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
7882 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
7883 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
7885 // The test does the following.
7886 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
7887 // ivec4 result = ivec4(smod(ints.x, ints.y), smod(ints.y, ints.z), smod(ints.z, ints.x), 255);
7888 // return float(result + 128) / 255.0;
7889 fragments["testfun"] =
7890 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7891 "%param1 = OpFunctionParameter %v4f32\n"
7892 "%label_testfun = OpLabel\n"
7893 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
7894 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
7895 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
7896 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
7897 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
7898 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
7899 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
7900 "%x_out = OpSMod %i32 %x_in %y_in\n"
7901 "%y_out = OpSMod %i32 %y_in %z_in\n"
7902 "%z_out = OpSMod %i32 %z_in %x_in\n"
7903 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
7904 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
7905 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
7906 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
7907 "OpReturnValue %float_out\n"
7910 const struct CaseParams
7913 const char* failMessageTemplate; // customized status message
7914 qpTestResult failResult; // override status on failure
7915 int operands[4][3]; // four (x, y, z) vectors of operands
7916 int results[4][3]; // four (x, y, z) vectors of results
7922 QP_TEST_RESULT_FAIL,
7923 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
7924 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
7928 "Inconsistent results, but within specification: ${reason}",
7929 negFailResult, // negative operands, not required by the spec
7930 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
7931 { { 5, -5, 3 }, { 0, 2, -3 }, { 3, -73, 69 }, { -35, 40, 0 } }, // results
7934 // If either operand is negative the result is undefined. Some implementations may still return correct values.
7936 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
7938 const CaseParams& params = cases[caseNdx];
7939 RGBA inputColors[4];
7940 RGBA outputColors[4];
7942 for (int i = 0; i < 4; ++i)
7944 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
7945 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
7948 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
7950 return testGroup.release();
7953 enum ConversionDataType
7955 DATA_TYPE_SIGNED_16,
7956 DATA_TYPE_SIGNED_32,
7957 DATA_TYPE_SIGNED_64,
7958 DATA_TYPE_UNSIGNED_16,
7959 DATA_TYPE_UNSIGNED_32,
7960 DATA_TYPE_UNSIGNED_64,
7963 DATA_TYPE_VEC2_SIGNED_16,
7964 DATA_TYPE_VEC2_SIGNED_32
7967 const string getBitWidthStr (ConversionDataType type)
7971 case DATA_TYPE_SIGNED_16:
7972 case DATA_TYPE_UNSIGNED_16:
7975 case DATA_TYPE_SIGNED_32:
7976 case DATA_TYPE_UNSIGNED_32:
7977 case DATA_TYPE_FLOAT_32:
7978 case DATA_TYPE_VEC2_SIGNED_16:
7981 case DATA_TYPE_SIGNED_64:
7982 case DATA_TYPE_UNSIGNED_64:
7983 case DATA_TYPE_FLOAT_64:
7984 case DATA_TYPE_VEC2_SIGNED_32:
7993 const string getByteWidthStr (ConversionDataType type)
7997 case DATA_TYPE_SIGNED_16:
7998 case DATA_TYPE_UNSIGNED_16:
8001 case DATA_TYPE_SIGNED_32:
8002 case DATA_TYPE_UNSIGNED_32:
8003 case DATA_TYPE_FLOAT_32:
8004 case DATA_TYPE_VEC2_SIGNED_16:
8007 case DATA_TYPE_SIGNED_64:
8008 case DATA_TYPE_UNSIGNED_64:
8009 case DATA_TYPE_FLOAT_64:
8010 case DATA_TYPE_VEC2_SIGNED_32:
8019 bool isSigned (ConversionDataType type)
8023 case DATA_TYPE_SIGNED_16:
8024 case DATA_TYPE_SIGNED_32:
8025 case DATA_TYPE_SIGNED_64:
8026 case DATA_TYPE_FLOAT_32:
8027 case DATA_TYPE_FLOAT_64:
8028 case DATA_TYPE_VEC2_SIGNED_16:
8029 case DATA_TYPE_VEC2_SIGNED_32:
8032 case DATA_TYPE_UNSIGNED_16:
8033 case DATA_TYPE_UNSIGNED_32:
8034 case DATA_TYPE_UNSIGNED_64:
8043 bool isInt (ConversionDataType type)
8047 case DATA_TYPE_SIGNED_16:
8048 case DATA_TYPE_SIGNED_32:
8049 case DATA_TYPE_SIGNED_64:
8050 case DATA_TYPE_UNSIGNED_16:
8051 case DATA_TYPE_UNSIGNED_32:
8052 case DATA_TYPE_UNSIGNED_64:
8055 case DATA_TYPE_FLOAT_32:
8056 case DATA_TYPE_FLOAT_64:
8057 case DATA_TYPE_VEC2_SIGNED_16:
8058 case DATA_TYPE_VEC2_SIGNED_32:
8067 bool isFloat (ConversionDataType type)
8071 case DATA_TYPE_SIGNED_16:
8072 case DATA_TYPE_SIGNED_32:
8073 case DATA_TYPE_SIGNED_64:
8074 case DATA_TYPE_UNSIGNED_16:
8075 case DATA_TYPE_UNSIGNED_32:
8076 case DATA_TYPE_UNSIGNED_64:
8077 case DATA_TYPE_VEC2_SIGNED_16:
8078 case DATA_TYPE_VEC2_SIGNED_32:
8081 case DATA_TYPE_FLOAT_32:
8082 case DATA_TYPE_FLOAT_64:
8091 const string getTypeName (ConversionDataType type)
8093 string prefix = isSigned(type) ? "" : "u";
8095 if (isInt(type)) return prefix + "int" + getBitWidthStr(type);
8096 else if (isFloat(type)) return prefix + "float" + getBitWidthStr(type);
8097 else if (type == DATA_TYPE_VEC2_SIGNED_16) return "i16vec2";
8098 else if (type == DATA_TYPE_VEC2_SIGNED_32) return "i32vec2";
8099 else DE_ASSERT(false);
8104 const string getTestName (ConversionDataType from, ConversionDataType to)
8106 return getTypeName(from) + "_to_" + getTypeName(to);
8109 const string getAsmTypeName (ConversionDataType type)
8113 if (isInt(type)) prefix = isSigned(type) ? "i" : "u";
8114 else if (isFloat(type)) prefix = "f";
8115 else if (type == DATA_TYPE_VEC2_SIGNED_16) return "i16vec2";
8116 else if (type == DATA_TYPE_VEC2_SIGNED_32) return "v2i32";
8117 else DE_ASSERT(false);
8119 return prefix + getBitWidthStr(type);
8122 template<typename T>
8123 BufferSp getSpecializedBuffer (deInt64 number)
8125 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
8128 BufferSp getBuffer (ConversionDataType type, deInt64 number)
8132 case DATA_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
8133 case DATA_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
8134 case DATA_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
8135 case DATA_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
8136 case DATA_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
8137 case DATA_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
8138 case DATA_TYPE_FLOAT_32: return getSpecializedBuffer<deUint32>(number);
8139 case DATA_TYPE_FLOAT_64: return getSpecializedBuffer<deUint64>(number);
8140 case DATA_TYPE_VEC2_SIGNED_16: return getSpecializedBuffer<deUint32>(number);
8141 case DATA_TYPE_VEC2_SIGNED_32: return getSpecializedBuffer<deUint64>(number);
8143 default: DE_ASSERT(false);
8144 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
8148 bool usesInt16 (ConversionDataType from, ConversionDataType to)
8150 return (from == DATA_TYPE_SIGNED_16 || from == DATA_TYPE_UNSIGNED_16
8151 || to == DATA_TYPE_SIGNED_16 || to == DATA_TYPE_UNSIGNED_16
8152 || from == DATA_TYPE_VEC2_SIGNED_16 || to == DATA_TYPE_VEC2_SIGNED_16);
8155 bool usesInt32 (ConversionDataType from, ConversionDataType to)
8157 return (from == DATA_TYPE_SIGNED_32 || from == DATA_TYPE_UNSIGNED_32
8158 || to == DATA_TYPE_SIGNED_32 || to == DATA_TYPE_UNSIGNED_32
8159 || from == DATA_TYPE_VEC2_SIGNED_32 || to == DATA_TYPE_VEC2_SIGNED_32);
8162 bool usesInt64 (ConversionDataType from, ConversionDataType to)
8164 return (from == DATA_TYPE_SIGNED_64 || from == DATA_TYPE_UNSIGNED_64
8165 || to == DATA_TYPE_SIGNED_64 || to == DATA_TYPE_UNSIGNED_64);
8168 bool usesFloat64 (ConversionDataType from, ConversionDataType to)
8170 return (from == DATA_TYPE_FLOAT_64 || to == DATA_TYPE_FLOAT_64);
8174 ComputeTestFeatures getConversionUsedFeatures (ConversionDataType from, ConversionDataType to)
8176 if (usesInt16(from, to) && usesInt64(from, to)) return COMPUTE_TEST_USES_INT16_INT64;
8177 else if (usesInt16(from, to) && usesInt32(from, to)) return COMPUTE_TEST_USES_NONE;
8178 else if (usesInt16(from, to)) return COMPUTE_TEST_USES_INT16; // This is not set for int16<-->int32 only conversions
8179 else if (usesInt64(from, to)) return COMPUTE_TEST_USES_INT64;
8180 else if (usesFloat64(from, to)) return COMPUTE_TEST_USES_FLOAT64;
8181 else return COMPUTE_TEST_USES_NONE;
8184 vector<string> getFeatureStringVector (ComputeTestFeatures computeTestFeatures)
8186 vector<string> features;
8187 if (computeTestFeatures == COMPUTE_TEST_USES_INT16_INT64)
8189 features.push_back("shaderInt16");
8190 features.push_back("shaderInt64");
8192 else if (computeTestFeatures == COMPUTE_TEST_USES_INT16) features.push_back("shaderInt16");
8193 else if (computeTestFeatures == COMPUTE_TEST_USES_INT64) features.push_back("shaderInt64");
8194 else if (computeTestFeatures == COMPUTE_TEST_USES_FLOAT64) features.push_back("shaderFloat64");
8195 else if (computeTestFeatures == COMPUTE_TEST_USES_NONE) {}
8196 else DE_ASSERT(false);
8203 ConvertCase (ConversionDataType from, ConversionDataType to, deInt64 number, bool separateOutput = false, deInt64 outputNumber = 0)
8206 , m_features (getConversionUsedFeatures(from, to))
8207 , m_name (getTestName(from, to))
8208 , m_inputBuffer (getBuffer(from, number))
8210 m_asmTypes["inputType"] = getAsmTypeName(from);
8211 m_asmTypes["outputType"] = getAsmTypeName(to);
8214 m_outputBuffer = getBuffer(to, outputNumber);
8216 m_outputBuffer = getBuffer(to, number);
8218 if (m_features == COMPUTE_TEST_USES_INT16)
8220 m_asmTypes["datatype_capabilities"] = "OpCapability Int16\n"
8221 "OpCapability StorageUniformBufferBlock16\n"
8222 "OpCapability StorageUniform16\n";
8223 m_asmTypes["datatype_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
8224 "%u16 = OpTypeInt 16 0\n"
8225 "%i16vec2 = OpTypeVector %i16 2\n";
8226 m_asmTypes["datatype_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
8228 else if (m_features == COMPUTE_TEST_USES_INT64)
8230 m_asmTypes["datatype_capabilities"] = "OpCapability Int64\n";
8231 m_asmTypes["datatype_additional_decl"] = "%i64 = OpTypeInt 64 1\n"
8232 "%u64 = OpTypeInt 64 0\n";
8233 m_asmTypes["datatype_extensions"] = "";
8235 else if (m_features == COMPUTE_TEST_USES_INT16_INT64)
8237 m_asmTypes["datatype_capabilities"] = "OpCapability Int16\n"
8238 "OpCapability StorageUniformBufferBlock16\n"
8239 "OpCapability StorageUniform16\n"
8240 "OpCapability Int64\n";
8241 m_asmTypes["datatype_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
8242 "%u16 = OpTypeInt 16 0\n"
8243 "%i64 = OpTypeInt 64 1\n"
8244 "%u64 = OpTypeInt 64 0\n";
8245 m_asmTypes["datatype_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
8247 else if (m_features == COMPUTE_TEST_USES_FLOAT64)
8249 m_asmTypes["datatype_capabilities"] = "OpCapability Float64\n";
8250 m_asmTypes["datatype_additional_decl"] = "%f64 = OpTypeFloat 64\n";
8252 else if (usesInt16(from, to) && usesInt32(from, to))
8254 m_asmTypes["datatype_capabilities"] = "OpCapability StorageUniformBufferBlock16\n"
8255 "OpCapability StorageUniform16\n";
8256 m_asmTypes["datatype_additional_decl"] = "%i16 = OpTypeInt 16 1\n"
8257 "%u16 = OpTypeInt 16 0\n"
8258 "%i16vec2 = OpTypeVector %i16 2\n";
8259 m_asmTypes["datatype_extensions"] = "OpExtension \"SPV_KHR_16bit_storage\"\n";
8267 ConversionDataType m_fromType;
8268 ConversionDataType m_toType;
8269 ComputeTestFeatures m_features;
8271 map<string, string> m_asmTypes;
8272 BufferSp m_inputBuffer;
8273 BufferSp m_outputBuffer;
8276 const string getConvertCaseShaderStr (const string& instruction, const ConvertCase& convertCase)
8278 map<string, string> params = convertCase.m_asmTypes;
8280 params["instruction"] = instruction;
8281 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
8282 params["outDecorator"] = getByteWidthStr(convertCase.m_toType);
8284 const StringTemplate shader (
8285 "OpCapability Shader\n"
8286 "${datatype_capabilities}"
8287 "${datatype_extensions:opt}"
8288 "OpMemoryModel Logical GLSL450\n"
8289 "OpEntryPoint GLCompute %main \"main\"\n"
8290 "OpExecutionMode %main LocalSize 1 1 1\n"
8291 "OpSource GLSL 430\n"
8292 "OpName %main \"main\"\n"
8294 "OpDecorate %indata DescriptorSet 0\n"
8295 "OpDecorate %indata Binding 0\n"
8296 "OpDecorate %outdata DescriptorSet 0\n"
8297 "OpDecorate %outdata Binding 1\n"
8298 "OpDecorate %in_buf BufferBlock\n"
8299 "OpDecorate %out_buf BufferBlock\n"
8300 "OpMemberDecorate %in_buf 0 Offset 0\n"
8301 "OpMemberDecorate %out_buf 0 Offset 0\n"
8303 "%void = OpTypeVoid\n"
8304 "%voidf = OpTypeFunction %void\n"
8305 "%u32 = OpTypeInt 32 0\n"
8306 "%i32 = OpTypeInt 32 1\n"
8307 "%f32 = OpTypeFloat 32\n"
8308 "%v2i32 = OpTypeVector %i32 2\n"
8309 "${datatype_additional_decl}"
8310 "%uvec3 = OpTypeVector %u32 3\n"
8312 "%in_ptr = OpTypePointer Uniform %${inputType}\n"
8313 "%out_ptr = OpTypePointer Uniform %${outputType}\n"
8314 "%in_buf = OpTypeStruct %${inputType}\n"
8315 "%out_buf = OpTypeStruct %${outputType}\n"
8316 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
8317 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
8318 "%indata = OpVariable %in_bufptr Uniform\n"
8319 "%outdata = OpVariable %out_bufptr Uniform\n"
8321 "%zero = OpConstant %i32 0\n"
8323 "%main = OpFunction %void None %voidf\n"
8324 "%label = OpLabel\n"
8325 "%inloc = OpAccessChain %in_ptr %indata %zero\n"
8326 "%outloc = OpAccessChain %out_ptr %outdata %zero\n"
8327 "%inval = OpLoad %${inputType} %inloc\n"
8328 "%conv = ${instruction} %${outputType} %inval\n"
8329 " OpStore %outloc %conv\n"
8334 return shader.specialize(params);
8337 void createConvertCases (vector<ConvertCase>& testCases, const string& instruction)
8339 if (instruction == "OpUConvert")
8341 // Convert unsigned int to unsigned int
8342 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_16, DATA_TYPE_UNSIGNED_32, 60653));
8343 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_16, DATA_TYPE_UNSIGNED_64, 17991));
8344 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_32, DATA_TYPE_UNSIGNED_64, 904256275));
8345 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_32, DATA_TYPE_UNSIGNED_16, 6275));
8346 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_64, DATA_TYPE_UNSIGNED_32, 701256243));
8347 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_64, DATA_TYPE_UNSIGNED_16, 4741));
8349 else if (instruction == "OpSConvert")
8351 // Sign extension int->int
8352 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_16, DATA_TYPE_SIGNED_32, 14669));
8353 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_16, DATA_TYPE_SIGNED_64, -3341));
8354 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_32, DATA_TYPE_SIGNED_64, 973610259));
8356 // Truncate for int->int
8357 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_32, DATA_TYPE_SIGNED_16, 12382));
8358 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_64, DATA_TYPE_SIGNED_32, -972812359));
8359 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_64, DATA_TYPE_SIGNED_16, -1067742499291926803ll, true, -4371));
8361 // Sign extension for int->uint
8362 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_16, DATA_TYPE_UNSIGNED_32, 14669));
8363 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_16, DATA_TYPE_UNSIGNED_64, -3341, true, 18446744073709548275ull));
8364 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_32, DATA_TYPE_UNSIGNED_64, 973610259));
8366 // Truncate for int->uint
8367 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_32, DATA_TYPE_UNSIGNED_16, 12382));
8368 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_64, DATA_TYPE_UNSIGNED_32, -972812359, true, 3322154937u));
8369 testCases.push_back(ConvertCase(DATA_TYPE_SIGNED_64, DATA_TYPE_UNSIGNED_16, -1067742499291926803ll, true, 61165));
8371 // Sign extension for uint->int
8372 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_16, DATA_TYPE_SIGNED_32, 14669));
8373 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_16, DATA_TYPE_SIGNED_64, 62195, true, -3341));
8374 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_32, DATA_TYPE_SIGNED_64, 973610259));
8376 // Truncate for uint->int
8377 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_32, DATA_TYPE_SIGNED_16, 12382));
8378 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_64, DATA_TYPE_SIGNED_32, 18446744072736739257ull, true, -972812359));
8379 testCases.push_back(ConvertCase(DATA_TYPE_UNSIGNED_64, DATA_TYPE_SIGNED_16, 17379001574417624813ull, true, -4371));
8381 // Convert i16vec2 to i32vec2 and vice versa
8382 // Unsigned values are used here to represent negative signed values and to allow defined shifting behaviour.
8383 // The actual signed value -32123 is used here as uint16 value 33413 and uint32 value 4294935173
8384 testCases.push_back(ConvertCase(DATA_TYPE_VEC2_SIGNED_16, DATA_TYPE_VEC2_SIGNED_32, (33413u << 16) | 27593, true, (4294935173ull << 32) | 27593));
8385 testCases.push_back(ConvertCase(DATA_TYPE_VEC2_SIGNED_32, DATA_TYPE_VEC2_SIGNED_16, (4294935173ull << 32) | 27593, true, (33413u << 16) | 27593));
8387 else if (instruction == "OpFConvert")
8389 // All hexadecimal values below represent 1024.0 as 32/64-bit IEEE 754 float
8390 testCases.push_back(ConvertCase(DATA_TYPE_FLOAT_32, DATA_TYPE_FLOAT_64, 0x449a4000, true, 0x4093480000000000));
8391 testCases.push_back(ConvertCase(DATA_TYPE_FLOAT_64, DATA_TYPE_FLOAT_32, 0x4093480000000000, true, 0x449a4000));
8394 DE_FATAL("Unknown instruction");
8397 const map<string, string> getConvertCaseFragments (string instruction, const ConvertCase& convertCase)
8399 map<string, string> params = convertCase.m_asmTypes;
8400 map<string, string> fragments;
8402 params["instruction"] = instruction;
8403 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
8405 const StringTemplate decoration (
8406 " OpDecorate %SSBOi DescriptorSet 0\n"
8407 " OpDecorate %SSBOo DescriptorSet 0\n"
8408 " OpDecorate %SSBOi Binding 0\n"
8409 " OpDecorate %SSBOo Binding 1\n"
8410 " OpDecorate %s_SSBOi Block\n"
8411 " OpDecorate %s_SSBOo Block\n"
8412 "OpMemberDecorate %s_SSBOi 0 Offset 0\n"
8413 "OpMemberDecorate %s_SSBOo 0 Offset 0\n");
8415 const StringTemplate pre_main (
8416 "${datatype_additional_decl:opt}"
8417 " %ptr_in = OpTypePointer StorageBuffer %${inputType}\n"
8418 " %ptr_out = OpTypePointer StorageBuffer %${outputType}\n"
8419 " %s_SSBOi = OpTypeStruct %${inputType}\n"
8420 " %s_SSBOo = OpTypeStruct %${outputType}\n"
8421 " %ptr_SSBOi = OpTypePointer StorageBuffer %s_SSBOi\n"
8422 " %ptr_SSBOo = OpTypePointer StorageBuffer %s_SSBOo\n"
8423 " %SSBOi = OpVariable %ptr_SSBOi StorageBuffer\n"
8424 " %SSBOo = OpVariable %ptr_SSBOo StorageBuffer\n");
8426 const StringTemplate testfun (
8427 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
8428 "%param = OpFunctionParameter %v4f32\n"
8429 "%label = OpLabel\n"
8430 "%iLoc = OpAccessChain %ptr_in %SSBOi %c_u32_0\n"
8431 "%oLoc = OpAccessChain %ptr_out %SSBOo %c_u32_0\n"
8432 "%valIn = OpLoad %${inputType} %iLoc\n"
8433 "%valOut = ${instruction} %${outputType} %valIn\n"
8434 " OpStore %oLoc %valOut\n"
8435 " OpReturnValue %param\n"
8436 " OpFunctionEnd\n");
8438 params["datatype_extensions"] =
8439 params["datatype_extensions"] +
8440 "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n";
8442 fragments["capability"] = params["datatype_capabilities"];
8443 fragments["extension"] = params["datatype_extensions"];
8444 fragments["decoration"] = decoration.specialize(params);
8445 fragments["pre_main"] = pre_main.specialize(params);
8446 fragments["testfun"] = testfun.specialize(params);
8451 // Test for OpSConvert, OpUConvert and OpFConvert in compute shaders
8452 tcu::TestCaseGroup* createConvertComputeTests (tcu::TestContext& testCtx, const string& instruction, const string& name)
8454 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, name.c_str(), instruction.c_str()));
8455 vector<ConvertCase> testCases;
8456 createConvertCases(testCases, instruction);
8458 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8460 ComputeShaderSpec spec;
8461 spec.assembly = getConvertCaseShaderStr(instruction, *test);
8462 spec.numWorkGroups = IVec3(1, 1, 1);
8463 spec.inputs.push_back (test->m_inputBuffer);
8464 spec.outputs.push_back (test->m_outputBuffer);
8466 if (test->m_features == COMPUTE_TEST_USES_INT16 || test->m_features == COMPUTE_TEST_USES_INT16_INT64 || usesInt16(test->m_fromType, test->m_toType))
8467 spec.extensions.push_back("VK_KHR_16bit_storage");
8469 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "", spec, test->m_features));
8471 return group.release();
8474 // Test for OpSConvert, OpUConvert and OpFConvert in graphics shaders
8475 tcu::TestCaseGroup* createConvertGraphicsTests (tcu::TestContext& testCtx, const string& instruction, const string& name)
8477 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, name.c_str(), instruction.c_str()));
8478 vector<ConvertCase> testCases;
8479 createConvertCases(testCases, instruction);
8481 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8483 map<string, string> fragments = getConvertCaseFragments(instruction, *test);
8484 vector<string> features = getFeatureStringVector(test->m_features);
8485 GraphicsResources resources;
8486 vector<string> extensions;
8487 vector<deInt32> noSpecConstants;
8488 PushConstants noPushConstants;
8489 GraphicsInterfaces noInterfaces;
8490 tcu::RGBA defaultColors[4];
8492 getDefaultColors (defaultColors);
8493 resources.inputs.push_back (std::make_pair(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, test->m_inputBuffer));
8494 resources.outputs.push_back (std::make_pair(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, test->m_outputBuffer));
8495 extensions.push_back ("VK_KHR_storage_buffer_storage_class");
8497 if (test->m_features == COMPUTE_TEST_USES_INT16 || test->m_features == COMPUTE_TEST_USES_INT16_INT64 || usesInt16(test->m_fromType, test->m_toType))
8498 extensions.push_back("VK_KHR_16bit_storage");
8500 createTestsForAllStages(
8501 test->m_name, defaultColors, defaultColors, fragments, noSpecConstants,
8502 noPushConstants, resources, noInterfaces, extensions, features, VulkanFeatures(), group.get());
8504 return group.release();
8507 const string getNumberTypeName (const NumberType type)
8509 if (type == NUMBERTYPE_INT32)
8513 else if (type == NUMBERTYPE_UINT32)
8517 else if (type == NUMBERTYPE_FLOAT32)
8528 deInt32 getInt(de::Random& rnd)
8530 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
8533 const string repeatString (const string& str, int times)
8536 for (int i = 0; i < times; ++i)
8543 const string getRandomConstantString (const NumberType type, de::Random& rnd)
8545 if (type == NUMBERTYPE_INT32)
8547 return numberToString<deInt32>(getInt(rnd));
8549 else if (type == NUMBERTYPE_UINT32)
8551 return numberToString<deUint32>(rnd.getUint32());
8553 else if (type == NUMBERTYPE_FLOAT32)
8555 return numberToString<float>(rnd.getFloat());
8564 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8566 map<string, string> params;
8569 for (int width = 2; width <= 4; ++width)
8571 const string randomConst = numberToString(getInt(rnd));
8572 const string widthStr = numberToString(width);
8573 const string composite_type = "${customType}vec" + widthStr;
8574 const int index = rnd.getInt(0, width-1);
8576 params["type"] = "vec";
8577 params["name"] = params["type"] + "_" + widthStr;
8578 params["compositeDecl"] = composite_type + " = OpTypeVector ${customType} " + widthStr +"\n";
8579 params["compositeType"] = composite_type;
8580 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8581 params["compositeConstruct"] = "%instance = OpCompositeConstruct " + composite_type + repeatString(" %filler", width) + "\n";
8582 params["indexes"] = numberToString(index);
8583 testCases.push_back(params);
8587 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8589 const int limit = 10;
8590 map<string, string> params;
8592 for (int width = 2; width <= limit; ++width)
8594 string randomConst = numberToString(getInt(rnd));
8595 string widthStr = numberToString(width);
8596 int index = rnd.getInt(0, width-1);
8598 params["type"] = "array";
8599 params["name"] = params["type"] + "_" + widthStr;
8600 params["compositeDecl"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
8601 + "%composite = OpTypeArray ${customType} %arraywidth\n";
8602 params["compositeType"] = "%composite";
8603 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8604 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8605 params["indexes"] = numberToString(index);
8606 testCases.push_back(params);
8610 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8612 const int limit = 10;
8613 map<string, string> params;
8615 for (int width = 2; width <= limit; ++width)
8617 string randomConst = numberToString(getInt(rnd));
8618 int index = rnd.getInt(0, width-1);
8620 params["type"] = "struct";
8621 params["name"] = params["type"] + "_" + numberToString(width);
8622 params["compositeDecl"] = "%composite = OpTypeStruct" + repeatString(" ${customType}", width) + "\n";
8623 params["compositeType"] = "%composite";
8624 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n";
8625 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
8626 params["indexes"] = numberToString(index);
8627 testCases.push_back(params);
8631 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8633 map<string, string> params;
8636 for (int width = 2; width <= 4; ++width)
8638 string widthStr = numberToString(width);
8640 for (int column = 2 ; column <= 4; ++column)
8642 int index_0 = rnd.getInt(0, column-1);
8643 int index_1 = rnd.getInt(0, width-1);
8644 string columnStr = numberToString(column);
8646 params["type"] = "matrix";
8647 params["name"] = params["type"] + "_" + widthStr + "x" + columnStr;
8648 params["compositeDecl"] = string("%vectype = OpTypeVector ${customType} " + widthStr + "\n")
8649 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
8650 params["compositeType"] = "%composite";
8652 params["filler"] = string("%filler = OpConstant ${customType} ") + getRandomConstantString(type, rnd) + "\n"
8653 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
8655 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
8656 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
8657 testCases.push_back(params);
8662 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
8664 createVectorCompositeCases(testCases, rnd, type);
8665 createArrayCompositeCases(testCases, rnd, type);
8666 createStructCompositeCases(testCases, rnd, type);
8667 // Matrix only supports float types
8668 if (type == NUMBERTYPE_FLOAT32)
8670 createMatrixCompositeCases(testCases, rnd, type);
8674 const string getAssemblyTypeDeclaration (const NumberType type)
8678 case NUMBERTYPE_INT32: return "OpTypeInt 32 1";
8679 case NUMBERTYPE_UINT32: return "OpTypeInt 32 0";
8680 case NUMBERTYPE_FLOAT32: return "OpTypeFloat 32";
8681 default: DE_ASSERT(false); return "";
8685 const string getAssemblyTypeName (const NumberType type)
8689 case NUMBERTYPE_INT32: return "%i32";
8690 case NUMBERTYPE_UINT32: return "%u32";
8691 case NUMBERTYPE_FLOAT32: return "%f32";
8692 default: DE_ASSERT(false); return "";
8696 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
8698 map<string, string> parameters(params);
8700 const string customType = getAssemblyTypeName(type);
8701 map<string, string> substCustomType;
8702 substCustomType["customType"] = customType;
8703 parameters["compositeDecl"] = StringTemplate(parameters.at("compositeDecl")).specialize(substCustomType);
8704 parameters["compositeType"] = StringTemplate(parameters.at("compositeType")).specialize(substCustomType);
8705 parameters["compositeConstruct"] = StringTemplate(parameters.at("compositeConstruct")).specialize(substCustomType);
8706 parameters["filler"] = StringTemplate(parameters.at("filler")).specialize(substCustomType);
8707 parameters["customType"] = customType;
8708 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
8710 if (parameters.at("compositeType") != "%u32vec3")
8712 parameters["u32vec3Decl"] = "%u32vec3 = OpTypeVector %u32 3\n";
8715 return StringTemplate(
8716 "OpCapability Shader\n"
8717 "OpCapability Matrix\n"
8718 "OpMemoryModel Logical GLSL450\n"
8719 "OpEntryPoint GLCompute %main \"main\" %id\n"
8720 "OpExecutionMode %main LocalSize 1 1 1\n"
8722 "OpSource GLSL 430\n"
8723 "OpName %main \"main\"\n"
8724 "OpName %id \"gl_GlobalInvocationID\"\n"
8727 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8728 "OpDecorate %buf BufferBlock\n"
8729 "OpDecorate %indata DescriptorSet 0\n"
8730 "OpDecorate %indata Binding 0\n"
8731 "OpDecorate %outdata DescriptorSet 0\n"
8732 "OpDecorate %outdata Binding 1\n"
8733 "OpDecorate %customarr ArrayStride 4\n"
8734 "${compositeDecorator}"
8735 "OpMemberDecorate %buf 0 Offset 0\n"
8738 "%void = OpTypeVoid\n"
8739 "%voidf = OpTypeFunction %void\n"
8740 "%u32 = OpTypeInt 32 0\n"
8741 "%i32 = OpTypeInt 32 1\n"
8742 "%f32 = OpTypeFloat 32\n"
8744 // Composite declaration
8750 "${u32vec3Decl:opt}"
8751 "%uvec3ptr = OpTypePointer Input %u32vec3\n"
8753 // Inherited from custom
8754 "%customptr = OpTypePointer Uniform ${customType}\n"
8755 "%customarr = OpTypeRuntimeArray ${customType}\n"
8756 "%buf = OpTypeStruct %customarr\n"
8757 "%bufptr = OpTypePointer Uniform %buf\n"
8759 "%indata = OpVariable %bufptr Uniform\n"
8760 "%outdata = OpVariable %bufptr Uniform\n"
8762 "%id = OpVariable %uvec3ptr Input\n"
8763 "%zero = OpConstant %i32 0\n"
8765 "%main = OpFunction %void None %voidf\n"
8766 "%label = OpLabel\n"
8767 "%idval = OpLoad %u32vec3 %id\n"
8768 "%x = OpCompositeExtract %u32 %idval 0\n"
8770 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
8771 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
8772 // Read the input value
8773 "%inval = OpLoad ${customType} %inloc\n"
8774 // Create the composite and fill it
8775 "${compositeConstruct}"
8776 // Insert the input value to a place
8777 "%instance2 = OpCompositeInsert ${compositeType} %inval %instance ${indexes}\n"
8778 // Read back the value from the position
8779 "%out_val = OpCompositeExtract ${customType} %instance2 ${indexes}\n"
8780 // Store it in the output position
8781 " OpStore %outloc %out_val\n"
8784 ).specialize(parameters);
8787 template<typename T>
8788 BufferSp createCompositeBuffer(T number)
8790 return BufferSp(new Buffer<T>(vector<T>(1, number)));
8793 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
8795 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
8796 de::Random rnd (deStringHash(group->getName()));
8798 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8800 NumberType numberType = NumberType(type);
8801 const string typeName = getNumberTypeName(numberType);
8802 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
8803 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
8804 vector<map<string, string> > testCases;
8806 createCompositeCases(testCases, rnd, numberType);
8808 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
8810 ComputeShaderSpec spec;
8812 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
8816 case NUMBERTYPE_INT32:
8818 deInt32 number = getInt(rnd);
8819 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
8820 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
8823 case NUMBERTYPE_UINT32:
8825 deUint32 number = rnd.getUint32();
8826 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
8827 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
8830 case NUMBERTYPE_FLOAT32:
8832 float number = rnd.getFloat();
8833 spec.inputs.push_back(createCompositeBuffer<float>(number));
8834 spec.outputs.push_back(createCompositeBuffer<float>(number));
8841 spec.numWorkGroups = IVec3(1, 1, 1);
8842 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
8844 group->addChild(subGroup.release());
8846 return group.release();
8849 struct AssemblyStructInfo
8851 AssemblyStructInfo (const deUint32 comp, const deUint32 idx)
8856 deUint32 components;
8860 const string specializeInBoundsShaderTemplate (const NumberType type, const AssemblyStructInfo& structInfo, const map<string, string>& params)
8862 // Create the full index string
8863 string fullIndex = numberToString(structInfo.index) + " " + params.at("indexes");
8864 // Convert it to list of indexes
8865 vector<string> indexes = de::splitString(fullIndex, ' ');
8867 map<string, string> parameters (params);
8868 parameters["structType"] = repeatString(" ${compositeType}", structInfo.components);
8869 parameters["structConstruct"] = repeatString(" %instance", structInfo.components);
8870 parameters["insertIndexes"] = fullIndex;
8872 // In matrix cases the last two index is the CompositeExtract indexes
8873 const deUint32 extractIndexes = (parameters["type"] == "matrix") ? 2 : 1;
8875 // Construct the extractIndex
8876 for (vector<string>::const_iterator index = indexes.end() - extractIndexes; index != indexes.end(); ++index)
8878 parameters["extractIndexes"] += " " + *index;
8881 // Remove the last 1 or 2 element depends on matrix case or not
8882 indexes.erase(indexes.end() - extractIndexes, indexes.end());
8885 // Generate AccessChain index expressions (except for the last one, because we use ptr to the composite)
8886 for (vector<string>::const_iterator index = indexes.begin(); index != indexes.end(); ++index)
8888 string indexId = "%index_" + numberToString(id++);
8889 parameters["accessChainConstDeclaration"] += indexId + " = OpConstant %u32 " + *index + "\n";
8890 parameters["accessChainIndexes"] += " " + indexId;
8893 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
8895 const string customType = getAssemblyTypeName(type);
8896 map<string, string> substCustomType;
8897 substCustomType["customType"] = customType;
8898 parameters["compositeDecl"] = StringTemplate(parameters.at("compositeDecl")).specialize(substCustomType);
8899 parameters["compositeType"] = StringTemplate(parameters.at("compositeType")).specialize(substCustomType);
8900 parameters["compositeConstruct"] = StringTemplate(parameters.at("compositeConstruct")).specialize(substCustomType);
8901 parameters["filler"] = StringTemplate(parameters.at("filler")).specialize(substCustomType);
8902 parameters["customType"] = customType;
8904 const string compositeType = parameters.at("compositeType");
8905 map<string, string> substCompositeType;
8906 substCompositeType["compositeType"] = compositeType;
8907 parameters["structType"] = StringTemplate(parameters.at("structType")).specialize(substCompositeType);
8908 if (compositeType != "%u32vec3")
8910 parameters["u32vec3Decl"] = "%u32vec3 = OpTypeVector %u32 3\n";
8913 return StringTemplate(
8914 "OpCapability Shader\n"
8915 "OpCapability Matrix\n"
8916 "OpMemoryModel Logical GLSL450\n"
8917 "OpEntryPoint GLCompute %main \"main\" %id\n"
8918 "OpExecutionMode %main LocalSize 1 1 1\n"
8920 "OpSource GLSL 430\n"
8921 "OpName %main \"main\"\n"
8922 "OpName %id \"gl_GlobalInvocationID\"\n"
8924 "OpDecorate %id BuiltIn GlobalInvocationId\n"
8925 "OpDecorate %buf BufferBlock\n"
8926 "OpDecorate %indata DescriptorSet 0\n"
8927 "OpDecorate %indata Binding 0\n"
8928 "OpDecorate %outdata DescriptorSet 0\n"
8929 "OpDecorate %outdata Binding 1\n"
8930 "OpDecorate %customarr ArrayStride 4\n"
8931 "${compositeDecorator}"
8932 "OpMemberDecorate %buf 0 Offset 0\n"
8934 "%void = OpTypeVoid\n"
8935 "%voidf = OpTypeFunction %void\n"
8936 "%i32 = OpTypeInt 32 1\n"
8937 "%u32 = OpTypeInt 32 0\n"
8938 "%f32 = OpTypeFloat 32\n"
8941 // %u32vec3 if not already declared in ${compositeDecl}
8942 "${u32vec3Decl:opt}"
8943 "%uvec3ptr = OpTypePointer Input %u32vec3\n"
8944 // Inherited from composite
8945 "%composite_p = OpTypePointer Function ${compositeType}\n"
8946 "%struct_t = OpTypeStruct${structType}\n"
8947 "%struct_p = OpTypePointer Function %struct_t\n"
8950 "${accessChainConstDeclaration}"
8951 // Inherited from custom
8952 "%customptr = OpTypePointer Uniform ${customType}\n"
8953 "%customarr = OpTypeRuntimeArray ${customType}\n"
8954 "%buf = OpTypeStruct %customarr\n"
8955 "%bufptr = OpTypePointer Uniform %buf\n"
8956 "%indata = OpVariable %bufptr Uniform\n"
8957 "%outdata = OpVariable %bufptr Uniform\n"
8959 "%id = OpVariable %uvec3ptr Input\n"
8960 "%zero = OpConstant %u32 0\n"
8961 "%main = OpFunction %void None %voidf\n"
8962 "%label = OpLabel\n"
8963 "%struct_v = OpVariable %struct_p Function\n"
8964 "%idval = OpLoad %u32vec3 %id\n"
8965 "%x = OpCompositeExtract %u32 %idval 0\n"
8966 // Create the input/output type
8967 "%inloc = OpInBoundsAccessChain %customptr %indata %zero %x\n"
8968 "%outloc = OpInBoundsAccessChain %customptr %outdata %zero %x\n"
8969 // Read the input value
8970 "%inval = OpLoad ${customType} %inloc\n"
8971 // Create the composite and fill it
8972 "${compositeConstruct}"
8973 // Create the struct and fill it with the composite
8974 "%struct = OpCompositeConstruct %struct_t${structConstruct}\n"
8976 "%comp_obj = OpCompositeInsert %struct_t %inval %struct ${insertIndexes}\n"
8978 " OpStore %struct_v %comp_obj\n"
8979 // Get deepest possible composite pointer
8980 "%inner_ptr = OpInBoundsAccessChain %composite_p %struct_v${accessChainIndexes}\n"
8981 "%read_obj = OpLoad ${compositeType} %inner_ptr\n"
8982 // Read back the stored value
8983 "%read_val = OpCompositeExtract ${customType} %read_obj${extractIndexes}\n"
8984 " OpStore %outloc %read_val\n"
8987 ).specialize(parameters);
8990 tcu::TestCaseGroup* createOpInBoundsAccessChainGroup (tcu::TestContext& testCtx)
8992 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opinboundsaccesschain", "Test the OpInBoundsAccessChain instruction"));
8993 de::Random rnd (deStringHash(group->getName()));
8995 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
8997 NumberType numberType = NumberType(type);
8998 const string typeName = getNumberTypeName(numberType);
8999 const string description = "Test the OpInBoundsAccessChain instruction with " + typeName + "s";
9000 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
9002 vector<map<string, string> > testCases;
9003 createCompositeCases(testCases, rnd, numberType);
9005 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
9007 ComputeShaderSpec spec;
9009 // Number of components inside of a struct
9010 deUint32 structComponents = rnd.getInt(2, 8);
9011 // Component index value
9012 deUint32 structIndex = rnd.getInt(0, structComponents - 1);
9013 AssemblyStructInfo structInfo(structComponents, structIndex);
9015 spec.assembly = specializeInBoundsShaderTemplate(numberType, structInfo, *test);
9019 case NUMBERTYPE_INT32:
9021 deInt32 number = getInt(rnd);
9022 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
9023 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
9026 case NUMBERTYPE_UINT32:
9028 deUint32 number = rnd.getUint32();
9029 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
9030 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
9033 case NUMBERTYPE_FLOAT32:
9035 float number = rnd.getFloat();
9036 spec.inputs.push_back(createCompositeBuffer<float>(number));
9037 spec.outputs.push_back(createCompositeBuffer<float>(number));
9043 spec.numWorkGroups = IVec3(1, 1, 1);
9044 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpInBoundsAccessChain test", spec));
9046 group->addChild(subGroup.release());
9048 return group.release();
9051 // If the params missing, uninitialized case
9052 const string specializeDefaultOutputShaderTemplate (const NumberType type, const map<string, string>& params = map<string, string>())
9054 map<string, string> parameters(params);
9056 parameters["customType"] = getAssemblyTypeName(type);
9058 // Declare the const value, and use it in the initializer
9059 if (params.find("constValue") != params.end())
9061 parameters["variableInitializer"] = " %const";
9063 // Uninitialized case
9066 parameters["commentDecl"] = ";";
9069 return StringTemplate(
9070 "OpCapability Shader\n"
9071 "OpMemoryModel Logical GLSL450\n"
9072 "OpEntryPoint GLCompute %main \"main\" %id\n"
9073 "OpExecutionMode %main LocalSize 1 1 1\n"
9074 "OpSource GLSL 430\n"
9075 "OpName %main \"main\"\n"
9076 "OpName %id \"gl_GlobalInvocationID\"\n"
9078 "OpDecorate %id BuiltIn GlobalInvocationId\n"
9079 "OpDecorate %indata DescriptorSet 0\n"
9080 "OpDecorate %indata Binding 0\n"
9081 "OpDecorate %outdata DescriptorSet 0\n"
9082 "OpDecorate %outdata Binding 1\n"
9083 "OpDecorate %in_arr ArrayStride 4\n"
9084 "OpDecorate %in_buf BufferBlock\n"
9085 "OpMemberDecorate %in_buf 0 Offset 0\n"
9087 "%void = OpTypeVoid\n"
9088 "%voidf = OpTypeFunction %void\n"
9089 "%u32 = OpTypeInt 32 0\n"
9090 "%i32 = OpTypeInt 32 1\n"
9091 "%f32 = OpTypeFloat 32\n"
9092 "%uvec3 = OpTypeVector %u32 3\n"
9093 "%uvec3ptr = OpTypePointer Input %uvec3\n"
9094 "${commentDecl:opt}%const = OpConstant ${customType} ${constValue:opt}\n"
9096 "%in_ptr = OpTypePointer Uniform ${customType}\n"
9097 "%in_arr = OpTypeRuntimeArray ${customType}\n"
9098 "%in_buf = OpTypeStruct %in_arr\n"
9099 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
9100 "%indata = OpVariable %in_bufptr Uniform\n"
9101 "%outdata = OpVariable %in_bufptr Uniform\n"
9102 "%id = OpVariable %uvec3ptr Input\n"
9103 "%var_ptr = OpTypePointer Function ${customType}\n"
9105 "%zero = OpConstant %i32 0\n"
9107 "%main = OpFunction %void None %voidf\n"
9108 "%label = OpLabel\n"
9109 "%out_var = OpVariable %var_ptr Function${variableInitializer:opt}\n"
9110 "%idval = OpLoad %uvec3 %id\n"
9111 "%x = OpCompositeExtract %u32 %idval 0\n"
9112 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
9113 "%outloc = OpAccessChain %in_ptr %outdata %zero %x\n"
9115 "%outval = OpLoad ${customType} %out_var\n"
9116 " OpStore %outloc %outval\n"
9119 ).specialize(parameters);
9122 bool compareFloats (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
9124 DE_ASSERT(outputAllocs.size() != 0);
9125 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
9127 // Use custom epsilon because of the float->string conversion
9128 const float epsilon = 0.00001f;
9130 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
9132 vector<deUint8> expectedBytes;
9136 expectedOutputs[outputNdx]->getBytes(expectedBytes);
9137 memcpy(&expected, &expectedBytes.front(), expectedBytes.size());
9138 memcpy(&actual, outputAllocs[outputNdx]->getHostPtr(), expectedBytes.size());
9140 // Test with epsilon
9141 if (fabs(expected - actual) > epsilon)
9143 log << TestLog::Message << "Error: The actual and expected values not matching."
9144 << " Expected: " << expected << " Actual: " << actual << " Epsilon: " << epsilon << TestLog::EndMessage;
9151 // Checks if the driver crash with uninitialized cases
9152 bool passthruVerify (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
9154 DE_ASSERT(outputAllocs.size() != 0);
9155 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
9157 // Copy and discard the result.
9158 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
9160 vector<deUint8> expectedBytes;
9161 expectedOutputs[outputNdx]->getBytes(expectedBytes);
9163 const size_t width = expectedBytes.size();
9164 vector<char> data (width);
9166 memcpy(&data[0], outputAllocs[outputNdx]->getHostPtr(), width);
9171 tcu::TestCaseGroup* createShaderDefaultOutputGroup (tcu::TestContext& testCtx)
9173 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "shader_default_output", "Test shader default output."));
9174 de::Random rnd (deStringHash(group->getName()));
9176 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
9178 NumberType numberType = NumberType(type);
9179 const string typeName = getNumberTypeName(numberType);
9180 const string description = "Test the OpVariable initializer with " + typeName + ".";
9181 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
9183 // 2 similar subcases (initialized and uninitialized)
9184 for (int subCase = 0; subCase < 2; ++subCase)
9186 ComputeShaderSpec spec;
9187 spec.numWorkGroups = IVec3(1, 1, 1);
9189 map<string, string> params;
9193 case NUMBERTYPE_INT32:
9195 deInt32 number = getInt(rnd);
9196 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
9197 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
9198 params["constValue"] = numberToString(number);
9201 case NUMBERTYPE_UINT32:
9203 deUint32 number = rnd.getUint32();
9204 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
9205 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
9206 params["constValue"] = numberToString(number);
9209 case NUMBERTYPE_FLOAT32:
9211 float number = rnd.getFloat();
9212 spec.inputs.push_back(createCompositeBuffer<float>(number));
9213 spec.outputs.push_back(createCompositeBuffer<float>(number));
9214 spec.verifyIO = &compareFloats;
9215 params["constValue"] = numberToString(number);
9222 // Initialized subcase
9225 spec.assembly = specializeDefaultOutputShaderTemplate(numberType, params);
9226 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "initialized", "OpVariable initializer tests.", spec));
9228 // Uninitialized subcase
9231 spec.assembly = specializeDefaultOutputShaderTemplate(numberType);
9232 spec.verifyIO = &passthruVerify;
9233 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "uninitialized", "OpVariable initializer tests.", spec));
9236 group->addChild(subGroup.release());
9238 return group.release();
9241 tcu::TestCaseGroup* createOpNopTests (tcu::TestContext& testCtx)
9243 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
9244 RGBA defaultColors[4];
9245 map<string, string> opNopFragments;
9247 getDefaultColors(defaultColors);
9249 opNopFragments["testfun"] =
9250 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
9251 "%param1 = OpFunctionParameter %v4f32\n"
9252 "%label_testfun = OpLabel\n"
9261 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
9262 "%b = OpFAdd %f32 %a %a\n"
9264 "%c = OpFSub %f32 %b %a\n"
9265 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
9268 "OpReturnValue %ret\n"
9271 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, testGroup.get());
9273 return testGroup.release();
9276 tcu::TestCaseGroup* createOpNameTests (tcu::TestContext& testCtx)
9278 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opname","Test OpName"));
9279 RGBA defaultColors[4];
9280 map<string, string> opNameFragments;
9282 getDefaultColors(defaultColors);
9284 opNameFragments["debug"] =
9285 "OpName %BP_main \"not_main\"";
9287 opNameFragments["testfun"] =
9288 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
9289 "%param1 = OpFunctionParameter %v4f32\n"
9290 "%label_func = OpLabel\n"
9291 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
9292 "%b = OpFAdd %f32 %a %a\n"
9293 "%c = OpFSub %f32 %b %a\n"
9294 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
9295 "OpReturnValue %ret\n"
9298 createTestsForAllStages("opname", defaultColors, defaultColors, opNameFragments, testGroup.get());
9300 return testGroup.release();
9303 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
9305 const bool testComputePipeline = true;
9307 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
9308 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
9309 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
9311 computeTests->addChild(createSpivVersionCheckTests(testCtx, testComputePipeline));
9312 computeTests->addChild(createLocalSizeGroup(testCtx));
9313 computeTests->addChild(createOpNopGroup(testCtx));
9314 computeTests->addChild(createOpFUnordGroup(testCtx));
9315 computeTests->addChild(createOpAtomicGroup(testCtx, false));
9316 computeTests->addChild(createOpAtomicGroup(testCtx, true)); // Using new StorageBuffer decoration
9317 computeTests->addChild(createOpLineGroup(testCtx));
9318 computeTests->addChild(createOpModuleProcessedGroup(testCtx));
9319 computeTests->addChild(createOpNoLineGroup(testCtx));
9320 computeTests->addChild(createOpConstantNullGroup(testCtx));
9321 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
9322 computeTests->addChild(createOpConstantUsageGroup(testCtx));
9323 computeTests->addChild(createSpecConstantGroup(testCtx));
9324 computeTests->addChild(createOpSourceGroup(testCtx));
9325 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
9326 computeTests->addChild(createDecorationGroupGroup(testCtx));
9327 computeTests->addChild(createOpPhiGroup(testCtx));
9328 computeTests->addChild(createLoopControlGroup(testCtx));
9329 computeTests->addChild(createFunctionControlGroup(testCtx));
9330 computeTests->addChild(createSelectionControlGroup(testCtx));
9331 computeTests->addChild(createBlockOrderGroup(testCtx));
9332 computeTests->addChild(createMultipleShaderGroup(testCtx));
9333 computeTests->addChild(createMemoryAccessGroup(testCtx));
9334 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
9335 computeTests->addChild(createOpCopyObjectGroup(testCtx));
9336 computeTests->addChild(createNoContractionGroup(testCtx));
9337 computeTests->addChild(createOpUndefGroup(testCtx));
9338 computeTests->addChild(createOpUnreachableGroup(testCtx));
9339 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
9340 computeTests ->addChild(createOpFRemGroup(testCtx));
9341 computeTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_PASS));
9342 computeTests->addChild(createOpSRemComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
9343 computeTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_PASS));
9344 computeTests->addChild(createOpSModComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
9345 computeTests->addChild(createConvertComputeTests(testCtx, "OpSConvert", "sconvert"));
9346 computeTests->addChild(createConvertComputeTests(testCtx, "OpUConvert", "uconvert"));
9347 computeTests->addChild(createConvertComputeTests(testCtx, "OpFConvert", "fconvert"));
9348 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
9349 computeTests->addChild(createOpInBoundsAccessChainGroup(testCtx));
9350 computeTests->addChild(createShaderDefaultOutputGroup(testCtx));
9351 computeTests->addChild(createOpNMinGroup(testCtx));
9352 computeTests->addChild(createOpNMaxGroup(testCtx));
9353 computeTests->addChild(createOpNClampGroup(testCtx));
9355 de::MovePtr<tcu::TestCaseGroup> computeAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
9357 computeAndroidTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9358 computeAndroidTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9360 computeTests->addChild(computeAndroidTests.release());
9362 computeTests->addChild(create16BitStorageComputeGroup(testCtx));
9363 computeTests->addChild(createUboMatrixPaddingComputeGroup(testCtx));
9364 computeTests->addChild(createVariableInitComputeGroup(testCtx));
9365 computeTests->addChild(createConditionalBranchComputeGroup(testCtx));
9366 computeTests->addChild(createIndexingComputeGroup(testCtx));
9367 computeTests->addChild(createVariablePointersComputeGroup(testCtx));
9368 computeTests->addChild(createImageSamplerComputeGroup(testCtx));
9369 computeTests->addChild(createOpNameGroup(testCtx));
9370 graphicsTests->addChild(createCrossStageInterfaceTests(testCtx));
9371 graphicsTests->addChild(createSpivVersionCheckTests(testCtx, !testComputePipeline));
9372 graphicsTests->addChild(createOpNopTests(testCtx));
9373 graphicsTests->addChild(createOpSourceTests(testCtx));
9374 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
9375 graphicsTests->addChild(createOpModuleProcessedTests(testCtx));
9376 graphicsTests->addChild(createOpLineTests(testCtx));
9377 graphicsTests->addChild(createOpNoLineTests(testCtx));
9378 graphicsTests->addChild(createOpConstantNullTests(testCtx));
9379 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
9380 graphicsTests->addChild(createMemoryAccessTests(testCtx));
9381 graphicsTests->addChild(createOpUndefTests(testCtx));
9382 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
9383 graphicsTests->addChild(createModuleTests(testCtx));
9384 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
9385 graphicsTests->addChild(createOpPhiTests(testCtx));
9386 graphicsTests->addChild(createNoContractionTests(testCtx));
9387 graphicsTests->addChild(createOpQuantizeTests(testCtx));
9388 graphicsTests->addChild(createLoopTests(testCtx));
9389 graphicsTests->addChild(createSpecConstantTests(testCtx));
9390 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
9391 graphicsTests->addChild(createBarrierTests(testCtx));
9392 graphicsTests->addChild(createDecorationGroupTests(testCtx));
9393 graphicsTests->addChild(createFRemTests(testCtx));
9394 graphicsTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
9395 graphicsTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
9398 de::MovePtr<tcu::TestCaseGroup> graphicsAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
9400 graphicsAndroidTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9401 graphicsAndroidTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
9403 graphicsTests->addChild(graphicsAndroidTests.release());
9405 graphicsTests->addChild(createOpNameTests(testCtx));
9407 graphicsTests->addChild(create16BitStorageGraphicsGroup(testCtx));
9408 graphicsTests->addChild(createUboMatrixPaddingGraphicsGroup(testCtx));
9409 graphicsTests->addChild(createVariableInitGraphicsGroup(testCtx));
9410 graphicsTests->addChild(createConditionalBranchGraphicsGroup(testCtx));
9411 graphicsTests->addChild(createIndexingGraphicsGroup(testCtx));
9412 graphicsTests->addChild(createVariablePointersGraphicsGroup(testCtx));
9413 graphicsTests->addChild(createImageSamplerGraphicsGroup(testCtx));
9414 graphicsTests->addChild(createConvertGraphicsTests(testCtx, "OpSConvert", "sconvert"));
9415 graphicsTests->addChild(createConvertGraphicsTests(testCtx, "OpUConvert", "uconvert"));
9416 graphicsTests->addChild(createConvertGraphicsTests(testCtx, "OpFConvert", "fconvert"));
9418 instructionTests->addChild(computeTests.release());
9419 instructionTests->addChild(graphicsTests.release());
9421 return instructionTests.release();