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 "deRandom.hpp"
48 #include "deStringUtil.hpp"
49 #include "deUniquePtr.hpp"
51 #include "tcuStringTemplate.hpp"
53 #include "vktSpvAsm16bitStorageTests.hpp"
54 #include "vktSpvAsmComputeShaderCase.hpp"
55 #include "vktSpvAsmComputeShaderTestUtil.hpp"
56 #include "vktSpvAsmGraphicsShaderTestUtil.hpp"
57 #include "vktSpvAsmVariablePointersTests.hpp"
58 #include "vktTestCaseUtil.hpp"
69 namespace SpirVAssembly
83 using tcu::TestStatus;
86 using tcu::StringTemplate;
90 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0)
92 T* const typedPtr = (T*)dst;
93 for (int ndx = 0; ndx < numValues; ndx++)
94 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue);
97 // Filter is a function that returns true if a value should pass, false otherwise.
98 template<typename T, typename FilterT>
99 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, FilterT filter, int offset = 0)
101 T* const typedPtr = (T*)dst;
103 for (int ndx = 0; ndx < numValues; ndx++)
106 value = randomScalar<T>(rnd, minValue, maxValue);
107 while (!filter(value));
109 typedPtr[offset + ndx] = value;
113 // Gets a 64-bit integer with a more logarithmic distribution
114 deInt64 randomInt64LogDistributed (de::Random& rnd)
116 deInt64 val = rnd.getUint64();
117 val &= (1ull << rnd.getInt(1, 63)) - 1;
123 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues)
125 for (int ndx = 0; ndx < numValues; ndx++)
126 dst[ndx] = randomInt64LogDistributed(rnd);
129 template<typename FilterT>
130 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues, FilterT filter)
132 for (int ndx = 0; ndx < numValues; ndx++)
136 value = randomInt64LogDistributed(rnd);
137 } while (!filter(value));
142 inline bool filterNonNegative (const deInt64 value)
147 inline bool filterPositive (const deInt64 value)
152 inline bool filterNotZero (const deInt64 value)
157 static void floorAll (vector<float>& values)
159 for (size_t i = 0; i < values.size(); i++)
160 values[i] = deFloatFloor(values[i]);
163 static void floorAll (vector<Vec4>& values)
165 for (size_t i = 0; i < values.size(); i++)
166 values[i] = floor(values[i]);
174 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {}
177 // Assembly code used for testing OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code:
181 // layout(std140, set = 0, binding = 0) readonly buffer Input {
184 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
188 // layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
191 // uint x = gl_GlobalInvocationID.x;
192 // output_data.elements[x] = -input_data.elements[x];
195 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
197 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
198 ComputeShaderSpec spec;
199 de::Random rnd (deStringHash(group->getName()));
200 const int numElements = 100;
201 vector<float> positiveFloats (numElements, 0);
202 vector<float> negativeFloats (numElements, 0);
204 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
206 for (size_t ndx = 0; ndx < numElements; ++ndx)
207 negativeFloats[ndx] = -positiveFloats[ndx];
210 string(getComputeAsmShaderPreamble()) +
212 "OpSource GLSL 430\n"
213 "OpName %main \"main\"\n"
214 "OpName %id \"gl_GlobalInvocationID\"\n"
216 "OpDecorate %id BuiltIn GlobalInvocationId\n"
218 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes())
220 + string(getComputeAsmInputOutputBuffer()) +
222 "%id = OpVariable %uvec3ptr Input\n"
223 "%zero = OpConstant %i32 0\n"
225 "%main = OpFunction %void None %voidf\n"
227 "%idval = OpLoad %uvec3 %id\n"
228 "%x = OpCompositeExtract %u32 %idval 0\n"
230 " OpNop\n" // Inside a function body
232 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
233 "%inval = OpLoad %f32 %inloc\n"
234 "%neg = OpFNegate %f32 %inval\n"
235 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
236 " OpStore %outloc %neg\n"
239 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
240 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
241 spec.numWorkGroups = IVec3(numElements, 1, 1);
243 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
245 return group.release();
248 bool compareFUnord (const std::vector<BufferSp>& inputs, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
250 if (outputAllocs.size() != 1)
253 const BufferSp& expectedOutput = expectedOutputs[0];
254 const deInt32* expectedOutputAsInt = static_cast<const deInt32*>(expectedOutputs[0]->data());
255 const deInt32* outputAsInt = static_cast<const deInt32*>(outputAllocs[0]->getHostPtr());
256 const float* input1AsFloat = static_cast<const float*>(inputs[0]->data());
257 const float* input2AsFloat = static_cast<const float*>(inputs[1]->data());
258 bool returnValue = true;
260 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(deInt32); ++idx)
262 if (outputAsInt[idx] != expectedOutputAsInt[idx])
264 log << TestLog::Message << "ERROR: Sub-case failed. inputs: " << input1AsFloat[idx] << "," << input2AsFloat[idx] << " output: " << outputAsInt[idx]<< " expected output: " << expectedOutputAsInt[idx] << TestLog::EndMessage;
271 typedef VkBool32 (*compareFuncType) (float, float);
277 compareFuncType compareFunc;
279 OpFUnordCase (const char* _name, const char* _opCode, compareFuncType _compareFunc)
282 , compareFunc (_compareFunc) {}
285 #define ADD_OPFUNORD_CASE(NAME, OPCODE, OPERATOR) \
287 struct compare_##NAME { static VkBool32 compare(float x, float y) { return (x OPERATOR y) ? VK_TRUE : VK_FALSE; } }; \
288 cases.push_back(OpFUnordCase(#NAME, OPCODE, compare_##NAME::compare)); \
289 } while (deGetFalse())
291 tcu::TestCaseGroup* createOpFUnordGroup (tcu::TestContext& testCtx)
293 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfunord", "Test the OpFUnord* opcodes"));
294 de::Random rnd (deStringHash(group->getName()));
295 const int numElements = 100;
296 vector<OpFUnordCase> cases;
298 const StringTemplate shaderTemplate (
300 string(getComputeAsmShaderPreamble()) +
302 "OpSource GLSL 430\n"
303 "OpName %main \"main\"\n"
304 "OpName %id \"gl_GlobalInvocationID\"\n"
306 "OpDecorate %id BuiltIn GlobalInvocationId\n"
308 "OpDecorate %buf BufferBlock\n"
309 "OpDecorate %buf2 BufferBlock\n"
310 "OpDecorate %indata1 DescriptorSet 0\n"
311 "OpDecorate %indata1 Binding 0\n"
312 "OpDecorate %indata2 DescriptorSet 0\n"
313 "OpDecorate %indata2 Binding 1\n"
314 "OpDecorate %outdata DescriptorSet 0\n"
315 "OpDecorate %outdata Binding 2\n"
316 "OpDecorate %f32arr ArrayStride 4\n"
317 "OpDecorate %i32arr ArrayStride 4\n"
318 "OpMemberDecorate %buf 0 Offset 0\n"
319 "OpMemberDecorate %buf2 0 Offset 0\n"
321 + string(getComputeAsmCommonTypes()) +
323 "%buf = OpTypeStruct %f32arr\n"
324 "%bufptr = OpTypePointer Uniform %buf\n"
325 "%indata1 = OpVariable %bufptr Uniform\n"
326 "%indata2 = OpVariable %bufptr Uniform\n"
328 "%buf2 = OpTypeStruct %i32arr\n"
329 "%buf2ptr = OpTypePointer Uniform %buf2\n"
330 "%outdata = OpVariable %buf2ptr Uniform\n"
332 "%id = OpVariable %uvec3ptr Input\n"
333 "%zero = OpConstant %i32 0\n"
334 "%consti1 = OpConstant %i32 1\n"
335 "%constf1 = OpConstant %f32 1.0\n"
337 "%main = OpFunction %void None %voidf\n"
339 "%idval = OpLoad %uvec3 %id\n"
340 "%x = OpCompositeExtract %u32 %idval 0\n"
342 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
343 "%inval1 = OpLoad %f32 %inloc1\n"
344 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
345 "%inval2 = OpLoad %f32 %inloc2\n"
346 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
348 "%result = ${OPCODE} %bool %inval1 %inval2\n"
349 "%int_res = OpSelect %i32 %result %consti1 %zero\n"
350 " OpStore %outloc %int_res\n"
355 ADD_OPFUNORD_CASE(equal, "OpFUnordEqual", ==);
356 ADD_OPFUNORD_CASE(less, "OpFUnordLessThan", <);
357 ADD_OPFUNORD_CASE(lessequal, "OpFUnordLessThanEqual", <=);
358 ADD_OPFUNORD_CASE(greater, "OpFUnordGreaterThan", >);
359 ADD_OPFUNORD_CASE(greaterequal, "OpFUnordGreaterThanEqual", >=);
360 ADD_OPFUNORD_CASE(notequal, "OpFUnordNotEqual", !=);
362 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
364 map<string, string> specializations;
365 ComputeShaderSpec spec;
366 const float NaN = std::numeric_limits<float>::quiet_NaN();
367 vector<float> inputFloats1 (numElements, 0);
368 vector<float> inputFloats2 (numElements, 0);
369 vector<deInt32> expectedInts (numElements, 0);
371 specializations["OPCODE"] = cases[caseNdx].opCode;
372 spec.assembly = shaderTemplate.specialize(specializations);
374 fillRandomScalars(rnd, 1.f, 100.f, &inputFloats1[0], numElements);
375 for (size_t ndx = 0; ndx < numElements; ++ndx)
379 case 0: inputFloats2[ndx] = inputFloats1[ndx] + 1.0f; break;
380 case 1: inputFloats2[ndx] = inputFloats1[ndx] - 1.0f; break;
381 case 2: inputFloats2[ndx] = inputFloats1[ndx]; break;
382 case 3: inputFloats2[ndx] = NaN; break;
383 case 4: inputFloats2[ndx] = inputFloats1[ndx]; inputFloats1[ndx] = NaN; break;
384 case 5: inputFloats2[ndx] = NaN; inputFloats1[ndx] = NaN; break;
386 expectedInts[ndx] = tcu::Float32(inputFloats1[ndx]).isNaN() || tcu::Float32(inputFloats2[ndx]).isNaN() || cases[caseNdx].compareFunc(inputFloats1[ndx], inputFloats2[ndx]);
389 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
390 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
391 spec.outputs.push_back(BufferSp(new Int32Buffer(expectedInts)));
392 spec.numWorkGroups = IVec3(numElements, 1, 1);
393 spec.verifyIO = &compareFUnord;
394 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
397 return group.release();
403 const char* assembly;
404 void (*calculateExpected)(deInt32&, deInt32);
405 deInt32 numOutputElements;
407 OpAtomicCase (const char* _name, const char* _assembly, void (*_calculateExpected)(deInt32&, deInt32), deInt32 _numOutputElements)
409 , assembly (_assembly)
410 , calculateExpected (_calculateExpected)
411 , numOutputElements (_numOutputElements) {}
414 tcu::TestCaseGroup* createOpAtomicGroup (tcu::TestContext& testCtx, bool useStorageBuffer)
416 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx,
417 useStorageBuffer ? "opatomic_storage_buffer" : "opatomic",
418 "Test the OpAtomic* opcodes"));
419 de::Random rnd (deStringHash(group->getName()));
420 const int numElements = 1000000;
421 vector<OpAtomicCase> cases;
423 const StringTemplate shaderTemplate (
425 string("OpCapability Shader\n") +
426 (useStorageBuffer ? "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" : "") +
427 "OpMemoryModel Logical GLSL450\n"
428 "OpEntryPoint GLCompute %main \"main\" %id\n"
429 "OpExecutionMode %main LocalSize 1 1 1\n" +
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 ${BLOCK_DECORATION}\n"
438 "OpDecorate %indata DescriptorSet 0\n"
439 "OpDecorate %indata Binding 0\n"
440 "OpDecorate %i32arr ArrayStride 4\n"
441 "OpMemberDecorate %buf 0 Offset 0\n"
443 "OpDecorate %sumbuf ${BLOCK_DECORATION}\n"
444 "OpDecorate %sum DescriptorSet 0\n"
445 "OpDecorate %sum Binding 1\n"
446 "OpMemberDecorate %sumbuf 0 Coherent\n"
447 "OpMemberDecorate %sumbuf 0 Offset 0\n"
449 + string(getComputeAsmCommonTypes()) +
450 "%void = OpTypeVoid\n"
451 "%voidf = OpTypeFunction %void\n"
452 "%u32 = OpTypeInt 32 0\n"
453 "%i32 = OpTypeInt 32 1\n"
454 "%uvec3 = OpTypeVector %u32 3\n"
455 "%uvec3ptr = OpTypePointer Input %uvec3\n"
456 "%i32ptr = OpTypePointer ${BLOCK_POINTER_TYPE} %i32\n"
457 "%i32arr = OpTypeRuntimeArray %i32\n"
459 "%buf = OpTypeStruct %i32arr\n"
460 "%bufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %buf\n"
461 "%indata = OpVariable %bufptr ${BLOCK_POINTER_TYPE}\n"
463 "%sumbuf = OpTypeStruct %i32arr\n"
464 "%sumbufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %sumbuf\n"
465 "%sum = OpVariable %sumbufptr ${BLOCK_POINTER_TYPE}\n"
467 "%id = OpVariable %uvec3ptr Input\n"
468 "%minusone = OpConstant %i32 -1\n"
469 "%zero = OpConstant %i32 0\n"
470 "%one = OpConstant %u32 1\n"
471 "%two = OpConstant %i32 2\n"
473 "%main = OpFunction %void None %voidf\n"
475 "%idval = OpLoad %uvec3 %id\n"
476 "%x = OpCompositeExtract %u32 %idval 0\n"
478 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
479 "%inval = OpLoad %i32 %inloc\n"
481 "%outloc = OpAccessChain %i32ptr %sum %zero ${INDEX}\n"
487 #define ADD_OPATOMIC_CASE(NAME, ASSEMBLY, CALCULATE_EXPECTED, NUM_OUTPUT_ELEMENTS) \
489 DE_STATIC_ASSERT((NUM_OUTPUT_ELEMENTS) == 1 || (NUM_OUTPUT_ELEMENTS) == numElements); \
490 struct calculateExpected_##NAME { static void calculateExpected(deInt32& expected, deInt32 input) CALCULATE_EXPECTED }; /* NOLINT(CALCULATE_EXPECTED) */ \
491 cases.push_back(OpAtomicCase(#NAME, ASSEMBLY, calculateExpected_##NAME::calculateExpected, NUM_OUTPUT_ELEMENTS)); \
492 } while (deGetFalse())
493 #define ADD_OPATOMIC_CASE_1(NAME, ASSEMBLY, CALCULATE_EXPECTED) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, CALCULATE_EXPECTED, 1)
494 #define ADD_OPATOMIC_CASE_N(NAME, ASSEMBLY, CALCULATE_EXPECTED) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, CALCULATE_EXPECTED, numElements)
496 ADD_OPATOMIC_CASE_1(iadd, "%unused = OpAtomicIAdd %i32 %outloc %one %zero %inval\n", { expected += input; } );
497 ADD_OPATOMIC_CASE_1(isub, "%unused = OpAtomicISub %i32 %outloc %one %zero %inval\n", { expected -= input; } );
498 ADD_OPATOMIC_CASE_1(iinc, "%unused = OpAtomicIIncrement %i32 %outloc %one %zero\n", { ++expected; (void)input;} );
499 ADD_OPATOMIC_CASE_1(idec, "%unused = OpAtomicIDecrement %i32 %outloc %one %zero\n", { --expected; (void)input;} );
500 ADD_OPATOMIC_CASE_N(load, "%inval2 = OpAtomicLoad %i32 %inloc %zero %zero\n"
501 " OpStore %outloc %inval2\n", { expected = input;} );
502 ADD_OPATOMIC_CASE_N(store, " OpAtomicStore %outloc %zero %zero %inval\n", { expected = input;} );
503 ADD_OPATOMIC_CASE_N(compex, "%even = OpSMod %i32 %inval %two\n"
504 " OpStore %outloc %even\n"
505 "%unused = OpAtomicCompareExchange %i32 %outloc %one %zero %zero %minusone %zero\n", { expected = (input % 2) == 0 ? -1 : 1;} );
507 #undef ADD_OPATOMIC_CASE
508 #undef ADD_OPATOMIC_CASE_1
509 #undef ADD_OPATOMIC_CASE_N
511 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
513 map<string, string> specializations;
514 ComputeShaderSpec spec;
515 vector<deInt32> inputInts (numElements, 0);
516 vector<deInt32> expected (cases[caseNdx].numOutputElements, -1);
518 specializations["INDEX"] = (cases[caseNdx].numOutputElements == 1) ? "%zero" : "%x";
519 specializations["INSTRUCTION"] = cases[caseNdx].assembly;
520 specializations["BLOCK_DECORATION"] = useStorageBuffer ? "Block" : "BufferBlock";
521 specializations["BLOCK_POINTER_TYPE"] = useStorageBuffer ? "StorageBuffer" : "Uniform";
522 spec.assembly = shaderTemplate.specialize(specializations);
524 if (useStorageBuffer)
525 spec.extensions.push_back("VK_KHR_storage_buffer_storage_class");
527 fillRandomScalars(rnd, 1, 100, &inputInts[0], numElements);
528 for (size_t ndx = 0; ndx < numElements; ++ndx)
530 cases[caseNdx].calculateExpected((cases[caseNdx].numOutputElements == 1) ? expected[0] : expected[ndx], inputInts[ndx]);
533 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
534 spec.outputs.push_back(BufferSp(new Int32Buffer(expected)));
535 spec.numWorkGroups = IVec3(numElements, 1, 1);
536 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
539 return group.release();
542 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
544 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
545 ComputeShaderSpec spec;
546 de::Random rnd (deStringHash(group->getName()));
547 const int numElements = 100;
548 vector<float> positiveFloats (numElements, 0);
549 vector<float> negativeFloats (numElements, 0);
551 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
553 for (size_t ndx = 0; ndx < numElements; ++ndx)
554 negativeFloats[ndx] = -positiveFloats[ndx];
557 string(getComputeAsmShaderPreamble()) +
559 "%fname1 = OpString \"negateInputs.comp\"\n"
560 "%fname2 = OpString \"negateInputs\"\n"
562 "OpSource GLSL 430\n"
563 "OpName %main \"main\"\n"
564 "OpName %id \"gl_GlobalInvocationID\"\n"
566 "OpDecorate %id BuiltIn GlobalInvocationId\n"
568 + string(getComputeAsmInputOutputBufferTraits()) +
570 "OpLine %fname1 0 0\n" // At the earliest possible position
572 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
574 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
575 "OpLine %fname2 1 0\n" // Different filenames
576 "OpLine %fname1 1000 100000\n"
578 "%id = OpVariable %uvec3ptr Input\n"
579 "%zero = OpConstant %i32 0\n"
581 "OpLine %fname1 1 1\n" // Before a function
583 "%main = OpFunction %void None %voidf\n"
586 "OpLine %fname1 1 1\n" // In a function
588 "%idval = OpLoad %uvec3 %id\n"
589 "%x = OpCompositeExtract %u32 %idval 0\n"
590 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
591 "%inval = OpLoad %f32 %inloc\n"
592 "%neg = OpFNegate %f32 %inval\n"
593 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
594 " OpStore %outloc %neg\n"
597 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
598 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
599 spec.numWorkGroups = IVec3(numElements, 1, 1);
601 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
603 return group.release();
606 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
608 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
609 ComputeShaderSpec spec;
610 de::Random rnd (deStringHash(group->getName()));
611 const int numElements = 100;
612 vector<float> positiveFloats (numElements, 0);
613 vector<float> negativeFloats (numElements, 0);
615 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
617 for (size_t ndx = 0; ndx < numElements; ++ndx)
618 negativeFloats[ndx] = -positiveFloats[ndx];
621 string(getComputeAsmShaderPreamble()) +
623 "%fname = OpString \"negateInputs.comp\"\n"
625 "OpSource GLSL 430\n"
626 "OpName %main \"main\"\n"
627 "OpName %id \"gl_GlobalInvocationID\"\n"
629 "OpDecorate %id BuiltIn GlobalInvocationId\n"
631 + string(getComputeAsmInputOutputBufferTraits()) +
633 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
635 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
637 "OpLine %fname 0 1\n"
638 "OpNoLine\n" // Immediately following a preceding OpLine
640 "OpLine %fname 1000 1\n"
642 "%id = OpVariable %uvec3ptr Input\n"
643 "%zero = OpConstant %i32 0\n"
645 "OpNoLine\n" // Contents after the previous OpLine
647 "%main = OpFunction %void None %voidf\n"
649 "%idval = OpLoad %uvec3 %id\n"
650 "%x = OpCompositeExtract %u32 %idval 0\n"
652 "OpNoLine\n" // Multiple OpNoLine
656 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
657 "%inval = OpLoad %f32 %inloc\n"
658 "%neg = OpFNegate %f32 %inval\n"
659 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
660 " OpStore %outloc %neg\n"
663 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
664 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
665 spec.numWorkGroups = IVec3(numElements, 1, 1);
667 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
669 return group.release();
672 // Compare instruction for the contraction compute case.
673 // Returns true if the output is what is expected from the test case.
674 bool compareNoContractCase(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
676 if (outputAllocs.size() != 1)
679 // We really just need this for size because we are not comparing the exact values.
680 const BufferSp& expectedOutput = expectedOutputs[0];
681 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
683 for(size_t i = 0; i < expectedOutput->getNumBytes() / sizeof(float); ++i) {
684 if (outputAsFloat[i] != 0.f &&
685 outputAsFloat[i] != -ldexp(1, -24)) {
693 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
695 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
696 vector<CaseParameter> cases;
697 const int numElements = 100;
698 vector<float> inputFloats1 (numElements, 0);
699 vector<float> inputFloats2 (numElements, 0);
700 vector<float> outputFloats (numElements, 0);
701 const StringTemplate shaderTemplate (
702 string(getComputeAsmShaderPreamble()) +
704 "OpName %main \"main\"\n"
705 "OpName %id \"gl_GlobalInvocationID\"\n"
707 "OpDecorate %id BuiltIn GlobalInvocationId\n"
711 "OpDecorate %buf BufferBlock\n"
712 "OpDecorate %indata1 DescriptorSet 0\n"
713 "OpDecorate %indata1 Binding 0\n"
714 "OpDecorate %indata2 DescriptorSet 0\n"
715 "OpDecorate %indata2 Binding 1\n"
716 "OpDecorate %outdata DescriptorSet 0\n"
717 "OpDecorate %outdata Binding 2\n"
718 "OpDecorate %f32arr ArrayStride 4\n"
719 "OpMemberDecorate %buf 0 Offset 0\n"
721 + string(getComputeAsmCommonTypes()) +
723 "%buf = OpTypeStruct %f32arr\n"
724 "%bufptr = OpTypePointer Uniform %buf\n"
725 "%indata1 = OpVariable %bufptr Uniform\n"
726 "%indata2 = OpVariable %bufptr Uniform\n"
727 "%outdata = OpVariable %bufptr Uniform\n"
729 "%id = OpVariable %uvec3ptr Input\n"
730 "%zero = OpConstant %i32 0\n"
731 "%c_f_m1 = OpConstant %f32 -1.\n"
733 "%main = OpFunction %void None %voidf\n"
735 "%idval = OpLoad %uvec3 %id\n"
736 "%x = OpCompositeExtract %u32 %idval 0\n"
737 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
738 "%inval1 = OpLoad %f32 %inloc1\n"
739 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
740 "%inval2 = OpLoad %f32 %inloc2\n"
741 "%mul = OpFMul %f32 %inval1 %inval2\n"
742 "%add = OpFAdd %f32 %mul %c_f_m1\n"
743 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
744 " OpStore %outloc %add\n"
748 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
749 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
750 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
752 for (size_t ndx = 0; ndx < numElements; ++ndx)
754 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
755 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
756 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
757 // conducted separately and the result is rounded to 1, or 0x1.fffffcp-1
758 // So the final result will be 0.f or 0x1p-24.
759 // If the operation is combined into a precise fused multiply-add, then the result would be
760 // 2^-46 (0xa8800000).
761 outputFloats[ndx] = 0.f;
764 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
766 map<string, string> specializations;
767 ComputeShaderSpec spec;
769 specializations["DECORATION"] = cases[caseNdx].param;
770 spec.assembly = shaderTemplate.specialize(specializations);
771 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
772 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
773 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
774 spec.numWorkGroups = IVec3(numElements, 1, 1);
775 // Check against the two possible answers based on rounding mode.
776 spec.verifyIO = &compareNoContractCase;
778 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
780 return group.release();
783 bool compareFRem(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
785 if (outputAllocs.size() != 1)
788 const BufferSp& expectedOutput = expectedOutputs[0];
789 const float *expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
790 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
792 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
794 const float f0 = expectedOutputAsFloat[idx];
795 const float f1 = outputAsFloat[idx];
796 // \todo relative error needs to be fairly high because FRem may be implemented as
797 // (roughly) frac(a/b)*b, so LSB errors can be magnified. But this should be fine for now.
798 if (deFloatAbs((f1 - f0) / f0) > 0.02)
805 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
807 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
808 ComputeShaderSpec spec;
809 de::Random rnd (deStringHash(group->getName()));
810 const int numElements = 200;
811 vector<float> inputFloats1 (numElements, 0);
812 vector<float> inputFloats2 (numElements, 0);
813 vector<float> outputFloats (numElements, 0);
815 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
816 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
818 for (size_t ndx = 0; ndx < numElements; ++ndx)
820 // Guard against divisors near zero.
821 if (std::fabs(inputFloats2[ndx]) < 1e-3)
822 inputFloats2[ndx] = 8.f;
824 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
825 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
829 string(getComputeAsmShaderPreamble()) +
831 "OpName %main \"main\"\n"
832 "OpName %id \"gl_GlobalInvocationID\"\n"
834 "OpDecorate %id BuiltIn GlobalInvocationId\n"
836 "OpDecorate %buf BufferBlock\n"
837 "OpDecorate %indata1 DescriptorSet 0\n"
838 "OpDecorate %indata1 Binding 0\n"
839 "OpDecorate %indata2 DescriptorSet 0\n"
840 "OpDecorate %indata2 Binding 1\n"
841 "OpDecorate %outdata DescriptorSet 0\n"
842 "OpDecorate %outdata Binding 2\n"
843 "OpDecorate %f32arr ArrayStride 4\n"
844 "OpMemberDecorate %buf 0 Offset 0\n"
846 + string(getComputeAsmCommonTypes()) +
848 "%buf = OpTypeStruct %f32arr\n"
849 "%bufptr = OpTypePointer Uniform %buf\n"
850 "%indata1 = OpVariable %bufptr Uniform\n"
851 "%indata2 = OpVariable %bufptr Uniform\n"
852 "%outdata = OpVariable %bufptr Uniform\n"
854 "%id = OpVariable %uvec3ptr Input\n"
855 "%zero = OpConstant %i32 0\n"
857 "%main = OpFunction %void None %voidf\n"
859 "%idval = OpLoad %uvec3 %id\n"
860 "%x = OpCompositeExtract %u32 %idval 0\n"
861 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
862 "%inval1 = OpLoad %f32 %inloc1\n"
863 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
864 "%inval2 = OpLoad %f32 %inloc2\n"
865 "%rem = OpFRem %f32 %inval1 %inval2\n"
866 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
867 " OpStore %outloc %rem\n"
871 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
872 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
873 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
874 spec.numWorkGroups = IVec3(numElements, 1, 1);
875 spec.verifyIO = &compareFRem;
877 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
879 return group.release();
882 bool compareNMin (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
884 if (outputAllocs.size() != 1)
887 const BufferSp& expectedOutput = expectedOutputs[0];
888 const float* const expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
889 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
891 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
893 const float f0 = expectedOutputAsFloat[idx];
894 const float f1 = outputAsFloat[idx];
896 // For NMin, we accept NaN as output if both inputs were NaN.
897 // Otherwise the NaN is the wrong choise, as on architectures that
898 // do not handle NaN, those are huge values.
899 if (!(tcu::Float32(f1).isNaN() && tcu::Float32(f0).isNaN()) && deFloatAbs(f1 - f0) > 0.00001f)
906 tcu::TestCaseGroup* createOpNMinGroup (tcu::TestContext& testCtx)
908 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnmin", "Test the OpNMin instruction"));
909 ComputeShaderSpec spec;
910 de::Random rnd (deStringHash(group->getName()));
911 const int numElements = 200;
912 vector<float> inputFloats1 (numElements, 0);
913 vector<float> inputFloats2 (numElements, 0);
914 vector<float> outputFloats (numElements, 0);
916 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
917 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
919 // Make the first case a full-NAN case.
920 inputFloats1[0] = TCU_NAN;
921 inputFloats2[0] = TCU_NAN;
923 for (size_t ndx = 0; ndx < numElements; ++ndx)
925 // By default, pick the smallest
926 outputFloats[ndx] = std::min(inputFloats1[ndx], inputFloats2[ndx]);
928 // Make half of the cases NaN cases
931 // Alternate between the NaN operand
934 outputFloats[ndx] = inputFloats2[ndx];
935 inputFloats1[ndx] = TCU_NAN;
939 outputFloats[ndx] = inputFloats1[ndx];
940 inputFloats2[ndx] = TCU_NAN;
946 "OpCapability Shader\n"
947 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
948 "OpMemoryModel Logical GLSL450\n"
949 "OpEntryPoint GLCompute %main \"main\" %id\n"
950 "OpExecutionMode %main LocalSize 1 1 1\n"
952 "OpName %main \"main\"\n"
953 "OpName %id \"gl_GlobalInvocationID\"\n"
955 "OpDecorate %id BuiltIn GlobalInvocationId\n"
957 "OpDecorate %buf BufferBlock\n"
958 "OpDecorate %indata1 DescriptorSet 0\n"
959 "OpDecorate %indata1 Binding 0\n"
960 "OpDecorate %indata2 DescriptorSet 0\n"
961 "OpDecorate %indata2 Binding 1\n"
962 "OpDecorate %outdata DescriptorSet 0\n"
963 "OpDecorate %outdata Binding 2\n"
964 "OpDecorate %f32arr ArrayStride 4\n"
965 "OpMemberDecorate %buf 0 Offset 0\n"
967 + string(getComputeAsmCommonTypes()) +
969 "%buf = OpTypeStruct %f32arr\n"
970 "%bufptr = OpTypePointer Uniform %buf\n"
971 "%indata1 = OpVariable %bufptr Uniform\n"
972 "%indata2 = OpVariable %bufptr Uniform\n"
973 "%outdata = OpVariable %bufptr Uniform\n"
975 "%id = OpVariable %uvec3ptr Input\n"
976 "%zero = OpConstant %i32 0\n"
978 "%main = OpFunction %void None %voidf\n"
980 "%idval = OpLoad %uvec3 %id\n"
981 "%x = OpCompositeExtract %u32 %idval 0\n"
982 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
983 "%inval1 = OpLoad %f32 %inloc1\n"
984 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
985 "%inval2 = OpLoad %f32 %inloc2\n"
986 "%rem = OpExtInst %f32 %std450 NMin %inval1 %inval2\n"
987 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
988 " OpStore %outloc %rem\n"
992 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
993 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
994 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
995 spec.numWorkGroups = IVec3(numElements, 1, 1);
996 spec.verifyIO = &compareNMin;
998 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1000 return group.release();
1003 bool compareNMax (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1005 if (outputAllocs.size() != 1)
1008 const BufferSp& expectedOutput = expectedOutputs[0];
1009 const float* const expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
1010 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
1012 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
1014 const float f0 = expectedOutputAsFloat[idx];
1015 const float f1 = outputAsFloat[idx];
1017 // For NMax, NaN is considered acceptable result, since in
1018 // architectures that do not handle NaNs, those are huge values.
1019 if (!tcu::Float32(f1).isNaN() && deFloatAbs(f1 - f0) > 0.00001f)
1026 tcu::TestCaseGroup* createOpNMaxGroup (tcu::TestContext& testCtx)
1028 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, "opnmax", "Test the OpNMax instruction"));
1029 ComputeShaderSpec spec;
1030 de::Random rnd (deStringHash(group->getName()));
1031 const int numElements = 200;
1032 vector<float> inputFloats1 (numElements, 0);
1033 vector<float> inputFloats2 (numElements, 0);
1034 vector<float> outputFloats (numElements, 0);
1036 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1037 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1039 // Make the first case a full-NAN case.
1040 inputFloats1[0] = TCU_NAN;
1041 inputFloats2[0] = TCU_NAN;
1043 for (size_t ndx = 0; ndx < numElements; ++ndx)
1045 // By default, pick the biggest
1046 outputFloats[ndx] = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1048 // Make half of the cases NaN cases
1051 // Alternate between the NaN operand
1054 outputFloats[ndx] = inputFloats2[ndx];
1055 inputFloats1[ndx] = TCU_NAN;
1059 outputFloats[ndx] = inputFloats1[ndx];
1060 inputFloats2[ndx] = TCU_NAN;
1066 "OpCapability Shader\n"
1067 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1068 "OpMemoryModel Logical GLSL450\n"
1069 "OpEntryPoint GLCompute %main \"main\" %id\n"
1070 "OpExecutionMode %main LocalSize 1 1 1\n"
1072 "OpName %main \"main\"\n"
1073 "OpName %id \"gl_GlobalInvocationID\"\n"
1075 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1077 "OpDecorate %buf BufferBlock\n"
1078 "OpDecorate %indata1 DescriptorSet 0\n"
1079 "OpDecorate %indata1 Binding 0\n"
1080 "OpDecorate %indata2 DescriptorSet 0\n"
1081 "OpDecorate %indata2 Binding 1\n"
1082 "OpDecorate %outdata DescriptorSet 0\n"
1083 "OpDecorate %outdata Binding 2\n"
1084 "OpDecorate %f32arr ArrayStride 4\n"
1085 "OpMemberDecorate %buf 0 Offset 0\n"
1087 + string(getComputeAsmCommonTypes()) +
1089 "%buf = OpTypeStruct %f32arr\n"
1090 "%bufptr = OpTypePointer Uniform %buf\n"
1091 "%indata1 = OpVariable %bufptr Uniform\n"
1092 "%indata2 = OpVariable %bufptr Uniform\n"
1093 "%outdata = OpVariable %bufptr Uniform\n"
1095 "%id = OpVariable %uvec3ptr Input\n"
1096 "%zero = OpConstant %i32 0\n"
1098 "%main = OpFunction %void None %voidf\n"
1099 "%label = OpLabel\n"
1100 "%idval = OpLoad %uvec3 %id\n"
1101 "%x = OpCompositeExtract %u32 %idval 0\n"
1102 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1103 "%inval1 = OpLoad %f32 %inloc1\n"
1104 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1105 "%inval2 = OpLoad %f32 %inloc2\n"
1106 "%rem = OpExtInst %f32 %std450 NMax %inval1 %inval2\n"
1107 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1108 " OpStore %outloc %rem\n"
1112 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1113 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1114 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1115 spec.numWorkGroups = IVec3(numElements, 1, 1);
1116 spec.verifyIO = &compareNMax;
1118 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1120 return group.release();
1123 bool compareNClamp (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1125 if (outputAllocs.size() != 1)
1128 const BufferSp& expectedOutput = expectedOutputs[0];
1129 const float* const expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
1130 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
1132 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float) / 2; ++idx)
1134 const float e0 = expectedOutputAsFloat[idx * 2];
1135 const float e1 = expectedOutputAsFloat[idx * 2 + 1];
1136 const float res = outputAsFloat[idx];
1138 // For NClamp, we have two possible outcomes based on
1139 // whether NaNs are handled or not.
1140 // If either min or max value is NaN, the result is undefined,
1141 // so this test doesn't stress those. If the clamped value is
1142 // NaN, and NaNs are handled, the result is min; if NaNs are not
1143 // handled, they are big values that result in max.
1144 // If all three parameters are NaN, the result should be NaN.
1145 if (!((tcu::Float32(e0).isNaN() && tcu::Float32(res).isNaN()) ||
1146 (deFloatAbs(e0 - res) < 0.00001f) ||
1147 (deFloatAbs(e1 - res) < 0.00001f)))
1154 tcu::TestCaseGroup* createOpNClampGroup (tcu::TestContext& testCtx)
1156 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnclamp", "Test the OpNClamp instruction"));
1157 ComputeShaderSpec spec;
1158 de::Random rnd (deStringHash(group->getName()));
1159 const int numElements = 200;
1160 vector<float> inputFloats1 (numElements, 0);
1161 vector<float> inputFloats2 (numElements, 0);
1162 vector<float> inputFloats3 (numElements, 0);
1163 vector<float> outputFloats (numElements * 2, 0);
1165 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1166 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1167 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats3[0], numElements);
1169 for (size_t ndx = 0; ndx < numElements; ++ndx)
1171 // Results are only defined if max value is bigger than min value.
1172 if (inputFloats2[ndx] > inputFloats3[ndx])
1174 float t = inputFloats2[ndx];
1175 inputFloats2[ndx] = inputFloats3[ndx];
1176 inputFloats3[ndx] = t;
1179 // By default, do the clamp, setting both possible answers
1180 float defaultRes = std::min(std::max(inputFloats1[ndx], inputFloats2[ndx]), inputFloats3[ndx]);
1182 float maxResA = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1183 float maxResB = maxResA;
1185 // Alternate between the NaN cases
1188 inputFloats1[ndx] = TCU_NAN;
1189 // If NaN is handled, the result should be same as the clamp minimum.
1190 // If NaN is not handled, the result should clamp to the clamp maximum.
1191 maxResA = inputFloats2[ndx];
1192 maxResB = inputFloats3[ndx];
1196 // Not a NaN case - only one legal result.
1197 maxResA = defaultRes;
1198 maxResB = defaultRes;
1201 outputFloats[ndx * 2] = maxResA;
1202 outputFloats[ndx * 2 + 1] = maxResB;
1205 // Make the first case a full-NAN case.
1206 inputFloats1[0] = TCU_NAN;
1207 inputFloats2[0] = TCU_NAN;
1208 inputFloats3[0] = TCU_NAN;
1209 outputFloats[0] = TCU_NAN;
1210 outputFloats[1] = TCU_NAN;
1213 "OpCapability Shader\n"
1214 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1215 "OpMemoryModel Logical GLSL450\n"
1216 "OpEntryPoint GLCompute %main \"main\" %id\n"
1217 "OpExecutionMode %main LocalSize 1 1 1\n"
1219 "OpName %main \"main\"\n"
1220 "OpName %id \"gl_GlobalInvocationID\"\n"
1222 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1224 "OpDecorate %buf BufferBlock\n"
1225 "OpDecorate %indata1 DescriptorSet 0\n"
1226 "OpDecorate %indata1 Binding 0\n"
1227 "OpDecorate %indata2 DescriptorSet 0\n"
1228 "OpDecorate %indata2 Binding 1\n"
1229 "OpDecorate %indata3 DescriptorSet 0\n"
1230 "OpDecorate %indata3 Binding 2\n"
1231 "OpDecorate %outdata DescriptorSet 0\n"
1232 "OpDecorate %outdata Binding 3\n"
1233 "OpDecorate %f32arr ArrayStride 4\n"
1234 "OpMemberDecorate %buf 0 Offset 0\n"
1236 + string(getComputeAsmCommonTypes()) +
1238 "%buf = OpTypeStruct %f32arr\n"
1239 "%bufptr = OpTypePointer Uniform %buf\n"
1240 "%indata1 = OpVariable %bufptr Uniform\n"
1241 "%indata2 = OpVariable %bufptr Uniform\n"
1242 "%indata3 = OpVariable %bufptr Uniform\n"
1243 "%outdata = OpVariable %bufptr Uniform\n"
1245 "%id = OpVariable %uvec3ptr Input\n"
1246 "%zero = OpConstant %i32 0\n"
1248 "%main = OpFunction %void None %voidf\n"
1249 "%label = OpLabel\n"
1250 "%idval = OpLoad %uvec3 %id\n"
1251 "%x = OpCompositeExtract %u32 %idval 0\n"
1252 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1253 "%inval1 = OpLoad %f32 %inloc1\n"
1254 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1255 "%inval2 = OpLoad %f32 %inloc2\n"
1256 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1257 "%inval3 = OpLoad %f32 %inloc3\n"
1258 "%rem = OpExtInst %f32 %std450 NClamp %inval1 %inval2 %inval3\n"
1259 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1260 " OpStore %outloc %rem\n"
1264 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1265 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1266 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1267 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1268 spec.numWorkGroups = IVec3(numElements, 1, 1);
1269 spec.verifyIO = &compareNClamp;
1271 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1273 return group.release();
1276 tcu::TestCaseGroup* createOpSRemComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1278 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem", "Test the OpSRem instruction"));
1279 de::Random rnd (deStringHash(group->getName()));
1280 const int numElements = 200;
1282 const struct CaseParams
1285 const char* failMessage; // customized status message
1286 qpTestResult failResult; // override status on failure
1287 int op1Min, op1Max; // operand ranges
1291 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1292 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1294 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1296 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1298 const CaseParams& params = cases[caseNdx];
1299 ComputeShaderSpec spec;
1300 vector<deInt32> inputInts1 (numElements, 0);
1301 vector<deInt32> inputInts2 (numElements, 0);
1302 vector<deInt32> outputInts (numElements, 0);
1304 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1305 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1307 for (int ndx = 0; ndx < numElements; ++ndx)
1309 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1310 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1314 string(getComputeAsmShaderPreamble()) +
1316 "OpName %main \"main\"\n"
1317 "OpName %id \"gl_GlobalInvocationID\"\n"
1319 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1321 "OpDecorate %buf BufferBlock\n"
1322 "OpDecorate %indata1 DescriptorSet 0\n"
1323 "OpDecorate %indata1 Binding 0\n"
1324 "OpDecorate %indata2 DescriptorSet 0\n"
1325 "OpDecorate %indata2 Binding 1\n"
1326 "OpDecorate %outdata DescriptorSet 0\n"
1327 "OpDecorate %outdata Binding 2\n"
1328 "OpDecorate %i32arr ArrayStride 4\n"
1329 "OpMemberDecorate %buf 0 Offset 0\n"
1331 + string(getComputeAsmCommonTypes()) +
1333 "%buf = OpTypeStruct %i32arr\n"
1334 "%bufptr = OpTypePointer Uniform %buf\n"
1335 "%indata1 = OpVariable %bufptr Uniform\n"
1336 "%indata2 = OpVariable %bufptr Uniform\n"
1337 "%outdata = OpVariable %bufptr Uniform\n"
1339 "%id = OpVariable %uvec3ptr Input\n"
1340 "%zero = OpConstant %i32 0\n"
1342 "%main = OpFunction %void None %voidf\n"
1343 "%label = OpLabel\n"
1344 "%idval = OpLoad %uvec3 %id\n"
1345 "%x = OpCompositeExtract %u32 %idval 0\n"
1346 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1347 "%inval1 = OpLoad %i32 %inloc1\n"
1348 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1349 "%inval2 = OpLoad %i32 %inloc2\n"
1350 "%rem = OpSRem %i32 %inval1 %inval2\n"
1351 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1352 " OpStore %outloc %rem\n"
1356 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1357 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1358 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1359 spec.numWorkGroups = IVec3(numElements, 1, 1);
1360 spec.failResult = params.failResult;
1361 spec.failMessage = params.failMessage;
1363 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1366 return group.release();
1369 tcu::TestCaseGroup* createOpSRemComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1371 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem64", "Test the 64-bit OpSRem instruction"));
1372 de::Random rnd (deStringHash(group->getName()));
1373 const int numElements = 200;
1375 const struct CaseParams
1378 const char* failMessage; // customized status message
1379 qpTestResult failResult; // override status on failure
1383 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1384 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1386 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1388 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1390 const CaseParams& params = cases[caseNdx];
1391 ComputeShaderSpec spec;
1392 vector<deInt64> inputInts1 (numElements, 0);
1393 vector<deInt64> inputInts2 (numElements, 0);
1394 vector<deInt64> outputInts (numElements, 0);
1396 if (params.positive)
1398 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1399 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1403 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1404 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1407 for (int ndx = 0; ndx < numElements; ++ndx)
1409 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1410 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1414 "OpCapability Int64\n"
1416 + string(getComputeAsmShaderPreamble()) +
1418 "OpName %main \"main\"\n"
1419 "OpName %id \"gl_GlobalInvocationID\"\n"
1421 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1423 "OpDecorate %buf BufferBlock\n"
1424 "OpDecorate %indata1 DescriptorSet 0\n"
1425 "OpDecorate %indata1 Binding 0\n"
1426 "OpDecorate %indata2 DescriptorSet 0\n"
1427 "OpDecorate %indata2 Binding 1\n"
1428 "OpDecorate %outdata DescriptorSet 0\n"
1429 "OpDecorate %outdata Binding 2\n"
1430 "OpDecorate %i64arr ArrayStride 8\n"
1431 "OpMemberDecorate %buf 0 Offset 0\n"
1433 + string(getComputeAsmCommonTypes())
1434 + string(getComputeAsmCommonInt64Types()) +
1436 "%buf = OpTypeStruct %i64arr\n"
1437 "%bufptr = OpTypePointer Uniform %buf\n"
1438 "%indata1 = OpVariable %bufptr Uniform\n"
1439 "%indata2 = OpVariable %bufptr Uniform\n"
1440 "%outdata = OpVariable %bufptr Uniform\n"
1442 "%id = OpVariable %uvec3ptr Input\n"
1443 "%zero = OpConstant %i64 0\n"
1445 "%main = OpFunction %void None %voidf\n"
1446 "%label = OpLabel\n"
1447 "%idval = OpLoad %uvec3 %id\n"
1448 "%x = OpCompositeExtract %u32 %idval 0\n"
1449 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1450 "%inval1 = OpLoad %i64 %inloc1\n"
1451 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1452 "%inval2 = OpLoad %i64 %inloc2\n"
1453 "%rem = OpSRem %i64 %inval1 %inval2\n"
1454 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1455 " OpStore %outloc %rem\n"
1459 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1460 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1461 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1462 spec.numWorkGroups = IVec3(numElements, 1, 1);
1463 spec.failResult = params.failResult;
1464 spec.failMessage = params.failMessage;
1466 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1469 return group.release();
1472 tcu::TestCaseGroup* createOpSModComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1474 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod", "Test the OpSMod instruction"));
1475 de::Random rnd (deStringHash(group->getName()));
1476 const int numElements = 200;
1478 const struct CaseParams
1481 const char* failMessage; // customized status message
1482 qpTestResult failResult; // override status on failure
1483 int op1Min, op1Max; // operand ranges
1487 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1488 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1490 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1492 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1494 const CaseParams& params = cases[caseNdx];
1496 ComputeShaderSpec spec;
1497 vector<deInt32> inputInts1 (numElements, 0);
1498 vector<deInt32> inputInts2 (numElements, 0);
1499 vector<deInt32> outputInts (numElements, 0);
1501 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1502 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1504 for (int ndx = 0; ndx < numElements; ++ndx)
1506 deInt32 rem = inputInts1[ndx] % inputInts2[ndx];
1509 outputInts[ndx] = 0;
1511 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1513 // They have the same sign
1514 outputInts[ndx] = rem;
1518 // They have opposite sign. The remainder operation takes the
1519 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1520 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1521 // the result has the correct sign and that it is still
1522 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1524 // See also http://mathforum.org/library/drmath/view/52343.html
1525 outputInts[ndx] = rem + inputInts2[ndx];
1530 string(getComputeAsmShaderPreamble()) +
1532 "OpName %main \"main\"\n"
1533 "OpName %id \"gl_GlobalInvocationID\"\n"
1535 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1537 "OpDecorate %buf BufferBlock\n"
1538 "OpDecorate %indata1 DescriptorSet 0\n"
1539 "OpDecorate %indata1 Binding 0\n"
1540 "OpDecorate %indata2 DescriptorSet 0\n"
1541 "OpDecorate %indata2 Binding 1\n"
1542 "OpDecorate %outdata DescriptorSet 0\n"
1543 "OpDecorate %outdata Binding 2\n"
1544 "OpDecorate %i32arr ArrayStride 4\n"
1545 "OpMemberDecorate %buf 0 Offset 0\n"
1547 + string(getComputeAsmCommonTypes()) +
1549 "%buf = OpTypeStruct %i32arr\n"
1550 "%bufptr = OpTypePointer Uniform %buf\n"
1551 "%indata1 = OpVariable %bufptr Uniform\n"
1552 "%indata2 = OpVariable %bufptr Uniform\n"
1553 "%outdata = OpVariable %bufptr Uniform\n"
1555 "%id = OpVariable %uvec3ptr Input\n"
1556 "%zero = OpConstant %i32 0\n"
1558 "%main = OpFunction %void None %voidf\n"
1559 "%label = OpLabel\n"
1560 "%idval = OpLoad %uvec3 %id\n"
1561 "%x = OpCompositeExtract %u32 %idval 0\n"
1562 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1563 "%inval1 = OpLoad %i32 %inloc1\n"
1564 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1565 "%inval2 = OpLoad %i32 %inloc2\n"
1566 "%rem = OpSMod %i32 %inval1 %inval2\n"
1567 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1568 " OpStore %outloc %rem\n"
1572 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1573 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1574 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1575 spec.numWorkGroups = IVec3(numElements, 1, 1);
1576 spec.failResult = params.failResult;
1577 spec.failMessage = params.failMessage;
1579 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1582 return group.release();
1585 tcu::TestCaseGroup* createOpSModComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1587 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod64", "Test the OpSMod instruction"));
1588 de::Random rnd (deStringHash(group->getName()));
1589 const int numElements = 200;
1591 const struct CaseParams
1594 const char* failMessage; // customized status message
1595 qpTestResult failResult; // override status on failure
1599 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1600 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1602 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1604 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1606 const CaseParams& params = cases[caseNdx];
1608 ComputeShaderSpec spec;
1609 vector<deInt64> inputInts1 (numElements, 0);
1610 vector<deInt64> inputInts2 (numElements, 0);
1611 vector<deInt64> outputInts (numElements, 0);
1614 if (params.positive)
1616 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1617 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1621 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1622 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1625 for (int ndx = 0; ndx < numElements; ++ndx)
1627 deInt64 rem = inputInts1[ndx] % inputInts2[ndx];
1630 outputInts[ndx] = 0;
1632 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1634 // They have the same sign
1635 outputInts[ndx] = rem;
1639 // They have opposite sign. The remainder operation takes the
1640 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1641 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1642 // the result has the correct sign and that it is still
1643 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1645 // See also http://mathforum.org/library/drmath/view/52343.html
1646 outputInts[ndx] = rem + inputInts2[ndx];
1651 "OpCapability Int64\n"
1653 + string(getComputeAsmShaderPreamble()) +
1655 "OpName %main \"main\"\n"
1656 "OpName %id \"gl_GlobalInvocationID\"\n"
1658 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1660 "OpDecorate %buf BufferBlock\n"
1661 "OpDecorate %indata1 DescriptorSet 0\n"
1662 "OpDecorate %indata1 Binding 0\n"
1663 "OpDecorate %indata2 DescriptorSet 0\n"
1664 "OpDecorate %indata2 Binding 1\n"
1665 "OpDecorate %outdata DescriptorSet 0\n"
1666 "OpDecorate %outdata Binding 2\n"
1667 "OpDecorate %i64arr ArrayStride 8\n"
1668 "OpMemberDecorate %buf 0 Offset 0\n"
1670 + string(getComputeAsmCommonTypes())
1671 + string(getComputeAsmCommonInt64Types()) +
1673 "%buf = OpTypeStruct %i64arr\n"
1674 "%bufptr = OpTypePointer Uniform %buf\n"
1675 "%indata1 = OpVariable %bufptr Uniform\n"
1676 "%indata2 = OpVariable %bufptr Uniform\n"
1677 "%outdata = OpVariable %bufptr Uniform\n"
1679 "%id = OpVariable %uvec3ptr Input\n"
1680 "%zero = OpConstant %i64 0\n"
1682 "%main = OpFunction %void None %voidf\n"
1683 "%label = OpLabel\n"
1684 "%idval = OpLoad %uvec3 %id\n"
1685 "%x = OpCompositeExtract %u32 %idval 0\n"
1686 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1687 "%inval1 = OpLoad %i64 %inloc1\n"
1688 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1689 "%inval2 = OpLoad %i64 %inloc2\n"
1690 "%rem = OpSMod %i64 %inval1 %inval2\n"
1691 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1692 " OpStore %outloc %rem\n"
1696 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1697 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1698 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1699 spec.numWorkGroups = IVec3(numElements, 1, 1);
1700 spec.failResult = params.failResult;
1701 spec.failMessage = params.failMessage;
1703 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1706 return group.release();
1709 // Copy contents in the input buffer to the output buffer.
1710 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
1712 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
1713 de::Random rnd (deStringHash(group->getName()));
1714 const int numElements = 100;
1716 // 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.
1717 ComputeShaderSpec spec1;
1718 vector<Vec4> inputFloats1 (numElements);
1719 vector<Vec4> outputFloats1 (numElements);
1721 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
1723 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1724 floorAll(inputFloats1);
1726 for (size_t ndx = 0; ndx < numElements; ++ndx)
1727 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
1730 string(getComputeAsmShaderPreamble()) +
1732 "OpName %main \"main\"\n"
1733 "OpName %id \"gl_GlobalInvocationID\"\n"
1735 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1736 "OpDecorate %vec4arr ArrayStride 16\n"
1738 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1740 "%vec4 = OpTypeVector %f32 4\n"
1741 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
1742 "%vec4ptr_f = OpTypePointer Function %vec4\n"
1743 "%vec4arr = OpTypeRuntimeArray %vec4\n"
1744 "%buf = OpTypeStruct %vec4arr\n"
1745 "%bufptr = OpTypePointer Uniform %buf\n"
1746 "%indata = OpVariable %bufptr Uniform\n"
1747 "%outdata = OpVariable %bufptr Uniform\n"
1749 "%id = OpVariable %uvec3ptr Input\n"
1750 "%zero = OpConstant %i32 0\n"
1751 "%c_f_0 = OpConstant %f32 0.\n"
1752 "%c_f_0_5 = OpConstant %f32 0.5\n"
1753 "%c_f_1_5 = OpConstant %f32 1.5\n"
1754 "%c_f_2_5 = OpConstant %f32 2.5\n"
1755 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
1757 "%main = OpFunction %void None %voidf\n"
1758 "%label = OpLabel\n"
1759 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
1760 "%idval = OpLoad %uvec3 %id\n"
1761 "%x = OpCompositeExtract %u32 %idval 0\n"
1762 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
1763 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
1764 " OpCopyMemory %v_vec4 %inloc\n"
1765 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
1766 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
1767 " OpStore %outloc %add\n"
1771 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
1772 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
1773 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1775 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
1777 // The following case copies a float[100] variable from the input buffer to the output buffer.
1778 ComputeShaderSpec spec2;
1779 vector<float> inputFloats2 (numElements);
1780 vector<float> outputFloats2 (numElements);
1782 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
1784 for (size_t ndx = 0; ndx < numElements; ++ndx)
1785 outputFloats2[ndx] = inputFloats2[ndx];
1788 string(getComputeAsmShaderPreamble()) +
1790 "OpName %main \"main\"\n"
1791 "OpName %id \"gl_GlobalInvocationID\"\n"
1793 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1794 "OpDecorate %f32arr100 ArrayStride 4\n"
1796 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1798 "%hundred = OpConstant %u32 100\n"
1799 "%f32arr100 = OpTypeArray %f32 %hundred\n"
1800 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
1801 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
1802 "%buf = OpTypeStruct %f32arr100\n"
1803 "%bufptr = OpTypePointer Uniform %buf\n"
1804 "%indata = OpVariable %bufptr Uniform\n"
1805 "%outdata = OpVariable %bufptr Uniform\n"
1807 "%id = OpVariable %uvec3ptr Input\n"
1808 "%zero = OpConstant %i32 0\n"
1810 "%main = OpFunction %void None %voidf\n"
1811 "%label = OpLabel\n"
1812 "%var = OpVariable %f32arr100ptr_f Function\n"
1813 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
1814 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
1815 " OpCopyMemory %var %inarr\n"
1816 " OpCopyMemory %outarr %var\n"
1820 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1821 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1822 spec2.numWorkGroups = IVec3(1, 1, 1);
1824 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
1826 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
1827 ComputeShaderSpec spec3;
1828 vector<float> inputFloats3 (16);
1829 vector<float> outputFloats3 (16);
1831 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
1833 for (size_t ndx = 0; ndx < 16; ++ndx)
1834 outputFloats3[ndx] = inputFloats3[ndx];
1837 string(getComputeAsmShaderPreamble()) +
1839 "OpName %main \"main\"\n"
1840 "OpName %id \"gl_GlobalInvocationID\"\n"
1842 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1843 "OpMemberDecorate %buf 0 Offset 0\n"
1844 "OpMemberDecorate %buf 1 Offset 16\n"
1845 "OpMemberDecorate %buf 2 Offset 32\n"
1846 "OpMemberDecorate %buf 3 Offset 48\n"
1848 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1850 "%vec4 = OpTypeVector %f32 4\n"
1851 "%buf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
1852 "%bufptr = OpTypePointer Uniform %buf\n"
1853 "%indata = OpVariable %bufptr Uniform\n"
1854 "%outdata = OpVariable %bufptr Uniform\n"
1855 "%vec4stptr = OpTypePointer Function %buf\n"
1857 "%id = OpVariable %uvec3ptr Input\n"
1858 "%zero = OpConstant %i32 0\n"
1860 "%main = OpFunction %void None %voidf\n"
1861 "%label = OpLabel\n"
1862 "%var = OpVariable %vec4stptr Function\n"
1863 " OpCopyMemory %var %indata\n"
1864 " OpCopyMemory %outdata %var\n"
1868 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1869 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
1870 spec3.numWorkGroups = IVec3(1, 1, 1);
1872 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
1874 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
1875 ComputeShaderSpec spec4;
1876 vector<float> inputFloats4 (numElements);
1877 vector<float> outputFloats4 (numElements);
1879 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
1881 for (size_t ndx = 0; ndx < numElements; ++ndx)
1882 outputFloats4[ndx] = -inputFloats4[ndx];
1885 string(getComputeAsmShaderPreamble()) +
1887 "OpName %main \"main\"\n"
1888 "OpName %id \"gl_GlobalInvocationID\"\n"
1890 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1892 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
1894 "%f32ptr_f = OpTypePointer Function %f32\n"
1895 "%id = OpVariable %uvec3ptr Input\n"
1896 "%zero = OpConstant %i32 0\n"
1898 "%main = OpFunction %void None %voidf\n"
1899 "%label = OpLabel\n"
1900 "%var = OpVariable %f32ptr_f Function\n"
1901 "%idval = OpLoad %uvec3 %id\n"
1902 "%x = OpCompositeExtract %u32 %idval 0\n"
1903 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1904 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1905 " OpCopyMemory %var %inloc\n"
1906 "%val = OpLoad %f32 %var\n"
1907 "%neg = OpFNegate %f32 %val\n"
1908 " OpStore %outloc %neg\n"
1912 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
1913 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
1914 spec4.numWorkGroups = IVec3(numElements, 1, 1);
1916 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
1918 return group.release();
1921 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
1923 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
1924 ComputeShaderSpec spec;
1925 de::Random rnd (deStringHash(group->getName()));
1926 const int numElements = 100;
1927 vector<float> inputFloats (numElements, 0);
1928 vector<float> outputFloats (numElements, 0);
1930 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
1932 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1933 floorAll(inputFloats);
1935 for (size_t ndx = 0; ndx < numElements; ++ndx)
1936 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
1939 string(getComputeAsmShaderPreamble()) +
1941 "OpName %main \"main\"\n"
1942 "OpName %id \"gl_GlobalInvocationID\"\n"
1944 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1946 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1948 "%fmat = OpTypeMatrix %fvec3 3\n"
1949 "%three = OpConstant %u32 3\n"
1950 "%farr = OpTypeArray %f32 %three\n"
1951 "%fst = OpTypeStruct %f32 %f32\n"
1953 + string(getComputeAsmInputOutputBuffer()) +
1955 "%id = OpVariable %uvec3ptr Input\n"
1956 "%zero = OpConstant %i32 0\n"
1957 "%c_f = OpConstant %f32 1.5\n"
1958 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
1959 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
1960 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
1961 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
1963 "%main = OpFunction %void None %voidf\n"
1964 "%label = OpLabel\n"
1965 "%c_f_copy = OpCopyObject %f32 %c_f\n"
1966 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
1967 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
1968 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
1969 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
1970 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
1971 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
1972 "%farr_elem = OpCompositeExtract %f32 %c_farr_copy 2\n"
1973 "%fst_elem = OpCompositeExtract %f32 %c_fst_copy 1\n"
1974 // Add up. 1.5 * 5 = 7.5.
1975 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
1976 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
1977 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
1978 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
1980 "%idval = OpLoad %uvec3 %id\n"
1981 "%x = OpCompositeExtract %u32 %idval 0\n"
1982 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1983 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1984 "%inval = OpLoad %f32 %inloc\n"
1985 "%add = OpFAdd %f32 %add4 %inval\n"
1986 " OpStore %outloc %add\n"
1989 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1990 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1991 spec.numWorkGroups = IVec3(numElements, 1, 1);
1993 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
1995 return group.release();
1997 // Assembly code used for testing OpUnreachable is based on GLSL source code:
2001 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2002 // float elements[];
2004 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2005 // float elements[];
2008 // void not_called_func() {
2009 // // place OpUnreachable here
2012 // uint modulo4(uint val) {
2013 // switch (val % uint(4)) {
2014 // case 0: return 3;
2015 // case 1: return 2;
2016 // case 2: return 1;
2017 // case 3: return 0;
2018 // default: return 100; // place OpUnreachable here
2024 // // place OpUnreachable here
2028 // uint x = gl_GlobalInvocationID.x;
2029 // if (const5() > modulo4(1000)) {
2030 // output_data.elements[x] = -input_data.elements[x];
2032 // // place OpUnreachable here
2033 // output_data.elements[x] = input_data.elements[x];
2037 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
2039 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
2040 ComputeShaderSpec spec;
2041 de::Random rnd (deStringHash(group->getName()));
2042 const int numElements = 100;
2043 vector<float> positiveFloats (numElements, 0);
2044 vector<float> negativeFloats (numElements, 0);
2046 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2048 for (size_t ndx = 0; ndx < numElements; ++ndx)
2049 negativeFloats[ndx] = -positiveFloats[ndx];
2052 string(getComputeAsmShaderPreamble()) +
2054 "OpSource GLSL 430\n"
2055 "OpName %main \"main\"\n"
2056 "OpName %func_not_called_func \"not_called_func(\"\n"
2057 "OpName %func_modulo4 \"modulo4(u1;\"\n"
2058 "OpName %func_const5 \"const5(\"\n"
2059 "OpName %id \"gl_GlobalInvocationID\"\n"
2061 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2063 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2065 "%u32ptr = OpTypePointer Function %u32\n"
2066 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
2067 "%unitf = OpTypeFunction %u32\n"
2069 "%id = OpVariable %uvec3ptr Input\n"
2070 "%zero = OpConstant %u32 0\n"
2071 "%one = OpConstant %u32 1\n"
2072 "%two = OpConstant %u32 2\n"
2073 "%three = OpConstant %u32 3\n"
2074 "%four = OpConstant %u32 4\n"
2075 "%five = OpConstant %u32 5\n"
2076 "%hundred = OpConstant %u32 100\n"
2077 "%thousand = OpConstant %u32 1000\n"
2079 + string(getComputeAsmInputOutputBuffer()) +
2082 "%main = OpFunction %void None %voidf\n"
2083 "%main_entry = OpLabel\n"
2084 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
2085 "%idval = OpLoad %uvec3 %id\n"
2086 "%x = OpCompositeExtract %u32 %idval 0\n"
2087 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2088 "%inval = OpLoad %f32 %inloc\n"
2089 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2090 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
2091 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
2092 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
2093 " OpSelectionMerge %if_end None\n"
2094 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2095 "%if_true = OpLabel\n"
2096 "%negate = OpFNegate %f32 %inval\n"
2097 " OpStore %outloc %negate\n"
2098 " OpBranch %if_end\n"
2099 "%if_false = OpLabel\n"
2100 " OpUnreachable\n" // Unreachable else branch for if statement
2101 "%if_end = OpLabel\n"
2105 // not_called_function()
2106 "%func_not_called_func = OpFunction %void None %voidf\n"
2107 "%not_called_func_entry = OpLabel\n"
2108 " OpUnreachable\n" // Unreachable entry block in not called static function
2112 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
2113 "%valptr = OpFunctionParameter %u32ptr\n"
2114 "%modulo4_entry = OpLabel\n"
2115 "%val = OpLoad %u32 %valptr\n"
2116 "%modulo = OpUMod %u32 %val %four\n"
2117 " OpSelectionMerge %switch_merge None\n"
2118 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
2119 "%case0 = OpLabel\n"
2120 " OpReturnValue %three\n"
2121 "%case1 = OpLabel\n"
2122 " OpReturnValue %two\n"
2123 "%case2 = OpLabel\n"
2124 " OpReturnValue %one\n"
2125 "%case3 = OpLabel\n"
2126 " OpReturnValue %zero\n"
2127 "%default = OpLabel\n"
2128 " OpUnreachable\n" // Unreachable default case for switch statement
2129 "%switch_merge = OpLabel\n"
2130 " OpUnreachable\n" // Unreachable merge block for switch statement
2134 "%func_const5 = OpFunction %u32 None %unitf\n"
2135 "%const5_entry = OpLabel\n"
2136 " OpReturnValue %five\n"
2137 "%unreachable = OpLabel\n"
2138 " OpUnreachable\n" // Unreachable block in function
2140 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2141 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2142 spec.numWorkGroups = IVec3(numElements, 1, 1);
2144 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
2146 return group.release();
2149 // Assembly code used for testing decoration group is based on GLSL source code:
2153 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
2154 // float elements[];
2156 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
2157 // float elements[];
2159 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
2160 // float elements[];
2162 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
2163 // float elements[];
2165 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
2166 // float elements[];
2168 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
2169 // float elements[];
2173 // uint x = gl_GlobalInvocationID.x;
2174 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
2176 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
2178 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
2179 ComputeShaderSpec spec;
2180 de::Random rnd (deStringHash(group->getName()));
2181 const int numElements = 100;
2182 vector<float> inputFloats0 (numElements, 0);
2183 vector<float> inputFloats1 (numElements, 0);
2184 vector<float> inputFloats2 (numElements, 0);
2185 vector<float> inputFloats3 (numElements, 0);
2186 vector<float> inputFloats4 (numElements, 0);
2187 vector<float> outputFloats (numElements, 0);
2189 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
2190 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
2191 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
2192 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
2193 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
2195 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2196 floorAll(inputFloats0);
2197 floorAll(inputFloats1);
2198 floorAll(inputFloats2);
2199 floorAll(inputFloats3);
2200 floorAll(inputFloats4);
2202 for (size_t ndx = 0; ndx < numElements; ++ndx)
2203 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
2206 string(getComputeAsmShaderPreamble()) +
2208 "OpSource GLSL 430\n"
2209 "OpName %main \"main\"\n"
2210 "OpName %id \"gl_GlobalInvocationID\"\n"
2212 // Not using group decoration on variable.
2213 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2214 // Not using group decoration on type.
2215 "OpDecorate %f32arr ArrayStride 4\n"
2217 "OpDecorate %groups BufferBlock\n"
2218 "OpDecorate %groupm Offset 0\n"
2219 "%groups = OpDecorationGroup\n"
2220 "%groupm = OpDecorationGroup\n"
2222 // Group decoration on multiple structs.
2223 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
2224 // Group decoration on multiple struct members.
2225 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
2227 "OpDecorate %group1 DescriptorSet 0\n"
2228 "OpDecorate %group3 DescriptorSet 0\n"
2229 "OpDecorate %group3 NonWritable\n"
2230 "OpDecorate %group3 Restrict\n"
2231 "%group0 = OpDecorationGroup\n"
2232 "%group1 = OpDecorationGroup\n"
2233 "%group3 = OpDecorationGroup\n"
2235 // Applying the same decoration group multiple times.
2236 "OpGroupDecorate %group1 %outdata\n"
2237 "OpGroupDecorate %group1 %outdata\n"
2238 "OpGroupDecorate %group1 %outdata\n"
2239 "OpDecorate %outdata DescriptorSet 0\n"
2240 "OpDecorate %outdata Binding 5\n"
2241 // Applying decoration group containing nothing.
2242 "OpGroupDecorate %group0 %indata0\n"
2243 "OpDecorate %indata0 DescriptorSet 0\n"
2244 "OpDecorate %indata0 Binding 0\n"
2245 // Applying decoration group containing one decoration.
2246 "OpGroupDecorate %group1 %indata1\n"
2247 "OpDecorate %indata1 Binding 1\n"
2248 // Applying decoration group containing multiple decorations.
2249 "OpGroupDecorate %group3 %indata2 %indata3\n"
2250 "OpDecorate %indata2 Binding 2\n"
2251 "OpDecorate %indata3 Binding 3\n"
2252 // Applying multiple decoration groups (with overlapping).
2253 "OpGroupDecorate %group0 %indata4\n"
2254 "OpGroupDecorate %group1 %indata4\n"
2255 "OpGroupDecorate %group3 %indata4\n"
2256 "OpDecorate %indata4 Binding 4\n"
2258 + string(getComputeAsmCommonTypes()) +
2260 "%id = OpVariable %uvec3ptr Input\n"
2261 "%zero = OpConstant %i32 0\n"
2263 "%outbuf = OpTypeStruct %f32arr\n"
2264 "%outbufptr = OpTypePointer Uniform %outbuf\n"
2265 "%outdata = OpVariable %outbufptr Uniform\n"
2266 "%inbuf0 = OpTypeStruct %f32arr\n"
2267 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
2268 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
2269 "%inbuf1 = OpTypeStruct %f32arr\n"
2270 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
2271 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
2272 "%inbuf2 = OpTypeStruct %f32arr\n"
2273 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
2274 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
2275 "%inbuf3 = OpTypeStruct %f32arr\n"
2276 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
2277 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
2278 "%inbuf4 = OpTypeStruct %f32arr\n"
2279 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
2280 "%indata4 = OpVariable %inbufptr Uniform\n"
2282 "%main = OpFunction %void None %voidf\n"
2283 "%label = OpLabel\n"
2284 "%idval = OpLoad %uvec3 %id\n"
2285 "%x = OpCompositeExtract %u32 %idval 0\n"
2286 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
2287 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
2288 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
2289 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
2290 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
2291 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2292 "%inval0 = OpLoad %f32 %inloc0\n"
2293 "%inval1 = OpLoad %f32 %inloc1\n"
2294 "%inval2 = OpLoad %f32 %inloc2\n"
2295 "%inval3 = OpLoad %f32 %inloc3\n"
2296 "%inval4 = OpLoad %f32 %inloc4\n"
2297 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
2298 "%add1 = OpFAdd %f32 %add0 %inval2\n"
2299 "%add2 = OpFAdd %f32 %add1 %inval3\n"
2300 "%add = OpFAdd %f32 %add2 %inval4\n"
2301 " OpStore %outloc %add\n"
2304 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
2305 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
2306 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2307 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2308 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2309 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2310 spec.numWorkGroups = IVec3(numElements, 1, 1);
2312 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
2314 return group.release();
2317 struct SpecConstantTwoIntCase
2319 const char* caseName;
2320 const char* scDefinition0;
2321 const char* scDefinition1;
2322 const char* scResultType;
2323 const char* scOperation;
2324 deInt32 scActualValue0;
2325 deInt32 scActualValue1;
2326 const char* resultOperation;
2327 vector<deInt32> expectedOutput;
2329 SpecConstantTwoIntCase (const char* name,
2330 const char* definition0,
2331 const char* definition1,
2332 const char* resultType,
2333 const char* operation,
2336 const char* resultOp,
2337 const vector<deInt32>& output)
2339 , scDefinition0 (definition0)
2340 , scDefinition1 (definition1)
2341 , scResultType (resultType)
2342 , scOperation (operation)
2343 , scActualValue0 (value0)
2344 , scActualValue1 (value1)
2345 , resultOperation (resultOp)
2346 , expectedOutput (output) {}
2349 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
2351 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
2352 vector<SpecConstantTwoIntCase> cases;
2353 de::Random rnd (deStringHash(group->getName()));
2354 const int numElements = 100;
2355 vector<deInt32> inputInts (numElements, 0);
2356 vector<deInt32> outputInts1 (numElements, 0);
2357 vector<deInt32> outputInts2 (numElements, 0);
2358 vector<deInt32> outputInts3 (numElements, 0);
2359 vector<deInt32> outputInts4 (numElements, 0);
2360 const StringTemplate shaderTemplate (
2361 string(getComputeAsmShaderPreamble()) +
2363 "OpName %main \"main\"\n"
2364 "OpName %id \"gl_GlobalInvocationID\"\n"
2366 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2367 "OpDecorate %sc_0 SpecId 0\n"
2368 "OpDecorate %sc_1 SpecId 1\n"
2369 "OpDecorate %i32arr ArrayStride 4\n"
2371 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2373 "%buf = OpTypeStruct %i32arr\n"
2374 "%bufptr = OpTypePointer Uniform %buf\n"
2375 "%indata = OpVariable %bufptr Uniform\n"
2376 "%outdata = OpVariable %bufptr Uniform\n"
2378 "%id = OpVariable %uvec3ptr Input\n"
2379 "%zero = OpConstant %i32 0\n"
2381 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
2382 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
2383 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
2385 "%main = OpFunction %void None %voidf\n"
2386 "%label = OpLabel\n"
2387 "%idval = OpLoad %uvec3 %id\n"
2388 "%x = OpCompositeExtract %u32 %idval 0\n"
2389 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2390 "%inval = OpLoad %i32 %inloc\n"
2391 "%final = ${GEN_RESULT}\n"
2392 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2393 " OpStore %outloc %final\n"
2395 " OpFunctionEnd\n");
2397 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
2399 for (size_t ndx = 0; ndx < numElements; ++ndx)
2401 outputInts1[ndx] = inputInts[ndx] + 42;
2402 outputInts2[ndx] = inputInts[ndx];
2403 outputInts3[ndx] = inputInts[ndx] - 11200;
2404 outputInts4[ndx] = inputInts[ndx] + 1;
2407 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
2408 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
2409 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
2411 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
2412 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
2413 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
2414 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
2415 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
2416 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2417 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2418 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
2419 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
2420 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
2421 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
2422 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2423 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2424 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
2425 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
2426 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
2427 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2428 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
2429 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
2430 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
2431 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2432 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
2433 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
2434 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2435 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2436 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2437 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2438 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
2439 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
2440 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
2441 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
2442 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
2444 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2446 map<string, string> specializations;
2447 ComputeShaderSpec spec;
2449 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
2450 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
2451 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
2452 specializations["SC_OP"] = cases[caseNdx].scOperation;
2453 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
2455 spec.assembly = shaderTemplate.specialize(specializations);
2456 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2457 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
2458 spec.numWorkGroups = IVec3(numElements, 1, 1);
2459 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
2460 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
2462 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec));
2465 ComputeShaderSpec spec;
2468 string(getComputeAsmShaderPreamble()) +
2470 "OpName %main \"main\"\n"
2471 "OpName %id \"gl_GlobalInvocationID\"\n"
2473 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2474 "OpDecorate %sc_0 SpecId 0\n"
2475 "OpDecorate %sc_1 SpecId 1\n"
2476 "OpDecorate %sc_2 SpecId 2\n"
2477 "OpDecorate %i32arr ArrayStride 4\n"
2479 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2481 "%ivec3 = OpTypeVector %i32 3\n"
2482 "%buf = OpTypeStruct %i32arr\n"
2483 "%bufptr = OpTypePointer Uniform %buf\n"
2484 "%indata = OpVariable %bufptr Uniform\n"
2485 "%outdata = OpVariable %bufptr Uniform\n"
2487 "%id = OpVariable %uvec3ptr Input\n"
2488 "%zero = OpConstant %i32 0\n"
2489 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
2490 "%vec3_undef = OpUndef %ivec3\n"
2492 "%sc_0 = OpSpecConstant %i32 0\n"
2493 "%sc_1 = OpSpecConstant %i32 0\n"
2494 "%sc_2 = OpSpecConstant %i32 0\n"
2495 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
2496 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
2497 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
2498 "%sc_vec3_0_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
2499 "%sc_vec3_1_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
2500 "%sc_vec3_2_s = OpSpecConstantOp %ivec3 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
2501 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
2502 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
2503 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
2504 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
2505 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
2506 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
2507 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
2509 "%main = OpFunction %void None %voidf\n"
2510 "%label = OpLabel\n"
2511 "%idval = OpLoad %uvec3 %id\n"
2512 "%x = OpCompositeExtract %u32 %idval 0\n"
2513 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2514 "%inval = OpLoad %i32 %inloc\n"
2515 "%final = OpIAdd %i32 %inval %sc_final\n"
2516 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2517 " OpStore %outloc %final\n"
2520 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2521 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
2522 spec.numWorkGroups = IVec3(numElements, 1, 1);
2523 spec.specConstants.push_back(123);
2524 spec.specConstants.push_back(56);
2525 spec.specConstants.push_back(-77);
2527 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
2529 return group.release();
2532 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
2534 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
2535 ComputeShaderSpec spec1;
2536 ComputeShaderSpec spec2;
2537 ComputeShaderSpec spec3;
2538 de::Random rnd (deStringHash(group->getName()));
2539 const int numElements = 100;
2540 vector<float> inputFloats (numElements, 0);
2541 vector<float> outputFloats1 (numElements, 0);
2542 vector<float> outputFloats2 (numElements, 0);
2543 vector<float> outputFloats3 (numElements, 0);
2545 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
2547 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2548 floorAll(inputFloats);
2550 for (size_t ndx = 0; ndx < numElements; ++ndx)
2554 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
2555 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
2556 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
2559 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
2560 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
2564 string(getComputeAsmShaderPreamble()) +
2566 "OpSource GLSL 430\n"
2567 "OpName %main \"main\"\n"
2568 "OpName %id \"gl_GlobalInvocationID\"\n"
2570 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2572 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2574 "%id = OpVariable %uvec3ptr Input\n"
2575 "%zero = OpConstant %i32 0\n"
2576 "%three = OpConstant %u32 3\n"
2577 "%constf5p5 = OpConstant %f32 5.5\n"
2578 "%constf20p5 = OpConstant %f32 20.5\n"
2579 "%constf1p75 = OpConstant %f32 1.75\n"
2580 "%constf8p5 = OpConstant %f32 8.5\n"
2581 "%constf6p5 = OpConstant %f32 6.5\n"
2583 "%main = OpFunction %void None %voidf\n"
2584 "%entry = OpLabel\n"
2585 "%idval = OpLoad %uvec3 %id\n"
2586 "%x = OpCompositeExtract %u32 %idval 0\n"
2587 "%selector = OpUMod %u32 %x %three\n"
2588 " OpSelectionMerge %phi None\n"
2589 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
2591 // Case 1 before OpPhi.
2592 "%case1 = OpLabel\n"
2595 "%default = OpLabel\n"
2599 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
2600 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2601 "%inval = OpLoad %f32 %inloc\n"
2602 "%add = OpFAdd %f32 %inval %operand\n"
2603 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2604 " OpStore %outloc %add\n"
2607 // Case 0 after OpPhi.
2608 "%case0 = OpLabel\n"
2612 // Case 2 after OpPhi.
2613 "%case2 = OpLabel\n"
2617 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2618 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
2619 spec1.numWorkGroups = IVec3(numElements, 1, 1);
2621 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
2624 string(getComputeAsmShaderPreamble()) +
2626 "OpName %main \"main\"\n"
2627 "OpName %id \"gl_GlobalInvocationID\"\n"
2629 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2631 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2633 "%id = OpVariable %uvec3ptr Input\n"
2634 "%zero = OpConstant %i32 0\n"
2635 "%one = OpConstant %i32 1\n"
2636 "%three = OpConstant %i32 3\n"
2637 "%constf6p5 = OpConstant %f32 6.5\n"
2639 "%main = OpFunction %void None %voidf\n"
2640 "%entry = OpLabel\n"
2641 "%idval = OpLoad %uvec3 %id\n"
2642 "%x = OpCompositeExtract %u32 %idval 0\n"
2643 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2644 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2645 "%inval = OpLoad %f32 %inloc\n"
2649 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
2650 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
2651 "%step_next = OpIAdd %i32 %step %one\n"
2652 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
2653 "%still_loop = OpSLessThan %bool %step %three\n"
2654 " OpLoopMerge %exit %phi None\n"
2655 " OpBranchConditional %still_loop %phi %exit\n"
2658 " OpStore %outloc %accum\n"
2661 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2662 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2663 spec2.numWorkGroups = IVec3(numElements, 1, 1);
2665 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
2668 string(getComputeAsmShaderPreamble()) +
2670 "OpName %main \"main\"\n"
2671 "OpName %id \"gl_GlobalInvocationID\"\n"
2673 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2675 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2677 "%f32ptr_f = OpTypePointer Function %f32\n"
2678 "%id = OpVariable %uvec3ptr Input\n"
2679 "%true = OpConstantTrue %bool\n"
2680 "%false = OpConstantFalse %bool\n"
2681 "%zero = OpConstant %i32 0\n"
2682 "%constf8p5 = OpConstant %f32 8.5\n"
2684 "%main = OpFunction %void None %voidf\n"
2685 "%entry = OpLabel\n"
2686 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
2687 "%idval = OpLoad %uvec3 %id\n"
2688 "%x = OpCompositeExtract %u32 %idval 0\n"
2689 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2690 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2691 "%a_init = OpLoad %f32 %inloc\n"
2692 "%b_init = OpLoad %f32 %b\n"
2696 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
2697 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
2698 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
2699 " OpLoopMerge %exit %phi None\n"
2700 " OpBranchConditional %still_loop %phi %exit\n"
2703 "%sub = OpFSub %f32 %a_next %b_next\n"
2704 " OpStore %outloc %sub\n"
2707 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2708 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
2709 spec3.numWorkGroups = IVec3(numElements, 1, 1);
2711 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
2713 return group.release();
2716 // Assembly code used for testing block order is based on GLSL source code:
2720 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2721 // float elements[];
2723 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2724 // float elements[];
2728 // uint x = gl_GlobalInvocationID.x;
2729 // output_data.elements[x] = input_data.elements[x];
2730 // if (x > uint(50)) {
2731 // switch (x % uint(3)) {
2732 // case 0: output_data.elements[x] += 1.5f; break;
2733 // case 1: output_data.elements[x] += 42.f; break;
2734 // case 2: output_data.elements[x] -= 27.f; break;
2738 // output_data.elements[x] = -input_data.elements[x];
2741 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
2743 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
2744 ComputeShaderSpec spec;
2745 de::Random rnd (deStringHash(group->getName()));
2746 const int numElements = 100;
2747 vector<float> inputFloats (numElements, 0);
2748 vector<float> outputFloats (numElements, 0);
2750 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2752 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2753 floorAll(inputFloats);
2755 for (size_t ndx = 0; ndx <= 50; ++ndx)
2756 outputFloats[ndx] = -inputFloats[ndx];
2758 for (size_t ndx = 51; ndx < numElements; ++ndx)
2762 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
2763 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
2764 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
2770 string(getComputeAsmShaderPreamble()) +
2772 "OpSource GLSL 430\n"
2773 "OpName %main \"main\"\n"
2774 "OpName %id \"gl_GlobalInvocationID\"\n"
2776 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2778 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2780 "%u32ptr = OpTypePointer Function %u32\n"
2781 "%u32ptr_input = OpTypePointer Input %u32\n"
2783 + string(getComputeAsmInputOutputBuffer()) +
2785 "%id = OpVariable %uvec3ptr Input\n"
2786 "%zero = OpConstant %i32 0\n"
2787 "%const3 = OpConstant %u32 3\n"
2788 "%const50 = OpConstant %u32 50\n"
2789 "%constf1p5 = OpConstant %f32 1.5\n"
2790 "%constf27 = OpConstant %f32 27.0\n"
2791 "%constf42 = OpConstant %f32 42.0\n"
2793 "%main = OpFunction %void None %voidf\n"
2796 "%entry = OpLabel\n"
2798 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
2799 "%xvar = OpVariable %u32ptr Function\n"
2800 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
2801 "%x = OpLoad %u32 %xptr\n"
2802 " OpStore %xvar %x\n"
2804 "%cmp = OpUGreaterThan %bool %x %const50\n"
2805 " OpSelectionMerge %if_merge None\n"
2806 " OpBranchConditional %cmp %if_true %if_false\n"
2808 // False branch for if-statement: placed in the middle of switch cases and before true branch.
2809 "%if_false = OpLabel\n"
2810 "%x_f = OpLoad %u32 %xvar\n"
2811 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
2812 "%inval_f = OpLoad %f32 %inloc_f\n"
2813 "%negate = OpFNegate %f32 %inval_f\n"
2814 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
2815 " OpStore %outloc_f %negate\n"
2816 " OpBranch %if_merge\n"
2818 // Merge block for if-statement: placed in the middle of true and false branch.
2819 "%if_merge = OpLabel\n"
2822 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
2823 "%if_true = OpLabel\n"
2824 "%xval_t = OpLoad %u32 %xvar\n"
2825 "%mod = OpUMod %u32 %xval_t %const3\n"
2826 " OpSelectionMerge %switch_merge None\n"
2827 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
2829 // Merge block for switch-statement: placed before the case
2830 // bodies. But it must follow OpSwitch which dominates it.
2831 "%switch_merge = OpLabel\n"
2832 " OpBranch %if_merge\n"
2834 // Case 1 for switch-statement: placed before case 0.
2835 // It must follow the OpSwitch that dominates it.
2836 "%case1 = OpLabel\n"
2837 "%x_1 = OpLoad %u32 %xvar\n"
2838 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
2839 "%inval_1 = OpLoad %f32 %inloc_1\n"
2840 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
2841 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
2842 " OpStore %outloc_1 %addf42\n"
2843 " OpBranch %switch_merge\n"
2845 // Case 2 for switch-statement.
2846 "%case2 = OpLabel\n"
2847 "%x_2 = OpLoad %u32 %xvar\n"
2848 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
2849 "%inval_2 = OpLoad %f32 %inloc_2\n"
2850 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
2851 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
2852 " OpStore %outloc_2 %subf27\n"
2853 " OpBranch %switch_merge\n"
2855 // Default case for switch-statement: placed in the middle of normal cases.
2856 "%default = OpLabel\n"
2857 " OpBranch %switch_merge\n"
2859 // Case 0 for switch-statement: out of order.
2860 "%case0 = OpLabel\n"
2861 "%x_0 = OpLoad %u32 %xvar\n"
2862 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
2863 "%inval_0 = OpLoad %f32 %inloc_0\n"
2864 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
2865 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
2866 " OpStore %outloc_0 %addf1p5\n"
2867 " OpBranch %switch_merge\n"
2870 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2871 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2872 spec.numWorkGroups = IVec3(numElements, 1, 1);
2874 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
2876 return group.release();
2879 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
2881 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
2882 ComputeShaderSpec spec1;
2883 ComputeShaderSpec spec2;
2884 de::Random rnd (deStringHash(group->getName()));
2885 const int numElements = 100;
2886 vector<float> inputFloats (numElements, 0);
2887 vector<float> outputFloats1 (numElements, 0);
2888 vector<float> outputFloats2 (numElements, 0);
2889 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
2891 for (size_t ndx = 0; ndx < numElements; ++ndx)
2893 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
2894 outputFloats2[ndx] = -inputFloats[ndx];
2897 const string assembly(
2898 "OpCapability Shader\n"
2899 "OpCapability ClipDistance\n"
2900 "OpMemoryModel Logical GLSL450\n"
2901 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
2902 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
2903 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
2904 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
2905 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
2906 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
2908 "OpName %comp_main1 \"entrypoint1\"\n"
2909 "OpName %comp_main2 \"entrypoint2\"\n"
2910 "OpName %vert_main \"entrypoint2\"\n"
2911 "OpName %id \"gl_GlobalInvocationID\"\n"
2912 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
2913 "OpName %vertexIndex \"gl_VertexIndex\"\n"
2914 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
2915 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
2916 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
2917 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
2919 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2920 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
2921 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
2922 "OpDecorate %vert_builtin_st Block\n"
2923 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
2924 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
2925 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
2927 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2929 "%zero = OpConstant %i32 0\n"
2930 "%one = OpConstant %u32 1\n"
2931 "%c_f32_1 = OpConstant %f32 1\n"
2933 "%i32inputptr = OpTypePointer Input %i32\n"
2934 "%vec4 = OpTypeVector %f32 4\n"
2935 "%vec4ptr = OpTypePointer Output %vec4\n"
2936 "%f32arr1 = OpTypeArray %f32 %one\n"
2937 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
2938 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
2939 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
2941 "%id = OpVariable %uvec3ptr Input\n"
2942 "%vertexIndex = OpVariable %i32inputptr Input\n"
2943 "%instanceIndex = OpVariable %i32inputptr Input\n"
2944 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
2946 // gl_Position = vec4(1.);
2947 "%vert_main = OpFunction %void None %voidf\n"
2948 "%vert_entry = OpLabel\n"
2949 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
2950 " OpStore %position %c_vec4_1\n"
2955 "%comp_main1 = OpFunction %void None %voidf\n"
2956 "%comp1_entry = OpLabel\n"
2957 "%idval1 = OpLoad %uvec3 %id\n"
2958 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
2959 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
2960 "%inval1 = OpLoad %f32 %inloc1\n"
2961 "%add = OpFAdd %f32 %inval1 %inval1\n"
2962 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
2963 " OpStore %outloc1 %add\n"
2968 "%comp_main2 = OpFunction %void None %voidf\n"
2969 "%comp2_entry = OpLabel\n"
2970 "%idval2 = OpLoad %uvec3 %id\n"
2971 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
2972 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
2973 "%inval2 = OpLoad %f32 %inloc2\n"
2974 "%neg = OpFNegate %f32 %inval2\n"
2975 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
2976 " OpStore %outloc2 %neg\n"
2978 " OpFunctionEnd\n");
2980 spec1.assembly = assembly;
2981 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2982 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
2983 spec1.numWorkGroups = IVec3(numElements, 1, 1);
2984 spec1.entryPoint = "entrypoint1";
2986 spec2.assembly = assembly;
2987 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2988 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2989 spec2.numWorkGroups = IVec3(numElements, 1, 1);
2990 spec2.entryPoint = "entrypoint2";
2992 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
2993 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
2995 return group.release();
2998 inline std::string makeLongUTF8String (size_t num4ByteChars)
3000 // An example of a longest valid UTF-8 character. Be explicit about the
3001 // character type because Microsoft compilers can otherwise interpret the
3002 // character string as being over wide (16-bit) characters. Ideally, we
3003 // would just use a C++11 UTF-8 string literal, but we want to support older
3004 // Microsoft compilers.
3005 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
3006 std::string longString;
3007 longString.reserve(num4ByteChars * 4);
3008 for (size_t count = 0; count < num4ByteChars; count++)
3010 longString += earthAfrica;
3015 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
3017 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
3018 vector<CaseParameter> cases;
3019 de::Random rnd (deStringHash(group->getName()));
3020 const int numElements = 100;
3021 vector<float> positiveFloats (numElements, 0);
3022 vector<float> negativeFloats (numElements, 0);
3023 const StringTemplate shaderTemplate (
3024 "OpCapability Shader\n"
3025 "OpMemoryModel Logical GLSL450\n"
3027 "OpEntryPoint GLCompute %main \"main\" %id\n"
3028 "OpExecutionMode %main LocalSize 1 1 1\n"
3032 "OpName %main \"main\"\n"
3033 "OpName %id \"gl_GlobalInvocationID\"\n"
3035 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3037 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3039 "%id = OpVariable %uvec3ptr Input\n"
3040 "%zero = OpConstant %i32 0\n"
3042 "%main = OpFunction %void None %voidf\n"
3043 "%label = OpLabel\n"
3044 "%idval = OpLoad %uvec3 %id\n"
3045 "%x = OpCompositeExtract %u32 %idval 0\n"
3046 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3047 "%inval = OpLoad %f32 %inloc\n"
3048 "%neg = OpFNegate %f32 %inval\n"
3049 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3050 " OpStore %outloc %neg\n"
3052 " OpFunctionEnd\n");
3054 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
3055 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
3056 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
3057 "OpSource GLSL 430 %fname"));
3058 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
3059 "OpSource GLSL 430 %fname"));
3060 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
3061 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
3062 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
3063 "OpSource GLSL 430 %fname \"\""));
3064 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
3065 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
3066 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
3067 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
3068 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
3069 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
3070 "OpSourceContinued \"id main() {}\""));
3071 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
3072 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3073 "OpSourceContinued \"\""));
3074 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
3075 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3076 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
3077 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
3078 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3079 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
3080 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
3081 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
3082 "OpSourceContinued \"void\"\n"
3083 "OpSourceContinued \"main()\"\n"
3084 "OpSourceContinued \"{}\""));
3085 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
3086 "OpSource GLSL 430 %fname \"\"\n"
3087 "OpSourceContinued \"#version 430\nvoid main() {}\""));
3089 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3091 for (size_t ndx = 0; ndx < numElements; ++ndx)
3092 negativeFloats[ndx] = -positiveFloats[ndx];
3094 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3096 map<string, string> specializations;
3097 ComputeShaderSpec spec;
3099 specializations["SOURCE"] = cases[caseNdx].param;
3100 spec.assembly = shaderTemplate.specialize(specializations);
3101 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3102 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3103 spec.numWorkGroups = IVec3(numElements, 1, 1);
3105 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3108 return group.release();
3111 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
3113 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
3114 vector<CaseParameter> cases;
3115 de::Random rnd (deStringHash(group->getName()));
3116 const int numElements = 100;
3117 vector<float> inputFloats (numElements, 0);
3118 vector<float> outputFloats (numElements, 0);
3119 const StringTemplate shaderTemplate (
3120 string(getComputeAsmShaderPreamble()) +
3122 "OpSourceExtension \"${EXTENSION}\"\n"
3124 "OpName %main \"main\"\n"
3125 "OpName %id \"gl_GlobalInvocationID\"\n"
3127 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3129 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3131 "%id = OpVariable %uvec3ptr Input\n"
3132 "%zero = OpConstant %i32 0\n"
3134 "%main = OpFunction %void None %voidf\n"
3135 "%label = OpLabel\n"
3136 "%idval = OpLoad %uvec3 %id\n"
3137 "%x = OpCompositeExtract %u32 %idval 0\n"
3138 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3139 "%inval = OpLoad %f32 %inloc\n"
3140 "%neg = OpFNegate %f32 %inval\n"
3141 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3142 " OpStore %outloc %neg\n"
3144 " OpFunctionEnd\n");
3146 cases.push_back(CaseParameter("empty_extension", ""));
3147 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
3148 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
3149 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
3150 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
3152 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
3154 for (size_t ndx = 0; ndx < numElements; ++ndx)
3155 outputFloats[ndx] = -inputFloats[ndx];
3157 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3159 map<string, string> specializations;
3160 ComputeShaderSpec spec;
3162 specializations["EXTENSION"] = cases[caseNdx].param;
3163 spec.assembly = shaderTemplate.specialize(specializations);
3164 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3165 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3166 spec.numWorkGroups = IVec3(numElements, 1, 1);
3168 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3171 return group.release();
3174 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
3175 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
3177 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
3178 vector<CaseParameter> cases;
3179 de::Random rnd (deStringHash(group->getName()));
3180 const int numElements = 100;
3181 vector<float> positiveFloats (numElements, 0);
3182 vector<float> negativeFloats (numElements, 0);
3183 const StringTemplate shaderTemplate (
3184 string(getComputeAsmShaderPreamble()) +
3186 "OpSource GLSL 430\n"
3187 "OpName %main \"main\"\n"
3188 "OpName %id \"gl_GlobalInvocationID\"\n"
3190 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3192 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3193 "%uvec2 = OpTypeVector %u32 2\n"
3194 "%bvec3 = OpTypeVector %bool 3\n"
3195 "%fvec4 = OpTypeVector %f32 4\n"
3196 "%fmat33 = OpTypeMatrix %fvec3 3\n"
3197 "%const100 = OpConstant %u32 100\n"
3198 "%uarr100 = OpTypeArray %i32 %const100\n"
3199 "%struct = OpTypeStruct %f32 %i32 %u32\n"
3200 "%pointer = OpTypePointer Function %i32\n"
3201 + string(getComputeAsmInputOutputBuffer()) +
3203 "%null = OpConstantNull ${TYPE}\n"
3205 "%id = OpVariable %uvec3ptr Input\n"
3206 "%zero = OpConstant %i32 0\n"
3208 "%main = OpFunction %void None %voidf\n"
3209 "%label = OpLabel\n"
3210 "%idval = OpLoad %uvec3 %id\n"
3211 "%x = OpCompositeExtract %u32 %idval 0\n"
3212 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3213 "%inval = OpLoad %f32 %inloc\n"
3214 "%neg = OpFNegate %f32 %inval\n"
3215 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3216 " OpStore %outloc %neg\n"
3218 " OpFunctionEnd\n");
3220 cases.push_back(CaseParameter("bool", "%bool"));
3221 cases.push_back(CaseParameter("sint32", "%i32"));
3222 cases.push_back(CaseParameter("uint32", "%u32"));
3223 cases.push_back(CaseParameter("float32", "%f32"));
3224 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
3225 cases.push_back(CaseParameter("vec3bool", "%bvec3"));
3226 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
3227 cases.push_back(CaseParameter("matrix", "%fmat33"));
3228 cases.push_back(CaseParameter("array", "%uarr100"));
3229 cases.push_back(CaseParameter("struct", "%struct"));
3230 cases.push_back(CaseParameter("pointer", "%pointer"));
3232 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3234 for (size_t ndx = 0; ndx < numElements; ++ndx)
3235 negativeFloats[ndx] = -positiveFloats[ndx];
3237 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3239 map<string, string> specializations;
3240 ComputeShaderSpec spec;
3242 specializations["TYPE"] = cases[caseNdx].param;
3243 spec.assembly = shaderTemplate.specialize(specializations);
3244 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3245 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3246 spec.numWorkGroups = IVec3(numElements, 1, 1);
3248 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3251 return group.release();
3254 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
3255 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
3257 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
3258 vector<CaseParameter> cases;
3259 de::Random rnd (deStringHash(group->getName()));
3260 const int numElements = 100;
3261 vector<float> positiveFloats (numElements, 0);
3262 vector<float> negativeFloats (numElements, 0);
3263 const StringTemplate shaderTemplate (
3264 string(getComputeAsmShaderPreamble()) +
3266 "OpSource GLSL 430\n"
3267 "OpName %main \"main\"\n"
3268 "OpName %id \"gl_GlobalInvocationID\"\n"
3270 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3272 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3274 "%id = OpVariable %uvec3ptr Input\n"
3275 "%zero = OpConstant %i32 0\n"
3279 "%main = OpFunction %void None %voidf\n"
3280 "%label = OpLabel\n"
3281 "%idval = OpLoad %uvec3 %id\n"
3282 "%x = OpCompositeExtract %u32 %idval 0\n"
3283 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3284 "%inval = OpLoad %f32 %inloc\n"
3285 "%neg = OpFNegate %f32 %inval\n"
3286 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3287 " OpStore %outloc %neg\n"
3289 " OpFunctionEnd\n");
3291 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
3292 "%const = OpConstantComposite %uvec3 %five %zero %five"));
3293 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
3294 "%ten = OpConstant %f32 10.\n"
3295 "%fzero = OpConstant %f32 0.\n"
3296 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
3297 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
3298 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
3299 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
3300 "%fzero = OpConstant %f32 0.\n"
3301 "%one = OpConstant %f32 1.\n"
3302 "%point5 = OpConstant %f32 0.5\n"
3303 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
3304 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
3305 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
3306 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
3307 "%st2 = OpTypeStruct %i32 %i32\n"
3308 "%struct = OpTypeStruct %st1 %st2\n"
3309 "%point5 = OpConstant %f32 0.5\n"
3310 "%one = OpConstant %u32 1\n"
3311 "%ten = OpConstant %i32 10\n"
3312 "%st1val = OpConstantComposite %st1 %one %point5\n"
3313 "%st2val = OpConstantComposite %st2 %ten %ten\n"
3314 "%const = OpConstantComposite %struct %st1val %st2val"));
3316 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3318 for (size_t ndx = 0; ndx < numElements; ++ndx)
3319 negativeFloats[ndx] = -positiveFloats[ndx];
3321 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3323 map<string, string> specializations;
3324 ComputeShaderSpec spec;
3326 specializations["CONSTANT"] = cases[caseNdx].param;
3327 spec.assembly = shaderTemplate.specialize(specializations);
3328 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3329 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3330 spec.numWorkGroups = IVec3(numElements, 1, 1);
3332 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3335 return group.release();
3338 // Creates a floating point number with the given exponent, and significand
3339 // bits set. It can only create normalized numbers. Only the least significant
3340 // 24 bits of the significand will be examined. The final bit of the
3341 // significand will also be ignored. This allows alignment to be written
3342 // similarly to C99 hex-floats.
3343 // For example if you wanted to write 0x1.7f34p-12 you would call
3344 // constructNormalizedFloat(-12, 0x7f3400)
3345 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
3349 for (deInt32 idx = 0; idx < 23; ++idx)
3351 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
3355 return std::ldexp(f, exponent);
3358 // Compare instruction for the OpQuantizeF16 compute exact case.
3359 // Returns true if the output is what is expected from the test case.
3360 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
3362 if (outputAllocs.size() != 1)
3365 // We really just need this for size because we cannot compare Nans.
3366 const BufferSp& expectedOutput = expectedOutputs[0];
3367 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
3369 if (expectedOutput->getNumBytes() != 4*sizeof(float)) {
3373 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
3374 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
3379 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
3380 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
3385 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
3386 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
3391 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
3392 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
3399 // Checks that every output from a test-case is a float NaN.
3400 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
3402 if (outputAllocs.size() != 1)
3405 // We really just need this for size because we cannot compare Nans.
3406 const BufferSp& expectedOutput = expectedOutputs[0];
3407 const float* output_as_float = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
3409 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
3411 if (!deFloatIsNaN(output_as_float[idx]))
3420 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
3421 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
3423 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
3425 const std::string shader (
3426 string(getComputeAsmShaderPreamble()) +
3428 "OpSource GLSL 430\n"
3429 "OpName %main \"main\"\n"
3430 "OpName %id \"gl_GlobalInvocationID\"\n"
3432 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3434 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3436 "%id = OpVariable %uvec3ptr Input\n"
3437 "%zero = OpConstant %i32 0\n"
3439 "%main = OpFunction %void None %voidf\n"
3440 "%label = OpLabel\n"
3441 "%idval = OpLoad %uvec3 %id\n"
3442 "%x = OpCompositeExtract %u32 %idval 0\n"
3443 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3444 "%inval = OpLoad %f32 %inloc\n"
3445 "%quant = OpQuantizeToF16 %f32 %inval\n"
3446 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3447 " OpStore %outloc %quant\n"
3449 " OpFunctionEnd\n");
3452 ComputeShaderSpec spec;
3453 const deUint32 numElements = 100;
3454 vector<float> infinities;
3455 vector<float> results;
3457 infinities.reserve(numElements);
3458 results.reserve(numElements);
3460 for (size_t idx = 0; idx < numElements; ++idx)
3465 infinities.push_back(std::numeric_limits<float>::infinity());
3466 results.push_back(std::numeric_limits<float>::infinity());
3469 infinities.push_back(-std::numeric_limits<float>::infinity());
3470 results.push_back(-std::numeric_limits<float>::infinity());
3473 infinities.push_back(std::ldexp(1.0f, 16));
3474 results.push_back(std::numeric_limits<float>::infinity());
3477 infinities.push_back(std::ldexp(-1.0f, 32));
3478 results.push_back(-std::numeric_limits<float>::infinity());
3483 spec.assembly = shader;
3484 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
3485 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
3486 spec.numWorkGroups = IVec3(numElements, 1, 1);
3488 group->addChild(new SpvAsmComputeShaderCase(
3489 testCtx, "infinities", "Check that infinities propagated and created", spec));
3493 ComputeShaderSpec spec;
3495 const deUint32 numElements = 100;
3497 nans.reserve(numElements);
3499 for (size_t idx = 0; idx < numElements; ++idx)
3503 nans.push_back(std::numeric_limits<float>::quiet_NaN());
3507 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
3511 spec.assembly = shader;
3512 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
3513 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
3514 spec.numWorkGroups = IVec3(numElements, 1, 1);
3515 spec.verifyIO = &compareNan;
3517 group->addChild(new SpvAsmComputeShaderCase(
3518 testCtx, "propagated_nans", "Check that nans are propagated", spec));
3522 ComputeShaderSpec spec;
3523 vector<float> small;
3524 vector<float> zeros;
3525 const deUint32 numElements = 100;
3527 small.reserve(numElements);
3528 zeros.reserve(numElements);
3530 for (size_t idx = 0; idx < numElements; ++idx)
3535 small.push_back(0.f);
3536 zeros.push_back(0.f);
3539 small.push_back(-0.f);
3540 zeros.push_back(-0.f);
3543 small.push_back(std::ldexp(1.0f, -16));
3544 zeros.push_back(0.f);
3547 small.push_back(std::ldexp(-1.0f, -32));
3548 zeros.push_back(-0.f);
3551 small.push_back(std::ldexp(1.0f, -127));
3552 zeros.push_back(0.f);
3555 small.push_back(-std::ldexp(1.0f, -128));
3556 zeros.push_back(-0.f);
3561 spec.assembly = shader;
3562 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
3563 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
3564 spec.numWorkGroups = IVec3(numElements, 1, 1);
3566 group->addChild(new SpvAsmComputeShaderCase(
3567 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
3571 ComputeShaderSpec spec;
3572 vector<float> exact;
3573 const deUint32 numElements = 200;
3575 exact.reserve(numElements);
3577 for (size_t idx = 0; idx < numElements; ++idx)
3578 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
3580 spec.assembly = shader;
3581 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
3582 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
3583 spec.numWorkGroups = IVec3(numElements, 1, 1);
3585 group->addChild(new SpvAsmComputeShaderCase(
3586 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
3590 ComputeShaderSpec spec;
3591 vector<float> inputs;
3592 const deUint32 numElements = 4;
3594 inputs.push_back(constructNormalizedFloat(8, 0x300300));
3595 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
3596 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
3597 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
3599 spec.assembly = shader;
3600 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
3601 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3602 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
3603 spec.numWorkGroups = IVec3(numElements, 1, 1);
3605 group->addChild(new SpvAsmComputeShaderCase(
3606 testCtx, "rounded", "Check that are rounded when needed", spec));
3609 return group.release();
3612 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
3614 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
3616 const std::string shader (
3617 string(getComputeAsmShaderPreamble()) +
3619 "OpName %main \"main\"\n"
3620 "OpName %id \"gl_GlobalInvocationID\"\n"
3622 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3624 "OpDecorate %sc_0 SpecId 0\n"
3625 "OpDecorate %sc_1 SpecId 1\n"
3626 "OpDecorate %sc_2 SpecId 2\n"
3627 "OpDecorate %sc_3 SpecId 3\n"
3628 "OpDecorate %sc_4 SpecId 4\n"
3629 "OpDecorate %sc_5 SpecId 5\n"
3631 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3633 "%id = OpVariable %uvec3ptr Input\n"
3634 "%zero = OpConstant %i32 0\n"
3635 "%c_u32_6 = OpConstant %u32 6\n"
3637 "%sc_0 = OpSpecConstant %f32 0.\n"
3638 "%sc_1 = OpSpecConstant %f32 0.\n"
3639 "%sc_2 = OpSpecConstant %f32 0.\n"
3640 "%sc_3 = OpSpecConstant %f32 0.\n"
3641 "%sc_4 = OpSpecConstant %f32 0.\n"
3642 "%sc_5 = OpSpecConstant %f32 0.\n"
3644 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
3645 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
3646 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
3647 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
3648 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
3649 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
3651 "%main = OpFunction %void None %voidf\n"
3652 "%label = OpLabel\n"
3653 "%idval = OpLoad %uvec3 %id\n"
3654 "%x = OpCompositeExtract %u32 %idval 0\n"
3655 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3656 "%selector = OpUMod %u32 %x %c_u32_6\n"
3657 " OpSelectionMerge %exit None\n"
3658 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
3660 "%case0 = OpLabel\n"
3661 " OpStore %outloc %sc_0_quant\n"
3664 "%case1 = OpLabel\n"
3665 " OpStore %outloc %sc_1_quant\n"
3668 "%case2 = OpLabel\n"
3669 " OpStore %outloc %sc_2_quant\n"
3672 "%case3 = OpLabel\n"
3673 " OpStore %outloc %sc_3_quant\n"
3676 "%case4 = OpLabel\n"
3677 " OpStore %outloc %sc_4_quant\n"
3680 "%case5 = OpLabel\n"
3681 " OpStore %outloc %sc_5_quant\n"
3687 " OpFunctionEnd\n");
3690 ComputeShaderSpec spec;
3691 const deUint8 numCases = 4;
3692 vector<float> inputs (numCases, 0.f);
3693 vector<float> outputs;
3695 spec.assembly = shader;
3696 spec.numWorkGroups = IVec3(numCases, 1, 1);
3698 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
3699 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
3700 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
3701 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
3703 outputs.push_back(std::numeric_limits<float>::infinity());
3704 outputs.push_back(-std::numeric_limits<float>::infinity());
3705 outputs.push_back(std::numeric_limits<float>::infinity());
3706 outputs.push_back(-std::numeric_limits<float>::infinity());
3708 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3709 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3711 group->addChild(new SpvAsmComputeShaderCase(
3712 testCtx, "infinities", "Check that infinities propagated and created", spec));
3716 ComputeShaderSpec spec;
3717 const deUint8 numCases = 2;
3718 vector<float> inputs (numCases, 0.f);
3719 vector<float> outputs;
3721 spec.assembly = shader;
3722 spec.numWorkGroups = IVec3(numCases, 1, 1);
3723 spec.verifyIO = &compareNan;
3725 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
3726 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
3728 for (deUint8 idx = 0; idx < numCases; ++idx)
3729 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
3731 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3732 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3734 group->addChild(new SpvAsmComputeShaderCase(
3735 testCtx, "propagated_nans", "Check that nans are propagated", spec));
3739 ComputeShaderSpec spec;
3740 const deUint8 numCases = 6;
3741 vector<float> inputs (numCases, 0.f);
3742 vector<float> outputs;
3744 spec.assembly = shader;
3745 spec.numWorkGroups = IVec3(numCases, 1, 1);
3747 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
3748 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
3749 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
3750 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
3751 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
3752 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
3754 outputs.push_back(0.f);
3755 outputs.push_back(-0.f);
3756 outputs.push_back(0.f);
3757 outputs.push_back(-0.f);
3758 outputs.push_back(0.f);
3759 outputs.push_back(-0.f);
3761 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3762 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3764 group->addChild(new SpvAsmComputeShaderCase(
3765 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
3769 ComputeShaderSpec spec;
3770 const deUint8 numCases = 6;
3771 vector<float> inputs (numCases, 0.f);
3772 vector<float> outputs;
3774 spec.assembly = shader;
3775 spec.numWorkGroups = IVec3(numCases, 1, 1);
3777 for (deUint8 idx = 0; idx < 6; ++idx)
3779 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
3780 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
3781 outputs.push_back(f);
3784 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3785 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3787 group->addChild(new SpvAsmComputeShaderCase(
3788 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
3792 ComputeShaderSpec spec;
3793 const deUint8 numCases = 4;
3794 vector<float> inputs (numCases, 0.f);
3795 vector<float> outputs;
3797 spec.assembly = shader;
3798 spec.numWorkGroups = IVec3(numCases, 1, 1);
3799 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
3801 outputs.push_back(constructNormalizedFloat(8, 0x300300));
3802 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
3803 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
3804 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
3806 for (deUint8 idx = 0; idx < numCases; ++idx)
3807 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
3809 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3810 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3812 group->addChild(new SpvAsmComputeShaderCase(
3813 testCtx, "rounded", "Check that are rounded when needed", spec));
3816 return group.release();
3819 // Checks that constant null/composite values can be used in computation.
3820 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
3822 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
3823 ComputeShaderSpec spec;
3824 de::Random rnd (deStringHash(group->getName()));
3825 const int numElements = 100;
3826 vector<float> positiveFloats (numElements, 0);
3827 vector<float> negativeFloats (numElements, 0);
3829 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3831 for (size_t ndx = 0; ndx < numElements; ++ndx)
3832 negativeFloats[ndx] = -positiveFloats[ndx];
3835 "OpCapability Shader\n"
3836 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
3837 "OpMemoryModel Logical GLSL450\n"
3838 "OpEntryPoint GLCompute %main \"main\" %id\n"
3839 "OpExecutionMode %main LocalSize 1 1 1\n"
3841 "OpSource GLSL 430\n"
3842 "OpName %main \"main\"\n"
3843 "OpName %id \"gl_GlobalInvocationID\"\n"
3845 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3847 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3849 "%fmat = OpTypeMatrix %fvec3 3\n"
3850 "%ten = OpConstant %u32 10\n"
3851 "%f32arr10 = OpTypeArray %f32 %ten\n"
3852 "%fst = OpTypeStruct %f32 %f32\n"
3854 + string(getComputeAsmInputOutputBuffer()) +
3856 "%id = OpVariable %uvec3ptr Input\n"
3857 "%zero = OpConstant %i32 0\n"
3859 // Create a bunch of null values
3860 "%unull = OpConstantNull %u32\n"
3861 "%fnull = OpConstantNull %f32\n"
3862 "%vnull = OpConstantNull %fvec3\n"
3863 "%mnull = OpConstantNull %fmat\n"
3864 "%anull = OpConstantNull %f32arr10\n"
3865 "%snull = OpConstantComposite %fst %fnull %fnull\n"
3867 "%main = OpFunction %void None %voidf\n"
3868 "%label = OpLabel\n"
3869 "%idval = OpLoad %uvec3 %id\n"
3870 "%x = OpCompositeExtract %u32 %idval 0\n"
3871 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3872 "%inval = OpLoad %f32 %inloc\n"
3873 "%neg = OpFNegate %f32 %inval\n"
3875 // Get the abs() of (a certain element of) those null values
3876 "%unull_cov = OpConvertUToF %f32 %unull\n"
3877 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
3878 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
3879 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
3880 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
3881 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
3882 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
3883 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
3884 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
3885 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
3886 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
3889 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
3890 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
3891 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
3892 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
3893 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
3894 "%final = OpFAdd %f32 %add5 %snull_abs\n"
3896 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3897 " OpStore %outloc %final\n" // write to output
3900 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3901 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3902 spec.numWorkGroups = IVec3(numElements, 1, 1);
3904 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
3906 return group.release();
3909 // Assembly code used for testing loop control is based on GLSL source code:
3912 // layout(std140, set = 0, binding = 0) readonly buffer Input {
3913 // float elements[];
3915 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
3916 // float elements[];
3920 // uint x = gl_GlobalInvocationID.x;
3921 // output_data.elements[x] = input_data.elements[x];
3922 // for (uint i = 0; i < 4; ++i)
3923 // output_data.elements[x] += 1.f;
3925 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
3927 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
3928 vector<CaseParameter> cases;
3929 de::Random rnd (deStringHash(group->getName()));
3930 const int numElements = 100;
3931 vector<float> inputFloats (numElements, 0);
3932 vector<float> outputFloats (numElements, 0);
3933 const StringTemplate shaderTemplate (
3934 string(getComputeAsmShaderPreamble()) +
3936 "OpSource GLSL 430\n"
3937 "OpName %main \"main\"\n"
3938 "OpName %id \"gl_GlobalInvocationID\"\n"
3940 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3942 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3944 "%u32ptr = OpTypePointer Function %u32\n"
3946 "%id = OpVariable %uvec3ptr Input\n"
3947 "%zero = OpConstant %i32 0\n"
3948 "%uzero = OpConstant %u32 0\n"
3949 "%one = OpConstant %i32 1\n"
3950 "%constf1 = OpConstant %f32 1.0\n"
3951 "%four = OpConstant %u32 4\n"
3953 "%main = OpFunction %void None %voidf\n"
3954 "%entry = OpLabel\n"
3955 "%i = OpVariable %u32ptr Function\n"
3956 " OpStore %i %uzero\n"
3958 "%idval = OpLoad %uvec3 %id\n"
3959 "%x = OpCompositeExtract %u32 %idval 0\n"
3960 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3961 "%inval = OpLoad %f32 %inloc\n"
3962 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3963 " OpStore %outloc %inval\n"
3964 " OpBranch %loop_entry\n"
3966 "%loop_entry = OpLabel\n"
3967 "%i_val = OpLoad %u32 %i\n"
3968 "%cmp_lt = OpULessThan %bool %i_val %four\n"
3969 " OpLoopMerge %loop_merge %loop_body ${CONTROL}\n"
3970 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
3971 "%loop_body = OpLabel\n"
3972 "%outval = OpLoad %f32 %outloc\n"
3973 "%addf1 = OpFAdd %f32 %outval %constf1\n"
3974 " OpStore %outloc %addf1\n"
3975 "%new_i = OpIAdd %u32 %i_val %one\n"
3976 " OpStore %i %new_i\n"
3977 " OpBranch %loop_entry\n"
3978 "%loop_merge = OpLabel\n"
3980 " OpFunctionEnd\n");
3982 cases.push_back(CaseParameter("none", "None"));
3983 cases.push_back(CaseParameter("unroll", "Unroll"));
3984 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
3985 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
3987 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3989 for (size_t ndx = 0; ndx < numElements; ++ndx)
3990 outputFloats[ndx] = inputFloats[ndx] + 4.f;
3992 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3994 map<string, string> specializations;
3995 ComputeShaderSpec spec;
3997 specializations["CONTROL"] = cases[caseNdx].param;
3998 spec.assembly = shaderTemplate.specialize(specializations);
3999 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4000 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4001 spec.numWorkGroups = IVec3(numElements, 1, 1);
4003 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4006 return group.release();
4009 // Assembly code used for testing selection control is based on GLSL source code:
4012 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4013 // float elements[];
4015 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4016 // float elements[];
4020 // uint x = gl_GlobalInvocationID.x;
4021 // float val = input_data.elements[x];
4023 // output_data.elements[x] = val + 1.f;
4025 // output_data.elements[x] = val - 1.f;
4027 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
4029 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
4030 vector<CaseParameter> cases;
4031 de::Random rnd (deStringHash(group->getName()));
4032 const int numElements = 100;
4033 vector<float> inputFloats (numElements, 0);
4034 vector<float> outputFloats (numElements, 0);
4035 const StringTemplate shaderTemplate (
4036 string(getComputeAsmShaderPreamble()) +
4038 "OpSource GLSL 430\n"
4039 "OpName %main \"main\"\n"
4040 "OpName %id \"gl_GlobalInvocationID\"\n"
4042 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4044 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4046 "%id = OpVariable %uvec3ptr Input\n"
4047 "%zero = OpConstant %i32 0\n"
4048 "%constf1 = OpConstant %f32 1.0\n"
4049 "%constf10 = OpConstant %f32 10.0\n"
4051 "%main = OpFunction %void None %voidf\n"
4052 "%entry = OpLabel\n"
4053 "%idval = OpLoad %uvec3 %id\n"
4054 "%x = OpCompositeExtract %u32 %idval 0\n"
4055 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4056 "%inval = OpLoad %f32 %inloc\n"
4057 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4058 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
4060 " OpSelectionMerge %if_end ${CONTROL}\n"
4061 " OpBranchConditional %cmp_gt %if_true %if_false\n"
4062 "%if_true = OpLabel\n"
4063 "%addf1 = OpFAdd %f32 %inval %constf1\n"
4064 " OpStore %outloc %addf1\n"
4065 " OpBranch %if_end\n"
4066 "%if_false = OpLabel\n"
4067 "%subf1 = OpFSub %f32 %inval %constf1\n"
4068 " OpStore %outloc %subf1\n"
4069 " OpBranch %if_end\n"
4070 "%if_end = OpLabel\n"
4072 " OpFunctionEnd\n");
4074 cases.push_back(CaseParameter("none", "None"));
4075 cases.push_back(CaseParameter("flatten", "Flatten"));
4076 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
4077 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
4079 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4081 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4082 floorAll(inputFloats);
4084 for (size_t ndx = 0; ndx < numElements; ++ndx)
4085 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
4087 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4089 map<string, string> specializations;
4090 ComputeShaderSpec spec;
4092 specializations["CONTROL"] = cases[caseNdx].param;
4093 spec.assembly = shaderTemplate.specialize(specializations);
4094 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4095 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4096 spec.numWorkGroups = IVec3(numElements, 1, 1);
4098 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4101 return group.release();
4104 // Assembly code used for testing function control is based on GLSL source code:
4108 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4109 // float elements[];
4111 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4112 // float elements[];
4115 // float const10() { return 10.f; }
4118 // uint x = gl_GlobalInvocationID.x;
4119 // output_data.elements[x] = input_data.elements[x] + const10();
4121 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
4123 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
4124 vector<CaseParameter> cases;
4125 de::Random rnd (deStringHash(group->getName()));
4126 const int numElements = 100;
4127 vector<float> inputFloats (numElements, 0);
4128 vector<float> outputFloats (numElements, 0);
4129 const StringTemplate shaderTemplate (
4130 string(getComputeAsmShaderPreamble()) +
4132 "OpSource GLSL 430\n"
4133 "OpName %main \"main\"\n"
4134 "OpName %func_const10 \"const10(\"\n"
4135 "OpName %id \"gl_GlobalInvocationID\"\n"
4137 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4139 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4141 "%f32f = OpTypeFunction %f32\n"
4142 "%id = OpVariable %uvec3ptr Input\n"
4143 "%zero = OpConstant %i32 0\n"
4144 "%constf10 = OpConstant %f32 10.0\n"
4146 "%main = OpFunction %void None %voidf\n"
4147 "%entry = OpLabel\n"
4148 "%idval = OpLoad %uvec3 %id\n"
4149 "%x = OpCompositeExtract %u32 %idval 0\n"
4150 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4151 "%inval = OpLoad %f32 %inloc\n"
4152 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
4153 "%fadd = OpFAdd %f32 %inval %ret_10\n"
4154 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4155 " OpStore %outloc %fadd\n"
4159 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
4160 "%label = OpLabel\n"
4161 " OpReturnValue %constf10\n"
4162 " OpFunctionEnd\n");
4164 cases.push_back(CaseParameter("none", "None"));
4165 cases.push_back(CaseParameter("inline", "Inline"));
4166 cases.push_back(CaseParameter("dont_inline", "DontInline"));
4167 cases.push_back(CaseParameter("pure", "Pure"));
4168 cases.push_back(CaseParameter("const", "Const"));
4169 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
4170 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
4171 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
4172 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
4174 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4176 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4177 floorAll(inputFloats);
4179 for (size_t ndx = 0; ndx < numElements; ++ndx)
4180 outputFloats[ndx] = inputFloats[ndx] + 10.f;
4182 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4184 map<string, string> specializations;
4185 ComputeShaderSpec spec;
4187 specializations["CONTROL"] = cases[caseNdx].param;
4188 spec.assembly = shaderTemplate.specialize(specializations);
4189 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4190 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4191 spec.numWorkGroups = IVec3(numElements, 1, 1);
4193 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4196 return group.release();
4199 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
4201 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
4202 vector<CaseParameter> cases;
4203 de::Random rnd (deStringHash(group->getName()));
4204 const int numElements = 100;
4205 vector<float> inputFloats (numElements, 0);
4206 vector<float> outputFloats (numElements, 0);
4207 const StringTemplate shaderTemplate (
4208 string(getComputeAsmShaderPreamble()) +
4210 "OpSource GLSL 430\n"
4211 "OpName %main \"main\"\n"
4212 "OpName %id \"gl_GlobalInvocationID\"\n"
4214 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4216 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4218 "%f32ptr_f = OpTypePointer Function %f32\n"
4220 "%id = OpVariable %uvec3ptr Input\n"
4221 "%zero = OpConstant %i32 0\n"
4222 "%four = OpConstant %i32 4\n"
4224 "%main = OpFunction %void None %voidf\n"
4225 "%label = OpLabel\n"
4226 "%copy = OpVariable %f32ptr_f Function\n"
4227 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
4228 "%x = OpCompositeExtract %u32 %idval 0\n"
4229 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4230 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4231 " OpCopyMemory %copy %inloc ${ACCESS}\n"
4232 "%val1 = OpLoad %f32 %copy\n"
4233 "%val2 = OpLoad %f32 %inloc\n"
4234 "%add = OpFAdd %f32 %val1 %val2\n"
4235 " OpStore %outloc %add ${ACCESS}\n"
4237 " OpFunctionEnd\n");
4239 cases.push_back(CaseParameter("null", ""));
4240 cases.push_back(CaseParameter("none", "None"));
4241 cases.push_back(CaseParameter("volatile", "Volatile"));
4242 cases.push_back(CaseParameter("aligned", "Aligned 4"));
4243 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
4244 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
4245 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
4247 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4249 for (size_t ndx = 0; ndx < numElements; ++ndx)
4250 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
4252 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4254 map<string, string> specializations;
4255 ComputeShaderSpec spec;
4257 specializations["ACCESS"] = cases[caseNdx].param;
4258 spec.assembly = shaderTemplate.specialize(specializations);
4259 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4260 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4261 spec.numWorkGroups = IVec3(numElements, 1, 1);
4263 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4266 return group.release();
4269 // Checks that we can get undefined values for various types, without exercising a computation with it.
4270 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
4272 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
4273 vector<CaseParameter> cases;
4274 de::Random rnd (deStringHash(group->getName()));
4275 const int numElements = 100;
4276 vector<float> positiveFloats (numElements, 0);
4277 vector<float> negativeFloats (numElements, 0);
4278 const StringTemplate shaderTemplate (
4279 string(getComputeAsmShaderPreamble()) +
4281 "OpSource GLSL 430\n"
4282 "OpName %main \"main\"\n"
4283 "OpName %id \"gl_GlobalInvocationID\"\n"
4285 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4287 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4288 "%uvec2 = OpTypeVector %u32 2\n"
4289 "%fvec4 = OpTypeVector %f32 4\n"
4290 "%fmat33 = OpTypeMatrix %fvec3 3\n"
4291 "%image = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
4292 "%sampler = OpTypeSampler\n"
4293 "%simage = OpTypeSampledImage %image\n"
4294 "%const100 = OpConstant %u32 100\n"
4295 "%uarr100 = OpTypeArray %i32 %const100\n"
4296 "%struct = OpTypeStruct %f32 %i32 %u32\n"
4297 "%pointer = OpTypePointer Function %i32\n"
4298 + string(getComputeAsmInputOutputBuffer()) +
4300 "%id = OpVariable %uvec3ptr Input\n"
4301 "%zero = OpConstant %i32 0\n"
4303 "%main = OpFunction %void None %voidf\n"
4304 "%label = OpLabel\n"
4306 "%undef = OpUndef ${TYPE}\n"
4308 "%idval = OpLoad %uvec3 %id\n"
4309 "%x = OpCompositeExtract %u32 %idval 0\n"
4311 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4312 "%inval = OpLoad %f32 %inloc\n"
4313 "%neg = OpFNegate %f32 %inval\n"
4314 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4315 " OpStore %outloc %neg\n"
4317 " OpFunctionEnd\n");
4319 cases.push_back(CaseParameter("bool", "%bool"));
4320 cases.push_back(CaseParameter("sint32", "%i32"));
4321 cases.push_back(CaseParameter("uint32", "%u32"));
4322 cases.push_back(CaseParameter("float32", "%f32"));
4323 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
4324 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
4325 cases.push_back(CaseParameter("matrix", "%fmat33"));
4326 cases.push_back(CaseParameter("image", "%image"));
4327 cases.push_back(CaseParameter("sampler", "%sampler"));
4328 cases.push_back(CaseParameter("sampledimage", "%simage"));
4329 cases.push_back(CaseParameter("array", "%uarr100"));
4330 cases.push_back(CaseParameter("runtimearray", "%f32arr"));
4331 cases.push_back(CaseParameter("struct", "%struct"));
4332 cases.push_back(CaseParameter("pointer", "%pointer"));
4334 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4336 for (size_t ndx = 0; ndx < numElements; ++ndx)
4337 negativeFloats[ndx] = -positiveFloats[ndx];
4339 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4341 map<string, string> specializations;
4342 ComputeShaderSpec spec;
4344 specializations["TYPE"] = cases[caseNdx].param;
4345 spec.assembly = shaderTemplate.specialize(specializations);
4346 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4347 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4348 spec.numWorkGroups = IVec3(numElements, 1, 1);
4350 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4353 return group.release();
4358 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
4360 struct NameCodePair { string name, code; };
4361 RGBA defaultColors[4];
4362 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
4363 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
4364 map<string, string> fragments = passthruFragments();
4365 const NameCodePair tests[] =
4367 {"unknown", "OpSource Unknown 321"},
4368 {"essl", "OpSource ESSL 310"},
4369 {"glsl", "OpSource GLSL 450"},
4370 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
4371 {"opencl_c", "OpSource OpenCL_C 120"},
4372 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
4373 {"file", opsourceGLSLWithFile},
4374 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
4375 // Longest possible source string: SPIR-V limits instructions to 65535
4376 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
4377 // contain 65530 UTF8 characters (one word each) plus one last word
4378 // containing 3 ASCII characters and \0.
4379 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
4382 getDefaultColors(defaultColors);
4383 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
4385 fragments["debug"] = tests[testNdx].code;
4386 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
4389 return opSourceTests.release();
4392 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
4394 struct NameCodePair { string name, code; };
4395 RGBA defaultColors[4];
4396 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
4397 map<string, string> fragments = passthruFragments();
4398 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
4399 const NameCodePair tests[] =
4401 {"empty", opsource + "OpSourceContinued \"\""},
4402 {"short", opsource + "OpSourceContinued \"abcde\""},
4403 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
4404 // Longest possible source string: SPIR-V limits instructions to 65535
4405 // words, of which the first one is OpSourceContinued/length; the rest
4406 // will contain 65533 UTF8 characters (one word each) plus one last word
4407 // containing 3 ASCII characters and \0.
4408 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
4411 getDefaultColors(defaultColors);
4412 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
4414 fragments["debug"] = tests[testNdx].code;
4415 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
4418 return opSourceTests.release();
4421 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
4423 RGBA defaultColors[4];
4424 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
4425 map<string, string> fragments;
4426 getDefaultColors(defaultColors);
4427 fragments["debug"] =
4428 "%name = OpString \"name\"\n";
4430 fragments["pre_main"] =
4433 "OpLine %name 1 1\n"
4435 "OpLine %name 1 1\n"
4436 "OpLine %name 1 1\n"
4437 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
4439 "OpLine %name 1 1\n"
4441 "OpLine %name 1 1\n"
4442 "OpLine %name 1 1\n"
4443 "%second_param1 = OpFunctionParameter %v4f32\n"
4446 "%label_secondfunction = OpLabel\n"
4448 "OpReturnValue %second_param1\n"
4453 fragments["testfun"] =
4454 // A %test_code function that returns its argument unchanged.
4457 "OpLine %name 1 1\n"
4458 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4460 "%param1 = OpFunctionParameter %v4f32\n"
4463 "%label_testfun = OpLabel\n"
4465 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
4466 "OpReturnValue %val1\n"
4468 "OpLine %name 1 1\n"
4471 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
4473 return opLineTests.release();
4477 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
4479 RGBA defaultColors[4];
4480 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
4481 map<string, string> fragments;
4482 std::vector<std::pair<std::string, std::string> > problemStrings;
4484 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
4485 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
4486 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
4487 getDefaultColors(defaultColors);
4489 fragments["debug"] =
4490 "%other_name = OpString \"other_name\"\n";
4492 fragments["pre_main"] =
4493 "OpLine %file_name 32 0\n"
4494 "OpLine %file_name 32 32\n"
4495 "OpLine %file_name 32 40\n"
4496 "OpLine %other_name 32 40\n"
4497 "OpLine %other_name 0 100\n"
4498 "OpLine %other_name 0 4294967295\n"
4499 "OpLine %other_name 4294967295 0\n"
4500 "OpLine %other_name 32 40\n"
4501 "OpLine %file_name 0 0\n"
4502 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
4503 "OpLine %file_name 1 0\n"
4504 "%second_param1 = OpFunctionParameter %v4f32\n"
4505 "OpLine %file_name 1 3\n"
4506 "OpLine %file_name 1 2\n"
4507 "%label_secondfunction = OpLabel\n"
4508 "OpLine %file_name 0 2\n"
4509 "OpReturnValue %second_param1\n"
4511 "OpLine %file_name 0 2\n"
4512 "OpLine %file_name 0 2\n";
4514 fragments["testfun"] =
4515 // A %test_code function that returns its argument unchanged.
4516 "OpLine %file_name 1 0\n"
4517 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4518 "OpLine %file_name 16 330\n"
4519 "%param1 = OpFunctionParameter %v4f32\n"
4520 "OpLine %file_name 14 442\n"
4521 "%label_testfun = OpLabel\n"
4522 "OpLine %file_name 11 1024\n"
4523 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
4524 "OpLine %file_name 2 97\n"
4525 "OpReturnValue %val1\n"
4527 "OpLine %file_name 5 32\n";
4529 for (size_t i = 0; i < problemStrings.size(); ++i)
4531 map<string, string> testFragments = fragments;
4532 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
4533 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
4536 return opLineTests.release();
4539 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
4541 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
4545 const char functionStart[] =
4546 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4547 "%param1 = OpFunctionParameter %v4f32\n"
4550 const char functionEnd[] =
4551 "OpReturnValue %transformed_param\n"
4554 struct NameConstantsCode
4561 NameConstantsCode tests[] =
4565 "%cnull = OpConstantNull %v4f32\n",
4566 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
4570 "%cnull = OpConstantNull %f32\n",
4571 "%vp = OpVariable %fp_v4f32 Function\n"
4572 "%v = OpLoad %v4f32 %vp\n"
4573 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
4574 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
4575 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
4576 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
4577 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
4581 "%cnull = OpConstantNull %bool\n",
4582 "%v = OpVariable %fp_v4f32 Function\n"
4583 " OpStore %v %param1\n"
4584 " OpSelectionMerge %false_label None\n"
4585 " OpBranchConditional %cnull %true_label %false_label\n"
4586 "%true_label = OpLabel\n"
4587 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
4588 " OpBranch %false_label\n"
4589 "%false_label = OpLabel\n"
4590 "%transformed_param = OpLoad %v4f32 %v\n"
4594 "%cnull = OpConstantNull %i32\n",
4595 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
4596 "%b = OpIEqual %bool %cnull %c_i32_0\n"
4597 " OpSelectionMerge %false_label None\n"
4598 " OpBranchConditional %b %true_label %false_label\n"
4599 "%true_label = OpLabel\n"
4600 " OpStore %v %param1\n"
4601 " OpBranch %false_label\n"
4602 "%false_label = OpLabel\n"
4603 "%transformed_param = OpLoad %v4f32 %v\n"
4607 "%stype = OpTypeStruct %f32 %v4f32\n"
4608 "%fp_stype = OpTypePointer Function %stype\n"
4609 "%cnull = OpConstantNull %stype\n",
4610 "%v = OpVariable %fp_stype Function %cnull\n"
4611 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
4612 "%f_val = OpLoad %v4f32 %f\n"
4613 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
4617 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
4618 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
4619 "%cnull = OpConstantNull %a4_v4f32\n",
4620 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
4621 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
4622 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
4623 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
4624 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
4625 "%f_val = OpLoad %v4f32 %f\n"
4626 "%f1_val = OpLoad %v4f32 %f1\n"
4627 "%f2_val = OpLoad %v4f32 %f2\n"
4628 "%f3_val = OpLoad %v4f32 %f3\n"
4629 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
4630 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
4631 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
4632 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
4636 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
4637 "%cnull = OpConstantNull %mat4x4_f32\n",
4638 // Our null matrix * any vector should result in a zero vector.
4639 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
4640 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
4644 getHalfColorsFullAlpha(colors);
4646 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
4648 map<string, string> fragments;
4649 fragments["pre_main"] = tests[testNdx].constants;
4650 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
4651 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
4653 return opConstantNullTests.release();
4655 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
4657 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
4658 RGBA inputColors[4];
4659 RGBA outputColors[4];
4662 const char functionStart[] =
4663 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4664 "%param1 = OpFunctionParameter %v4f32\n"
4667 const char functionEnd[] =
4668 "OpReturnValue %transformed_param\n"
4671 struct NameConstantsCode
4678 NameConstantsCode tests[] =
4683 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
4684 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
4689 "%stype = OpTypeStruct %v4f32 %f32\n"
4690 "%fp_stype = OpTypePointer Function %stype\n"
4691 "%f32_n_1 = OpConstant %f32 -1.0\n"
4692 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
4693 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
4694 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
4696 "%v = OpVariable %fp_stype Function %cval\n"
4697 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
4698 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
4699 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
4700 "%f32_val = OpLoad %f32 %f32_ptr\n"
4701 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
4702 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
4703 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
4706 // [1|0|0|0.5] [x] = x + 0.5
4707 // [0|1|0|0.5] [y] = y + 0.5
4708 // [0|0|1|0.5] [z] = z + 0.5
4709 // [0|0|0|1 ] [1] = 1
4712 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
4713 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
4714 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
4715 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
4716 "%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"
4717 "%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",
4719 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
4724 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4725 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
4726 "%f32_n_1 = OpConstant %f32 -1.0\n"
4727 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
4728 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
4730 "%v = OpVariable %fp_a4f32 Function %carr\n"
4731 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
4732 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
4733 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
4734 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
4735 "%f_val = OpLoad %f32 %f\n"
4736 "%f1_val = OpLoad %f32 %f1\n"
4737 "%f2_val = OpLoad %f32 %f2\n"
4738 "%f3_val = OpLoad %f32 %f3\n"
4739 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
4740 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
4741 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
4742 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
4743 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
4750 // [ 1.0, 1.0, 1.0, 1.0]
4754 // [ 0.0, 0.5, 0.0, 0.0]
4758 // [ 1.0, 1.0, 1.0, 1.0]
4761 "array_of_struct_of_array",
4763 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4764 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
4765 "%stype = OpTypeStruct %f32 %a4f32\n"
4766 "%a3stype = OpTypeArray %stype %c_u32_3\n"
4767 "%fp_a3stype = OpTypePointer Function %a3stype\n"
4768 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
4769 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
4770 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
4771 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
4772 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
4774 "%v = OpVariable %fp_a3stype Function %carr\n"
4775 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
4776 "%f_l = OpLoad %f32 %f\n"
4777 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
4778 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
4782 getHalfColorsFullAlpha(inputColors);
4783 outputColors[0] = RGBA(255, 255, 255, 255);
4784 outputColors[1] = RGBA(255, 127, 127, 255);
4785 outputColors[2] = RGBA(127, 255, 127, 255);
4786 outputColors[3] = RGBA(127, 127, 255, 255);
4788 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
4790 map<string, string> fragments;
4791 fragments["pre_main"] = tests[testNdx].constants;
4792 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
4793 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
4795 return opConstantCompositeTests.release();
4798 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
4800 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
4801 RGBA inputColors[4];
4802 RGBA outputColors[4];
4803 map<string, string> fragments;
4805 // vec4 test_code(vec4 param) {
4806 // vec4 result = param;
4807 // for (int i = 0; i < 4; ++i) {
4808 // if (i == 0) result[i] = 0.;
4809 // else result[i] = 1. - result[i];
4813 const char function[] =
4814 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4815 "%param1 = OpFunctionParameter %v4f32\n"
4817 "%iptr = OpVariable %fp_i32 Function\n"
4818 "%result = OpVariable %fp_v4f32 Function\n"
4819 " OpStore %iptr %c_i32_0\n"
4820 " OpStore %result %param1\n"
4823 // Loop entry block.
4825 "%ival = OpLoad %i32 %iptr\n"
4826 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
4827 " OpLoopMerge %exit %if_entry None\n"
4828 " OpBranchConditional %lt_4 %if_entry %exit\n"
4830 // Merge block for loop.
4832 "%ret = OpLoad %v4f32 %result\n"
4833 " OpReturnValue %ret\n"
4835 // If-statement entry block.
4836 "%if_entry = OpLabel\n"
4837 "%loc = OpAccessChain %fp_f32 %result %ival\n"
4838 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
4839 " OpSelectionMerge %if_exit None\n"
4840 " OpBranchConditional %eq_0 %if_true %if_false\n"
4842 // False branch for if-statement.
4843 "%if_false = OpLabel\n"
4844 "%val = OpLoad %f32 %loc\n"
4845 "%sub = OpFSub %f32 %c_f32_1 %val\n"
4846 " OpStore %loc %sub\n"
4847 " OpBranch %if_exit\n"
4849 // Merge block for if-statement.
4850 "%if_exit = OpLabel\n"
4851 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
4852 " OpStore %iptr %ival_next\n"
4855 // True branch for if-statement.
4856 "%if_true = OpLabel\n"
4857 " OpStore %loc %c_f32_0\n"
4858 " OpBranch %if_exit\n"
4862 fragments["testfun"] = function;
4864 inputColors[0] = RGBA(127, 127, 127, 0);
4865 inputColors[1] = RGBA(127, 0, 0, 0);
4866 inputColors[2] = RGBA(0, 127, 0, 0);
4867 inputColors[3] = RGBA(0, 0, 127, 0);
4869 outputColors[0] = RGBA(0, 128, 128, 255);
4870 outputColors[1] = RGBA(0, 255, 255, 255);
4871 outputColors[2] = RGBA(0, 128, 255, 255);
4872 outputColors[3] = RGBA(0, 255, 128, 255);
4874 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
4876 return group.release();
4879 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
4881 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
4882 RGBA inputColors[4];
4883 RGBA outputColors[4];
4884 map<string, string> fragments;
4886 const char typesAndConstants[] =
4887 "%c_f32_p2 = OpConstant %f32 0.2\n"
4888 "%c_f32_p4 = OpConstant %f32 0.4\n"
4889 "%c_f32_p6 = OpConstant %f32 0.6\n"
4890 "%c_f32_p8 = OpConstant %f32 0.8\n";
4892 // vec4 test_code(vec4 param) {
4893 // vec4 result = param;
4894 // for (int i = 0; i < 4; ++i) {
4896 // case 0: result[i] += .2; break;
4897 // case 1: result[i] += .6; break;
4898 // case 2: result[i] += .4; break;
4899 // case 3: result[i] += .8; break;
4900 // default: break; // unreachable
4905 const char function[] =
4906 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4907 "%param1 = OpFunctionParameter %v4f32\n"
4909 "%iptr = OpVariable %fp_i32 Function\n"
4910 "%result = OpVariable %fp_v4f32 Function\n"
4911 " OpStore %iptr %c_i32_0\n"
4912 " OpStore %result %param1\n"
4915 // Loop entry block.
4917 "%ival = OpLoad %i32 %iptr\n"
4918 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
4919 " OpLoopMerge %exit %switch_exit None\n"
4920 " OpBranchConditional %lt_4 %switch_entry %exit\n"
4922 // Merge block for loop.
4924 "%ret = OpLoad %v4f32 %result\n"
4925 " OpReturnValue %ret\n"
4927 // Switch-statement entry block.
4928 "%switch_entry = OpLabel\n"
4929 "%loc = OpAccessChain %fp_f32 %result %ival\n"
4930 "%val = OpLoad %f32 %loc\n"
4931 " OpSelectionMerge %switch_exit None\n"
4932 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
4934 "%case2 = OpLabel\n"
4935 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
4936 " OpStore %loc %addp4\n"
4937 " OpBranch %switch_exit\n"
4939 "%switch_default = OpLabel\n"
4942 "%case3 = OpLabel\n"
4943 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
4944 " OpStore %loc %addp8\n"
4945 " OpBranch %switch_exit\n"
4947 "%case0 = OpLabel\n"
4948 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
4949 " OpStore %loc %addp2\n"
4950 " OpBranch %switch_exit\n"
4952 // Merge block for switch-statement.
4953 "%switch_exit = OpLabel\n"
4954 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
4955 " OpStore %iptr %ival_next\n"
4958 "%case1 = OpLabel\n"
4959 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
4960 " OpStore %loc %addp6\n"
4961 " OpBranch %switch_exit\n"
4965 fragments["pre_main"] = typesAndConstants;
4966 fragments["testfun"] = function;
4968 inputColors[0] = RGBA(127, 27, 127, 51);
4969 inputColors[1] = RGBA(127, 0, 0, 51);
4970 inputColors[2] = RGBA(0, 27, 0, 51);
4971 inputColors[3] = RGBA(0, 0, 127, 51);
4973 outputColors[0] = RGBA(178, 180, 229, 255);
4974 outputColors[1] = RGBA(178, 153, 102, 255);
4975 outputColors[2] = RGBA(51, 180, 102, 255);
4976 outputColors[3] = RGBA(51, 153, 229, 255);
4978 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
4980 return group.release();
4983 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
4985 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
4986 RGBA inputColors[4];
4987 RGBA outputColors[4];
4988 map<string, string> fragments;
4990 const char decorations[] =
4991 "OpDecorate %array_group ArrayStride 4\n"
4992 "OpDecorate %struct_member_group Offset 0\n"
4993 "%array_group = OpDecorationGroup\n"
4994 "%struct_member_group = OpDecorationGroup\n"
4996 "OpDecorate %group1 RelaxedPrecision\n"
4997 "OpDecorate %group3 RelaxedPrecision\n"
4998 "OpDecorate %group3 Invariant\n"
4999 "OpDecorate %group3 Restrict\n"
5000 "%group0 = OpDecorationGroup\n"
5001 "%group1 = OpDecorationGroup\n"
5002 "%group3 = OpDecorationGroup\n";
5004 const char typesAndConstants[] =
5005 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
5006 "%struct1 = OpTypeStruct %a3f32\n"
5007 "%struct2 = OpTypeStruct %a3f32\n"
5008 "%fp_struct1 = OpTypePointer Function %struct1\n"
5009 "%fp_struct2 = OpTypePointer Function %struct2\n"
5010 "%c_f32_2 = OpConstant %f32 2.\n"
5011 "%c_f32_n2 = OpConstant %f32 -2.\n"
5013 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
5014 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
5015 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
5016 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
5018 const char function[] =
5019 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5020 "%param = OpFunctionParameter %v4f32\n"
5021 "%entry = OpLabel\n"
5022 "%result = OpVariable %fp_v4f32 Function\n"
5023 "%v_struct1 = OpVariable %fp_struct1 Function\n"
5024 "%v_struct2 = OpVariable %fp_struct2 Function\n"
5025 " OpStore %result %param\n"
5026 " OpStore %v_struct1 %c_struct1\n"
5027 " OpStore %v_struct2 %c_struct2\n"
5028 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
5029 "%val1 = OpLoad %f32 %ptr1\n"
5030 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
5031 "%val2 = OpLoad %f32 %ptr2\n"
5032 "%addvalues = OpFAdd %f32 %val1 %val2\n"
5033 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
5034 "%val = OpLoad %f32 %ptr\n"
5035 "%addresult = OpFAdd %f32 %addvalues %val\n"
5036 " OpStore %ptr %addresult\n"
5037 "%ret = OpLoad %v4f32 %result\n"
5038 " OpReturnValue %ret\n"
5041 struct CaseNameDecoration
5047 CaseNameDecoration tests[] =
5050 "same_decoration_group_on_multiple_types",
5051 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
5054 "empty_decoration_group",
5055 "OpGroupDecorate %group0 %a3f32\n"
5056 "OpGroupDecorate %group0 %result\n"
5059 "one_element_decoration_group",
5060 "OpGroupDecorate %array_group %a3f32\n"
5063 "multiple_elements_decoration_group",
5064 "OpGroupDecorate %group3 %v_struct1\n"
5067 "multiple_decoration_groups_on_same_variable",
5068 "OpGroupDecorate %group0 %v_struct2\n"
5069 "OpGroupDecorate %group1 %v_struct2\n"
5070 "OpGroupDecorate %group3 %v_struct2\n"
5073 "same_decoration_group_multiple_times",
5074 "OpGroupDecorate %group1 %addvalues\n"
5075 "OpGroupDecorate %group1 %addvalues\n"
5076 "OpGroupDecorate %group1 %addvalues\n"
5081 getHalfColorsFullAlpha(inputColors);
5082 getHalfColorsFullAlpha(outputColors);
5084 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
5086 fragments["decoration"] = decorations + tests[idx].decoration;
5087 fragments["pre_main"] = typesAndConstants;
5088 fragments["testfun"] = function;
5090 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
5093 return group.release();
5096 struct SpecConstantTwoIntGraphicsCase
5098 const char* caseName;
5099 const char* scDefinition0;
5100 const char* scDefinition1;
5101 const char* scResultType;
5102 const char* scOperation;
5103 deInt32 scActualValue0;
5104 deInt32 scActualValue1;
5105 const char* resultOperation;
5106 RGBA expectedColors[4];
5108 SpecConstantTwoIntGraphicsCase (const char* name,
5109 const char* definition0,
5110 const char* definition1,
5111 const char* resultType,
5112 const char* operation,
5115 const char* resultOp,
5116 const RGBA (&output)[4])
5118 , scDefinition0 (definition0)
5119 , scDefinition1 (definition1)
5120 , scResultType (resultType)
5121 , scOperation (operation)
5122 , scActualValue0 (value0)
5123 , scActualValue1 (value1)
5124 , resultOperation (resultOp)
5126 expectedColors[0] = output[0];
5127 expectedColors[1] = output[1];
5128 expectedColors[2] = output[2];
5129 expectedColors[3] = output[3];
5133 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
5135 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
5136 vector<SpecConstantTwoIntGraphicsCase> cases;
5137 RGBA inputColors[4];
5138 RGBA outputColors0[4];
5139 RGBA outputColors1[4];
5140 RGBA outputColors2[4];
5142 const char decorations1[] =
5143 "OpDecorate %sc_0 SpecId 0\n"
5144 "OpDecorate %sc_1 SpecId 1\n";
5146 const char typesAndConstants1[] =
5147 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
5148 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
5149 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
5151 const char function1[] =
5152 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5153 "%param = OpFunctionParameter %v4f32\n"
5154 "%label = OpLabel\n"
5155 "%result = OpVariable %fp_v4f32 Function\n"
5156 " OpStore %result %param\n"
5157 "%gen = ${GEN_RESULT}\n"
5158 "%index = OpIAdd %i32 %gen %c_i32_1\n"
5159 "%loc = OpAccessChain %fp_f32 %result %index\n"
5160 "%val = OpLoad %f32 %loc\n"
5161 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
5162 " OpStore %loc %add\n"
5163 "%ret = OpLoad %v4f32 %result\n"
5164 " OpReturnValue %ret\n"
5167 inputColors[0] = RGBA(127, 127, 127, 255);
5168 inputColors[1] = RGBA(127, 0, 0, 255);
5169 inputColors[2] = RGBA(0, 127, 0, 255);
5170 inputColors[3] = RGBA(0, 0, 127, 255);
5172 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
5173 outputColors0[0] = RGBA(255, 127, 127, 255);
5174 outputColors0[1] = RGBA(255, 0, 0, 255);
5175 outputColors0[2] = RGBA(128, 127, 0, 255);
5176 outputColors0[3] = RGBA(128, 0, 127, 255);
5178 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
5179 outputColors1[0] = RGBA(127, 255, 127, 255);
5180 outputColors1[1] = RGBA(127, 128, 0, 255);
5181 outputColors1[2] = RGBA(0, 255, 0, 255);
5182 outputColors1[3] = RGBA(0, 128, 127, 255);
5184 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
5185 outputColors2[0] = RGBA(127, 127, 255, 255);
5186 outputColors2[1] = RGBA(127, 0, 128, 255);
5187 outputColors2[2] = RGBA(0, 127, 128, 255);
5188 outputColors2[3] = RGBA(0, 0, 255, 255);
5190 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
5191 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
5192 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
5194 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
5195 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
5196 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
5197 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
5198 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
5199 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
5200 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
5201 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
5202 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
5203 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
5204 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
5205 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
5206 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
5207 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
5208 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
5209 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
5210 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
5211 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
5212 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
5213 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
5214 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
5215 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
5216 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
5217 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
5218 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
5219 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
5220 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
5221 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
5222 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
5223 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
5224 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
5225 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
5226 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
5228 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5230 map<string, string> specializations;
5231 map<string, string> fragments;
5232 vector<deInt32> specConstants;
5234 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
5235 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
5236 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
5237 specializations["SC_OP"] = cases[caseNdx].scOperation;
5238 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
5240 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
5241 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
5242 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
5244 specConstants.push_back(cases[caseNdx].scActualValue0);
5245 specConstants.push_back(cases[caseNdx].scActualValue1);
5247 createTestsForAllStages(cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants, group.get());
5250 const char decorations2[] =
5251 "OpDecorate %sc_0 SpecId 0\n"
5252 "OpDecorate %sc_1 SpecId 1\n"
5253 "OpDecorate %sc_2 SpecId 2\n";
5255 const char typesAndConstants2[] =
5256 "%v3i32 = OpTypeVector %i32 3\n"
5257 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
5258 "%vec3_undef = OpUndef %v3i32\n"
5260 "%sc_0 = OpSpecConstant %i32 0\n"
5261 "%sc_1 = OpSpecConstant %i32 0\n"
5262 "%sc_2 = OpSpecConstant %i32 0\n"
5263 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
5264 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
5265 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
5266 "%sc_vec3_0_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
5267 "%sc_vec3_1_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
5268 "%sc_vec3_2_s = OpSpecConstantOp %v3i32 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
5269 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
5270 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
5271 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
5272 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
5273 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
5274 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
5275 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
5277 const char function2[] =
5278 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5279 "%param = OpFunctionParameter %v4f32\n"
5280 "%label = OpLabel\n"
5281 "%result = OpVariable %fp_v4f32 Function\n"
5282 " OpStore %result %param\n"
5283 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
5284 "%val = OpLoad %f32 %loc\n"
5285 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
5286 " OpStore %loc %add\n"
5287 "%ret = OpLoad %v4f32 %result\n"
5288 " OpReturnValue %ret\n"
5291 map<string, string> fragments;
5292 vector<deInt32> specConstants;
5294 fragments["decoration"] = decorations2;
5295 fragments["pre_main"] = typesAndConstants2;
5296 fragments["testfun"] = function2;
5298 specConstants.push_back(56789);
5299 specConstants.push_back(-2);
5300 specConstants.push_back(56788);
5302 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
5304 return group.release();
5307 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
5309 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
5310 RGBA inputColors[4];
5311 RGBA outputColors1[4];
5312 RGBA outputColors2[4];
5313 RGBA outputColors3[4];
5314 map<string, string> fragments1;
5315 map<string, string> fragments2;
5316 map<string, string> fragments3;
5318 const char typesAndConstants1[] =
5319 "%c_f32_p2 = OpConstant %f32 0.2\n"
5320 "%c_f32_p4 = OpConstant %f32 0.4\n"
5321 "%c_f32_p5 = OpConstant %f32 0.5\n"
5322 "%c_f32_p8 = OpConstant %f32 0.8\n";
5324 // vec4 test_code(vec4 param) {
5325 // vec4 result = param;
5326 // for (int i = 0; i < 4; ++i) {
5329 // case 0: operand = .2; break;
5330 // case 1: operand = .5; break;
5331 // case 2: operand = .4; break;
5332 // case 3: operand = .0; break;
5333 // default: break; // unreachable
5335 // result[i] += operand;
5339 const char function1[] =
5340 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5341 "%param1 = OpFunctionParameter %v4f32\n"
5343 "%iptr = OpVariable %fp_i32 Function\n"
5344 "%result = OpVariable %fp_v4f32 Function\n"
5345 " OpStore %iptr %c_i32_0\n"
5346 " OpStore %result %param1\n"
5350 "%ival = OpLoad %i32 %iptr\n"
5351 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5352 " OpLoopMerge %exit %phi None\n"
5353 " OpBranchConditional %lt_4 %entry %exit\n"
5355 "%entry = OpLabel\n"
5356 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5357 "%val = OpLoad %f32 %loc\n"
5358 " OpSelectionMerge %phi None\n"
5359 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5361 "%case0 = OpLabel\n"
5363 "%case1 = OpLabel\n"
5365 "%case2 = OpLabel\n"
5367 "%case3 = OpLabel\n"
5370 "%default = OpLabel\n"
5374 "%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
5375 "%add = OpFAdd %f32 %val %operand\n"
5376 " OpStore %loc %add\n"
5377 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5378 " OpStore %iptr %ival_next\n"
5382 "%ret = OpLoad %v4f32 %result\n"
5383 " OpReturnValue %ret\n"
5387 fragments1["pre_main"] = typesAndConstants1;
5388 fragments1["testfun"] = function1;
5390 getHalfColorsFullAlpha(inputColors);
5392 outputColors1[0] = RGBA(178, 255, 229, 255);
5393 outputColors1[1] = RGBA(178, 127, 102, 255);
5394 outputColors1[2] = RGBA(51, 255, 102, 255);
5395 outputColors1[3] = RGBA(51, 127, 229, 255);
5397 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
5399 const char typesAndConstants2[] =
5400 "%c_f32_p2 = OpConstant %f32 0.2\n";
5402 // Add .4 to the second element of the given parameter.
5403 const char function2[] =
5404 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5405 "%param = OpFunctionParameter %v4f32\n"
5406 "%entry = OpLabel\n"
5407 "%result = OpVariable %fp_v4f32 Function\n"
5408 " OpStore %result %param\n"
5409 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
5410 "%val = OpLoad %f32 %loc\n"
5414 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
5415 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
5416 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
5417 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
5418 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
5419 " OpLoopMerge %exit %phi None\n"
5420 " OpBranchConditional %still_loop %phi %exit\n"
5423 " OpStore %loc %accum\n"
5424 "%ret = OpLoad %v4f32 %result\n"
5425 " OpReturnValue %ret\n"
5429 fragments2["pre_main"] = typesAndConstants2;
5430 fragments2["testfun"] = function2;
5432 outputColors2[0] = RGBA(127, 229, 127, 255);
5433 outputColors2[1] = RGBA(127, 102, 0, 255);
5434 outputColors2[2] = RGBA(0, 229, 0, 255);
5435 outputColors2[3] = RGBA(0, 102, 127, 255);
5437 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
5439 const char typesAndConstants3[] =
5440 "%true = OpConstantTrue %bool\n"
5441 "%false = OpConstantFalse %bool\n"
5442 "%c_f32_p2 = OpConstant %f32 0.2\n";
5444 // Swap the second and the third element of the given parameter.
5445 const char function3[] =
5446 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5447 "%param = OpFunctionParameter %v4f32\n"
5448 "%entry = OpLabel\n"
5449 "%result = OpVariable %fp_v4f32 Function\n"
5450 " OpStore %result %param\n"
5451 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
5452 "%a_init = OpLoad %f32 %a_loc\n"
5453 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
5454 "%b_init = OpLoad %f32 %b_loc\n"
5458 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
5459 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
5460 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
5461 " OpLoopMerge %exit %phi None\n"
5462 " OpBranchConditional %still_loop %phi %exit\n"
5465 " OpStore %a_loc %a_next\n"
5466 " OpStore %b_loc %b_next\n"
5467 "%ret = OpLoad %v4f32 %result\n"
5468 " OpReturnValue %ret\n"
5472 fragments3["pre_main"] = typesAndConstants3;
5473 fragments3["testfun"] = function3;
5475 outputColors3[0] = RGBA(127, 127, 127, 255);
5476 outputColors3[1] = RGBA(127, 0, 0, 255);
5477 outputColors3[2] = RGBA(0, 0, 127, 255);
5478 outputColors3[3] = RGBA(0, 127, 0, 255);
5480 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
5482 return group.release();
5485 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
5487 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
5488 RGBA inputColors[4];
5489 RGBA outputColors[4];
5491 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
5492 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
5493 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
5494 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
5495 const char constantsAndTypes[] =
5496 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
5497 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5498 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
5499 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
5500 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n"
5503 const char function[] =
5504 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5505 "%param = OpFunctionParameter %v4f32\n"
5506 "%label = OpLabel\n"
5507 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
5508 "%var2 = OpVariable %fp_f32 Function\n"
5509 "%red = OpCompositeExtract %f32 %param 0\n"
5510 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
5511 " OpStore %var2 %plus_red\n"
5512 "%val1 = OpLoad %f32 %var1\n"
5513 "%val2 = OpLoad %f32 %var2\n"
5514 "%mul = OpFMul %f32 %val1 %val2\n"
5515 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
5516 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
5517 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
5518 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
5519 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
5520 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
5521 " OpReturnValue %ret\n"
5524 struct CaseNameDecoration
5531 CaseNameDecoration tests[] = {
5532 {"multiplication", "OpDecorate %mul NoContraction"},
5533 {"addition", "OpDecorate %add NoContraction"},
5534 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
5537 getHalfColorsFullAlpha(inputColors);
5539 for (deUint8 idx = 0; idx < 4; ++idx)
5541 inputColors[idx].setRed(0);
5542 outputColors[idx] = RGBA(0, 0, 0, 255);
5545 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
5547 map<string, string> fragments;
5549 fragments["decoration"] = tests[testNdx].decoration;
5550 fragments["pre_main"] = constantsAndTypes;
5551 fragments["testfun"] = function;
5553 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
5556 return group.release();
5559 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
5561 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
5564 const char constantsAndTypes[] =
5565 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
5566 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
5567 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
5568 "%fp_stype = OpTypePointer Function %stype\n";
5570 const char function[] =
5571 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5572 "%param1 = OpFunctionParameter %v4f32\n"
5574 "%v1 = OpVariable %fp_v4f32 Function\n"
5575 "%v2 = OpVariable %fp_a2f32 Function\n"
5576 "%v3 = OpVariable %fp_f32 Function\n"
5577 "%v = OpVariable %fp_stype Function\n"
5578 "%vv = OpVariable %fp_stype Function\n"
5579 "%vvv = OpVariable %fp_f32 Function\n"
5581 " OpStore %v1 %c_v4f32_1_1_1_1\n"
5582 " OpStore %v2 %c_a2f32_1\n"
5583 " OpStore %v3 %c_f32_1\n"
5585 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5586 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
5587 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5588 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
5589 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
5590 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
5592 " OpStore %p_v4f32 %v1_v ${access_type}\n"
5593 " OpStore %p_a2f32 %v2_v ${access_type}\n"
5594 " OpStore %p_f32 %v3_v ${access_type}\n"
5596 " OpCopyMemory %vv %v ${access_type}\n"
5597 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
5599 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
5600 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
5601 "%v_f32_3 = OpLoad %f32 %vvv\n"
5603 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
5604 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
5605 " OpReturnValue %ret2\n"
5608 struct NameMemoryAccess
5615 NameMemoryAccess tests[] =
5618 { "volatile", "Volatile" },
5619 { "aligned", "Aligned 1" },
5620 { "volatile_aligned", "Volatile|Aligned 1" },
5621 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
5622 { "volatile_nontemporal", "Volatile|Nontemporal" },
5623 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
5626 getHalfColorsFullAlpha(colors);
5628 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
5630 map<string, string> fragments;
5631 map<string, string> memoryAccess;
5632 memoryAccess["access_type"] = tests[testNdx].accessType;
5634 fragments["pre_main"] = constantsAndTypes;
5635 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
5636 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
5638 return memoryAccessTests.release();
5640 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
5642 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
5643 RGBA defaultColors[4];
5644 map<string, string> fragments;
5645 getDefaultColors(defaultColors);
5647 // First, simple cases that don't do anything with the OpUndef result.
5648 struct NameCodePair { string name, decl, type; };
5649 const NameCodePair tests[] =
5651 {"bool", "", "%bool"},
5652 {"vec2uint32", "%type = OpTypeVector %u32 2", "%type"},
5653 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown", "%type"},
5654 {"sampler", "%type = OpTypeSampler", "%type"},
5655 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img", "%type"},
5656 {"pointer", "", "%fp_i32"},
5657 {"runtimearray", "%type = OpTypeRuntimeArray %f32", "%type"},
5658 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100", "%type"},
5659 {"struct", "%type = OpTypeStruct %f32 %i32 %u32", "%type"}};
5660 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5662 fragments["undef_type"] = tests[testNdx].type;
5663 fragments["testfun"] = StringTemplate(
5664 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5665 "%param1 = OpFunctionParameter %v4f32\n"
5666 "%label_testfun = OpLabel\n"
5667 "%undef = OpUndef ${undef_type}\n"
5668 "OpReturnValue %param1\n"
5669 "OpFunctionEnd\n").specialize(fragments);
5670 fragments["pre_main"] = tests[testNdx].decl;
5671 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
5675 fragments["testfun"] =
5676 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5677 "%param1 = OpFunctionParameter %v4f32\n"
5678 "%label_testfun = OpLabel\n"
5679 "%undef = OpUndef %f32\n"
5680 "%zero = OpFMul %f32 %undef %c_f32_0\n"
5681 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
5682 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
5683 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5684 "%b = OpFAdd %f32 %a %actually_zero\n"
5685 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
5686 "OpReturnValue %ret\n"
5689 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
5691 fragments["testfun"] =
5692 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5693 "%param1 = OpFunctionParameter %v4f32\n"
5694 "%label_testfun = OpLabel\n"
5695 "%undef = OpUndef %i32\n"
5696 "%zero = OpIMul %i32 %undef %c_i32_0\n"
5697 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
5698 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
5699 "OpReturnValue %ret\n"
5702 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
5704 fragments["testfun"] =
5705 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5706 "%param1 = OpFunctionParameter %v4f32\n"
5707 "%label_testfun = OpLabel\n"
5708 "%undef = OpUndef %u32\n"
5709 "%zero = OpIMul %u32 %undef %c_i32_0\n"
5710 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
5711 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
5712 "OpReturnValue %ret\n"
5715 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
5717 fragments["testfun"] =
5718 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5719 "%param1 = OpFunctionParameter %v4f32\n"
5720 "%label_testfun = OpLabel\n"
5721 "%undef = OpUndef %v4f32\n"
5722 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
5723 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
5724 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
5725 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
5726 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
5727 "%is_nan_0 = OpIsNan %bool %zero_0\n"
5728 "%is_nan_1 = OpIsNan %bool %zero_1\n"
5729 "%is_nan_2 = OpIsNan %bool %zero_2\n"
5730 "%is_nan_3 = OpIsNan %bool %zero_3\n"
5731 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
5732 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
5733 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
5734 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
5735 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5736 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
5737 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
5738 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
5739 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
5740 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
5741 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
5742 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
5743 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
5744 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
5745 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
5746 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
5747 "OpReturnValue %ret\n"
5750 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
5752 fragments["pre_main"] =
5753 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
5754 fragments["testfun"] =
5755 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5756 "%param1 = OpFunctionParameter %v4f32\n"
5757 "%label_testfun = OpLabel\n"
5758 "%undef = OpUndef %m2x2f32\n"
5759 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
5760 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
5761 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
5762 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
5763 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
5764 "%is_nan_0 = OpIsNan %bool %zero_0\n"
5765 "%is_nan_1 = OpIsNan %bool %zero_1\n"
5766 "%is_nan_2 = OpIsNan %bool %zero_2\n"
5767 "%is_nan_3 = OpIsNan %bool %zero_3\n"
5768 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
5769 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
5770 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
5771 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
5772 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5773 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
5774 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
5775 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
5776 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
5777 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
5778 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
5779 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
5780 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
5781 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
5782 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
5783 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
5784 "OpReturnValue %ret\n"
5787 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
5789 return opUndefTests.release();
5792 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
5794 const RGBA inputColors[4] =
5797 RGBA(0, 0, 255, 255),
5798 RGBA(0, 255, 0, 255),
5799 RGBA(0, 255, 255, 255)
5802 const RGBA expectedColors[4] =
5804 RGBA(255, 0, 0, 255),
5805 RGBA(255, 0, 0, 255),
5806 RGBA(255, 0, 0, 255),
5807 RGBA(255, 0, 0, 255)
5810 const struct SingleFP16Possibility
5813 const char* constant; // Value to assign to %test_constant.
5815 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
5821 -constructNormalizedFloat(1, 0x300000),
5822 "%cond = OpFOrdEqual %bool %c %test_constant\n"
5827 constructNormalizedFloat(7, 0x000000),
5828 "%cond = OpFOrdEqual %bool %c %test_constant\n"
5830 // SPIR-V requires that OpQuantizeToF16 flushes
5831 // any numbers that would end up denormalized in F16 to zero.
5835 std::ldexp(1.5f, -140),
5836 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5841 -std::ldexp(1.5f, -140),
5842 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5847 std::ldexp(1.0f, -16),
5848 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5849 }, // too small positive
5851 "negative_too_small",
5853 -std::ldexp(1.0f, -32),
5854 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5855 }, // too small negative
5859 -std::ldexp(1.0f, 128),
5861 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
5862 "%inf = OpIsInf %bool %c\n"
5863 "%cond = OpLogicalAnd %bool %gz %inf\n"
5868 std::ldexp(1.0f, 128),
5870 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
5871 "%inf = OpIsInf %bool %c\n"
5872 "%cond = OpLogicalAnd %bool %gz %inf\n"
5875 "round_to_negative_inf",
5877 -std::ldexp(1.0f, 32),
5879 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
5880 "%inf = OpIsInf %bool %c\n"
5881 "%cond = OpLogicalAnd %bool %gz %inf\n"
5886 std::ldexp(1.0f, 16),
5888 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
5889 "%inf = OpIsInf %bool %c\n"
5890 "%cond = OpLogicalAnd %bool %gz %inf\n"
5895 std::numeric_limits<float>::quiet_NaN(),
5897 // Test for any NaN value, as NaNs are not preserved
5898 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
5899 "%cond = OpIsNan %bool %direct_quant\n"
5904 std::numeric_limits<float>::quiet_NaN(),
5906 // Test for any NaN value, as NaNs are not preserved
5907 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
5908 "%cond = OpIsNan %bool %direct_quant\n"
5911 const char* constants =
5912 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
5914 StringTemplate function (
5915 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5916 "%param1 = OpFunctionParameter %v4f32\n"
5917 "%label_testfun = OpLabel\n"
5918 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5919 "%b = OpFAdd %f32 %test_constant %a\n"
5920 "%c = OpQuantizeToF16 %f32 %b\n"
5922 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
5923 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
5924 " OpReturnValue %retval\n"
5928 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
5929 const char* specConstants =
5930 "%test_constant = OpSpecConstant %f32 0.\n"
5931 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
5933 StringTemplate specConstantFunction(
5934 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5935 "%param1 = OpFunctionParameter %v4f32\n"
5936 "%label_testfun = OpLabel\n"
5938 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
5939 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
5940 " OpReturnValue %retval\n"
5944 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
5946 map<string, string> codeSpecialization;
5947 map<string, string> fragments;
5948 codeSpecialization["condition"] = tests[idx].condition;
5949 fragments["testfun"] = function.specialize(codeSpecialization);
5950 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
5951 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
5954 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
5956 map<string, string> codeSpecialization;
5957 map<string, string> fragments;
5958 vector<deInt32> passConstants;
5959 deInt32 specConstant;
5961 codeSpecialization["condition"] = tests[idx].condition;
5962 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
5963 fragments["decoration"] = specDecorations;
5964 fragments["pre_main"] = specConstants;
5966 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
5967 passConstants.push_back(specConstant);
5969 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
5973 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
5975 RGBA inputColors[4] = {
5977 RGBA(0, 0, 255, 255),
5978 RGBA(0, 255, 0, 255),
5979 RGBA(0, 255, 255, 255)
5982 RGBA expectedColors[4] =
5984 RGBA(255, 0, 0, 255),
5985 RGBA(255, 0, 0, 255),
5986 RGBA(255, 0, 0, 255),
5987 RGBA(255, 0, 0, 255)
5990 struct DualFP16Possibility
5995 const char* possibleOutput1;
5996 const char* possibleOutput2;
5999 "positive_round_up_or_round_down",
6001 constructNormalizedFloat(8, 0x300300),
6006 "negative_round_up_or_round_down",
6008 -constructNormalizedFloat(-7, 0x600800),
6015 constructNormalizedFloat(2, 0x01e000),
6020 "carry_to_exponent",
6022 constructNormalizedFloat(1, 0xffe000),
6027 StringTemplate constants (
6028 "%input_const = OpConstant %f32 ${input}\n"
6029 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6030 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6033 StringTemplate specConstants (
6034 "%input_const = OpSpecConstant %f32 0.\n"
6035 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6036 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6039 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
6041 const char* function =
6042 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6043 "%param1 = OpFunctionParameter %v4f32\n"
6044 "%label_testfun = OpLabel\n"
6045 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6046 // For the purposes of this test we assume that 0.f will always get
6047 // faithfully passed through the pipeline stages.
6048 "%b = OpFAdd %f32 %input_const %a\n"
6049 "%c = OpQuantizeToF16 %f32 %b\n"
6050 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
6051 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
6052 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
6053 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6054 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
6055 " OpReturnValue %retval\n"
6058 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6059 map<string, string> fragments;
6060 map<string, string> constantSpecialization;
6062 constantSpecialization["input"] = tests[idx].input;
6063 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6064 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6065 fragments["testfun"] = function;
6066 fragments["pre_main"] = constants.specialize(constantSpecialization);
6067 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6070 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6071 map<string, string> fragments;
6072 map<string, string> constantSpecialization;
6073 vector<deInt32> passConstants;
6074 deInt32 specConstant;
6076 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6077 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6078 fragments["testfun"] = function;
6079 fragments["decoration"] = specDecorations;
6080 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
6082 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
6083 passConstants.push_back(specConstant);
6085 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6089 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
6091 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
6092 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
6093 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
6094 return opQuantizeTests.release();
6097 struct ShaderPermutation
6099 deUint8 vertexPermutation;
6100 deUint8 geometryPermutation;
6101 deUint8 tesscPermutation;
6102 deUint8 tessePermutation;
6103 deUint8 fragmentPermutation;
6106 ShaderPermutation getShaderPermutation(deUint8 inputValue)
6108 ShaderPermutation permutation =
6110 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
6111 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
6112 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
6113 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
6114 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
6119 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
6121 RGBA defaultColors[4];
6122 RGBA invertedColors[4];
6123 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
6125 const ShaderElement combinedPipeline[] =
6127 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
6128 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
6129 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
6130 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
6131 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
6134 getDefaultColors(defaultColors);
6135 getInvertedDefaultColors(invertedColors);
6136 addFunctionCaseWithPrograms<InstanceContext>(
6137 moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline,
6138 createInstanceContext(combinedPipeline, map<string, string>()));
6140 const char* numbers[] =
6145 for (deInt8 idx = 0; idx < 32; ++idx)
6147 ShaderPermutation permutation = getShaderPermutation(idx);
6148 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
6149 const ShaderElement pipeline[] =
6151 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
6152 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
6153 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
6154 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
6155 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
6158 // If there are an even number of swaps, then it should be no-op.
6159 // If there are an odd number, the color should be flipped.
6160 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
6162 addFunctionCaseWithPrograms<InstanceContext>(
6163 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
6164 createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
6168 addFunctionCaseWithPrograms<InstanceContext>(
6169 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
6170 createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
6173 return moduleTests.release();
6176 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
6178 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
6179 RGBA defaultColors[4];
6180 getDefaultColors(defaultColors);
6181 map<string, string> fragments;
6182 fragments["pre_main"] =
6183 "%c_f32_5 = OpConstant %f32 5.\n";
6185 // A loop with a single block. The Continue Target is the loop block
6186 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
6187 // -- the "continue construct" forms the entire loop.
6188 fragments["testfun"] =
6189 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6190 "%param1 = OpFunctionParameter %v4f32\n"
6192 "%entry = OpLabel\n"
6193 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6196 ";adds and subtracts 1.0 to %val in alternate iterations\n"
6198 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
6199 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
6200 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
6201 "%val = OpFAdd %f32 %val1 %delta\n"
6202 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
6203 "%count__ = OpISub %i32 %count %c_i32_1\n"
6204 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6205 "OpLoopMerge %exit %loop None\n"
6206 "OpBranchConditional %again %loop %exit\n"
6209 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
6210 "OpReturnValue %result\n"
6214 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
6216 // Body comprised of multiple basic blocks.
6217 const StringTemplate multiBlock(
6218 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6219 "%param1 = OpFunctionParameter %v4f32\n"
6221 "%entry = OpLabel\n"
6222 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6225 ";adds and subtracts 1.0 to %val in alternate iterations\n"
6227 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
6228 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
6229 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
6230 // There are several possibilities for the Continue Target below. Each
6231 // will be specialized into a separate test case.
6232 "OpLoopMerge %exit ${continue_target} None\n"
6236 ";delta_next = (delta > 0) ? -1 : 1;\n"
6237 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
6238 "OpSelectionMerge %gather DontFlatten\n"
6239 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
6242 "OpBranch %gather\n"
6245 "OpBranch %gather\n"
6247 "%gather = OpLabel\n"
6248 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
6249 "%val = OpFAdd %f32 %val1 %delta\n"
6250 "%count__ = OpISub %i32 %count %c_i32_1\n"
6251 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6252 "OpBranchConditional %again %loop %exit\n"
6255 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
6256 "OpReturnValue %result\n"
6260 map<string, string> continue_target;
6262 // The Continue Target is the loop block itself.
6263 continue_target["continue_target"] = "%loop";
6264 fragments["testfun"] = multiBlock.specialize(continue_target);
6265 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
6267 // The Continue Target is at the end of the loop.
6268 continue_target["continue_target"] = "%gather";
6269 fragments["testfun"] = multiBlock.specialize(continue_target);
6270 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
6272 // A loop with continue statement.
6273 fragments["testfun"] =
6274 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6275 "%param1 = OpFunctionParameter %v4f32\n"
6277 "%entry = OpLabel\n"
6278 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6281 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
6283 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6284 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
6285 "OpLoopMerge %exit %continue None\n"
6289 ";skip if %count==2\n"
6290 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
6291 "OpSelectionMerge %continue DontFlatten\n"
6292 "OpBranchConditional %eq2 %continue %body\n"
6295 "%fcount = OpConvertSToF %f32 %count\n"
6296 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6297 "OpBranch %continue\n"
6299 "%continue = OpLabel\n"
6300 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
6301 "%count__ = OpISub %i32 %count %c_i32_1\n"
6302 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6303 "OpBranchConditional %again %loop %exit\n"
6306 "%same = OpFSub %f32 %val %c_f32_8\n"
6307 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6308 "OpReturnValue %result\n"
6310 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
6312 // A loop with break.
6313 fragments["testfun"] =
6314 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6315 "%param1 = OpFunctionParameter %v4f32\n"
6317 "%entry = OpLabel\n"
6318 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6319 "%dot = OpDot %f32 %param1 %param1\n"
6320 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6321 "%zero = OpConvertFToU %u32 %div\n"
6322 "%two = OpIAdd %i32 %zero %c_i32_2\n"
6323 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6326 ";adds 4 and 3 to %val0 (exits early)\n"
6328 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6329 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
6330 "OpLoopMerge %exit %continue None\n"
6334 ";end loop if %count==%two\n"
6335 "%above2 = OpSGreaterThan %bool %count %two\n"
6336 "OpSelectionMerge %continue DontFlatten\n"
6337 "OpBranchConditional %above2 %body %exit\n"
6340 "%fcount = OpConvertSToF %f32 %count\n"
6341 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6342 "OpBranch %continue\n"
6344 "%continue = OpLabel\n"
6345 "%count__ = OpISub %i32 %count %c_i32_1\n"
6346 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6347 "OpBranchConditional %again %loop %exit\n"
6350 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
6351 "%same = OpFSub %f32 %val_post %c_f32_7\n"
6352 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6353 "OpReturnValue %result\n"
6355 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
6357 // A loop with return.
6358 fragments["testfun"] =
6359 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6360 "%param1 = OpFunctionParameter %v4f32\n"
6362 "%entry = OpLabel\n"
6363 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6364 "%dot = OpDot %f32 %param1 %param1\n"
6365 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6366 "%zero = OpConvertFToU %u32 %div\n"
6367 "%two = OpIAdd %i32 %zero %c_i32_2\n"
6368 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6371 ";returns early without modifying %param1\n"
6373 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6374 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
6375 "OpLoopMerge %exit %continue None\n"
6379 ";return if %count==%two\n"
6380 "%above2 = OpSGreaterThan %bool %count %two\n"
6381 "OpSelectionMerge %continue DontFlatten\n"
6382 "OpBranchConditional %above2 %body %early_exit\n"
6384 "%early_exit = OpLabel\n"
6385 "OpReturnValue %param1\n"
6388 "%fcount = OpConvertSToF %f32 %count\n"
6389 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6390 "OpBranch %continue\n"
6392 "%continue = OpLabel\n"
6393 "%count__ = OpISub %i32 %count %c_i32_1\n"
6394 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6395 "OpBranchConditional %again %loop %exit\n"
6398 ";should never get here, so return an incorrect result\n"
6399 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
6400 "OpReturnValue %result\n"
6402 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
6404 return testGroup.release();
6407 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
6408 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
6410 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
6411 map<string, string> fragments;
6413 // A barrier inside a function body.
6414 fragments["pre_main"] =
6415 "%Workgroup = OpConstant %i32 2\n"
6416 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
6417 fragments["testfun"] =
6418 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6419 "%param1 = OpFunctionParameter %v4f32\n"
6420 "%label_testfun = OpLabel\n"
6421 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6422 "OpReturnValue %param1\n"
6424 addTessCtrlTest(testGroup.get(), "in_function", fragments);
6426 // Common setup code for the following tests.
6427 fragments["pre_main"] =
6428 "%Workgroup = OpConstant %i32 2\n"
6429 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
6430 "%c_f32_5 = OpConstant %f32 5.\n";
6431 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
6432 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6433 "%param1 = OpFunctionParameter %v4f32\n"
6434 "%entry = OpLabel\n"
6435 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6436 "%dot = OpDot %f32 %param1 %param1\n"
6437 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6438 "%zero = OpConvertFToU %u32 %div\n";
6440 // Barriers inside OpSwitch branches.
6441 fragments["testfun"] =
6443 "OpSelectionMerge %switch_exit None\n"
6444 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
6446 "%case1 = OpLabel\n"
6447 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6448 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6449 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6450 "OpBranch %switch_exit\n"
6452 "%switch_default = OpLabel\n"
6453 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6454 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6455 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6456 "OpBranch %switch_exit\n"
6458 "%case0 = OpLabel\n"
6459 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6460 "OpBranch %switch_exit\n"
6462 "%switch_exit = OpLabel\n"
6463 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
6464 "OpReturnValue %ret\n"
6466 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
6468 // Barriers inside if-then-else.
6469 fragments["testfun"] =
6471 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
6472 "OpSelectionMerge %exit DontFlatten\n"
6473 "OpBranchConditional %eq0 %then %else\n"
6476 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6477 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6478 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6482 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6486 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
6487 "OpReturnValue %ret\n"
6489 addTessCtrlTest(testGroup.get(), "in_if", fragments);
6491 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
6492 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
6493 fragments["testfun"] =
6495 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
6496 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
6497 "OpSelectionMerge %exit DontFlatten\n"
6498 "OpBranchConditional %thread0 %then %else\n"
6501 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6505 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
6509 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
6510 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6511 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
6512 "OpReturnValue %ret\n"
6514 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
6516 // A barrier inside a loop.
6517 fragments["pre_main"] =
6518 "%Workgroup = OpConstant %i32 2\n"
6519 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
6520 "%c_f32_10 = OpConstant %f32 10.\n";
6521 fragments["testfun"] =
6522 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6523 "%param1 = OpFunctionParameter %v4f32\n"
6524 "%entry = OpLabel\n"
6525 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6528 ";adds 4, 3, 2, and 1 to %val0\n"
6530 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
6531 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
6532 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6533 "%fcount = OpConvertSToF %f32 %count\n"
6534 "%val = OpFAdd %f32 %val1 %fcount\n"
6535 "%count__ = OpISub %i32 %count %c_i32_1\n"
6536 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6537 "OpLoopMerge %exit %loop None\n"
6538 "OpBranchConditional %again %loop %exit\n"
6541 "%same = OpFSub %f32 %val %c_f32_10\n"
6542 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6543 "OpReturnValue %ret\n"
6545 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
6547 return testGroup.release();
6550 // Test for the OpFRem instruction.
6551 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
6553 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
6554 map<string, string> fragments;
6555 RGBA inputColors[4];
6556 RGBA outputColors[4];
6558 fragments["pre_main"] =
6559 "%c_f32_3 = OpConstant %f32 3.0\n"
6560 "%c_f32_n3 = OpConstant %f32 -3.0\n"
6561 "%c_f32_4 = OpConstant %f32 4.0\n"
6562 "%c_f32_p75 = OpConstant %f32 0.75\n"
6563 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
6564 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
6565 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
6567 // The test does the following.
6568 // vec4 result = (param1 * 8.0) - 4.0;
6569 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
6570 fragments["testfun"] =
6571 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6572 "%param1 = OpFunctionParameter %v4f32\n"
6573 "%label_testfun = OpLabel\n"
6574 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
6575 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
6576 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
6577 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
6578 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
6579 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
6580 "OpReturnValue %xy_0_1\n"
6584 inputColors[0] = RGBA(16, 16, 0, 255);
6585 inputColors[1] = RGBA(232, 232, 0, 255);
6586 inputColors[2] = RGBA(232, 16, 0, 255);
6587 inputColors[3] = RGBA(16, 232, 0, 255);
6589 outputColors[0] = RGBA(64, 64, 0, 255);
6590 outputColors[1] = RGBA(255, 255, 0, 255);
6591 outputColors[2] = RGBA(255, 64, 0, 255);
6592 outputColors[3] = RGBA(64, 255, 0, 255);
6594 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
6595 return testGroup.release();
6598 // Test for the OpSRem instruction.
6599 tcu::TestCaseGroup* createOpSRemGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
6601 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "srem", "OpSRem"));
6602 map<string, string> fragments;
6604 fragments["pre_main"] =
6605 "%c_f32_255 = OpConstant %f32 255.0\n"
6606 "%c_i32_128 = OpConstant %i32 128\n"
6607 "%c_i32_255 = OpConstant %i32 255\n"
6608 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
6609 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
6610 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
6612 // The test does the following.
6613 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
6614 // ivec4 result = ivec4(srem(ints.x, ints.y), srem(ints.y, ints.z), srem(ints.z, ints.x), 255);
6615 // return float(result + 128) / 255.0;
6616 fragments["testfun"] =
6617 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6618 "%param1 = OpFunctionParameter %v4f32\n"
6619 "%label_testfun = OpLabel\n"
6620 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
6621 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
6622 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
6623 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
6624 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
6625 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
6626 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
6627 "%x_out = OpSRem %i32 %x_in %y_in\n"
6628 "%y_out = OpSRem %i32 %y_in %z_in\n"
6629 "%z_out = OpSRem %i32 %z_in %x_in\n"
6630 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
6631 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
6632 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
6633 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
6634 "OpReturnValue %float_out\n"
6637 const struct CaseParams
6640 const char* failMessageTemplate; // customized status message
6641 qpTestResult failResult; // override status on failure
6642 int operands[4][3]; // four (x, y, z) vectors of operands
6643 int results[4][3]; // four (x, y, z) vectors of results
6649 QP_TEST_RESULT_FAIL,
6650 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
6651 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
6655 "Inconsistent results, but within specification: ${reason}",
6656 negFailResult, // negative operands, not required by the spec
6657 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
6658 { { 5, 12, -2 }, { 0, -5, 2 }, { 3, 8, -6 }, { 25, -60, 0 } }, // results
6661 // If either operand is negative the result is undefined. Some implementations may still return correct values.
6663 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
6665 const CaseParams& params = cases[caseNdx];
6666 RGBA inputColors[4];
6667 RGBA outputColors[4];
6669 for (int i = 0; i < 4; ++i)
6671 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
6672 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
6675 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
6678 return testGroup.release();
6681 // Test for the OpSMod instruction.
6682 tcu::TestCaseGroup* createOpSModGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
6684 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "smod", "OpSMod"));
6685 map<string, string> fragments;
6687 fragments["pre_main"] =
6688 "%c_f32_255 = OpConstant %f32 255.0\n"
6689 "%c_i32_128 = OpConstant %i32 128\n"
6690 "%c_i32_255 = OpConstant %i32 255\n"
6691 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
6692 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
6693 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
6695 // The test does the following.
6696 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
6697 // ivec4 result = ivec4(smod(ints.x, ints.y), smod(ints.y, ints.z), smod(ints.z, ints.x), 255);
6698 // return float(result + 128) / 255.0;
6699 fragments["testfun"] =
6700 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6701 "%param1 = OpFunctionParameter %v4f32\n"
6702 "%label_testfun = OpLabel\n"
6703 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
6704 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
6705 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
6706 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
6707 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
6708 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
6709 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
6710 "%x_out = OpSMod %i32 %x_in %y_in\n"
6711 "%y_out = OpSMod %i32 %y_in %z_in\n"
6712 "%z_out = OpSMod %i32 %z_in %x_in\n"
6713 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
6714 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
6715 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
6716 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
6717 "OpReturnValue %float_out\n"
6720 const struct CaseParams
6723 const char* failMessageTemplate; // customized status message
6724 qpTestResult failResult; // override status on failure
6725 int operands[4][3]; // four (x, y, z) vectors of operands
6726 int results[4][3]; // four (x, y, z) vectors of results
6732 QP_TEST_RESULT_FAIL,
6733 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
6734 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
6738 "Inconsistent results, but within specification: ${reason}",
6739 negFailResult, // negative operands, not required by the spec
6740 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
6741 { { 5, -5, 3 }, { 0, 2, -3 }, { 3, -73, 69 }, { -35, 40, 0 } }, // results
6744 // If either operand is negative the result is undefined. Some implementations may still return correct values.
6746 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
6748 const CaseParams& params = cases[caseNdx];
6749 RGBA inputColors[4];
6750 RGBA outputColors[4];
6752 for (int i = 0; i < 4; ++i)
6754 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
6755 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
6758 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
6761 return testGroup.release();
6766 INTEGER_TYPE_SIGNED_16,
6767 INTEGER_TYPE_SIGNED_32,
6768 INTEGER_TYPE_SIGNED_64,
6770 INTEGER_TYPE_UNSIGNED_16,
6771 INTEGER_TYPE_UNSIGNED_32,
6772 INTEGER_TYPE_UNSIGNED_64,
6775 const string getBitWidthStr (IntegerType type)
6779 case INTEGER_TYPE_SIGNED_16:
6780 case INTEGER_TYPE_UNSIGNED_16: return "16";
6782 case INTEGER_TYPE_SIGNED_32:
6783 case INTEGER_TYPE_UNSIGNED_32: return "32";
6785 case INTEGER_TYPE_SIGNED_64:
6786 case INTEGER_TYPE_UNSIGNED_64: return "64";
6788 default: DE_ASSERT(false);
6793 const string getByteWidthStr (IntegerType type)
6797 case INTEGER_TYPE_SIGNED_16:
6798 case INTEGER_TYPE_UNSIGNED_16: return "2";
6800 case INTEGER_TYPE_SIGNED_32:
6801 case INTEGER_TYPE_UNSIGNED_32: return "4";
6803 case INTEGER_TYPE_SIGNED_64:
6804 case INTEGER_TYPE_UNSIGNED_64: return "8";
6806 default: DE_ASSERT(false);
6811 bool isSigned (IntegerType type)
6813 return (type <= INTEGER_TYPE_SIGNED_64);
6816 const string getTypeName (IntegerType type)
6818 string prefix = isSigned(type) ? "" : "u";
6819 return prefix + "int" + getBitWidthStr(type);
6822 const string getTestName (IntegerType from, IntegerType to)
6824 return getTypeName(from) + "_to_" + getTypeName(to);
6827 const string getAsmTypeDeclaration (IntegerType type)
6829 string sign = isSigned(type) ? " 1" : " 0";
6830 return "OpTypeInt " + getBitWidthStr(type) + sign;
6833 template<typename T>
6834 BufferSp getSpecializedBuffer (deInt64 number)
6836 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
6839 BufferSp getBuffer (IntegerType type, deInt64 number)
6843 case INTEGER_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
6844 case INTEGER_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
6845 case INTEGER_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
6847 case INTEGER_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
6848 case INTEGER_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
6849 case INTEGER_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
6851 default: DE_ASSERT(false);
6852 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
6856 bool usesInt16 (IntegerType from, IntegerType to)
6858 return (from == INTEGER_TYPE_SIGNED_16 || from == INTEGER_TYPE_UNSIGNED_16
6859 || to == INTEGER_TYPE_SIGNED_16 || to == INTEGER_TYPE_UNSIGNED_16);
6862 bool usesInt64 (IntegerType from, IntegerType to)
6864 return (from == INTEGER_TYPE_SIGNED_64 || from == INTEGER_TYPE_UNSIGNED_64
6865 || to == INTEGER_TYPE_SIGNED_64 || to == INTEGER_TYPE_UNSIGNED_64);
6868 ComputeTestFeatures getConversionUsedFeatures (IntegerType from, IntegerType to)
6870 if (usesInt16(from, to))
6872 if (usesInt64(from, to))
6874 return COMPUTE_TEST_USES_INT16_INT64;
6878 return COMPUTE_TEST_USES_INT16;
6883 return COMPUTE_TEST_USES_INT64;
6889 ConvertCase (IntegerType from, IntegerType to, deInt64 number)
6892 , m_features (getConversionUsedFeatures(from, to))
6893 , m_name (getTestName(from, to))
6894 , m_inputBuffer (getBuffer(from, number))
6895 , m_outputBuffer (getBuffer(to, number))
6897 m_asmTypes["inputType"] = getAsmTypeDeclaration(from);
6898 m_asmTypes["outputType"] = getAsmTypeDeclaration(to);
6900 if (m_features == COMPUTE_TEST_USES_INT16)
6902 m_asmTypes["int_capabilities"] = "OpCapability Int16\n";
6904 else if (m_features == COMPUTE_TEST_USES_INT64)
6906 m_asmTypes["int_capabilities"] = "OpCapability Int64\n";
6908 else if (m_features == COMPUTE_TEST_USES_INT16_INT64)
6910 m_asmTypes["int_capabilities"] = string("OpCapability Int16\n") +
6911 "OpCapability Int64\n";
6919 IntegerType m_fromType;
6920 IntegerType m_toType;
6921 ComputeTestFeatures m_features;
6923 map<string, string> m_asmTypes;
6924 BufferSp m_inputBuffer;
6925 BufferSp m_outputBuffer;
6928 const string getConvertCaseShaderStr (const string& instruction, const ConvertCase& convertCase)
6930 map<string, string> params = convertCase.m_asmTypes;
6932 params["instruction"] = instruction;
6934 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
6935 params["outDecorator"] = getByteWidthStr(convertCase.m_toType);
6937 const StringTemplate shader (
6938 "OpCapability Shader\n"
6939 "${int_capabilities}"
6940 "OpMemoryModel Logical GLSL450\n"
6941 "OpEntryPoint GLCompute %main \"main\" %id\n"
6942 "OpExecutionMode %main LocalSize 1 1 1\n"
6943 "OpSource GLSL 430\n"
6944 "OpName %main \"main\"\n"
6945 "OpName %id \"gl_GlobalInvocationID\"\n"
6947 "OpDecorate %id BuiltIn GlobalInvocationId\n"
6948 "OpDecorate %indata DescriptorSet 0\n"
6949 "OpDecorate %indata Binding 0\n"
6950 "OpDecorate %outdata DescriptorSet 0\n"
6951 "OpDecorate %outdata Binding 1\n"
6952 "OpDecorate %in_arr ArrayStride ${inDecorator}\n"
6953 "OpDecorate %out_arr ArrayStride ${outDecorator}\n"
6954 "OpDecorate %in_buf BufferBlock\n"
6955 "OpDecorate %out_buf BufferBlock\n"
6956 "OpMemberDecorate %in_buf 0 Offset 0\n"
6957 "OpMemberDecorate %out_buf 0 Offset 0\n"
6959 "%void = OpTypeVoid\n"
6960 "%voidf = OpTypeFunction %void\n"
6961 "%u32 = OpTypeInt 32 0\n"
6962 "%i32 = OpTypeInt 32 1\n"
6963 "%uvec3 = OpTypeVector %u32 3\n"
6964 "%uvec3ptr = OpTypePointer Input %uvec3\n"
6966 "%in_type = ${inputType}\n"
6967 "%out_type = ${outputType}\n"
6969 "%in_ptr = OpTypePointer Uniform %in_type\n"
6970 "%out_ptr = OpTypePointer Uniform %out_type\n"
6971 "%in_arr = OpTypeRuntimeArray %in_type\n"
6972 "%out_arr = OpTypeRuntimeArray %out_type\n"
6973 "%in_buf = OpTypeStruct %in_arr\n"
6974 "%out_buf = OpTypeStruct %out_arr\n"
6975 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
6976 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
6977 "%indata = OpVariable %in_bufptr Uniform\n"
6978 "%outdata = OpVariable %out_bufptr Uniform\n"
6979 "%inputptr = OpTypePointer Input %in_type\n"
6980 "%id = OpVariable %uvec3ptr Input\n"
6982 "%zero = OpConstant %i32 0\n"
6984 "%main = OpFunction %void None %voidf\n"
6985 "%label = OpLabel\n"
6986 "%idval = OpLoad %uvec3 %id\n"
6987 "%x = OpCompositeExtract %u32 %idval 0\n"
6988 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
6989 "%outloc = OpAccessChain %out_ptr %outdata %zero %x\n"
6990 "%inval = OpLoad %in_type %inloc\n"
6991 "%conv = ${instruction} %out_type %inval\n"
6992 " OpStore %outloc %conv\n"
6997 return shader.specialize(params);
7000 void createSConvertCases (vector<ConvertCase>& testCases)
7002 // Convert int to int
7003 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_32, 14669));
7004 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_64, 3341));
7006 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_SIGNED_64, 973610259));
7008 // Convert int to unsigned int
7009 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_32, 9288));
7010 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_64, 15460));
7012 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_UNSIGNED_64, 346213461));
7015 // Test for the OpSConvert instruction.
7016 tcu::TestCaseGroup* createSConvertTests (tcu::TestContext& testCtx)
7018 const string instruction ("OpSConvert");
7019 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "sconvert", "OpSConvert"));
7020 vector<ConvertCase> testCases;
7021 createSConvertCases(testCases);
7023 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7025 ComputeShaderSpec spec;
7027 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7028 spec.inputs.push_back(test->m_inputBuffer);
7029 spec.outputs.push_back(test->m_outputBuffer);
7030 spec.numWorkGroups = IVec3(1, 1, 1);
7032 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpSConvert.", spec, test->m_features));
7035 return group.release();
7038 void createUConvertCases (vector<ConvertCase>& testCases)
7040 // Convert unsigned int to unsigned int
7041 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_32, 60653));
7042 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_64, 17991));
7044 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_UNSIGNED_64, 904256275));
7046 // Convert unsigned int to int
7047 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_32, 38002));
7048 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_64, 64921));
7050 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_SIGNED_64, 4294956295ll));
7053 // Test for the OpUConvert instruction.
7054 tcu::TestCaseGroup* createUConvertTests (tcu::TestContext& testCtx)
7056 const string instruction ("OpUConvert");
7057 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "uconvert", "OpUConvert"));
7058 vector<ConvertCase> testCases;
7059 createUConvertCases(testCases);
7061 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7063 ComputeShaderSpec spec;
7065 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7066 spec.inputs.push_back(test->m_inputBuffer);
7067 spec.outputs.push_back(test->m_outputBuffer);
7068 spec.numWorkGroups = IVec3(1, 1, 1);
7070 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpUConvert.", spec, test->m_features));
7072 return group.release();
7075 const string getNumberTypeName (const NumberType type)
7077 if (type == NUMBERTYPE_INT32)
7081 else if (type == NUMBERTYPE_UINT32)
7085 else if (type == NUMBERTYPE_FLOAT32)
7096 deInt32 getInt(de::Random& rnd)
7098 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
7101 const string repeatString (const string& str, int times)
7104 for (int i = 0; i < times; ++i)
7111 const string getRandomConstantString (const NumberType type, de::Random& rnd)
7113 if (type == NUMBERTYPE_INT32)
7115 return numberToString<deInt32>(getInt(rnd));
7117 else if (type == NUMBERTYPE_UINT32)
7119 return numberToString<deUint32>(rnd.getUint32());
7121 else if (type == NUMBERTYPE_FLOAT32)
7123 return numberToString<float>(rnd.getFloat());
7132 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7134 map<string, string> params;
7137 for (int width = 2; width <= 4; ++width)
7139 string randomConst = numberToString(getInt(rnd));
7140 string widthStr = numberToString(width);
7141 int index = rnd.getInt(0, width-1);
7143 params["type"] = "vec";
7144 params["name"] = params["type"] + "_" + widthStr;
7145 params["compositeType"] = "%composite = OpTypeVector %custom " + widthStr +"\n";
7146 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7147 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7148 params["indexes"] = numberToString(index);
7149 testCases.push_back(params);
7153 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7155 const int limit = 10;
7156 map<string, string> params;
7158 for (int width = 2; width <= limit; ++width)
7160 string randomConst = numberToString(getInt(rnd));
7161 string widthStr = numberToString(width);
7162 int index = rnd.getInt(0, width-1);
7164 params["type"] = "array";
7165 params["name"] = params["type"] + "_" + widthStr;
7166 params["compositeType"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
7167 + "%composite = OpTypeArray %custom %arraywidth\n";
7169 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7170 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7171 params["indexes"] = numberToString(index);
7172 testCases.push_back(params);
7176 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7178 const int limit = 10;
7179 map<string, string> params;
7181 for (int width = 2; width <= limit; ++width)
7183 string randomConst = numberToString(getInt(rnd));
7184 int index = rnd.getInt(0, width-1);
7186 params["type"] = "struct";
7187 params["name"] = params["type"] + "_" + numberToString(width);
7188 params["compositeType"] = "%composite = OpTypeStruct" + repeatString(" %custom", width) + "\n";
7189 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7190 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7191 params["indexes"] = numberToString(index);
7192 testCases.push_back(params);
7196 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7198 map<string, string> params;
7201 for (int width = 2; width <= 4; ++width)
7203 string widthStr = numberToString(width);
7205 for (int column = 2 ; column <= 4; ++column)
7207 int index_0 = rnd.getInt(0, column-1);
7208 int index_1 = rnd.getInt(0, width-1);
7209 string columnStr = numberToString(column);
7211 params["type"] = "matrix";
7212 params["name"] = params["type"] + "_" + widthStr + "x" + columnStr;
7213 params["compositeType"] = string("%vectype = OpTypeVector %custom " + widthStr + "\n")
7214 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
7216 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n"
7217 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
7219 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
7220 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
7221 testCases.push_back(params);
7226 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7228 createVectorCompositeCases(testCases, rnd, type);
7229 createArrayCompositeCases(testCases, rnd, type);
7230 createStructCompositeCases(testCases, rnd, type);
7231 // Matrix only supports float types
7232 if (type == NUMBERTYPE_FLOAT32)
7234 createMatrixCompositeCases(testCases, rnd, type);
7238 const string getAssemblyTypeDeclaration (const NumberType type)
7242 case NUMBERTYPE_INT32: return "OpTypeInt 32 1";
7243 case NUMBERTYPE_UINT32: return "OpTypeInt 32 0";
7244 case NUMBERTYPE_FLOAT32: return "OpTypeFloat 32";
7245 default: DE_ASSERT(false); return "";
7249 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
7251 map<string, string> parameters(params);
7253 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7255 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
7257 return StringTemplate (
7258 "OpCapability Shader\n"
7259 "OpCapability Matrix\n"
7260 "OpMemoryModel Logical GLSL450\n"
7261 "OpEntryPoint GLCompute %main \"main\" %id\n"
7262 "OpExecutionMode %main LocalSize 1 1 1\n"
7264 "OpSource GLSL 430\n"
7265 "OpName %main \"main\"\n"
7266 "OpName %id \"gl_GlobalInvocationID\"\n"
7269 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7270 "OpDecorate %buf BufferBlock\n"
7271 "OpDecorate %indata DescriptorSet 0\n"
7272 "OpDecorate %indata Binding 0\n"
7273 "OpDecorate %outdata DescriptorSet 0\n"
7274 "OpDecorate %outdata Binding 1\n"
7275 "OpDecorate %customarr ArrayStride 4\n"
7276 "${compositeDecorator}"
7277 "OpMemberDecorate %buf 0 Offset 0\n"
7280 "%void = OpTypeVoid\n"
7281 "%voidf = OpTypeFunction %void\n"
7282 "%u32 = OpTypeInt 32 0\n"
7283 "%i32 = OpTypeInt 32 1\n"
7284 "%uvec3 = OpTypeVector %u32 3\n"
7285 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7288 "%custom = ${typeDeclaration}\n"
7294 // Inherited from custom
7295 "%customptr = OpTypePointer Uniform %custom\n"
7296 "%customarr = OpTypeRuntimeArray %custom\n"
7297 "%buf = OpTypeStruct %customarr\n"
7298 "%bufptr = OpTypePointer Uniform %buf\n"
7300 "%indata = OpVariable %bufptr Uniform\n"
7301 "%outdata = OpVariable %bufptr Uniform\n"
7303 "%id = OpVariable %uvec3ptr Input\n"
7304 "%zero = OpConstant %i32 0\n"
7306 "%main = OpFunction %void None %voidf\n"
7307 "%label = OpLabel\n"
7308 "%idval = OpLoad %uvec3 %id\n"
7309 "%x = OpCompositeExtract %u32 %idval 0\n"
7311 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
7312 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
7313 // Read the input value
7314 "%inval = OpLoad %custom %inloc\n"
7315 // Create the composite and fill it
7316 "${compositeConstruct}"
7317 // Insert the input value to a place
7318 "%instance2 = OpCompositeInsert %composite %inval %instance ${indexes}\n"
7319 // Read back the value from the position
7320 "%out_val = OpCompositeExtract %custom %instance2 ${indexes}\n"
7321 // Store it in the output position
7322 " OpStore %outloc %out_val\n"
7325 ).specialize(parameters);
7328 template<typename T>
7329 BufferSp createCompositeBuffer(T number)
7331 return BufferSp(new Buffer<T>(vector<T>(1, number)));
7334 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
7336 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
7337 de::Random rnd (deStringHash(group->getName()));
7339 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7341 NumberType numberType = NumberType(type);
7342 const string typeName = getNumberTypeName(numberType);
7343 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
7344 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7345 vector<map<string, string> > testCases;
7347 createCompositeCases(testCases, rnd, numberType);
7349 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7351 ComputeShaderSpec spec;
7353 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
7357 case NUMBERTYPE_INT32:
7359 deInt32 number = getInt(rnd);
7360 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7361 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7364 case NUMBERTYPE_UINT32:
7366 deUint32 number = rnd.getUint32();
7367 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7368 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7371 case NUMBERTYPE_FLOAT32:
7373 float number = rnd.getFloat();
7374 spec.inputs.push_back(createCompositeBuffer<float>(number));
7375 spec.outputs.push_back(createCompositeBuffer<float>(number));
7382 spec.numWorkGroups = IVec3(1, 1, 1);
7383 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
7385 group->addChild(subGroup.release());
7387 return group.release();
7390 struct AssemblyStructInfo
7392 AssemblyStructInfo (const deUint32 comp, const deUint32 idx)
7397 deUint32 components;
7401 const string specializeInBoundsShaderTemplate (const NumberType type, const AssemblyStructInfo& structInfo, const map<string, string>& params)
7403 // Create the full index string
7404 string fullIndex = numberToString(structInfo.index) + " " + params.at("indexes");
7405 // Convert it to list of indexes
7406 vector<string> indexes = de::splitString(fullIndex, ' ');
7408 map<string, string> parameters (params);
7409 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7410 parameters["structType"] = repeatString(" %composite", structInfo.components);
7411 parameters["structConstruct"] = repeatString(" %instance", structInfo.components);
7412 parameters["insertIndexes"] = fullIndex;
7414 // In matrix cases the last two index is the CompositeExtract indexes
7415 const deUint32 extractIndexes = (parameters["type"] == "matrix") ? 2 : 1;
7417 // Construct the extractIndex
7418 for (vector<string>::const_iterator index = indexes.end() - extractIndexes; index != indexes.end(); ++index)
7420 parameters["extractIndexes"] += " " + *index;
7423 // Remove the last 1 or 2 element depends on matrix case or not
7424 indexes.erase(indexes.end() - extractIndexes, indexes.end());
7427 // Generate AccessChain index expressions (except for the last one, because we use ptr to the composite)
7428 for (vector<string>::const_iterator index = indexes.begin(); index != indexes.end(); ++index)
7430 string indexId = "%index_" + numberToString(id++);
7431 parameters["accessChainConstDeclaration"] += indexId + " = OpConstant %u32 " + *index + "\n";
7432 parameters["accessChainIndexes"] += " " + indexId;
7435 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
7437 return StringTemplate (
7438 "OpCapability Shader\n"
7439 "OpCapability Matrix\n"
7440 "OpMemoryModel Logical GLSL450\n"
7441 "OpEntryPoint GLCompute %main \"main\" %id\n"
7442 "OpExecutionMode %main LocalSize 1 1 1\n"
7444 "OpSource GLSL 430\n"
7445 "OpName %main \"main\"\n"
7446 "OpName %id \"gl_GlobalInvocationID\"\n"
7448 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7449 "OpDecorate %buf BufferBlock\n"
7450 "OpDecorate %indata DescriptorSet 0\n"
7451 "OpDecorate %indata Binding 0\n"
7452 "OpDecorate %outdata DescriptorSet 0\n"
7453 "OpDecorate %outdata Binding 1\n"
7454 "OpDecorate %customarr ArrayStride 4\n"
7455 "${compositeDecorator}"
7456 "OpMemberDecorate %buf 0 Offset 0\n"
7458 "%void = OpTypeVoid\n"
7459 "%voidf = OpTypeFunction %void\n"
7460 "%u32 = OpTypeInt 32 0\n"
7461 "%uvec3 = OpTypeVector %u32 3\n"
7462 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7464 "%custom = ${typeDeclaration}\n"
7467 // Inherited from composite
7468 "%composite_p = OpTypePointer Function %composite\n"
7469 "%struct_t = OpTypeStruct${structType}\n"
7470 "%struct_p = OpTypePointer Function %struct_t\n"
7473 "${accessChainConstDeclaration}"
7474 // Inherited from custom
7475 "%customptr = OpTypePointer Uniform %custom\n"
7476 "%customarr = OpTypeRuntimeArray %custom\n"
7477 "%buf = OpTypeStruct %customarr\n"
7478 "%bufptr = OpTypePointer Uniform %buf\n"
7479 "%indata = OpVariable %bufptr Uniform\n"
7480 "%outdata = OpVariable %bufptr Uniform\n"
7482 "%id = OpVariable %uvec3ptr Input\n"
7483 "%zero = OpConstant %u32 0\n"
7484 "%main = OpFunction %void None %voidf\n"
7485 "%label = OpLabel\n"
7486 "%struct_v = OpVariable %struct_p Function\n"
7487 "%idval = OpLoad %uvec3 %id\n"
7488 "%x = OpCompositeExtract %u32 %idval 0\n"
7489 // Create the input/output type
7490 "%inloc = OpInBoundsAccessChain %customptr %indata %zero %x\n"
7491 "%outloc = OpInBoundsAccessChain %customptr %outdata %zero %x\n"
7492 // Read the input value
7493 "%inval = OpLoad %custom %inloc\n"
7494 // Create the composite and fill it
7495 "${compositeConstruct}"
7496 // Create the struct and fill it with the composite
7497 "%struct = OpCompositeConstruct %struct_t${structConstruct}\n"
7499 "%comp_obj = OpCompositeInsert %struct_t %inval %struct ${insertIndexes}\n"
7501 " OpStore %struct_v %comp_obj\n"
7502 // Get deepest possible composite pointer
7503 "%inner_ptr = OpInBoundsAccessChain %composite_p %struct_v${accessChainIndexes}\n"
7504 "%read_obj = OpLoad %composite %inner_ptr\n"
7505 // Read back the stored value
7506 "%read_val = OpCompositeExtract %custom %read_obj${extractIndexes}\n"
7507 " OpStore %outloc %read_val\n"
7509 " OpFunctionEnd\n").specialize(parameters);
7512 tcu::TestCaseGroup* createOpInBoundsAccessChainGroup (tcu::TestContext& testCtx)
7514 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opinboundsaccesschain", "Test the OpInBoundsAccessChain instruction"));
7515 de::Random rnd (deStringHash(group->getName()));
7517 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7519 NumberType numberType = NumberType(type);
7520 const string typeName = getNumberTypeName(numberType);
7521 const string description = "Test the OpInBoundsAccessChain instruction with " + typeName + "s";
7522 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7524 vector<map<string, string> > testCases;
7525 createCompositeCases(testCases, rnd, numberType);
7527 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7529 ComputeShaderSpec spec;
7531 // Number of components inside of a struct
7532 deUint32 structComponents = rnd.getInt(2, 8);
7533 // Component index value
7534 deUint32 structIndex = rnd.getInt(0, structComponents - 1);
7535 AssemblyStructInfo structInfo(structComponents, structIndex);
7537 spec.assembly = specializeInBoundsShaderTemplate(numberType, structInfo, *test);
7541 case NUMBERTYPE_INT32:
7543 deInt32 number = getInt(rnd);
7544 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7545 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7548 case NUMBERTYPE_UINT32:
7550 deUint32 number = rnd.getUint32();
7551 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7552 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7555 case NUMBERTYPE_FLOAT32:
7557 float number = rnd.getFloat();
7558 spec.inputs.push_back(createCompositeBuffer<float>(number));
7559 spec.outputs.push_back(createCompositeBuffer<float>(number));
7565 spec.numWorkGroups = IVec3(1, 1, 1);
7566 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpInBoundsAccessChain test", spec));
7568 group->addChild(subGroup.release());
7570 return group.release();
7573 // If the params missing, uninitialized case
7574 const string specializeDefaultOutputShaderTemplate (const NumberType type, const map<string, string>& params = map<string, string>())
7576 map<string, string> parameters(params);
7578 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7580 // Declare the const value, and use it in the initializer
7581 if (params.find("constValue") != params.end())
7583 parameters["constDeclaration"] = "%const = OpConstant %in_type " + params.at("constValue") + "\n";
7584 parameters["variableInitializer"] = "%const";
7586 // Uninitialized case
7589 parameters["constDeclaration"] = "";
7590 parameters["variableInitializer"] = "";
7593 return StringTemplate(
7594 "OpCapability Shader\n"
7595 "OpMemoryModel Logical GLSL450\n"
7596 "OpEntryPoint GLCompute %main \"main\" %id\n"
7597 "OpExecutionMode %main LocalSize 1 1 1\n"
7598 "OpSource GLSL 430\n"
7599 "OpName %main \"main\"\n"
7600 "OpName %id \"gl_GlobalInvocationID\"\n"
7602 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7603 "OpDecorate %indata DescriptorSet 0\n"
7604 "OpDecorate %indata Binding 0\n"
7605 "OpDecorate %outdata DescriptorSet 0\n"
7606 "OpDecorate %outdata Binding 1\n"
7607 "OpDecorate %in_arr ArrayStride 4\n"
7608 "OpDecorate %in_buf BufferBlock\n"
7609 "OpMemberDecorate %in_buf 0 Offset 0\n"
7611 "%void = OpTypeVoid\n"
7612 "%voidf = OpTypeFunction %void\n"
7613 "%u32 = OpTypeInt 32 0\n"
7614 "%i32 = OpTypeInt 32 1\n"
7615 "%uvec3 = OpTypeVector %u32 3\n"
7616 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7618 "%in_type = ${typeDeclaration}\n"
7619 // "%const = OpConstant %in_type ${constValue}\n"
7620 "${constDeclaration}\n"
7622 "%in_ptr = OpTypePointer Uniform %in_type\n"
7623 "%in_arr = OpTypeRuntimeArray %in_type\n"
7624 "%in_buf = OpTypeStruct %in_arr\n"
7625 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
7626 "%indata = OpVariable %in_bufptr Uniform\n"
7627 "%outdata = OpVariable %in_bufptr Uniform\n"
7628 "%id = OpVariable %uvec3ptr Input\n"
7629 "%var_ptr = OpTypePointer Function %in_type\n"
7631 "%zero = OpConstant %i32 0\n"
7633 "%main = OpFunction %void None %voidf\n"
7634 "%label = OpLabel\n"
7635 "%out_var = OpVariable %var_ptr Function ${variableInitializer}\n"
7636 "%idval = OpLoad %uvec3 %id\n"
7637 "%x = OpCompositeExtract %u32 %idval 0\n"
7638 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
7639 "%outloc = OpAccessChain %in_ptr %outdata %zero %x\n"
7641 "%outval = OpLoad %in_type %out_var\n"
7642 " OpStore %outloc %outval\n"
7645 ).specialize(parameters);
7648 bool compareFloats (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
7650 DE_ASSERT(outputAllocs.size() != 0);
7651 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
7653 // Use custom epsilon because of the float->string conversion
7654 const float epsilon = 0.00001f;
7656 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
7659 memcpy(&expected, expectedOutputs[outputNdx]->data(), expectedOutputs[outputNdx]->getNumBytes());
7662 memcpy(&actual, outputAllocs[outputNdx]->getHostPtr(), expectedOutputs[outputNdx]->getNumBytes());
7664 // Test with epsilon
7665 if (fabs(expected - actual) > epsilon)
7667 log << TestLog::Message << "Error: The actual and expected values not matching."
7668 << " Expected: " << expected << " Actual: " << actual << " Epsilon: " << epsilon << TestLog::EndMessage;
7675 // Checks if the driver crash with uninitialized cases
7676 bool passthruVerify (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
7678 DE_ASSERT(outputAllocs.size() != 0);
7679 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
7681 // Copy and discard the result.
7682 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
7684 size_t width = expectedOutputs[outputNdx]->getNumBytes();
7686 vector<char> data(width);
7687 memcpy(&data[0], outputAllocs[outputNdx]->getHostPtr(), width);
7692 tcu::TestCaseGroup* createShaderDefaultOutputGroup (tcu::TestContext& testCtx)
7694 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "shader_default_output", "Test shader default output."));
7695 de::Random rnd (deStringHash(group->getName()));
7697 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7699 NumberType numberType = NumberType(type);
7700 const string typeName = getNumberTypeName(numberType);
7701 const string description = "Test the OpVariable initializer with " + typeName + ".";
7702 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7704 // 2 similar subcases (initialized and uninitialized)
7705 for (int subCase = 0; subCase < 2; ++subCase)
7707 ComputeShaderSpec spec;
7708 spec.numWorkGroups = IVec3(1, 1, 1);
7710 map<string, string> params;
7714 case NUMBERTYPE_INT32:
7716 deInt32 number = getInt(rnd);
7717 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7718 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7719 params["constValue"] = numberToString(number);
7722 case NUMBERTYPE_UINT32:
7724 deUint32 number = rnd.getUint32();
7725 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7726 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7727 params["constValue"] = numberToString(number);
7730 case NUMBERTYPE_FLOAT32:
7732 float number = rnd.getFloat();
7733 spec.inputs.push_back(createCompositeBuffer<float>(number));
7734 spec.outputs.push_back(createCompositeBuffer<float>(number));
7735 spec.verifyIO = &compareFloats;
7736 params["constValue"] = numberToString(number);
7743 // Initialized subcase
7746 spec.assembly = specializeDefaultOutputShaderTemplate(numberType, params);
7747 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "initialized", "OpVariable initializer tests.", spec));
7749 // Uninitialized subcase
7752 spec.assembly = specializeDefaultOutputShaderTemplate(numberType);
7753 spec.verifyIO = &passthruVerify;
7754 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "uninitialized", "OpVariable initializer tests.", spec));
7757 group->addChild(subGroup.release());
7759 return group.release();
7762 tcu::TestCaseGroup* createOpNopTests (tcu::TestContext& testCtx)
7764 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
7765 RGBA defaultColors[4];
7766 map<string, string> opNopFragments;
7768 getDefaultColors(defaultColors);
7770 opNopFragments["testfun"] =
7771 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7772 "%param1 = OpFunctionParameter %v4f32\n"
7773 "%label_testfun = OpLabel\n"
7782 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7783 "%b = OpFAdd %f32 %a %a\n"
7785 "%c = OpFSub %f32 %b %a\n"
7786 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
7789 "OpReturnValue %ret\n"
7792 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, testGroup.get());
7794 return testGroup.release();
7797 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
7799 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
7800 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
7801 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
7803 computeTests->addChild(createOpNopGroup(testCtx));
7804 computeTests->addChild(createOpFUnordGroup(testCtx));
7805 computeTests->addChild(createOpAtomicGroup(testCtx, false));
7806 computeTests->addChild(createOpAtomicGroup(testCtx, true)); // Using new StorageBuffer decoration
7807 computeTests->addChild(createOpLineGroup(testCtx));
7808 computeTests->addChild(createOpNoLineGroup(testCtx));
7809 computeTests->addChild(createOpConstantNullGroup(testCtx));
7810 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
7811 computeTests->addChild(createOpConstantUsageGroup(testCtx));
7812 computeTests->addChild(createSpecConstantGroup(testCtx));
7813 computeTests->addChild(createOpSourceGroup(testCtx));
7814 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
7815 computeTests->addChild(createDecorationGroupGroup(testCtx));
7816 computeTests->addChild(createOpPhiGroup(testCtx));
7817 computeTests->addChild(createLoopControlGroup(testCtx));
7818 computeTests->addChild(createFunctionControlGroup(testCtx));
7819 computeTests->addChild(createSelectionControlGroup(testCtx));
7820 computeTests->addChild(createBlockOrderGroup(testCtx));
7821 computeTests->addChild(createMultipleShaderGroup(testCtx));
7822 computeTests->addChild(createMemoryAccessGroup(testCtx));
7823 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
7824 computeTests->addChild(createOpCopyObjectGroup(testCtx));
7825 computeTests->addChild(createNoContractionGroup(testCtx));
7826 computeTests->addChild(createOpUndefGroup(testCtx));
7827 computeTests->addChild(createOpUnreachableGroup(testCtx));
7828 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
7829 computeTests ->addChild(createOpFRemGroup(testCtx));
7830 computeTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_PASS));
7831 computeTests->addChild(createOpSRemComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
7832 computeTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_PASS));
7833 computeTests->addChild(createOpSModComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
7834 computeTests->addChild(createSConvertTests(testCtx));
7835 computeTests->addChild(createUConvertTests(testCtx));
7836 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
7837 computeTests->addChild(createOpInBoundsAccessChainGroup(testCtx));
7838 computeTests->addChild(createShaderDefaultOutputGroup(testCtx));
7839 computeTests->addChild(createOpNMinGroup(testCtx));
7840 computeTests->addChild(createOpNMaxGroup(testCtx));
7841 computeTests->addChild(createOpNClampGroup(testCtx));
7843 de::MovePtr<tcu::TestCaseGroup> computeAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
7845 computeAndroidTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7846 computeAndroidTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7848 computeTests->addChild(computeAndroidTests.release());
7851 computeTests->addChild(create16BitStorageComputeGroup(testCtx));
7852 computeTests->addChild(createVariablePointersComputeGroup(testCtx));
7853 graphicsTests->addChild(createOpNopTests(testCtx));
7854 graphicsTests->addChild(createOpSourceTests(testCtx));
7855 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
7856 graphicsTests->addChild(createOpLineTests(testCtx));
7857 graphicsTests->addChild(createOpNoLineTests(testCtx));
7858 graphicsTests->addChild(createOpConstantNullTests(testCtx));
7859 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
7860 graphicsTests->addChild(createMemoryAccessTests(testCtx));
7861 graphicsTests->addChild(createOpUndefTests(testCtx));
7862 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
7863 graphicsTests->addChild(createModuleTests(testCtx));
7864 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
7865 graphicsTests->addChild(createOpPhiTests(testCtx));
7866 graphicsTests->addChild(createNoContractionTests(testCtx));
7867 graphicsTests->addChild(createOpQuantizeTests(testCtx));
7868 graphicsTests->addChild(createLoopTests(testCtx));
7869 graphicsTests->addChild(createSpecConstantTests(testCtx));
7870 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
7871 graphicsTests->addChild(createBarrierTests(testCtx));
7872 graphicsTests->addChild(createDecorationGroupTests(testCtx));
7873 graphicsTests->addChild(createFRemTests(testCtx));
7874 graphicsTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
7875 graphicsTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
7878 de::MovePtr<tcu::TestCaseGroup> graphicsAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
7880 graphicsAndroidTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7881 graphicsAndroidTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7883 graphicsTests->addChild(graphicsAndroidTests.release());
7886 graphicsTests->addChild(create16BitStorageGraphicsGroup(testCtx));
7887 graphicsTests->addChild(createVariablePointersGraphicsGroup(testCtx));
7889 instructionTests->addChild(computeTests.release());
7890 instructionTests->addChild(graphicsTests.release());
7892 return instructionTests.release();