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 "vktSpvAsmUboMatrixPaddingTests.hpp"
55 #include "vktSpvAsmConditionalBranchTests.hpp"
56 #include "vktSpvAsmComputeShaderCase.hpp"
57 #include "vktSpvAsmComputeShaderTestUtil.hpp"
58 #include "vktSpvAsmGraphicsShaderTestUtil.hpp"
59 #include "vktSpvAsmVariablePointersTests.hpp"
60 #include "vktTestCaseUtil.hpp"
71 namespace SpirVAssembly
85 using tcu::TestStatus;
88 using tcu::StringTemplate;
92 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0)
94 T* const typedPtr = (T*)dst;
95 for (int ndx = 0; ndx < numValues; ndx++)
96 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue);
99 // Filter is a function that returns true if a value should pass, false otherwise.
100 template<typename T, typename FilterT>
101 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, FilterT filter, int offset = 0)
103 T* const typedPtr = (T*)dst;
105 for (int ndx = 0; ndx < numValues; ndx++)
108 value = randomScalar<T>(rnd, minValue, maxValue);
109 while (!filter(value));
111 typedPtr[offset + ndx] = value;
115 // Gets a 64-bit integer with a more logarithmic distribution
116 deInt64 randomInt64LogDistributed (de::Random& rnd)
118 deInt64 val = rnd.getUint64();
119 val &= (1ull << rnd.getInt(1, 63)) - 1;
125 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues)
127 for (int ndx = 0; ndx < numValues; ndx++)
128 dst[ndx] = randomInt64LogDistributed(rnd);
131 template<typename FilterT>
132 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues, FilterT filter)
134 for (int ndx = 0; ndx < numValues; ndx++)
138 value = randomInt64LogDistributed(rnd);
139 } while (!filter(value));
144 inline bool filterNonNegative (const deInt64 value)
149 inline bool filterPositive (const deInt64 value)
154 inline bool filterNotZero (const deInt64 value)
159 static void floorAll (vector<float>& values)
161 for (size_t i = 0; i < values.size(); i++)
162 values[i] = deFloatFloor(values[i]);
165 static void floorAll (vector<Vec4>& values)
167 for (size_t i = 0; i < values.size(); i++)
168 values[i] = floor(values[i]);
176 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {}
179 // Assembly code used for testing OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code:
183 // layout(std140, set = 0, binding = 0) readonly buffer Input {
186 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
190 // layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
193 // uint x = gl_GlobalInvocationID.x;
194 // output_data.elements[x] = -input_data.elements[x];
197 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
199 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
200 ComputeShaderSpec spec;
201 de::Random rnd (deStringHash(group->getName()));
202 const int numElements = 100;
203 vector<float> positiveFloats (numElements, 0);
204 vector<float> negativeFloats (numElements, 0);
206 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
208 for (size_t ndx = 0; ndx < numElements; ++ndx)
209 negativeFloats[ndx] = -positiveFloats[ndx];
212 string(getComputeAsmShaderPreamble()) +
214 "OpSource GLSL 430\n"
215 "OpName %main \"main\"\n"
216 "OpName %id \"gl_GlobalInvocationID\"\n"
218 "OpDecorate %id BuiltIn GlobalInvocationId\n"
220 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes())
222 + string(getComputeAsmInputOutputBuffer()) +
224 "%id = OpVariable %uvec3ptr Input\n"
225 "%zero = OpConstant %i32 0\n"
227 "%main = OpFunction %void None %voidf\n"
229 "%idval = OpLoad %uvec3 %id\n"
230 "%x = OpCompositeExtract %u32 %idval 0\n"
232 " OpNop\n" // Inside a function body
234 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
235 "%inval = OpLoad %f32 %inloc\n"
236 "%neg = OpFNegate %f32 %inval\n"
237 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
238 " OpStore %outloc %neg\n"
241 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
242 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
243 spec.numWorkGroups = IVec3(numElements, 1, 1);
245 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
247 return group.release();
250 bool compareFUnord (const std::vector<BufferSp>& inputs, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
252 if (outputAllocs.size() != 1)
255 const BufferSp& expectedOutput = expectedOutputs[0];
256 const deInt32* expectedOutputAsInt = static_cast<const deInt32*>(expectedOutputs[0]->data());
257 const deInt32* outputAsInt = static_cast<const deInt32*>(outputAllocs[0]->getHostPtr());
258 const float* input1AsFloat = static_cast<const float*>(inputs[0]->data());
259 const float* input2AsFloat = static_cast<const float*>(inputs[1]->data());
260 bool returnValue = true;
262 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(deInt32); ++idx)
264 if (outputAsInt[idx] != expectedOutputAsInt[idx])
266 log << TestLog::Message << "ERROR: Sub-case failed. inputs: " << input1AsFloat[idx] << "," << input2AsFloat[idx] << " output: " << outputAsInt[idx]<< " expected output: " << expectedOutputAsInt[idx] << TestLog::EndMessage;
273 typedef VkBool32 (*compareFuncType) (float, float);
279 compareFuncType compareFunc;
281 OpFUnordCase (const char* _name, const char* _opCode, compareFuncType _compareFunc)
284 , compareFunc (_compareFunc) {}
287 #define ADD_OPFUNORD_CASE(NAME, OPCODE, OPERATOR) \
289 struct compare_##NAME { static VkBool32 compare(float x, float y) { return (x OPERATOR y) ? VK_TRUE : VK_FALSE; } }; \
290 cases.push_back(OpFUnordCase(#NAME, OPCODE, compare_##NAME::compare)); \
291 } while (deGetFalse())
293 tcu::TestCaseGroup* createOpFUnordGroup (tcu::TestContext& testCtx)
295 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfunord", "Test the OpFUnord* opcodes"));
296 de::Random rnd (deStringHash(group->getName()));
297 const int numElements = 100;
298 vector<OpFUnordCase> cases;
300 const StringTemplate shaderTemplate (
302 string(getComputeAsmShaderPreamble()) +
304 "OpSource GLSL 430\n"
305 "OpName %main \"main\"\n"
306 "OpName %id \"gl_GlobalInvocationID\"\n"
308 "OpDecorate %id BuiltIn GlobalInvocationId\n"
310 "OpDecorate %buf BufferBlock\n"
311 "OpDecorate %buf2 BufferBlock\n"
312 "OpDecorate %indata1 DescriptorSet 0\n"
313 "OpDecorate %indata1 Binding 0\n"
314 "OpDecorate %indata2 DescriptorSet 0\n"
315 "OpDecorate %indata2 Binding 1\n"
316 "OpDecorate %outdata DescriptorSet 0\n"
317 "OpDecorate %outdata Binding 2\n"
318 "OpDecorate %f32arr ArrayStride 4\n"
319 "OpDecorate %i32arr ArrayStride 4\n"
320 "OpMemberDecorate %buf 0 Offset 0\n"
321 "OpMemberDecorate %buf2 0 Offset 0\n"
323 + string(getComputeAsmCommonTypes()) +
325 "%buf = OpTypeStruct %f32arr\n"
326 "%bufptr = OpTypePointer Uniform %buf\n"
327 "%indata1 = OpVariable %bufptr Uniform\n"
328 "%indata2 = OpVariable %bufptr Uniform\n"
330 "%buf2 = OpTypeStruct %i32arr\n"
331 "%buf2ptr = OpTypePointer Uniform %buf2\n"
332 "%outdata = OpVariable %buf2ptr Uniform\n"
334 "%id = OpVariable %uvec3ptr Input\n"
335 "%zero = OpConstant %i32 0\n"
336 "%consti1 = OpConstant %i32 1\n"
337 "%constf1 = OpConstant %f32 1.0\n"
339 "%main = OpFunction %void None %voidf\n"
341 "%idval = OpLoad %uvec3 %id\n"
342 "%x = OpCompositeExtract %u32 %idval 0\n"
344 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
345 "%inval1 = OpLoad %f32 %inloc1\n"
346 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
347 "%inval2 = OpLoad %f32 %inloc2\n"
348 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
350 "%result = ${OPCODE} %bool %inval1 %inval2\n"
351 "%int_res = OpSelect %i32 %result %consti1 %zero\n"
352 " OpStore %outloc %int_res\n"
357 ADD_OPFUNORD_CASE(equal, "OpFUnordEqual", ==);
358 ADD_OPFUNORD_CASE(less, "OpFUnordLessThan", <);
359 ADD_OPFUNORD_CASE(lessequal, "OpFUnordLessThanEqual", <=);
360 ADD_OPFUNORD_CASE(greater, "OpFUnordGreaterThan", >);
361 ADD_OPFUNORD_CASE(greaterequal, "OpFUnordGreaterThanEqual", >=);
362 ADD_OPFUNORD_CASE(notequal, "OpFUnordNotEqual", !=);
364 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
366 map<string, string> specializations;
367 ComputeShaderSpec spec;
368 const float NaN = std::numeric_limits<float>::quiet_NaN();
369 vector<float> inputFloats1 (numElements, 0);
370 vector<float> inputFloats2 (numElements, 0);
371 vector<deInt32> expectedInts (numElements, 0);
373 specializations["OPCODE"] = cases[caseNdx].opCode;
374 spec.assembly = shaderTemplate.specialize(specializations);
376 fillRandomScalars(rnd, 1.f, 100.f, &inputFloats1[0], numElements);
377 for (size_t ndx = 0; ndx < numElements; ++ndx)
381 case 0: inputFloats2[ndx] = inputFloats1[ndx] + 1.0f; break;
382 case 1: inputFloats2[ndx] = inputFloats1[ndx] - 1.0f; break;
383 case 2: inputFloats2[ndx] = inputFloats1[ndx]; break;
384 case 3: inputFloats2[ndx] = NaN; break;
385 case 4: inputFloats2[ndx] = inputFloats1[ndx]; inputFloats1[ndx] = NaN; break;
386 case 5: inputFloats2[ndx] = NaN; inputFloats1[ndx] = NaN; break;
388 expectedInts[ndx] = tcu::Float32(inputFloats1[ndx]).isNaN() || tcu::Float32(inputFloats2[ndx]).isNaN() || cases[caseNdx].compareFunc(inputFloats1[ndx], inputFloats2[ndx]);
391 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
392 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
393 spec.outputs.push_back(BufferSp(new Int32Buffer(expectedInts)));
394 spec.numWorkGroups = IVec3(numElements, 1, 1);
395 spec.verifyIO = &compareFUnord;
396 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
399 return group.release();
405 const char* assembly;
406 void (*calculateExpected)(deInt32&, deInt32);
407 deInt32 numOutputElements;
409 OpAtomicCase (const char* _name, const char* _assembly, void (*_calculateExpected)(deInt32&, deInt32), deInt32 _numOutputElements)
411 , assembly (_assembly)
412 , calculateExpected (_calculateExpected)
413 , numOutputElements (_numOutputElements) {}
416 tcu::TestCaseGroup* createOpAtomicGroup (tcu::TestContext& testCtx, bool useStorageBuffer)
418 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx,
419 useStorageBuffer ? "opatomic_storage_buffer" : "opatomic",
420 "Test the OpAtomic* opcodes"));
421 de::Random rnd (deStringHash(group->getName()));
422 const int numElements = 65535;
423 vector<OpAtomicCase> cases;
425 const StringTemplate shaderTemplate (
427 string("OpCapability Shader\n") +
428 (useStorageBuffer ? "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" : "") +
429 "OpMemoryModel Logical GLSL450\n"
430 "OpEntryPoint GLCompute %main \"main\" %id\n"
431 "OpExecutionMode %main LocalSize 1 1 1\n" +
433 "OpSource GLSL 430\n"
434 "OpName %main \"main\"\n"
435 "OpName %id \"gl_GlobalInvocationID\"\n"
437 "OpDecorate %id BuiltIn GlobalInvocationId\n"
439 "OpDecorate %buf ${BLOCK_DECORATION}\n"
440 "OpDecorate %indata DescriptorSet 0\n"
441 "OpDecorate %indata Binding 0\n"
442 "OpDecorate %i32arr ArrayStride 4\n"
443 "OpMemberDecorate %buf 0 Offset 0\n"
445 "OpDecorate %sumbuf ${BLOCK_DECORATION}\n"
446 "OpDecorate %sum DescriptorSet 0\n"
447 "OpDecorate %sum Binding 1\n"
448 "OpMemberDecorate %sumbuf 0 Coherent\n"
449 "OpMemberDecorate %sumbuf 0 Offset 0\n"
451 + getComputeAsmCommonTypes("${BLOCK_POINTER_TYPE}") +
453 "%buf = OpTypeStruct %i32arr\n"
454 "%bufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %buf\n"
455 "%indata = OpVariable %bufptr ${BLOCK_POINTER_TYPE}\n"
457 "%sumbuf = OpTypeStruct %i32arr\n"
458 "%sumbufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %sumbuf\n"
459 "%sum = OpVariable %sumbufptr ${BLOCK_POINTER_TYPE}\n"
461 "%id = OpVariable %uvec3ptr Input\n"
462 "%minusone = OpConstant %i32 -1\n"
463 "%zero = OpConstant %i32 0\n"
464 "%one = OpConstant %u32 1\n"
465 "%two = OpConstant %i32 2\n"
467 "%main = OpFunction %void None %voidf\n"
469 "%idval = OpLoad %uvec3 %id\n"
470 "%x = OpCompositeExtract %u32 %idval 0\n"
472 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
473 "%inval = OpLoad %i32 %inloc\n"
475 "%outloc = OpAccessChain %i32ptr %sum %zero ${INDEX}\n"
481 #define ADD_OPATOMIC_CASE(NAME, ASSEMBLY, CALCULATE_EXPECTED, NUM_OUTPUT_ELEMENTS) \
483 DE_STATIC_ASSERT((NUM_OUTPUT_ELEMENTS) == 1 || (NUM_OUTPUT_ELEMENTS) == numElements); \
484 struct calculateExpected_##NAME { static void calculateExpected(deInt32& expected, deInt32 input) CALCULATE_EXPECTED }; /* NOLINT(CALCULATE_EXPECTED) */ \
485 cases.push_back(OpAtomicCase(#NAME, ASSEMBLY, calculateExpected_##NAME::calculateExpected, NUM_OUTPUT_ELEMENTS)); \
486 } while (deGetFalse())
487 #define ADD_OPATOMIC_CASE_1(NAME, ASSEMBLY, CALCULATE_EXPECTED) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, CALCULATE_EXPECTED, 1)
488 #define ADD_OPATOMIC_CASE_N(NAME, ASSEMBLY, CALCULATE_EXPECTED) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, CALCULATE_EXPECTED, numElements)
490 ADD_OPATOMIC_CASE_1(iadd, "%unused = OpAtomicIAdd %i32 %outloc %one %zero %inval\n", { expected += input; } );
491 ADD_OPATOMIC_CASE_1(isub, "%unused = OpAtomicISub %i32 %outloc %one %zero %inval\n", { expected -= input; } );
492 ADD_OPATOMIC_CASE_1(iinc, "%unused = OpAtomicIIncrement %i32 %outloc %one %zero\n", { ++expected; (void)input;} );
493 ADD_OPATOMIC_CASE_1(idec, "%unused = OpAtomicIDecrement %i32 %outloc %one %zero\n", { --expected; (void)input;} );
494 ADD_OPATOMIC_CASE_N(load, "%inval2 = OpAtomicLoad %i32 %inloc %zero %zero\n"
495 " OpStore %outloc %inval2\n", { expected = input;} );
496 ADD_OPATOMIC_CASE_N(store, " OpAtomicStore %outloc %zero %zero %inval\n", { expected = input;} );
497 ADD_OPATOMIC_CASE_N(compex, "%even = OpSMod %i32 %inval %two\n"
498 " OpStore %outloc %even\n"
499 "%unused = OpAtomicCompareExchange %i32 %outloc %one %zero %zero %minusone %zero\n", { expected = (input % 2) == 0 ? -1 : 1;} );
501 #undef ADD_OPATOMIC_CASE
502 #undef ADD_OPATOMIC_CASE_1
503 #undef ADD_OPATOMIC_CASE_N
505 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
507 map<string, string> specializations;
508 ComputeShaderSpec spec;
509 vector<deInt32> inputInts (numElements, 0);
510 vector<deInt32> expected (cases[caseNdx].numOutputElements, -1);
512 specializations["INDEX"] = (cases[caseNdx].numOutputElements == 1) ? "%zero" : "%x";
513 specializations["INSTRUCTION"] = cases[caseNdx].assembly;
514 specializations["BLOCK_DECORATION"] = useStorageBuffer ? "Block" : "BufferBlock";
515 specializations["BLOCK_POINTER_TYPE"] = useStorageBuffer ? "StorageBuffer" : "Uniform";
516 spec.assembly = shaderTemplate.specialize(specializations);
518 if (useStorageBuffer)
519 spec.extensions.push_back("VK_KHR_storage_buffer_storage_class");
521 fillRandomScalars(rnd, 1, 100, &inputInts[0], numElements);
522 for (size_t ndx = 0; ndx < numElements; ++ndx)
524 cases[caseNdx].calculateExpected((cases[caseNdx].numOutputElements == 1) ? expected[0] : expected[ndx], inputInts[ndx]);
527 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
528 spec.outputs.push_back(BufferSp(new Int32Buffer(expected)));
529 spec.numWorkGroups = IVec3(numElements, 1, 1);
530 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
533 return group.release();
536 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
538 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
539 ComputeShaderSpec spec;
540 de::Random rnd (deStringHash(group->getName()));
541 const int numElements = 100;
542 vector<float> positiveFloats (numElements, 0);
543 vector<float> negativeFloats (numElements, 0);
545 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
547 for (size_t ndx = 0; ndx < numElements; ++ndx)
548 negativeFloats[ndx] = -positiveFloats[ndx];
551 string(getComputeAsmShaderPreamble()) +
553 "%fname1 = OpString \"negateInputs.comp\"\n"
554 "%fname2 = OpString \"negateInputs\"\n"
556 "OpSource GLSL 430\n"
557 "OpName %main \"main\"\n"
558 "OpName %id \"gl_GlobalInvocationID\"\n"
560 "OpDecorate %id BuiltIn GlobalInvocationId\n"
562 + string(getComputeAsmInputOutputBufferTraits()) +
564 "OpLine %fname1 0 0\n" // At the earliest possible position
566 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
568 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
569 "OpLine %fname2 1 0\n" // Different filenames
570 "OpLine %fname1 1000 100000\n"
572 "%id = OpVariable %uvec3ptr Input\n"
573 "%zero = OpConstant %i32 0\n"
575 "OpLine %fname1 1 1\n" // Before a function
577 "%main = OpFunction %void None %voidf\n"
580 "OpLine %fname1 1 1\n" // In a function
582 "%idval = OpLoad %uvec3 %id\n"
583 "%x = OpCompositeExtract %u32 %idval 0\n"
584 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
585 "%inval = OpLoad %f32 %inloc\n"
586 "%neg = OpFNegate %f32 %inval\n"
587 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
588 " OpStore %outloc %neg\n"
591 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
592 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
593 spec.numWorkGroups = IVec3(numElements, 1, 1);
595 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
597 return group.release();
600 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
602 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
603 ComputeShaderSpec spec;
604 de::Random rnd (deStringHash(group->getName()));
605 const int numElements = 100;
606 vector<float> positiveFloats (numElements, 0);
607 vector<float> negativeFloats (numElements, 0);
609 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
611 for (size_t ndx = 0; ndx < numElements; ++ndx)
612 negativeFloats[ndx] = -positiveFloats[ndx];
615 string(getComputeAsmShaderPreamble()) +
617 "%fname = OpString \"negateInputs.comp\"\n"
619 "OpSource GLSL 430\n"
620 "OpName %main \"main\"\n"
621 "OpName %id \"gl_GlobalInvocationID\"\n"
623 "OpDecorate %id BuiltIn GlobalInvocationId\n"
625 + string(getComputeAsmInputOutputBufferTraits()) +
627 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
629 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
631 "OpLine %fname 0 1\n"
632 "OpNoLine\n" // Immediately following a preceding OpLine
634 "OpLine %fname 1000 1\n"
636 "%id = OpVariable %uvec3ptr Input\n"
637 "%zero = OpConstant %i32 0\n"
639 "OpNoLine\n" // Contents after the previous OpLine
641 "%main = OpFunction %void None %voidf\n"
643 "%idval = OpLoad %uvec3 %id\n"
644 "%x = OpCompositeExtract %u32 %idval 0\n"
646 "OpNoLine\n" // Multiple OpNoLine
650 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
651 "%inval = OpLoad %f32 %inloc\n"
652 "%neg = OpFNegate %f32 %inval\n"
653 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
654 " OpStore %outloc %neg\n"
657 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
658 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
659 spec.numWorkGroups = IVec3(numElements, 1, 1);
661 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
663 return group.release();
666 // Compare instruction for the contraction compute case.
667 // Returns true if the output is what is expected from the test case.
668 bool compareNoContractCase(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
670 if (outputAllocs.size() != 1)
673 // We really just need this for size because we are not comparing the exact values.
674 const BufferSp& expectedOutput = expectedOutputs[0];
675 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
677 for(size_t i = 0; i < expectedOutput->getNumBytes() / sizeof(float); ++i) {
678 if (outputAsFloat[i] != 0.f &&
679 outputAsFloat[i] != -ldexp(1, -24)) {
687 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
689 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
690 vector<CaseParameter> cases;
691 const int numElements = 100;
692 vector<float> inputFloats1 (numElements, 0);
693 vector<float> inputFloats2 (numElements, 0);
694 vector<float> outputFloats (numElements, 0);
695 const StringTemplate shaderTemplate (
696 string(getComputeAsmShaderPreamble()) +
698 "OpName %main \"main\"\n"
699 "OpName %id \"gl_GlobalInvocationID\"\n"
701 "OpDecorate %id BuiltIn GlobalInvocationId\n"
705 "OpDecorate %buf BufferBlock\n"
706 "OpDecorate %indata1 DescriptorSet 0\n"
707 "OpDecorate %indata1 Binding 0\n"
708 "OpDecorate %indata2 DescriptorSet 0\n"
709 "OpDecorate %indata2 Binding 1\n"
710 "OpDecorate %outdata DescriptorSet 0\n"
711 "OpDecorate %outdata Binding 2\n"
712 "OpDecorate %f32arr ArrayStride 4\n"
713 "OpMemberDecorate %buf 0 Offset 0\n"
715 + string(getComputeAsmCommonTypes()) +
717 "%buf = OpTypeStruct %f32arr\n"
718 "%bufptr = OpTypePointer Uniform %buf\n"
719 "%indata1 = OpVariable %bufptr Uniform\n"
720 "%indata2 = OpVariable %bufptr Uniform\n"
721 "%outdata = OpVariable %bufptr Uniform\n"
723 "%id = OpVariable %uvec3ptr Input\n"
724 "%zero = OpConstant %i32 0\n"
725 "%c_f_m1 = OpConstant %f32 -1.\n"
727 "%main = OpFunction %void None %voidf\n"
729 "%idval = OpLoad %uvec3 %id\n"
730 "%x = OpCompositeExtract %u32 %idval 0\n"
731 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
732 "%inval1 = OpLoad %f32 %inloc1\n"
733 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
734 "%inval2 = OpLoad %f32 %inloc2\n"
735 "%mul = OpFMul %f32 %inval1 %inval2\n"
736 "%add = OpFAdd %f32 %mul %c_f_m1\n"
737 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
738 " OpStore %outloc %add\n"
742 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
743 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
744 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
746 for (size_t ndx = 0; ndx < numElements; ++ndx)
748 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
749 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
750 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
751 // conducted separately and the result is rounded to 1, or 0x1.fffffcp-1
752 // So the final result will be 0.f or 0x1p-24.
753 // If the operation is combined into a precise fused multiply-add, then the result would be
754 // 2^-46 (0xa8800000).
755 outputFloats[ndx] = 0.f;
758 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
760 map<string, string> specializations;
761 ComputeShaderSpec spec;
763 specializations["DECORATION"] = cases[caseNdx].param;
764 spec.assembly = shaderTemplate.specialize(specializations);
765 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
766 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
767 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
768 spec.numWorkGroups = IVec3(numElements, 1, 1);
769 // Check against the two possible answers based on rounding mode.
770 spec.verifyIO = &compareNoContractCase;
772 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
774 return group.release();
777 bool compareFRem(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
779 if (outputAllocs.size() != 1)
782 const BufferSp& expectedOutput = expectedOutputs[0];
783 const float *expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
784 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
786 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
788 const float f0 = expectedOutputAsFloat[idx];
789 const float f1 = outputAsFloat[idx];
790 // \todo relative error needs to be fairly high because FRem may be implemented as
791 // (roughly) frac(a/b)*b, so LSB errors can be magnified. But this should be fine for now.
792 if (deFloatAbs((f1 - f0) / f0) > 0.02)
799 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
801 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
802 ComputeShaderSpec spec;
803 de::Random rnd (deStringHash(group->getName()));
804 const int numElements = 200;
805 vector<float> inputFloats1 (numElements, 0);
806 vector<float> inputFloats2 (numElements, 0);
807 vector<float> outputFloats (numElements, 0);
809 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
810 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
812 for (size_t ndx = 0; ndx < numElements; ++ndx)
814 // Guard against divisors near zero.
815 if (std::fabs(inputFloats2[ndx]) < 1e-3)
816 inputFloats2[ndx] = 8.f;
818 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
819 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
823 string(getComputeAsmShaderPreamble()) +
825 "OpName %main \"main\"\n"
826 "OpName %id \"gl_GlobalInvocationID\"\n"
828 "OpDecorate %id BuiltIn GlobalInvocationId\n"
830 "OpDecorate %buf BufferBlock\n"
831 "OpDecorate %indata1 DescriptorSet 0\n"
832 "OpDecorate %indata1 Binding 0\n"
833 "OpDecorate %indata2 DescriptorSet 0\n"
834 "OpDecorate %indata2 Binding 1\n"
835 "OpDecorate %outdata DescriptorSet 0\n"
836 "OpDecorate %outdata Binding 2\n"
837 "OpDecorate %f32arr ArrayStride 4\n"
838 "OpMemberDecorate %buf 0 Offset 0\n"
840 + string(getComputeAsmCommonTypes()) +
842 "%buf = OpTypeStruct %f32arr\n"
843 "%bufptr = OpTypePointer Uniform %buf\n"
844 "%indata1 = OpVariable %bufptr Uniform\n"
845 "%indata2 = OpVariable %bufptr Uniform\n"
846 "%outdata = OpVariable %bufptr Uniform\n"
848 "%id = OpVariable %uvec3ptr Input\n"
849 "%zero = OpConstant %i32 0\n"
851 "%main = OpFunction %void None %voidf\n"
853 "%idval = OpLoad %uvec3 %id\n"
854 "%x = OpCompositeExtract %u32 %idval 0\n"
855 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
856 "%inval1 = OpLoad %f32 %inloc1\n"
857 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
858 "%inval2 = OpLoad %f32 %inloc2\n"
859 "%rem = OpFRem %f32 %inval1 %inval2\n"
860 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
861 " OpStore %outloc %rem\n"
865 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
866 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
867 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
868 spec.numWorkGroups = IVec3(numElements, 1, 1);
869 spec.verifyIO = &compareFRem;
871 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
873 return group.release();
876 bool compareNMin (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
878 if (outputAllocs.size() != 1)
881 const BufferSp& expectedOutput = expectedOutputs[0];
882 const float* const expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
883 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
885 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
887 const float f0 = expectedOutputAsFloat[idx];
888 const float f1 = outputAsFloat[idx];
890 // For NMin, we accept NaN as output if both inputs were NaN.
891 // Otherwise the NaN is the wrong choise, as on architectures that
892 // do not handle NaN, those are huge values.
893 if (!(tcu::Float32(f1).isNaN() && tcu::Float32(f0).isNaN()) && deFloatAbs(f1 - f0) > 0.00001f)
900 tcu::TestCaseGroup* createOpNMinGroup (tcu::TestContext& testCtx)
902 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnmin", "Test the OpNMin instruction"));
903 ComputeShaderSpec spec;
904 de::Random rnd (deStringHash(group->getName()));
905 const int numElements = 200;
906 vector<float> inputFloats1 (numElements, 0);
907 vector<float> inputFloats2 (numElements, 0);
908 vector<float> outputFloats (numElements, 0);
910 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
911 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
913 // Make the first case a full-NAN case.
914 inputFloats1[0] = TCU_NAN;
915 inputFloats2[0] = TCU_NAN;
917 for (size_t ndx = 0; ndx < numElements; ++ndx)
919 // By default, pick the smallest
920 outputFloats[ndx] = std::min(inputFloats1[ndx], inputFloats2[ndx]);
922 // Make half of the cases NaN cases
925 // Alternate between the NaN operand
928 outputFloats[ndx] = inputFloats2[ndx];
929 inputFloats1[ndx] = TCU_NAN;
933 outputFloats[ndx] = inputFloats1[ndx];
934 inputFloats2[ndx] = TCU_NAN;
940 "OpCapability Shader\n"
941 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
942 "OpMemoryModel Logical GLSL450\n"
943 "OpEntryPoint GLCompute %main \"main\" %id\n"
944 "OpExecutionMode %main LocalSize 1 1 1\n"
946 "OpName %main \"main\"\n"
947 "OpName %id \"gl_GlobalInvocationID\"\n"
949 "OpDecorate %id BuiltIn GlobalInvocationId\n"
951 "OpDecorate %buf BufferBlock\n"
952 "OpDecorate %indata1 DescriptorSet 0\n"
953 "OpDecorate %indata1 Binding 0\n"
954 "OpDecorate %indata2 DescriptorSet 0\n"
955 "OpDecorate %indata2 Binding 1\n"
956 "OpDecorate %outdata DescriptorSet 0\n"
957 "OpDecorate %outdata Binding 2\n"
958 "OpDecorate %f32arr ArrayStride 4\n"
959 "OpMemberDecorate %buf 0 Offset 0\n"
961 + string(getComputeAsmCommonTypes()) +
963 "%buf = OpTypeStruct %f32arr\n"
964 "%bufptr = OpTypePointer Uniform %buf\n"
965 "%indata1 = OpVariable %bufptr Uniform\n"
966 "%indata2 = OpVariable %bufptr Uniform\n"
967 "%outdata = OpVariable %bufptr Uniform\n"
969 "%id = OpVariable %uvec3ptr Input\n"
970 "%zero = OpConstant %i32 0\n"
972 "%main = OpFunction %void None %voidf\n"
974 "%idval = OpLoad %uvec3 %id\n"
975 "%x = OpCompositeExtract %u32 %idval 0\n"
976 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
977 "%inval1 = OpLoad %f32 %inloc1\n"
978 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
979 "%inval2 = OpLoad %f32 %inloc2\n"
980 "%rem = OpExtInst %f32 %std450 NMin %inval1 %inval2\n"
981 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
982 " OpStore %outloc %rem\n"
986 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
987 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
988 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
989 spec.numWorkGroups = IVec3(numElements, 1, 1);
990 spec.verifyIO = &compareNMin;
992 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
994 return group.release();
997 bool compareNMax (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
999 if (outputAllocs.size() != 1)
1002 const BufferSp& expectedOutput = expectedOutputs[0];
1003 const float* const expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
1004 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
1006 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
1008 const float f0 = expectedOutputAsFloat[idx];
1009 const float f1 = outputAsFloat[idx];
1011 // For NMax, NaN is considered acceptable result, since in
1012 // architectures that do not handle NaNs, those are huge values.
1013 if (!tcu::Float32(f1).isNaN() && deFloatAbs(f1 - f0) > 0.00001f)
1020 tcu::TestCaseGroup* createOpNMaxGroup (tcu::TestContext& testCtx)
1022 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, "opnmax", "Test the OpNMax instruction"));
1023 ComputeShaderSpec spec;
1024 de::Random rnd (deStringHash(group->getName()));
1025 const int numElements = 200;
1026 vector<float> inputFloats1 (numElements, 0);
1027 vector<float> inputFloats2 (numElements, 0);
1028 vector<float> outputFloats (numElements, 0);
1030 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1031 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1033 // Make the first case a full-NAN case.
1034 inputFloats1[0] = TCU_NAN;
1035 inputFloats2[0] = TCU_NAN;
1037 for (size_t ndx = 0; ndx < numElements; ++ndx)
1039 // By default, pick the biggest
1040 outputFloats[ndx] = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1042 // Make half of the cases NaN cases
1045 // Alternate between the NaN operand
1048 outputFloats[ndx] = inputFloats2[ndx];
1049 inputFloats1[ndx] = TCU_NAN;
1053 outputFloats[ndx] = inputFloats1[ndx];
1054 inputFloats2[ndx] = TCU_NAN;
1060 "OpCapability Shader\n"
1061 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1062 "OpMemoryModel Logical GLSL450\n"
1063 "OpEntryPoint GLCompute %main \"main\" %id\n"
1064 "OpExecutionMode %main LocalSize 1 1 1\n"
1066 "OpName %main \"main\"\n"
1067 "OpName %id \"gl_GlobalInvocationID\"\n"
1069 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1071 "OpDecorate %buf BufferBlock\n"
1072 "OpDecorate %indata1 DescriptorSet 0\n"
1073 "OpDecorate %indata1 Binding 0\n"
1074 "OpDecorate %indata2 DescriptorSet 0\n"
1075 "OpDecorate %indata2 Binding 1\n"
1076 "OpDecorate %outdata DescriptorSet 0\n"
1077 "OpDecorate %outdata Binding 2\n"
1078 "OpDecorate %f32arr ArrayStride 4\n"
1079 "OpMemberDecorate %buf 0 Offset 0\n"
1081 + string(getComputeAsmCommonTypes()) +
1083 "%buf = OpTypeStruct %f32arr\n"
1084 "%bufptr = OpTypePointer Uniform %buf\n"
1085 "%indata1 = OpVariable %bufptr Uniform\n"
1086 "%indata2 = OpVariable %bufptr Uniform\n"
1087 "%outdata = OpVariable %bufptr Uniform\n"
1089 "%id = OpVariable %uvec3ptr Input\n"
1090 "%zero = OpConstant %i32 0\n"
1092 "%main = OpFunction %void None %voidf\n"
1093 "%label = OpLabel\n"
1094 "%idval = OpLoad %uvec3 %id\n"
1095 "%x = OpCompositeExtract %u32 %idval 0\n"
1096 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1097 "%inval1 = OpLoad %f32 %inloc1\n"
1098 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1099 "%inval2 = OpLoad %f32 %inloc2\n"
1100 "%rem = OpExtInst %f32 %std450 NMax %inval1 %inval2\n"
1101 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1102 " OpStore %outloc %rem\n"
1106 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1107 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1108 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1109 spec.numWorkGroups = IVec3(numElements, 1, 1);
1110 spec.verifyIO = &compareNMax;
1112 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1114 return group.release();
1117 bool compareNClamp (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1119 if (outputAllocs.size() != 1)
1122 const BufferSp& expectedOutput = expectedOutputs[0];
1123 const float* const expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
1124 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
1126 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float) / 2; ++idx)
1128 const float e0 = expectedOutputAsFloat[idx * 2];
1129 const float e1 = expectedOutputAsFloat[idx * 2 + 1];
1130 const float res = outputAsFloat[idx];
1132 // For NClamp, we have two possible outcomes based on
1133 // whether NaNs are handled or not.
1134 // If either min or max value is NaN, the result is undefined,
1135 // so this test doesn't stress those. If the clamped value is
1136 // NaN, and NaNs are handled, the result is min; if NaNs are not
1137 // handled, they are big values that result in max.
1138 // If all three parameters are NaN, the result should be NaN.
1139 if (!((tcu::Float32(e0).isNaN() && tcu::Float32(res).isNaN()) ||
1140 (deFloatAbs(e0 - res) < 0.00001f) ||
1141 (deFloatAbs(e1 - res) < 0.00001f)))
1148 tcu::TestCaseGroup* createOpNClampGroup (tcu::TestContext& testCtx)
1150 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnclamp", "Test the OpNClamp instruction"));
1151 ComputeShaderSpec spec;
1152 de::Random rnd (deStringHash(group->getName()));
1153 const int numElements = 200;
1154 vector<float> inputFloats1 (numElements, 0);
1155 vector<float> inputFloats2 (numElements, 0);
1156 vector<float> inputFloats3 (numElements, 0);
1157 vector<float> outputFloats (numElements * 2, 0);
1159 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1160 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1161 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats3[0], numElements);
1163 for (size_t ndx = 0; ndx < numElements; ++ndx)
1165 // Results are only defined if max value is bigger than min value.
1166 if (inputFloats2[ndx] > inputFloats3[ndx])
1168 float t = inputFloats2[ndx];
1169 inputFloats2[ndx] = inputFloats3[ndx];
1170 inputFloats3[ndx] = t;
1173 // By default, do the clamp, setting both possible answers
1174 float defaultRes = std::min(std::max(inputFloats1[ndx], inputFloats2[ndx]), inputFloats3[ndx]);
1176 float maxResA = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1177 float maxResB = maxResA;
1179 // Alternate between the NaN cases
1182 inputFloats1[ndx] = TCU_NAN;
1183 // If NaN is handled, the result should be same as the clamp minimum.
1184 // If NaN is not handled, the result should clamp to the clamp maximum.
1185 maxResA = inputFloats2[ndx];
1186 maxResB = inputFloats3[ndx];
1190 // Not a NaN case - only one legal result.
1191 maxResA = defaultRes;
1192 maxResB = defaultRes;
1195 outputFloats[ndx * 2] = maxResA;
1196 outputFloats[ndx * 2 + 1] = maxResB;
1199 // Make the first case a full-NAN case.
1200 inputFloats1[0] = TCU_NAN;
1201 inputFloats2[0] = TCU_NAN;
1202 inputFloats3[0] = TCU_NAN;
1203 outputFloats[0] = TCU_NAN;
1204 outputFloats[1] = TCU_NAN;
1207 "OpCapability Shader\n"
1208 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1209 "OpMemoryModel Logical GLSL450\n"
1210 "OpEntryPoint GLCompute %main \"main\" %id\n"
1211 "OpExecutionMode %main LocalSize 1 1 1\n"
1213 "OpName %main \"main\"\n"
1214 "OpName %id \"gl_GlobalInvocationID\"\n"
1216 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1218 "OpDecorate %buf BufferBlock\n"
1219 "OpDecorate %indata1 DescriptorSet 0\n"
1220 "OpDecorate %indata1 Binding 0\n"
1221 "OpDecorate %indata2 DescriptorSet 0\n"
1222 "OpDecorate %indata2 Binding 1\n"
1223 "OpDecorate %indata3 DescriptorSet 0\n"
1224 "OpDecorate %indata3 Binding 2\n"
1225 "OpDecorate %outdata DescriptorSet 0\n"
1226 "OpDecorate %outdata Binding 3\n"
1227 "OpDecorate %f32arr ArrayStride 4\n"
1228 "OpMemberDecorate %buf 0 Offset 0\n"
1230 + string(getComputeAsmCommonTypes()) +
1232 "%buf = OpTypeStruct %f32arr\n"
1233 "%bufptr = OpTypePointer Uniform %buf\n"
1234 "%indata1 = OpVariable %bufptr Uniform\n"
1235 "%indata2 = OpVariable %bufptr Uniform\n"
1236 "%indata3 = OpVariable %bufptr Uniform\n"
1237 "%outdata = OpVariable %bufptr Uniform\n"
1239 "%id = OpVariable %uvec3ptr Input\n"
1240 "%zero = OpConstant %i32 0\n"
1242 "%main = OpFunction %void None %voidf\n"
1243 "%label = OpLabel\n"
1244 "%idval = OpLoad %uvec3 %id\n"
1245 "%x = OpCompositeExtract %u32 %idval 0\n"
1246 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1247 "%inval1 = OpLoad %f32 %inloc1\n"
1248 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1249 "%inval2 = OpLoad %f32 %inloc2\n"
1250 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1251 "%inval3 = OpLoad %f32 %inloc3\n"
1252 "%rem = OpExtInst %f32 %std450 NClamp %inval1 %inval2 %inval3\n"
1253 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1254 " OpStore %outloc %rem\n"
1258 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1259 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1260 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1261 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1262 spec.numWorkGroups = IVec3(numElements, 1, 1);
1263 spec.verifyIO = &compareNClamp;
1265 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1267 return group.release();
1270 tcu::TestCaseGroup* createOpSRemComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1272 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem", "Test the OpSRem instruction"));
1273 de::Random rnd (deStringHash(group->getName()));
1274 const int numElements = 200;
1276 const struct CaseParams
1279 const char* failMessage; // customized status message
1280 qpTestResult failResult; // override status on failure
1281 int op1Min, op1Max; // operand ranges
1285 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1286 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1288 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1290 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1292 const CaseParams& params = cases[caseNdx];
1293 ComputeShaderSpec spec;
1294 vector<deInt32> inputInts1 (numElements, 0);
1295 vector<deInt32> inputInts2 (numElements, 0);
1296 vector<deInt32> outputInts (numElements, 0);
1298 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1299 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1301 for (int ndx = 0; ndx < numElements; ++ndx)
1303 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1304 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1308 string(getComputeAsmShaderPreamble()) +
1310 "OpName %main \"main\"\n"
1311 "OpName %id \"gl_GlobalInvocationID\"\n"
1313 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1315 "OpDecorate %buf BufferBlock\n"
1316 "OpDecorate %indata1 DescriptorSet 0\n"
1317 "OpDecorate %indata1 Binding 0\n"
1318 "OpDecorate %indata2 DescriptorSet 0\n"
1319 "OpDecorate %indata2 Binding 1\n"
1320 "OpDecorate %outdata DescriptorSet 0\n"
1321 "OpDecorate %outdata Binding 2\n"
1322 "OpDecorate %i32arr ArrayStride 4\n"
1323 "OpMemberDecorate %buf 0 Offset 0\n"
1325 + string(getComputeAsmCommonTypes()) +
1327 "%buf = OpTypeStruct %i32arr\n"
1328 "%bufptr = OpTypePointer Uniform %buf\n"
1329 "%indata1 = OpVariable %bufptr Uniform\n"
1330 "%indata2 = OpVariable %bufptr Uniform\n"
1331 "%outdata = OpVariable %bufptr Uniform\n"
1333 "%id = OpVariable %uvec3ptr Input\n"
1334 "%zero = OpConstant %i32 0\n"
1336 "%main = OpFunction %void None %voidf\n"
1337 "%label = OpLabel\n"
1338 "%idval = OpLoad %uvec3 %id\n"
1339 "%x = OpCompositeExtract %u32 %idval 0\n"
1340 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1341 "%inval1 = OpLoad %i32 %inloc1\n"
1342 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1343 "%inval2 = OpLoad %i32 %inloc2\n"
1344 "%rem = OpSRem %i32 %inval1 %inval2\n"
1345 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1346 " OpStore %outloc %rem\n"
1350 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1351 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1352 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1353 spec.numWorkGroups = IVec3(numElements, 1, 1);
1354 spec.failResult = params.failResult;
1355 spec.failMessage = params.failMessage;
1357 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1360 return group.release();
1363 tcu::TestCaseGroup* createOpSRemComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1365 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem64", "Test the 64-bit OpSRem instruction"));
1366 de::Random rnd (deStringHash(group->getName()));
1367 const int numElements = 200;
1369 const struct CaseParams
1372 const char* failMessage; // customized status message
1373 qpTestResult failResult; // override status on failure
1377 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1378 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1380 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1382 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1384 const CaseParams& params = cases[caseNdx];
1385 ComputeShaderSpec spec;
1386 vector<deInt64> inputInts1 (numElements, 0);
1387 vector<deInt64> inputInts2 (numElements, 0);
1388 vector<deInt64> outputInts (numElements, 0);
1390 if (params.positive)
1392 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1393 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1397 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1398 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1401 for (int ndx = 0; ndx < numElements; ++ndx)
1403 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1404 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1408 "OpCapability Int64\n"
1410 + string(getComputeAsmShaderPreamble()) +
1412 "OpName %main \"main\"\n"
1413 "OpName %id \"gl_GlobalInvocationID\"\n"
1415 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1417 "OpDecorate %buf BufferBlock\n"
1418 "OpDecorate %indata1 DescriptorSet 0\n"
1419 "OpDecorate %indata1 Binding 0\n"
1420 "OpDecorate %indata2 DescriptorSet 0\n"
1421 "OpDecorate %indata2 Binding 1\n"
1422 "OpDecorate %outdata DescriptorSet 0\n"
1423 "OpDecorate %outdata Binding 2\n"
1424 "OpDecorate %i64arr ArrayStride 8\n"
1425 "OpMemberDecorate %buf 0 Offset 0\n"
1427 + string(getComputeAsmCommonTypes())
1428 + string(getComputeAsmCommonInt64Types()) +
1430 "%buf = OpTypeStruct %i64arr\n"
1431 "%bufptr = OpTypePointer Uniform %buf\n"
1432 "%indata1 = OpVariable %bufptr Uniform\n"
1433 "%indata2 = OpVariable %bufptr Uniform\n"
1434 "%outdata = OpVariable %bufptr Uniform\n"
1436 "%id = OpVariable %uvec3ptr Input\n"
1437 "%zero = OpConstant %i64 0\n"
1439 "%main = OpFunction %void None %voidf\n"
1440 "%label = OpLabel\n"
1441 "%idval = OpLoad %uvec3 %id\n"
1442 "%x = OpCompositeExtract %u32 %idval 0\n"
1443 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1444 "%inval1 = OpLoad %i64 %inloc1\n"
1445 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1446 "%inval2 = OpLoad %i64 %inloc2\n"
1447 "%rem = OpSRem %i64 %inval1 %inval2\n"
1448 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1449 " OpStore %outloc %rem\n"
1453 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1454 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1455 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1456 spec.numWorkGroups = IVec3(numElements, 1, 1);
1457 spec.failResult = params.failResult;
1458 spec.failMessage = params.failMessage;
1460 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1463 return group.release();
1466 tcu::TestCaseGroup* createOpSModComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1468 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod", "Test the OpSMod instruction"));
1469 de::Random rnd (deStringHash(group->getName()));
1470 const int numElements = 200;
1472 const struct CaseParams
1475 const char* failMessage; // customized status message
1476 qpTestResult failResult; // override status on failure
1477 int op1Min, op1Max; // operand ranges
1481 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1482 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1484 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1486 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1488 const CaseParams& params = cases[caseNdx];
1490 ComputeShaderSpec spec;
1491 vector<deInt32> inputInts1 (numElements, 0);
1492 vector<deInt32> inputInts2 (numElements, 0);
1493 vector<deInt32> outputInts (numElements, 0);
1495 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1496 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1498 for (int ndx = 0; ndx < numElements; ++ndx)
1500 deInt32 rem = inputInts1[ndx] % inputInts2[ndx];
1503 outputInts[ndx] = 0;
1505 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1507 // They have the same sign
1508 outputInts[ndx] = rem;
1512 // They have opposite sign. The remainder operation takes the
1513 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1514 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1515 // the result has the correct sign and that it is still
1516 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1518 // See also http://mathforum.org/library/drmath/view/52343.html
1519 outputInts[ndx] = rem + inputInts2[ndx];
1524 string(getComputeAsmShaderPreamble()) +
1526 "OpName %main \"main\"\n"
1527 "OpName %id \"gl_GlobalInvocationID\"\n"
1529 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1531 "OpDecorate %buf BufferBlock\n"
1532 "OpDecorate %indata1 DescriptorSet 0\n"
1533 "OpDecorate %indata1 Binding 0\n"
1534 "OpDecorate %indata2 DescriptorSet 0\n"
1535 "OpDecorate %indata2 Binding 1\n"
1536 "OpDecorate %outdata DescriptorSet 0\n"
1537 "OpDecorate %outdata Binding 2\n"
1538 "OpDecorate %i32arr ArrayStride 4\n"
1539 "OpMemberDecorate %buf 0 Offset 0\n"
1541 + string(getComputeAsmCommonTypes()) +
1543 "%buf = OpTypeStruct %i32arr\n"
1544 "%bufptr = OpTypePointer Uniform %buf\n"
1545 "%indata1 = OpVariable %bufptr Uniform\n"
1546 "%indata2 = OpVariable %bufptr Uniform\n"
1547 "%outdata = OpVariable %bufptr Uniform\n"
1549 "%id = OpVariable %uvec3ptr Input\n"
1550 "%zero = OpConstant %i32 0\n"
1552 "%main = OpFunction %void None %voidf\n"
1553 "%label = OpLabel\n"
1554 "%idval = OpLoad %uvec3 %id\n"
1555 "%x = OpCompositeExtract %u32 %idval 0\n"
1556 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1557 "%inval1 = OpLoad %i32 %inloc1\n"
1558 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1559 "%inval2 = OpLoad %i32 %inloc2\n"
1560 "%rem = OpSMod %i32 %inval1 %inval2\n"
1561 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1562 " OpStore %outloc %rem\n"
1566 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1567 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1568 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1569 spec.numWorkGroups = IVec3(numElements, 1, 1);
1570 spec.failResult = params.failResult;
1571 spec.failMessage = params.failMessage;
1573 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1576 return group.release();
1579 tcu::TestCaseGroup* createOpSModComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1581 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod64", "Test the OpSMod instruction"));
1582 de::Random rnd (deStringHash(group->getName()));
1583 const int numElements = 200;
1585 const struct CaseParams
1588 const char* failMessage; // customized status message
1589 qpTestResult failResult; // override status on failure
1593 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1594 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1596 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1598 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1600 const CaseParams& params = cases[caseNdx];
1602 ComputeShaderSpec spec;
1603 vector<deInt64> inputInts1 (numElements, 0);
1604 vector<deInt64> inputInts2 (numElements, 0);
1605 vector<deInt64> outputInts (numElements, 0);
1608 if (params.positive)
1610 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1611 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1615 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1616 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1619 for (int ndx = 0; ndx < numElements; ++ndx)
1621 deInt64 rem = inputInts1[ndx] % inputInts2[ndx];
1624 outputInts[ndx] = 0;
1626 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1628 // They have the same sign
1629 outputInts[ndx] = rem;
1633 // They have opposite sign. The remainder operation takes the
1634 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1635 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1636 // the result has the correct sign and that it is still
1637 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1639 // See also http://mathforum.org/library/drmath/view/52343.html
1640 outputInts[ndx] = rem + inputInts2[ndx];
1645 "OpCapability Int64\n"
1647 + string(getComputeAsmShaderPreamble()) +
1649 "OpName %main \"main\"\n"
1650 "OpName %id \"gl_GlobalInvocationID\"\n"
1652 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1654 "OpDecorate %buf BufferBlock\n"
1655 "OpDecorate %indata1 DescriptorSet 0\n"
1656 "OpDecorate %indata1 Binding 0\n"
1657 "OpDecorate %indata2 DescriptorSet 0\n"
1658 "OpDecorate %indata2 Binding 1\n"
1659 "OpDecorate %outdata DescriptorSet 0\n"
1660 "OpDecorate %outdata Binding 2\n"
1661 "OpDecorate %i64arr ArrayStride 8\n"
1662 "OpMemberDecorate %buf 0 Offset 0\n"
1664 + string(getComputeAsmCommonTypes())
1665 + string(getComputeAsmCommonInt64Types()) +
1667 "%buf = OpTypeStruct %i64arr\n"
1668 "%bufptr = OpTypePointer Uniform %buf\n"
1669 "%indata1 = OpVariable %bufptr Uniform\n"
1670 "%indata2 = OpVariable %bufptr Uniform\n"
1671 "%outdata = OpVariable %bufptr Uniform\n"
1673 "%id = OpVariable %uvec3ptr Input\n"
1674 "%zero = OpConstant %i64 0\n"
1676 "%main = OpFunction %void None %voidf\n"
1677 "%label = OpLabel\n"
1678 "%idval = OpLoad %uvec3 %id\n"
1679 "%x = OpCompositeExtract %u32 %idval 0\n"
1680 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1681 "%inval1 = OpLoad %i64 %inloc1\n"
1682 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1683 "%inval2 = OpLoad %i64 %inloc2\n"
1684 "%rem = OpSMod %i64 %inval1 %inval2\n"
1685 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1686 " OpStore %outloc %rem\n"
1690 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1691 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1692 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1693 spec.numWorkGroups = IVec3(numElements, 1, 1);
1694 spec.failResult = params.failResult;
1695 spec.failMessage = params.failMessage;
1697 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1700 return group.release();
1703 // Copy contents in the input buffer to the output buffer.
1704 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
1706 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
1707 de::Random rnd (deStringHash(group->getName()));
1708 const int numElements = 100;
1710 // 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.
1711 ComputeShaderSpec spec1;
1712 vector<Vec4> inputFloats1 (numElements);
1713 vector<Vec4> outputFloats1 (numElements);
1715 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
1717 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1718 floorAll(inputFloats1);
1720 for (size_t ndx = 0; ndx < numElements; ++ndx)
1721 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
1724 string(getComputeAsmShaderPreamble()) +
1726 "OpName %main \"main\"\n"
1727 "OpName %id \"gl_GlobalInvocationID\"\n"
1729 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1730 "OpDecorate %vec4arr ArrayStride 16\n"
1732 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1734 "%vec4 = OpTypeVector %f32 4\n"
1735 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
1736 "%vec4ptr_f = OpTypePointer Function %vec4\n"
1737 "%vec4arr = OpTypeRuntimeArray %vec4\n"
1738 "%buf = OpTypeStruct %vec4arr\n"
1739 "%bufptr = OpTypePointer Uniform %buf\n"
1740 "%indata = OpVariable %bufptr Uniform\n"
1741 "%outdata = OpVariable %bufptr Uniform\n"
1743 "%id = OpVariable %uvec3ptr Input\n"
1744 "%zero = OpConstant %i32 0\n"
1745 "%c_f_0 = OpConstant %f32 0.\n"
1746 "%c_f_0_5 = OpConstant %f32 0.5\n"
1747 "%c_f_1_5 = OpConstant %f32 1.5\n"
1748 "%c_f_2_5 = OpConstant %f32 2.5\n"
1749 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
1751 "%main = OpFunction %void None %voidf\n"
1752 "%label = OpLabel\n"
1753 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
1754 "%idval = OpLoad %uvec3 %id\n"
1755 "%x = OpCompositeExtract %u32 %idval 0\n"
1756 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
1757 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
1758 " OpCopyMemory %v_vec4 %inloc\n"
1759 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
1760 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
1761 " OpStore %outloc %add\n"
1765 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
1766 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
1767 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1769 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
1771 // The following case copies a float[100] variable from the input buffer to the output buffer.
1772 ComputeShaderSpec spec2;
1773 vector<float> inputFloats2 (numElements);
1774 vector<float> outputFloats2 (numElements);
1776 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
1778 for (size_t ndx = 0; ndx < numElements; ++ndx)
1779 outputFloats2[ndx] = inputFloats2[ndx];
1782 string(getComputeAsmShaderPreamble()) +
1784 "OpName %main \"main\"\n"
1785 "OpName %id \"gl_GlobalInvocationID\"\n"
1787 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1788 "OpDecorate %f32arr100 ArrayStride 4\n"
1790 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1792 "%hundred = OpConstant %u32 100\n"
1793 "%f32arr100 = OpTypeArray %f32 %hundred\n"
1794 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
1795 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
1796 "%buf = OpTypeStruct %f32arr100\n"
1797 "%bufptr = OpTypePointer Uniform %buf\n"
1798 "%indata = OpVariable %bufptr Uniform\n"
1799 "%outdata = OpVariable %bufptr Uniform\n"
1801 "%id = OpVariable %uvec3ptr Input\n"
1802 "%zero = OpConstant %i32 0\n"
1804 "%main = OpFunction %void None %voidf\n"
1805 "%label = OpLabel\n"
1806 "%var = OpVariable %f32arr100ptr_f Function\n"
1807 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
1808 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
1809 " OpCopyMemory %var %inarr\n"
1810 " OpCopyMemory %outarr %var\n"
1814 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1815 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1816 spec2.numWorkGroups = IVec3(1, 1, 1);
1818 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
1820 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
1821 ComputeShaderSpec spec3;
1822 vector<float> inputFloats3 (16);
1823 vector<float> outputFloats3 (16);
1825 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
1827 for (size_t ndx = 0; ndx < 16; ++ndx)
1828 outputFloats3[ndx] = inputFloats3[ndx];
1831 string(getComputeAsmShaderPreamble()) +
1833 "OpName %main \"main\"\n"
1834 "OpName %id \"gl_GlobalInvocationID\"\n"
1836 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1837 "OpMemberDecorate %buf 0 Offset 0\n"
1838 "OpMemberDecorate %buf 1 Offset 16\n"
1839 "OpMemberDecorate %buf 2 Offset 32\n"
1840 "OpMemberDecorate %buf 3 Offset 48\n"
1842 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1844 "%vec4 = OpTypeVector %f32 4\n"
1845 "%buf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
1846 "%bufptr = OpTypePointer Uniform %buf\n"
1847 "%indata = OpVariable %bufptr Uniform\n"
1848 "%outdata = OpVariable %bufptr Uniform\n"
1849 "%vec4stptr = OpTypePointer Function %buf\n"
1851 "%id = OpVariable %uvec3ptr Input\n"
1852 "%zero = OpConstant %i32 0\n"
1854 "%main = OpFunction %void None %voidf\n"
1855 "%label = OpLabel\n"
1856 "%var = OpVariable %vec4stptr Function\n"
1857 " OpCopyMemory %var %indata\n"
1858 " OpCopyMemory %outdata %var\n"
1862 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1863 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
1864 spec3.numWorkGroups = IVec3(1, 1, 1);
1866 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
1868 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
1869 ComputeShaderSpec spec4;
1870 vector<float> inputFloats4 (numElements);
1871 vector<float> outputFloats4 (numElements);
1873 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
1875 for (size_t ndx = 0; ndx < numElements; ++ndx)
1876 outputFloats4[ndx] = -inputFloats4[ndx];
1879 string(getComputeAsmShaderPreamble()) +
1881 "OpName %main \"main\"\n"
1882 "OpName %id \"gl_GlobalInvocationID\"\n"
1884 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1886 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
1888 "%f32ptr_f = OpTypePointer Function %f32\n"
1889 "%id = OpVariable %uvec3ptr Input\n"
1890 "%zero = OpConstant %i32 0\n"
1892 "%main = OpFunction %void None %voidf\n"
1893 "%label = OpLabel\n"
1894 "%var = OpVariable %f32ptr_f Function\n"
1895 "%idval = OpLoad %uvec3 %id\n"
1896 "%x = OpCompositeExtract %u32 %idval 0\n"
1897 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1898 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1899 " OpCopyMemory %var %inloc\n"
1900 "%val = OpLoad %f32 %var\n"
1901 "%neg = OpFNegate %f32 %val\n"
1902 " OpStore %outloc %neg\n"
1906 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
1907 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
1908 spec4.numWorkGroups = IVec3(numElements, 1, 1);
1910 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
1912 return group.release();
1915 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
1917 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
1918 ComputeShaderSpec spec;
1919 de::Random rnd (deStringHash(group->getName()));
1920 const int numElements = 100;
1921 vector<float> inputFloats (numElements, 0);
1922 vector<float> outputFloats (numElements, 0);
1924 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
1926 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1927 floorAll(inputFloats);
1929 for (size_t ndx = 0; ndx < numElements; ++ndx)
1930 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
1933 string(getComputeAsmShaderPreamble()) +
1935 "OpName %main \"main\"\n"
1936 "OpName %id \"gl_GlobalInvocationID\"\n"
1938 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1940 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1942 "%fmat = OpTypeMatrix %fvec3 3\n"
1943 "%three = OpConstant %u32 3\n"
1944 "%farr = OpTypeArray %f32 %three\n"
1945 "%fst = OpTypeStruct %f32 %f32\n"
1947 + string(getComputeAsmInputOutputBuffer()) +
1949 "%id = OpVariable %uvec3ptr Input\n"
1950 "%zero = OpConstant %i32 0\n"
1951 "%c_f = OpConstant %f32 1.5\n"
1952 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
1953 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
1954 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
1955 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
1957 "%main = OpFunction %void None %voidf\n"
1958 "%label = OpLabel\n"
1959 "%c_f_copy = OpCopyObject %f32 %c_f\n"
1960 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
1961 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
1962 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
1963 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
1964 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
1965 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
1966 "%farr_elem = OpCompositeExtract %f32 %c_farr_copy 2\n"
1967 "%fst_elem = OpCompositeExtract %f32 %c_fst_copy 1\n"
1968 // Add up. 1.5 * 5 = 7.5.
1969 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
1970 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
1971 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
1972 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
1974 "%idval = OpLoad %uvec3 %id\n"
1975 "%x = OpCompositeExtract %u32 %idval 0\n"
1976 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1977 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1978 "%inval = OpLoad %f32 %inloc\n"
1979 "%add = OpFAdd %f32 %add4 %inval\n"
1980 " OpStore %outloc %add\n"
1983 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1984 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1985 spec.numWorkGroups = IVec3(numElements, 1, 1);
1987 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
1989 return group.release();
1991 // Assembly code used for testing OpUnreachable is based on GLSL source code:
1995 // layout(std140, set = 0, binding = 0) readonly buffer Input {
1996 // float elements[];
1998 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
1999 // float elements[];
2002 // void not_called_func() {
2003 // // place OpUnreachable here
2006 // uint modulo4(uint val) {
2007 // switch (val % uint(4)) {
2008 // case 0: return 3;
2009 // case 1: return 2;
2010 // case 2: return 1;
2011 // case 3: return 0;
2012 // default: return 100; // place OpUnreachable here
2018 // // place OpUnreachable here
2022 // uint x = gl_GlobalInvocationID.x;
2023 // if (const5() > modulo4(1000)) {
2024 // output_data.elements[x] = -input_data.elements[x];
2026 // // place OpUnreachable here
2027 // output_data.elements[x] = input_data.elements[x];
2031 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
2033 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
2034 ComputeShaderSpec spec;
2035 de::Random rnd (deStringHash(group->getName()));
2036 const int numElements = 100;
2037 vector<float> positiveFloats (numElements, 0);
2038 vector<float> negativeFloats (numElements, 0);
2040 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2042 for (size_t ndx = 0; ndx < numElements; ++ndx)
2043 negativeFloats[ndx] = -positiveFloats[ndx];
2046 string(getComputeAsmShaderPreamble()) +
2048 "OpSource GLSL 430\n"
2049 "OpName %main \"main\"\n"
2050 "OpName %func_not_called_func \"not_called_func(\"\n"
2051 "OpName %func_modulo4 \"modulo4(u1;\"\n"
2052 "OpName %func_const5 \"const5(\"\n"
2053 "OpName %id \"gl_GlobalInvocationID\"\n"
2055 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2057 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2059 "%u32ptr = OpTypePointer Function %u32\n"
2060 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
2061 "%unitf = OpTypeFunction %u32\n"
2063 "%id = OpVariable %uvec3ptr Input\n"
2064 "%zero = OpConstant %u32 0\n"
2065 "%one = OpConstant %u32 1\n"
2066 "%two = OpConstant %u32 2\n"
2067 "%three = OpConstant %u32 3\n"
2068 "%four = OpConstant %u32 4\n"
2069 "%five = OpConstant %u32 5\n"
2070 "%hundred = OpConstant %u32 100\n"
2071 "%thousand = OpConstant %u32 1000\n"
2073 + string(getComputeAsmInputOutputBuffer()) +
2076 "%main = OpFunction %void None %voidf\n"
2077 "%main_entry = OpLabel\n"
2078 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
2079 "%idval = OpLoad %uvec3 %id\n"
2080 "%x = OpCompositeExtract %u32 %idval 0\n"
2081 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2082 "%inval = OpLoad %f32 %inloc\n"
2083 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2084 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
2085 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
2086 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
2087 " OpSelectionMerge %if_end None\n"
2088 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2089 "%if_true = OpLabel\n"
2090 "%negate = OpFNegate %f32 %inval\n"
2091 " OpStore %outloc %negate\n"
2092 " OpBranch %if_end\n"
2093 "%if_false = OpLabel\n"
2094 " OpUnreachable\n" // Unreachable else branch for if statement
2095 "%if_end = OpLabel\n"
2099 // not_called_function()
2100 "%func_not_called_func = OpFunction %void None %voidf\n"
2101 "%not_called_func_entry = OpLabel\n"
2102 " OpUnreachable\n" // Unreachable entry block in not called static function
2106 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
2107 "%valptr = OpFunctionParameter %u32ptr\n"
2108 "%modulo4_entry = OpLabel\n"
2109 "%val = OpLoad %u32 %valptr\n"
2110 "%modulo = OpUMod %u32 %val %four\n"
2111 " OpSelectionMerge %switch_merge None\n"
2112 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
2113 "%case0 = OpLabel\n"
2114 " OpReturnValue %three\n"
2115 "%case1 = OpLabel\n"
2116 " OpReturnValue %two\n"
2117 "%case2 = OpLabel\n"
2118 " OpReturnValue %one\n"
2119 "%case3 = OpLabel\n"
2120 " OpReturnValue %zero\n"
2121 "%default = OpLabel\n"
2122 " OpUnreachable\n" // Unreachable default case for switch statement
2123 "%switch_merge = OpLabel\n"
2124 " OpUnreachable\n" // Unreachable merge block for switch statement
2128 "%func_const5 = OpFunction %u32 None %unitf\n"
2129 "%const5_entry = OpLabel\n"
2130 " OpReturnValue %five\n"
2131 "%unreachable = OpLabel\n"
2132 " OpUnreachable\n" // Unreachable block in function
2134 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2135 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2136 spec.numWorkGroups = IVec3(numElements, 1, 1);
2138 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
2140 return group.release();
2143 // Assembly code used for testing decoration group is based on GLSL source code:
2147 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
2148 // float elements[];
2150 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
2151 // float elements[];
2153 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
2154 // float elements[];
2156 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
2157 // float elements[];
2159 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
2160 // float elements[];
2162 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
2163 // float elements[];
2167 // uint x = gl_GlobalInvocationID.x;
2168 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
2170 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
2172 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
2173 ComputeShaderSpec spec;
2174 de::Random rnd (deStringHash(group->getName()));
2175 const int numElements = 100;
2176 vector<float> inputFloats0 (numElements, 0);
2177 vector<float> inputFloats1 (numElements, 0);
2178 vector<float> inputFloats2 (numElements, 0);
2179 vector<float> inputFloats3 (numElements, 0);
2180 vector<float> inputFloats4 (numElements, 0);
2181 vector<float> outputFloats (numElements, 0);
2183 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
2184 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
2185 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
2186 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
2187 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
2189 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2190 floorAll(inputFloats0);
2191 floorAll(inputFloats1);
2192 floorAll(inputFloats2);
2193 floorAll(inputFloats3);
2194 floorAll(inputFloats4);
2196 for (size_t ndx = 0; ndx < numElements; ++ndx)
2197 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
2200 string(getComputeAsmShaderPreamble()) +
2202 "OpSource GLSL 430\n"
2203 "OpName %main \"main\"\n"
2204 "OpName %id \"gl_GlobalInvocationID\"\n"
2206 // Not using group decoration on variable.
2207 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2208 // Not using group decoration on type.
2209 "OpDecorate %f32arr ArrayStride 4\n"
2211 "OpDecorate %groups BufferBlock\n"
2212 "OpDecorate %groupm Offset 0\n"
2213 "%groups = OpDecorationGroup\n"
2214 "%groupm = OpDecorationGroup\n"
2216 // Group decoration on multiple structs.
2217 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
2218 // Group decoration on multiple struct members.
2219 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
2221 "OpDecorate %group1 DescriptorSet 0\n"
2222 "OpDecorate %group3 DescriptorSet 0\n"
2223 "OpDecorate %group3 NonWritable\n"
2224 "OpDecorate %group3 Restrict\n"
2225 "%group0 = OpDecorationGroup\n"
2226 "%group1 = OpDecorationGroup\n"
2227 "%group3 = OpDecorationGroup\n"
2229 // Applying the same decoration group multiple times.
2230 "OpGroupDecorate %group1 %outdata\n"
2231 "OpGroupDecorate %group1 %outdata\n"
2232 "OpGroupDecorate %group1 %outdata\n"
2233 "OpDecorate %outdata DescriptorSet 0\n"
2234 "OpDecorate %outdata Binding 5\n"
2235 // Applying decoration group containing nothing.
2236 "OpGroupDecorate %group0 %indata0\n"
2237 "OpDecorate %indata0 DescriptorSet 0\n"
2238 "OpDecorate %indata0 Binding 0\n"
2239 // Applying decoration group containing one decoration.
2240 "OpGroupDecorate %group1 %indata1\n"
2241 "OpDecorate %indata1 Binding 1\n"
2242 // Applying decoration group containing multiple decorations.
2243 "OpGroupDecorate %group3 %indata2 %indata3\n"
2244 "OpDecorate %indata2 Binding 2\n"
2245 "OpDecorate %indata3 Binding 3\n"
2246 // Applying multiple decoration groups (with overlapping).
2247 "OpGroupDecorate %group0 %indata4\n"
2248 "OpGroupDecorate %group1 %indata4\n"
2249 "OpGroupDecorate %group3 %indata4\n"
2250 "OpDecorate %indata4 Binding 4\n"
2252 + string(getComputeAsmCommonTypes()) +
2254 "%id = OpVariable %uvec3ptr Input\n"
2255 "%zero = OpConstant %i32 0\n"
2257 "%outbuf = OpTypeStruct %f32arr\n"
2258 "%outbufptr = OpTypePointer Uniform %outbuf\n"
2259 "%outdata = OpVariable %outbufptr Uniform\n"
2260 "%inbuf0 = OpTypeStruct %f32arr\n"
2261 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
2262 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
2263 "%inbuf1 = OpTypeStruct %f32arr\n"
2264 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
2265 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
2266 "%inbuf2 = OpTypeStruct %f32arr\n"
2267 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
2268 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
2269 "%inbuf3 = OpTypeStruct %f32arr\n"
2270 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
2271 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
2272 "%inbuf4 = OpTypeStruct %f32arr\n"
2273 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
2274 "%indata4 = OpVariable %inbufptr Uniform\n"
2276 "%main = OpFunction %void None %voidf\n"
2277 "%label = OpLabel\n"
2278 "%idval = OpLoad %uvec3 %id\n"
2279 "%x = OpCompositeExtract %u32 %idval 0\n"
2280 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
2281 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
2282 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
2283 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
2284 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
2285 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2286 "%inval0 = OpLoad %f32 %inloc0\n"
2287 "%inval1 = OpLoad %f32 %inloc1\n"
2288 "%inval2 = OpLoad %f32 %inloc2\n"
2289 "%inval3 = OpLoad %f32 %inloc3\n"
2290 "%inval4 = OpLoad %f32 %inloc4\n"
2291 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
2292 "%add1 = OpFAdd %f32 %add0 %inval2\n"
2293 "%add2 = OpFAdd %f32 %add1 %inval3\n"
2294 "%add = OpFAdd %f32 %add2 %inval4\n"
2295 " OpStore %outloc %add\n"
2298 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
2299 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
2300 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2301 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2302 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2303 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2304 spec.numWorkGroups = IVec3(numElements, 1, 1);
2306 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
2308 return group.release();
2311 struct SpecConstantTwoIntCase
2313 const char* caseName;
2314 const char* scDefinition0;
2315 const char* scDefinition1;
2316 const char* scResultType;
2317 const char* scOperation;
2318 deInt32 scActualValue0;
2319 deInt32 scActualValue1;
2320 const char* resultOperation;
2321 vector<deInt32> expectedOutput;
2323 SpecConstantTwoIntCase (const char* name,
2324 const char* definition0,
2325 const char* definition1,
2326 const char* resultType,
2327 const char* operation,
2330 const char* resultOp,
2331 const vector<deInt32>& output)
2333 , scDefinition0 (definition0)
2334 , scDefinition1 (definition1)
2335 , scResultType (resultType)
2336 , scOperation (operation)
2337 , scActualValue0 (value0)
2338 , scActualValue1 (value1)
2339 , resultOperation (resultOp)
2340 , expectedOutput (output) {}
2343 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
2345 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
2346 vector<SpecConstantTwoIntCase> cases;
2347 de::Random rnd (deStringHash(group->getName()));
2348 const int numElements = 100;
2349 vector<deInt32> inputInts (numElements, 0);
2350 vector<deInt32> outputInts1 (numElements, 0);
2351 vector<deInt32> outputInts2 (numElements, 0);
2352 vector<deInt32> outputInts3 (numElements, 0);
2353 vector<deInt32> outputInts4 (numElements, 0);
2354 const StringTemplate shaderTemplate (
2355 string(getComputeAsmShaderPreamble()) +
2357 "OpName %main \"main\"\n"
2358 "OpName %id \"gl_GlobalInvocationID\"\n"
2360 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2361 "OpDecorate %sc_0 SpecId 0\n"
2362 "OpDecorate %sc_1 SpecId 1\n"
2363 "OpDecorate %i32arr ArrayStride 4\n"
2365 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2367 "%buf = OpTypeStruct %i32arr\n"
2368 "%bufptr = OpTypePointer Uniform %buf\n"
2369 "%indata = OpVariable %bufptr Uniform\n"
2370 "%outdata = OpVariable %bufptr Uniform\n"
2372 "%id = OpVariable %uvec3ptr Input\n"
2373 "%zero = OpConstant %i32 0\n"
2375 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
2376 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
2377 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
2379 "%main = OpFunction %void None %voidf\n"
2380 "%label = OpLabel\n"
2381 "%idval = OpLoad %uvec3 %id\n"
2382 "%x = OpCompositeExtract %u32 %idval 0\n"
2383 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2384 "%inval = OpLoad %i32 %inloc\n"
2385 "%final = ${GEN_RESULT}\n"
2386 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2387 " OpStore %outloc %final\n"
2389 " OpFunctionEnd\n");
2391 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
2393 for (size_t ndx = 0; ndx < numElements; ++ndx)
2395 outputInts1[ndx] = inputInts[ndx] + 42;
2396 outputInts2[ndx] = inputInts[ndx];
2397 outputInts3[ndx] = inputInts[ndx] - 11200;
2398 outputInts4[ndx] = inputInts[ndx] + 1;
2401 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
2402 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
2403 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
2405 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
2406 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
2407 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
2408 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
2409 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
2410 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2411 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2412 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
2413 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
2414 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
2415 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
2416 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2417 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2418 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
2419 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
2420 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
2421 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2422 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
2423 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
2424 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
2425 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2426 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
2427 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
2428 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2429 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2430 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2431 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2432 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
2433 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
2434 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
2435 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
2436 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
2438 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2440 map<string, string> specializations;
2441 ComputeShaderSpec spec;
2443 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
2444 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
2445 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
2446 specializations["SC_OP"] = cases[caseNdx].scOperation;
2447 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
2449 spec.assembly = shaderTemplate.specialize(specializations);
2450 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2451 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
2452 spec.numWorkGroups = IVec3(numElements, 1, 1);
2453 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
2454 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
2456 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec));
2459 ComputeShaderSpec spec;
2462 string(getComputeAsmShaderPreamble()) +
2464 "OpName %main \"main\"\n"
2465 "OpName %id \"gl_GlobalInvocationID\"\n"
2467 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2468 "OpDecorate %sc_0 SpecId 0\n"
2469 "OpDecorate %sc_1 SpecId 1\n"
2470 "OpDecorate %sc_2 SpecId 2\n"
2471 "OpDecorate %i32arr ArrayStride 4\n"
2473 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2475 "%ivec3 = OpTypeVector %i32 3\n"
2476 "%buf = OpTypeStruct %i32arr\n"
2477 "%bufptr = OpTypePointer Uniform %buf\n"
2478 "%indata = OpVariable %bufptr Uniform\n"
2479 "%outdata = OpVariable %bufptr Uniform\n"
2481 "%id = OpVariable %uvec3ptr Input\n"
2482 "%zero = OpConstant %i32 0\n"
2483 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
2484 "%vec3_undef = OpUndef %ivec3\n"
2486 "%sc_0 = OpSpecConstant %i32 0\n"
2487 "%sc_1 = OpSpecConstant %i32 0\n"
2488 "%sc_2 = OpSpecConstant %i32 0\n"
2489 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
2490 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
2491 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
2492 "%sc_vec3_0_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
2493 "%sc_vec3_1_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
2494 "%sc_vec3_2_s = OpSpecConstantOp %ivec3 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
2495 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
2496 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
2497 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
2498 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
2499 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
2500 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
2501 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
2503 "%main = OpFunction %void None %voidf\n"
2504 "%label = OpLabel\n"
2505 "%idval = OpLoad %uvec3 %id\n"
2506 "%x = OpCompositeExtract %u32 %idval 0\n"
2507 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2508 "%inval = OpLoad %i32 %inloc\n"
2509 "%final = OpIAdd %i32 %inval %sc_final\n"
2510 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2511 " OpStore %outloc %final\n"
2514 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2515 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
2516 spec.numWorkGroups = IVec3(numElements, 1, 1);
2517 spec.specConstants.push_back(123);
2518 spec.specConstants.push_back(56);
2519 spec.specConstants.push_back(-77);
2521 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
2523 return group.release();
2526 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
2528 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
2529 ComputeShaderSpec spec1;
2530 ComputeShaderSpec spec2;
2531 ComputeShaderSpec spec3;
2532 de::Random rnd (deStringHash(group->getName()));
2533 const int numElements = 100;
2534 vector<float> inputFloats (numElements, 0);
2535 vector<float> outputFloats1 (numElements, 0);
2536 vector<float> outputFloats2 (numElements, 0);
2537 vector<float> outputFloats3 (numElements, 0);
2539 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
2541 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2542 floorAll(inputFloats);
2544 for (size_t ndx = 0; ndx < numElements; ++ndx)
2548 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
2549 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
2550 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
2553 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
2554 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
2558 string(getComputeAsmShaderPreamble()) +
2560 "OpSource GLSL 430\n"
2561 "OpName %main \"main\"\n"
2562 "OpName %id \"gl_GlobalInvocationID\"\n"
2564 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2566 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2568 "%id = OpVariable %uvec3ptr Input\n"
2569 "%zero = OpConstant %i32 0\n"
2570 "%three = OpConstant %u32 3\n"
2571 "%constf5p5 = OpConstant %f32 5.5\n"
2572 "%constf20p5 = OpConstant %f32 20.5\n"
2573 "%constf1p75 = OpConstant %f32 1.75\n"
2574 "%constf8p5 = OpConstant %f32 8.5\n"
2575 "%constf6p5 = OpConstant %f32 6.5\n"
2577 "%main = OpFunction %void None %voidf\n"
2578 "%entry = OpLabel\n"
2579 "%idval = OpLoad %uvec3 %id\n"
2580 "%x = OpCompositeExtract %u32 %idval 0\n"
2581 "%selector = OpUMod %u32 %x %three\n"
2582 " OpSelectionMerge %phi None\n"
2583 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
2585 // Case 1 before OpPhi.
2586 "%case1 = OpLabel\n"
2589 "%default = OpLabel\n"
2593 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
2594 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2595 "%inval = OpLoad %f32 %inloc\n"
2596 "%add = OpFAdd %f32 %inval %operand\n"
2597 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2598 " OpStore %outloc %add\n"
2601 // Case 0 after OpPhi.
2602 "%case0 = OpLabel\n"
2606 // Case 2 after OpPhi.
2607 "%case2 = OpLabel\n"
2611 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2612 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
2613 spec1.numWorkGroups = IVec3(numElements, 1, 1);
2615 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
2618 string(getComputeAsmShaderPreamble()) +
2620 "OpName %main \"main\"\n"
2621 "OpName %id \"gl_GlobalInvocationID\"\n"
2623 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2625 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2627 "%id = OpVariable %uvec3ptr Input\n"
2628 "%zero = OpConstant %i32 0\n"
2629 "%one = OpConstant %i32 1\n"
2630 "%three = OpConstant %i32 3\n"
2631 "%constf6p5 = OpConstant %f32 6.5\n"
2633 "%main = OpFunction %void None %voidf\n"
2634 "%entry = OpLabel\n"
2635 "%idval = OpLoad %uvec3 %id\n"
2636 "%x = OpCompositeExtract %u32 %idval 0\n"
2637 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2638 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2639 "%inval = OpLoad %f32 %inloc\n"
2643 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
2644 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
2645 "%step_next = OpIAdd %i32 %step %one\n"
2646 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
2647 "%still_loop = OpSLessThan %bool %step %three\n"
2648 " OpLoopMerge %exit %phi None\n"
2649 " OpBranchConditional %still_loop %phi %exit\n"
2652 " OpStore %outloc %accum\n"
2655 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2656 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2657 spec2.numWorkGroups = IVec3(numElements, 1, 1);
2659 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
2662 string(getComputeAsmShaderPreamble()) +
2664 "OpName %main \"main\"\n"
2665 "OpName %id \"gl_GlobalInvocationID\"\n"
2667 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2669 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2671 "%f32ptr_f = OpTypePointer Function %f32\n"
2672 "%id = OpVariable %uvec3ptr Input\n"
2673 "%true = OpConstantTrue %bool\n"
2674 "%false = OpConstantFalse %bool\n"
2675 "%zero = OpConstant %i32 0\n"
2676 "%constf8p5 = OpConstant %f32 8.5\n"
2678 "%main = OpFunction %void None %voidf\n"
2679 "%entry = OpLabel\n"
2680 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
2681 "%idval = OpLoad %uvec3 %id\n"
2682 "%x = OpCompositeExtract %u32 %idval 0\n"
2683 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2684 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2685 "%a_init = OpLoad %f32 %inloc\n"
2686 "%b_init = OpLoad %f32 %b\n"
2690 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
2691 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
2692 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
2693 " OpLoopMerge %exit %phi None\n"
2694 " OpBranchConditional %still_loop %phi %exit\n"
2697 "%sub = OpFSub %f32 %a_next %b_next\n"
2698 " OpStore %outloc %sub\n"
2701 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2702 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
2703 spec3.numWorkGroups = IVec3(numElements, 1, 1);
2705 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
2707 return group.release();
2710 // Assembly code used for testing block order is based on GLSL source code:
2714 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2715 // float elements[];
2717 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2718 // float elements[];
2722 // uint x = gl_GlobalInvocationID.x;
2723 // output_data.elements[x] = input_data.elements[x];
2724 // if (x > uint(50)) {
2725 // switch (x % uint(3)) {
2726 // case 0: output_data.elements[x] += 1.5f; break;
2727 // case 1: output_data.elements[x] += 42.f; break;
2728 // case 2: output_data.elements[x] -= 27.f; break;
2732 // output_data.elements[x] = -input_data.elements[x];
2735 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
2737 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
2738 ComputeShaderSpec spec;
2739 de::Random rnd (deStringHash(group->getName()));
2740 const int numElements = 100;
2741 vector<float> inputFloats (numElements, 0);
2742 vector<float> outputFloats (numElements, 0);
2744 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2746 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2747 floorAll(inputFloats);
2749 for (size_t ndx = 0; ndx <= 50; ++ndx)
2750 outputFloats[ndx] = -inputFloats[ndx];
2752 for (size_t ndx = 51; ndx < numElements; ++ndx)
2756 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
2757 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
2758 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
2764 string(getComputeAsmShaderPreamble()) +
2766 "OpSource GLSL 430\n"
2767 "OpName %main \"main\"\n"
2768 "OpName %id \"gl_GlobalInvocationID\"\n"
2770 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2772 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2774 "%u32ptr = OpTypePointer Function %u32\n"
2775 "%u32ptr_input = OpTypePointer Input %u32\n"
2777 + string(getComputeAsmInputOutputBuffer()) +
2779 "%id = OpVariable %uvec3ptr Input\n"
2780 "%zero = OpConstant %i32 0\n"
2781 "%const3 = OpConstant %u32 3\n"
2782 "%const50 = OpConstant %u32 50\n"
2783 "%constf1p5 = OpConstant %f32 1.5\n"
2784 "%constf27 = OpConstant %f32 27.0\n"
2785 "%constf42 = OpConstant %f32 42.0\n"
2787 "%main = OpFunction %void None %voidf\n"
2790 "%entry = OpLabel\n"
2792 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
2793 "%xvar = OpVariable %u32ptr Function\n"
2794 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
2795 "%x = OpLoad %u32 %xptr\n"
2796 " OpStore %xvar %x\n"
2798 "%cmp = OpUGreaterThan %bool %x %const50\n"
2799 " OpSelectionMerge %if_merge None\n"
2800 " OpBranchConditional %cmp %if_true %if_false\n"
2802 // False branch for if-statement: placed in the middle of switch cases and before true branch.
2803 "%if_false = OpLabel\n"
2804 "%x_f = OpLoad %u32 %xvar\n"
2805 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
2806 "%inval_f = OpLoad %f32 %inloc_f\n"
2807 "%negate = OpFNegate %f32 %inval_f\n"
2808 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
2809 " OpStore %outloc_f %negate\n"
2810 " OpBranch %if_merge\n"
2812 // Merge block for if-statement: placed in the middle of true and false branch.
2813 "%if_merge = OpLabel\n"
2816 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
2817 "%if_true = OpLabel\n"
2818 "%xval_t = OpLoad %u32 %xvar\n"
2819 "%mod = OpUMod %u32 %xval_t %const3\n"
2820 " OpSelectionMerge %switch_merge None\n"
2821 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
2823 // Merge block for switch-statement: placed before the case
2824 // bodies. But it must follow OpSwitch which dominates it.
2825 "%switch_merge = OpLabel\n"
2826 " OpBranch %if_merge\n"
2828 // Case 1 for switch-statement: placed before case 0.
2829 // It must follow the OpSwitch that dominates it.
2830 "%case1 = OpLabel\n"
2831 "%x_1 = OpLoad %u32 %xvar\n"
2832 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
2833 "%inval_1 = OpLoad %f32 %inloc_1\n"
2834 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
2835 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
2836 " OpStore %outloc_1 %addf42\n"
2837 " OpBranch %switch_merge\n"
2839 // Case 2 for switch-statement.
2840 "%case2 = OpLabel\n"
2841 "%x_2 = OpLoad %u32 %xvar\n"
2842 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
2843 "%inval_2 = OpLoad %f32 %inloc_2\n"
2844 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
2845 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
2846 " OpStore %outloc_2 %subf27\n"
2847 " OpBranch %switch_merge\n"
2849 // Default case for switch-statement: placed in the middle of normal cases.
2850 "%default = OpLabel\n"
2851 " OpBranch %switch_merge\n"
2853 // Case 0 for switch-statement: out of order.
2854 "%case0 = OpLabel\n"
2855 "%x_0 = OpLoad %u32 %xvar\n"
2856 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
2857 "%inval_0 = OpLoad %f32 %inloc_0\n"
2858 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
2859 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
2860 " OpStore %outloc_0 %addf1p5\n"
2861 " OpBranch %switch_merge\n"
2864 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2865 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2866 spec.numWorkGroups = IVec3(numElements, 1, 1);
2868 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
2870 return group.release();
2873 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
2875 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
2876 ComputeShaderSpec spec1;
2877 ComputeShaderSpec spec2;
2878 de::Random rnd (deStringHash(group->getName()));
2879 const int numElements = 100;
2880 vector<float> inputFloats (numElements, 0);
2881 vector<float> outputFloats1 (numElements, 0);
2882 vector<float> outputFloats2 (numElements, 0);
2883 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
2885 for (size_t ndx = 0; ndx < numElements; ++ndx)
2887 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
2888 outputFloats2[ndx] = -inputFloats[ndx];
2891 const string assembly(
2892 "OpCapability Shader\n"
2893 "OpCapability ClipDistance\n"
2894 "OpMemoryModel Logical GLSL450\n"
2895 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
2896 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
2897 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
2898 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
2899 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
2900 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
2902 "OpName %comp_main1 \"entrypoint1\"\n"
2903 "OpName %comp_main2 \"entrypoint2\"\n"
2904 "OpName %vert_main \"entrypoint2\"\n"
2905 "OpName %id \"gl_GlobalInvocationID\"\n"
2906 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
2907 "OpName %vertexIndex \"gl_VertexIndex\"\n"
2908 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
2909 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
2910 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
2911 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
2913 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2914 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
2915 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
2916 "OpDecorate %vert_builtin_st Block\n"
2917 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
2918 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
2919 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
2921 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2923 "%zero = OpConstant %i32 0\n"
2924 "%one = OpConstant %u32 1\n"
2925 "%c_f32_1 = OpConstant %f32 1\n"
2927 "%i32inputptr = OpTypePointer Input %i32\n"
2928 "%vec4 = OpTypeVector %f32 4\n"
2929 "%vec4ptr = OpTypePointer Output %vec4\n"
2930 "%f32arr1 = OpTypeArray %f32 %one\n"
2931 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
2932 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
2933 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
2935 "%id = OpVariable %uvec3ptr Input\n"
2936 "%vertexIndex = OpVariable %i32inputptr Input\n"
2937 "%instanceIndex = OpVariable %i32inputptr Input\n"
2938 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
2940 // gl_Position = vec4(1.);
2941 "%vert_main = OpFunction %void None %voidf\n"
2942 "%vert_entry = OpLabel\n"
2943 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
2944 " OpStore %position %c_vec4_1\n"
2949 "%comp_main1 = OpFunction %void None %voidf\n"
2950 "%comp1_entry = OpLabel\n"
2951 "%idval1 = OpLoad %uvec3 %id\n"
2952 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
2953 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
2954 "%inval1 = OpLoad %f32 %inloc1\n"
2955 "%add = OpFAdd %f32 %inval1 %inval1\n"
2956 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
2957 " OpStore %outloc1 %add\n"
2962 "%comp_main2 = OpFunction %void None %voidf\n"
2963 "%comp2_entry = OpLabel\n"
2964 "%idval2 = OpLoad %uvec3 %id\n"
2965 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
2966 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
2967 "%inval2 = OpLoad %f32 %inloc2\n"
2968 "%neg = OpFNegate %f32 %inval2\n"
2969 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
2970 " OpStore %outloc2 %neg\n"
2972 " OpFunctionEnd\n");
2974 spec1.assembly = assembly;
2975 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2976 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
2977 spec1.numWorkGroups = IVec3(numElements, 1, 1);
2978 spec1.entryPoint = "entrypoint1";
2980 spec2.assembly = assembly;
2981 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2982 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2983 spec2.numWorkGroups = IVec3(numElements, 1, 1);
2984 spec2.entryPoint = "entrypoint2";
2986 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
2987 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
2989 return group.release();
2992 inline std::string makeLongUTF8String (size_t num4ByteChars)
2994 // An example of a longest valid UTF-8 character. Be explicit about the
2995 // character type because Microsoft compilers can otherwise interpret the
2996 // character string as being over wide (16-bit) characters. Ideally, we
2997 // would just use a C++11 UTF-8 string literal, but we want to support older
2998 // Microsoft compilers.
2999 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
3000 std::string longString;
3001 longString.reserve(num4ByteChars * 4);
3002 for (size_t count = 0; count < num4ByteChars; count++)
3004 longString += earthAfrica;
3009 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
3011 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
3012 vector<CaseParameter> cases;
3013 de::Random rnd (deStringHash(group->getName()));
3014 const int numElements = 100;
3015 vector<float> positiveFloats (numElements, 0);
3016 vector<float> negativeFloats (numElements, 0);
3017 const StringTemplate shaderTemplate (
3018 "OpCapability Shader\n"
3019 "OpMemoryModel Logical GLSL450\n"
3021 "OpEntryPoint GLCompute %main \"main\" %id\n"
3022 "OpExecutionMode %main LocalSize 1 1 1\n"
3026 "OpName %main \"main\"\n"
3027 "OpName %id \"gl_GlobalInvocationID\"\n"
3029 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3031 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3033 "%id = OpVariable %uvec3ptr Input\n"
3034 "%zero = OpConstant %i32 0\n"
3036 "%main = OpFunction %void None %voidf\n"
3037 "%label = OpLabel\n"
3038 "%idval = OpLoad %uvec3 %id\n"
3039 "%x = OpCompositeExtract %u32 %idval 0\n"
3040 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3041 "%inval = OpLoad %f32 %inloc\n"
3042 "%neg = OpFNegate %f32 %inval\n"
3043 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3044 " OpStore %outloc %neg\n"
3046 " OpFunctionEnd\n");
3048 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
3049 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
3050 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
3051 "OpSource GLSL 430 %fname"));
3052 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
3053 "OpSource GLSL 430 %fname"));
3054 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
3055 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
3056 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
3057 "OpSource GLSL 430 %fname \"\""));
3058 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
3059 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
3060 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
3061 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
3062 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
3063 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
3064 "OpSourceContinued \"id main() {}\""));
3065 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
3066 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3067 "OpSourceContinued \"\""));
3068 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
3069 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3070 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
3071 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
3072 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3073 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
3074 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
3075 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
3076 "OpSourceContinued \"void\"\n"
3077 "OpSourceContinued \"main()\"\n"
3078 "OpSourceContinued \"{}\""));
3079 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
3080 "OpSource GLSL 430 %fname \"\"\n"
3081 "OpSourceContinued \"#version 430\nvoid main() {}\""));
3083 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3085 for (size_t ndx = 0; ndx < numElements; ++ndx)
3086 negativeFloats[ndx] = -positiveFloats[ndx];
3088 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3090 map<string, string> specializations;
3091 ComputeShaderSpec spec;
3093 specializations["SOURCE"] = cases[caseNdx].param;
3094 spec.assembly = shaderTemplate.specialize(specializations);
3095 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3096 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3097 spec.numWorkGroups = IVec3(numElements, 1, 1);
3099 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3102 return group.release();
3105 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
3107 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
3108 vector<CaseParameter> cases;
3109 de::Random rnd (deStringHash(group->getName()));
3110 const int numElements = 100;
3111 vector<float> inputFloats (numElements, 0);
3112 vector<float> outputFloats (numElements, 0);
3113 const StringTemplate shaderTemplate (
3114 string(getComputeAsmShaderPreamble()) +
3116 "OpSourceExtension \"${EXTENSION}\"\n"
3118 "OpName %main \"main\"\n"
3119 "OpName %id \"gl_GlobalInvocationID\"\n"
3121 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3123 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3125 "%id = OpVariable %uvec3ptr Input\n"
3126 "%zero = OpConstant %i32 0\n"
3128 "%main = OpFunction %void None %voidf\n"
3129 "%label = OpLabel\n"
3130 "%idval = OpLoad %uvec3 %id\n"
3131 "%x = OpCompositeExtract %u32 %idval 0\n"
3132 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3133 "%inval = OpLoad %f32 %inloc\n"
3134 "%neg = OpFNegate %f32 %inval\n"
3135 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3136 " OpStore %outloc %neg\n"
3138 " OpFunctionEnd\n");
3140 cases.push_back(CaseParameter("empty_extension", ""));
3141 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
3142 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
3143 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
3144 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
3146 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
3148 for (size_t ndx = 0; ndx < numElements; ++ndx)
3149 outputFloats[ndx] = -inputFloats[ndx];
3151 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3153 map<string, string> specializations;
3154 ComputeShaderSpec spec;
3156 specializations["EXTENSION"] = cases[caseNdx].param;
3157 spec.assembly = shaderTemplate.specialize(specializations);
3158 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3159 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3160 spec.numWorkGroups = IVec3(numElements, 1, 1);
3162 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3165 return group.release();
3168 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
3169 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
3171 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
3172 vector<CaseParameter> cases;
3173 de::Random rnd (deStringHash(group->getName()));
3174 const int numElements = 100;
3175 vector<float> positiveFloats (numElements, 0);
3176 vector<float> negativeFloats (numElements, 0);
3177 const StringTemplate shaderTemplate (
3178 string(getComputeAsmShaderPreamble()) +
3180 "OpSource GLSL 430\n"
3181 "OpName %main \"main\"\n"
3182 "OpName %id \"gl_GlobalInvocationID\"\n"
3184 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3186 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3187 "%uvec2 = OpTypeVector %u32 2\n"
3188 "%bvec3 = OpTypeVector %bool 3\n"
3189 "%fvec4 = OpTypeVector %f32 4\n"
3190 "%fmat33 = OpTypeMatrix %fvec3 3\n"
3191 "%const100 = OpConstant %u32 100\n"
3192 "%uarr100 = OpTypeArray %i32 %const100\n"
3193 "%struct = OpTypeStruct %f32 %i32 %u32\n"
3194 "%pointer = OpTypePointer Function %i32\n"
3195 + string(getComputeAsmInputOutputBuffer()) +
3197 "%null = OpConstantNull ${TYPE}\n"
3199 "%id = OpVariable %uvec3ptr Input\n"
3200 "%zero = OpConstant %i32 0\n"
3202 "%main = OpFunction %void None %voidf\n"
3203 "%label = OpLabel\n"
3204 "%idval = OpLoad %uvec3 %id\n"
3205 "%x = OpCompositeExtract %u32 %idval 0\n"
3206 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3207 "%inval = OpLoad %f32 %inloc\n"
3208 "%neg = OpFNegate %f32 %inval\n"
3209 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3210 " OpStore %outloc %neg\n"
3212 " OpFunctionEnd\n");
3214 cases.push_back(CaseParameter("bool", "%bool"));
3215 cases.push_back(CaseParameter("sint32", "%i32"));
3216 cases.push_back(CaseParameter("uint32", "%u32"));
3217 cases.push_back(CaseParameter("float32", "%f32"));
3218 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
3219 cases.push_back(CaseParameter("vec3bool", "%bvec3"));
3220 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
3221 cases.push_back(CaseParameter("matrix", "%fmat33"));
3222 cases.push_back(CaseParameter("array", "%uarr100"));
3223 cases.push_back(CaseParameter("struct", "%struct"));
3224 cases.push_back(CaseParameter("pointer", "%pointer"));
3226 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3228 for (size_t ndx = 0; ndx < numElements; ++ndx)
3229 negativeFloats[ndx] = -positiveFloats[ndx];
3231 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3233 map<string, string> specializations;
3234 ComputeShaderSpec spec;
3236 specializations["TYPE"] = cases[caseNdx].param;
3237 spec.assembly = shaderTemplate.specialize(specializations);
3238 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3239 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3240 spec.numWorkGroups = IVec3(numElements, 1, 1);
3242 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3245 return group.release();
3248 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
3249 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
3251 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
3252 vector<CaseParameter> cases;
3253 de::Random rnd (deStringHash(group->getName()));
3254 const int numElements = 100;
3255 vector<float> positiveFloats (numElements, 0);
3256 vector<float> negativeFloats (numElements, 0);
3257 const StringTemplate shaderTemplate (
3258 string(getComputeAsmShaderPreamble()) +
3260 "OpSource GLSL 430\n"
3261 "OpName %main \"main\"\n"
3262 "OpName %id \"gl_GlobalInvocationID\"\n"
3264 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3266 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3268 "%id = OpVariable %uvec3ptr Input\n"
3269 "%zero = OpConstant %i32 0\n"
3273 "%main = OpFunction %void None %voidf\n"
3274 "%label = OpLabel\n"
3275 "%idval = OpLoad %uvec3 %id\n"
3276 "%x = OpCompositeExtract %u32 %idval 0\n"
3277 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3278 "%inval = OpLoad %f32 %inloc\n"
3279 "%neg = OpFNegate %f32 %inval\n"
3280 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3281 " OpStore %outloc %neg\n"
3283 " OpFunctionEnd\n");
3285 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
3286 "%const = OpConstantComposite %uvec3 %five %zero %five"));
3287 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
3288 "%ten = OpConstant %f32 10.\n"
3289 "%fzero = OpConstant %f32 0.\n"
3290 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
3291 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
3292 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
3293 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
3294 "%fzero = OpConstant %f32 0.\n"
3295 "%one = OpConstant %f32 1.\n"
3296 "%point5 = OpConstant %f32 0.5\n"
3297 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
3298 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
3299 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
3300 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
3301 "%st2 = OpTypeStruct %i32 %i32\n"
3302 "%struct = OpTypeStruct %st1 %st2\n"
3303 "%point5 = OpConstant %f32 0.5\n"
3304 "%one = OpConstant %u32 1\n"
3305 "%ten = OpConstant %i32 10\n"
3306 "%st1val = OpConstantComposite %st1 %one %point5\n"
3307 "%st2val = OpConstantComposite %st2 %ten %ten\n"
3308 "%const = OpConstantComposite %struct %st1val %st2val"));
3310 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3312 for (size_t ndx = 0; ndx < numElements; ++ndx)
3313 negativeFloats[ndx] = -positiveFloats[ndx];
3315 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3317 map<string, string> specializations;
3318 ComputeShaderSpec spec;
3320 specializations["CONSTANT"] = cases[caseNdx].param;
3321 spec.assembly = shaderTemplate.specialize(specializations);
3322 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3323 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3324 spec.numWorkGroups = IVec3(numElements, 1, 1);
3326 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3329 return group.release();
3332 // Creates a floating point number with the given exponent, and significand
3333 // bits set. It can only create normalized numbers. Only the least significant
3334 // 24 bits of the significand will be examined. The final bit of the
3335 // significand will also be ignored. This allows alignment to be written
3336 // similarly to C99 hex-floats.
3337 // For example if you wanted to write 0x1.7f34p-12 you would call
3338 // constructNormalizedFloat(-12, 0x7f3400)
3339 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
3343 for (deInt32 idx = 0; idx < 23; ++idx)
3345 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
3349 return std::ldexp(f, exponent);
3352 // Compare instruction for the OpQuantizeF16 compute exact case.
3353 // Returns true if the output is what is expected from the test case.
3354 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
3356 if (outputAllocs.size() != 1)
3359 // We really just need this for size because we cannot compare Nans.
3360 const BufferSp& expectedOutput = expectedOutputs[0];
3361 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
3363 if (expectedOutput->getNumBytes() != 4*sizeof(float)) {
3367 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
3368 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
3373 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
3374 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
3379 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
3380 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
3385 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
3386 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
3393 // Checks that every output from a test-case is a float NaN.
3394 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
3396 if (outputAllocs.size() != 1)
3399 // We really just need this for size because we cannot compare Nans.
3400 const BufferSp& expectedOutput = expectedOutputs[0];
3401 const float* output_as_float = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
3403 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
3405 if (!deFloatIsNaN(output_as_float[idx]))
3414 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
3415 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
3417 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
3419 const std::string shader (
3420 string(getComputeAsmShaderPreamble()) +
3422 "OpSource GLSL 430\n"
3423 "OpName %main \"main\"\n"
3424 "OpName %id \"gl_GlobalInvocationID\"\n"
3426 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3428 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3430 "%id = OpVariable %uvec3ptr Input\n"
3431 "%zero = OpConstant %i32 0\n"
3433 "%main = OpFunction %void None %voidf\n"
3434 "%label = OpLabel\n"
3435 "%idval = OpLoad %uvec3 %id\n"
3436 "%x = OpCompositeExtract %u32 %idval 0\n"
3437 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3438 "%inval = OpLoad %f32 %inloc\n"
3439 "%quant = OpQuantizeToF16 %f32 %inval\n"
3440 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3441 " OpStore %outloc %quant\n"
3443 " OpFunctionEnd\n");
3446 ComputeShaderSpec spec;
3447 const deUint32 numElements = 100;
3448 vector<float> infinities;
3449 vector<float> results;
3451 infinities.reserve(numElements);
3452 results.reserve(numElements);
3454 for (size_t idx = 0; idx < numElements; ++idx)
3459 infinities.push_back(std::numeric_limits<float>::infinity());
3460 results.push_back(std::numeric_limits<float>::infinity());
3463 infinities.push_back(-std::numeric_limits<float>::infinity());
3464 results.push_back(-std::numeric_limits<float>::infinity());
3467 infinities.push_back(std::ldexp(1.0f, 16));
3468 results.push_back(std::numeric_limits<float>::infinity());
3471 infinities.push_back(std::ldexp(-1.0f, 32));
3472 results.push_back(-std::numeric_limits<float>::infinity());
3477 spec.assembly = shader;
3478 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
3479 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
3480 spec.numWorkGroups = IVec3(numElements, 1, 1);
3482 group->addChild(new SpvAsmComputeShaderCase(
3483 testCtx, "infinities", "Check that infinities propagated and created", spec));
3487 ComputeShaderSpec spec;
3489 const deUint32 numElements = 100;
3491 nans.reserve(numElements);
3493 for (size_t idx = 0; idx < numElements; ++idx)
3497 nans.push_back(std::numeric_limits<float>::quiet_NaN());
3501 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
3505 spec.assembly = shader;
3506 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
3507 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
3508 spec.numWorkGroups = IVec3(numElements, 1, 1);
3509 spec.verifyIO = &compareNan;
3511 group->addChild(new SpvAsmComputeShaderCase(
3512 testCtx, "propagated_nans", "Check that nans are propagated", spec));
3516 ComputeShaderSpec spec;
3517 vector<float> small;
3518 vector<float> zeros;
3519 const deUint32 numElements = 100;
3521 small.reserve(numElements);
3522 zeros.reserve(numElements);
3524 for (size_t idx = 0; idx < numElements; ++idx)
3529 small.push_back(0.f);
3530 zeros.push_back(0.f);
3533 small.push_back(-0.f);
3534 zeros.push_back(-0.f);
3537 small.push_back(std::ldexp(1.0f, -16));
3538 zeros.push_back(0.f);
3541 small.push_back(std::ldexp(-1.0f, -32));
3542 zeros.push_back(-0.f);
3545 small.push_back(std::ldexp(1.0f, -127));
3546 zeros.push_back(0.f);
3549 small.push_back(-std::ldexp(1.0f, -128));
3550 zeros.push_back(-0.f);
3555 spec.assembly = shader;
3556 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
3557 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
3558 spec.numWorkGroups = IVec3(numElements, 1, 1);
3560 group->addChild(new SpvAsmComputeShaderCase(
3561 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
3565 ComputeShaderSpec spec;
3566 vector<float> exact;
3567 const deUint32 numElements = 200;
3569 exact.reserve(numElements);
3571 for (size_t idx = 0; idx < numElements; ++idx)
3572 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
3574 spec.assembly = shader;
3575 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
3576 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
3577 spec.numWorkGroups = IVec3(numElements, 1, 1);
3579 group->addChild(new SpvAsmComputeShaderCase(
3580 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
3584 ComputeShaderSpec spec;
3585 vector<float> inputs;
3586 const deUint32 numElements = 4;
3588 inputs.push_back(constructNormalizedFloat(8, 0x300300));
3589 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
3590 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
3591 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
3593 spec.assembly = shader;
3594 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
3595 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3596 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
3597 spec.numWorkGroups = IVec3(numElements, 1, 1);
3599 group->addChild(new SpvAsmComputeShaderCase(
3600 testCtx, "rounded", "Check that are rounded when needed", spec));
3603 return group.release();
3606 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
3608 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
3610 const std::string shader (
3611 string(getComputeAsmShaderPreamble()) +
3613 "OpName %main \"main\"\n"
3614 "OpName %id \"gl_GlobalInvocationID\"\n"
3616 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3618 "OpDecorate %sc_0 SpecId 0\n"
3619 "OpDecorate %sc_1 SpecId 1\n"
3620 "OpDecorate %sc_2 SpecId 2\n"
3621 "OpDecorate %sc_3 SpecId 3\n"
3622 "OpDecorate %sc_4 SpecId 4\n"
3623 "OpDecorate %sc_5 SpecId 5\n"
3625 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3627 "%id = OpVariable %uvec3ptr Input\n"
3628 "%zero = OpConstant %i32 0\n"
3629 "%c_u32_6 = OpConstant %u32 6\n"
3631 "%sc_0 = OpSpecConstant %f32 0.\n"
3632 "%sc_1 = OpSpecConstant %f32 0.\n"
3633 "%sc_2 = OpSpecConstant %f32 0.\n"
3634 "%sc_3 = OpSpecConstant %f32 0.\n"
3635 "%sc_4 = OpSpecConstant %f32 0.\n"
3636 "%sc_5 = OpSpecConstant %f32 0.\n"
3638 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
3639 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
3640 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
3641 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
3642 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
3643 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
3645 "%main = OpFunction %void None %voidf\n"
3646 "%label = OpLabel\n"
3647 "%idval = OpLoad %uvec3 %id\n"
3648 "%x = OpCompositeExtract %u32 %idval 0\n"
3649 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3650 "%selector = OpUMod %u32 %x %c_u32_6\n"
3651 " OpSelectionMerge %exit None\n"
3652 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
3654 "%case0 = OpLabel\n"
3655 " OpStore %outloc %sc_0_quant\n"
3658 "%case1 = OpLabel\n"
3659 " OpStore %outloc %sc_1_quant\n"
3662 "%case2 = OpLabel\n"
3663 " OpStore %outloc %sc_2_quant\n"
3666 "%case3 = OpLabel\n"
3667 " OpStore %outloc %sc_3_quant\n"
3670 "%case4 = OpLabel\n"
3671 " OpStore %outloc %sc_4_quant\n"
3674 "%case5 = OpLabel\n"
3675 " OpStore %outloc %sc_5_quant\n"
3681 " OpFunctionEnd\n");
3684 ComputeShaderSpec spec;
3685 const deUint8 numCases = 4;
3686 vector<float> inputs (numCases, 0.f);
3687 vector<float> outputs;
3689 spec.assembly = shader;
3690 spec.numWorkGroups = IVec3(numCases, 1, 1);
3692 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
3693 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
3694 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
3695 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
3697 outputs.push_back(std::numeric_limits<float>::infinity());
3698 outputs.push_back(-std::numeric_limits<float>::infinity());
3699 outputs.push_back(std::numeric_limits<float>::infinity());
3700 outputs.push_back(-std::numeric_limits<float>::infinity());
3702 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3703 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3705 group->addChild(new SpvAsmComputeShaderCase(
3706 testCtx, "infinities", "Check that infinities propagated and created", spec));
3710 ComputeShaderSpec spec;
3711 const deUint8 numCases = 2;
3712 vector<float> inputs (numCases, 0.f);
3713 vector<float> outputs;
3715 spec.assembly = shader;
3716 spec.numWorkGroups = IVec3(numCases, 1, 1);
3717 spec.verifyIO = &compareNan;
3719 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
3720 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
3722 for (deUint8 idx = 0; idx < numCases; ++idx)
3723 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
3725 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3726 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3728 group->addChild(new SpvAsmComputeShaderCase(
3729 testCtx, "propagated_nans", "Check that nans are propagated", spec));
3733 ComputeShaderSpec spec;
3734 const deUint8 numCases = 6;
3735 vector<float> inputs (numCases, 0.f);
3736 vector<float> outputs;
3738 spec.assembly = shader;
3739 spec.numWorkGroups = IVec3(numCases, 1, 1);
3741 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
3742 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
3743 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
3744 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
3745 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
3746 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
3748 outputs.push_back(0.f);
3749 outputs.push_back(-0.f);
3750 outputs.push_back(0.f);
3751 outputs.push_back(-0.f);
3752 outputs.push_back(0.f);
3753 outputs.push_back(-0.f);
3755 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3756 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3758 group->addChild(new SpvAsmComputeShaderCase(
3759 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
3763 ComputeShaderSpec spec;
3764 const deUint8 numCases = 6;
3765 vector<float> inputs (numCases, 0.f);
3766 vector<float> outputs;
3768 spec.assembly = shader;
3769 spec.numWorkGroups = IVec3(numCases, 1, 1);
3771 for (deUint8 idx = 0; idx < 6; ++idx)
3773 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
3774 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
3775 outputs.push_back(f);
3778 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3779 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3781 group->addChild(new SpvAsmComputeShaderCase(
3782 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
3786 ComputeShaderSpec spec;
3787 const deUint8 numCases = 4;
3788 vector<float> inputs (numCases, 0.f);
3789 vector<float> outputs;
3791 spec.assembly = shader;
3792 spec.numWorkGroups = IVec3(numCases, 1, 1);
3793 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
3795 outputs.push_back(constructNormalizedFloat(8, 0x300300));
3796 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
3797 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
3798 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
3800 for (deUint8 idx = 0; idx < numCases; ++idx)
3801 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
3803 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3804 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3806 group->addChild(new SpvAsmComputeShaderCase(
3807 testCtx, "rounded", "Check that are rounded when needed", spec));
3810 return group.release();
3813 // Checks that constant null/composite values can be used in computation.
3814 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
3816 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
3817 ComputeShaderSpec spec;
3818 de::Random rnd (deStringHash(group->getName()));
3819 const int numElements = 100;
3820 vector<float> positiveFloats (numElements, 0);
3821 vector<float> negativeFloats (numElements, 0);
3823 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3825 for (size_t ndx = 0; ndx < numElements; ++ndx)
3826 negativeFloats[ndx] = -positiveFloats[ndx];
3829 "OpCapability Shader\n"
3830 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
3831 "OpMemoryModel Logical GLSL450\n"
3832 "OpEntryPoint GLCompute %main \"main\" %id\n"
3833 "OpExecutionMode %main LocalSize 1 1 1\n"
3835 "OpSource GLSL 430\n"
3836 "OpName %main \"main\"\n"
3837 "OpName %id \"gl_GlobalInvocationID\"\n"
3839 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3841 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3843 "%fmat = OpTypeMatrix %fvec3 3\n"
3844 "%ten = OpConstant %u32 10\n"
3845 "%f32arr10 = OpTypeArray %f32 %ten\n"
3846 "%fst = OpTypeStruct %f32 %f32\n"
3848 + string(getComputeAsmInputOutputBuffer()) +
3850 "%id = OpVariable %uvec3ptr Input\n"
3851 "%zero = OpConstant %i32 0\n"
3853 // Create a bunch of null values
3854 "%unull = OpConstantNull %u32\n"
3855 "%fnull = OpConstantNull %f32\n"
3856 "%vnull = OpConstantNull %fvec3\n"
3857 "%mnull = OpConstantNull %fmat\n"
3858 "%anull = OpConstantNull %f32arr10\n"
3859 "%snull = OpConstantComposite %fst %fnull %fnull\n"
3861 "%main = OpFunction %void None %voidf\n"
3862 "%label = OpLabel\n"
3863 "%idval = OpLoad %uvec3 %id\n"
3864 "%x = OpCompositeExtract %u32 %idval 0\n"
3865 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3866 "%inval = OpLoad %f32 %inloc\n"
3867 "%neg = OpFNegate %f32 %inval\n"
3869 // Get the abs() of (a certain element of) those null values
3870 "%unull_cov = OpConvertUToF %f32 %unull\n"
3871 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
3872 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
3873 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
3874 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
3875 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
3876 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
3877 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
3878 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
3879 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
3880 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
3883 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
3884 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
3885 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
3886 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
3887 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
3888 "%final = OpFAdd %f32 %add5 %snull_abs\n"
3890 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3891 " OpStore %outloc %final\n" // write to output
3894 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3895 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3896 spec.numWorkGroups = IVec3(numElements, 1, 1);
3898 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
3900 return group.release();
3903 // Assembly code used for testing loop control is based on GLSL source code:
3906 // layout(std140, set = 0, binding = 0) readonly buffer Input {
3907 // float elements[];
3909 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
3910 // float elements[];
3914 // uint x = gl_GlobalInvocationID.x;
3915 // output_data.elements[x] = input_data.elements[x];
3916 // for (uint i = 0; i < 4; ++i)
3917 // output_data.elements[x] += 1.f;
3919 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
3921 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
3922 vector<CaseParameter> cases;
3923 de::Random rnd (deStringHash(group->getName()));
3924 const int numElements = 100;
3925 vector<float> inputFloats (numElements, 0);
3926 vector<float> outputFloats (numElements, 0);
3927 const StringTemplate shaderTemplate (
3928 string(getComputeAsmShaderPreamble()) +
3930 "OpSource GLSL 430\n"
3931 "OpName %main \"main\"\n"
3932 "OpName %id \"gl_GlobalInvocationID\"\n"
3934 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3936 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3938 "%u32ptr = OpTypePointer Function %u32\n"
3940 "%id = OpVariable %uvec3ptr Input\n"
3941 "%zero = OpConstant %i32 0\n"
3942 "%uzero = OpConstant %u32 0\n"
3943 "%one = OpConstant %i32 1\n"
3944 "%constf1 = OpConstant %f32 1.0\n"
3945 "%four = OpConstant %u32 4\n"
3947 "%main = OpFunction %void None %voidf\n"
3948 "%entry = OpLabel\n"
3949 "%i = OpVariable %u32ptr Function\n"
3950 " OpStore %i %uzero\n"
3952 "%idval = OpLoad %uvec3 %id\n"
3953 "%x = OpCompositeExtract %u32 %idval 0\n"
3954 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3955 "%inval = OpLoad %f32 %inloc\n"
3956 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3957 " OpStore %outloc %inval\n"
3958 " OpBranch %loop_entry\n"
3960 "%loop_entry = OpLabel\n"
3961 "%i_val = OpLoad %u32 %i\n"
3962 "%cmp_lt = OpULessThan %bool %i_val %four\n"
3963 " OpLoopMerge %loop_merge %loop_body ${CONTROL}\n"
3964 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
3965 "%loop_body = OpLabel\n"
3966 "%outval = OpLoad %f32 %outloc\n"
3967 "%addf1 = OpFAdd %f32 %outval %constf1\n"
3968 " OpStore %outloc %addf1\n"
3969 "%new_i = OpIAdd %u32 %i_val %one\n"
3970 " OpStore %i %new_i\n"
3971 " OpBranch %loop_entry\n"
3972 "%loop_merge = OpLabel\n"
3974 " OpFunctionEnd\n");
3976 cases.push_back(CaseParameter("none", "None"));
3977 cases.push_back(CaseParameter("unroll", "Unroll"));
3978 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
3979 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
3981 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3983 for (size_t ndx = 0; ndx < numElements; ++ndx)
3984 outputFloats[ndx] = inputFloats[ndx] + 4.f;
3986 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3988 map<string, string> specializations;
3989 ComputeShaderSpec spec;
3991 specializations["CONTROL"] = cases[caseNdx].param;
3992 spec.assembly = shaderTemplate.specialize(specializations);
3993 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3994 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3995 spec.numWorkGroups = IVec3(numElements, 1, 1);
3997 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4000 return group.release();
4003 // Assembly code used for testing selection control is based on GLSL source code:
4006 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4007 // float elements[];
4009 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4010 // float elements[];
4014 // uint x = gl_GlobalInvocationID.x;
4015 // float val = input_data.elements[x];
4017 // output_data.elements[x] = val + 1.f;
4019 // output_data.elements[x] = val - 1.f;
4021 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
4023 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
4024 vector<CaseParameter> cases;
4025 de::Random rnd (deStringHash(group->getName()));
4026 const int numElements = 100;
4027 vector<float> inputFloats (numElements, 0);
4028 vector<float> outputFloats (numElements, 0);
4029 const StringTemplate shaderTemplate (
4030 string(getComputeAsmShaderPreamble()) +
4032 "OpSource GLSL 430\n"
4033 "OpName %main \"main\"\n"
4034 "OpName %id \"gl_GlobalInvocationID\"\n"
4036 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4038 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4040 "%id = OpVariable %uvec3ptr Input\n"
4041 "%zero = OpConstant %i32 0\n"
4042 "%constf1 = OpConstant %f32 1.0\n"
4043 "%constf10 = OpConstant %f32 10.0\n"
4045 "%main = OpFunction %void None %voidf\n"
4046 "%entry = OpLabel\n"
4047 "%idval = OpLoad %uvec3 %id\n"
4048 "%x = OpCompositeExtract %u32 %idval 0\n"
4049 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4050 "%inval = OpLoad %f32 %inloc\n"
4051 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4052 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
4054 " OpSelectionMerge %if_end ${CONTROL}\n"
4055 " OpBranchConditional %cmp_gt %if_true %if_false\n"
4056 "%if_true = OpLabel\n"
4057 "%addf1 = OpFAdd %f32 %inval %constf1\n"
4058 " OpStore %outloc %addf1\n"
4059 " OpBranch %if_end\n"
4060 "%if_false = OpLabel\n"
4061 "%subf1 = OpFSub %f32 %inval %constf1\n"
4062 " OpStore %outloc %subf1\n"
4063 " OpBranch %if_end\n"
4064 "%if_end = OpLabel\n"
4066 " OpFunctionEnd\n");
4068 cases.push_back(CaseParameter("none", "None"));
4069 cases.push_back(CaseParameter("flatten", "Flatten"));
4070 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
4071 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
4073 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4075 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4076 floorAll(inputFloats);
4078 for (size_t ndx = 0; ndx < numElements; ++ndx)
4079 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
4081 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4083 map<string, string> specializations;
4084 ComputeShaderSpec spec;
4086 specializations["CONTROL"] = cases[caseNdx].param;
4087 spec.assembly = shaderTemplate.specialize(specializations);
4088 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4089 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4090 spec.numWorkGroups = IVec3(numElements, 1, 1);
4092 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4095 return group.release();
4098 // Assembly code used for testing function control is based on GLSL source code:
4102 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4103 // float elements[];
4105 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4106 // float elements[];
4109 // float const10() { return 10.f; }
4112 // uint x = gl_GlobalInvocationID.x;
4113 // output_data.elements[x] = input_data.elements[x] + const10();
4115 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
4117 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
4118 vector<CaseParameter> cases;
4119 de::Random rnd (deStringHash(group->getName()));
4120 const int numElements = 100;
4121 vector<float> inputFloats (numElements, 0);
4122 vector<float> outputFloats (numElements, 0);
4123 const StringTemplate shaderTemplate (
4124 string(getComputeAsmShaderPreamble()) +
4126 "OpSource GLSL 430\n"
4127 "OpName %main \"main\"\n"
4128 "OpName %func_const10 \"const10(\"\n"
4129 "OpName %id \"gl_GlobalInvocationID\"\n"
4131 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4133 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4135 "%f32f = OpTypeFunction %f32\n"
4136 "%id = OpVariable %uvec3ptr Input\n"
4137 "%zero = OpConstant %i32 0\n"
4138 "%constf10 = OpConstant %f32 10.0\n"
4140 "%main = OpFunction %void None %voidf\n"
4141 "%entry = OpLabel\n"
4142 "%idval = OpLoad %uvec3 %id\n"
4143 "%x = OpCompositeExtract %u32 %idval 0\n"
4144 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4145 "%inval = OpLoad %f32 %inloc\n"
4146 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
4147 "%fadd = OpFAdd %f32 %inval %ret_10\n"
4148 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4149 " OpStore %outloc %fadd\n"
4153 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
4154 "%label = OpLabel\n"
4155 " OpReturnValue %constf10\n"
4156 " OpFunctionEnd\n");
4158 cases.push_back(CaseParameter("none", "None"));
4159 cases.push_back(CaseParameter("inline", "Inline"));
4160 cases.push_back(CaseParameter("dont_inline", "DontInline"));
4161 cases.push_back(CaseParameter("pure", "Pure"));
4162 cases.push_back(CaseParameter("const", "Const"));
4163 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
4164 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
4165 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
4166 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
4168 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4170 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4171 floorAll(inputFloats);
4173 for (size_t ndx = 0; ndx < numElements; ++ndx)
4174 outputFloats[ndx] = inputFloats[ndx] + 10.f;
4176 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4178 map<string, string> specializations;
4179 ComputeShaderSpec spec;
4181 specializations["CONTROL"] = cases[caseNdx].param;
4182 spec.assembly = shaderTemplate.specialize(specializations);
4183 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4184 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4185 spec.numWorkGroups = IVec3(numElements, 1, 1);
4187 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4190 return group.release();
4193 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
4195 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
4196 vector<CaseParameter> cases;
4197 de::Random rnd (deStringHash(group->getName()));
4198 const int numElements = 100;
4199 vector<float> inputFloats (numElements, 0);
4200 vector<float> outputFloats (numElements, 0);
4201 const StringTemplate shaderTemplate (
4202 string(getComputeAsmShaderPreamble()) +
4204 "OpSource GLSL 430\n"
4205 "OpName %main \"main\"\n"
4206 "OpName %id \"gl_GlobalInvocationID\"\n"
4208 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4210 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4212 "%f32ptr_f = OpTypePointer Function %f32\n"
4214 "%id = OpVariable %uvec3ptr Input\n"
4215 "%zero = OpConstant %i32 0\n"
4216 "%four = OpConstant %i32 4\n"
4218 "%main = OpFunction %void None %voidf\n"
4219 "%label = OpLabel\n"
4220 "%copy = OpVariable %f32ptr_f Function\n"
4221 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
4222 "%x = OpCompositeExtract %u32 %idval 0\n"
4223 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4224 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4225 " OpCopyMemory %copy %inloc ${ACCESS}\n"
4226 "%val1 = OpLoad %f32 %copy\n"
4227 "%val2 = OpLoad %f32 %inloc\n"
4228 "%add = OpFAdd %f32 %val1 %val2\n"
4229 " OpStore %outloc %add ${ACCESS}\n"
4231 " OpFunctionEnd\n");
4233 cases.push_back(CaseParameter("null", ""));
4234 cases.push_back(CaseParameter("none", "None"));
4235 cases.push_back(CaseParameter("volatile", "Volatile"));
4236 cases.push_back(CaseParameter("aligned", "Aligned 4"));
4237 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
4238 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
4239 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
4241 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4243 for (size_t ndx = 0; ndx < numElements; ++ndx)
4244 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
4246 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4248 map<string, string> specializations;
4249 ComputeShaderSpec spec;
4251 specializations["ACCESS"] = cases[caseNdx].param;
4252 spec.assembly = shaderTemplate.specialize(specializations);
4253 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4254 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4255 spec.numWorkGroups = IVec3(numElements, 1, 1);
4257 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4260 return group.release();
4263 // Checks that we can get undefined values for various types, without exercising a computation with it.
4264 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
4266 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
4267 vector<CaseParameter> cases;
4268 de::Random rnd (deStringHash(group->getName()));
4269 const int numElements = 100;
4270 vector<float> positiveFloats (numElements, 0);
4271 vector<float> negativeFloats (numElements, 0);
4272 const StringTemplate shaderTemplate (
4273 string(getComputeAsmShaderPreamble()) +
4275 "OpSource GLSL 430\n"
4276 "OpName %main \"main\"\n"
4277 "OpName %id \"gl_GlobalInvocationID\"\n"
4279 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4281 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4282 "%uvec2 = OpTypeVector %u32 2\n"
4283 "%fvec4 = OpTypeVector %f32 4\n"
4284 "%fmat33 = OpTypeMatrix %fvec3 3\n"
4285 "%image = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
4286 "%sampler = OpTypeSampler\n"
4287 "%simage = OpTypeSampledImage %image\n"
4288 "%const100 = OpConstant %u32 100\n"
4289 "%uarr100 = OpTypeArray %i32 %const100\n"
4290 "%struct = OpTypeStruct %f32 %i32 %u32\n"
4291 "%pointer = OpTypePointer Function %i32\n"
4292 + string(getComputeAsmInputOutputBuffer()) +
4294 "%id = OpVariable %uvec3ptr Input\n"
4295 "%zero = OpConstant %i32 0\n"
4297 "%main = OpFunction %void None %voidf\n"
4298 "%label = OpLabel\n"
4300 "%undef = OpUndef ${TYPE}\n"
4302 "%idval = OpLoad %uvec3 %id\n"
4303 "%x = OpCompositeExtract %u32 %idval 0\n"
4305 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4306 "%inval = OpLoad %f32 %inloc\n"
4307 "%neg = OpFNegate %f32 %inval\n"
4308 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4309 " OpStore %outloc %neg\n"
4311 " OpFunctionEnd\n");
4313 cases.push_back(CaseParameter("bool", "%bool"));
4314 cases.push_back(CaseParameter("sint32", "%i32"));
4315 cases.push_back(CaseParameter("uint32", "%u32"));
4316 cases.push_back(CaseParameter("float32", "%f32"));
4317 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
4318 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
4319 cases.push_back(CaseParameter("matrix", "%fmat33"));
4320 cases.push_back(CaseParameter("image", "%image"));
4321 cases.push_back(CaseParameter("sampler", "%sampler"));
4322 cases.push_back(CaseParameter("sampledimage", "%simage"));
4323 cases.push_back(CaseParameter("array", "%uarr100"));
4324 cases.push_back(CaseParameter("runtimearray", "%f32arr"));
4325 cases.push_back(CaseParameter("struct", "%struct"));
4326 cases.push_back(CaseParameter("pointer", "%pointer"));
4328 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4330 for (size_t ndx = 0; ndx < numElements; ++ndx)
4331 negativeFloats[ndx] = -positiveFloats[ndx];
4333 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4335 map<string, string> specializations;
4336 ComputeShaderSpec spec;
4338 specializations["TYPE"] = cases[caseNdx].param;
4339 spec.assembly = shaderTemplate.specialize(specializations);
4340 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4341 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4342 spec.numWorkGroups = IVec3(numElements, 1, 1);
4344 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4347 return group.release();
4352 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
4354 struct NameCodePair { string name, code; };
4355 RGBA defaultColors[4];
4356 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
4357 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
4358 map<string, string> fragments = passthruFragments();
4359 const NameCodePair tests[] =
4361 {"unknown", "OpSource Unknown 321"},
4362 {"essl", "OpSource ESSL 310"},
4363 {"glsl", "OpSource GLSL 450"},
4364 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
4365 {"opencl_c", "OpSource OpenCL_C 120"},
4366 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
4367 {"file", opsourceGLSLWithFile},
4368 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
4369 // Longest possible source string: SPIR-V limits instructions to 65535
4370 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
4371 // contain 65530 UTF8 characters (one word each) plus one last word
4372 // containing 3 ASCII characters and \0.
4373 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
4376 getDefaultColors(defaultColors);
4377 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
4379 fragments["debug"] = tests[testNdx].code;
4380 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
4383 return opSourceTests.release();
4386 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
4388 struct NameCodePair { string name, code; };
4389 RGBA defaultColors[4];
4390 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
4391 map<string, string> fragments = passthruFragments();
4392 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
4393 const NameCodePair tests[] =
4395 {"empty", opsource + "OpSourceContinued \"\""},
4396 {"short", opsource + "OpSourceContinued \"abcde\""},
4397 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
4398 // Longest possible source string: SPIR-V limits instructions to 65535
4399 // words, of which the first one is OpSourceContinued/length; the rest
4400 // will contain 65533 UTF8 characters (one word each) plus one last word
4401 // containing 3 ASCII characters and \0.
4402 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
4405 getDefaultColors(defaultColors);
4406 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
4408 fragments["debug"] = tests[testNdx].code;
4409 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
4412 return opSourceTests.release();
4415 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
4417 RGBA defaultColors[4];
4418 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
4419 map<string, string> fragments;
4420 getDefaultColors(defaultColors);
4421 fragments["debug"] =
4422 "%name = OpString \"name\"\n";
4424 fragments["pre_main"] =
4427 "OpLine %name 1 1\n"
4429 "OpLine %name 1 1\n"
4430 "OpLine %name 1 1\n"
4431 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
4433 "OpLine %name 1 1\n"
4435 "OpLine %name 1 1\n"
4436 "OpLine %name 1 1\n"
4437 "%second_param1 = OpFunctionParameter %v4f32\n"
4440 "%label_secondfunction = OpLabel\n"
4442 "OpReturnValue %second_param1\n"
4447 fragments["testfun"] =
4448 // A %test_code function that returns its argument unchanged.
4451 "OpLine %name 1 1\n"
4452 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4454 "%param1 = OpFunctionParameter %v4f32\n"
4457 "%label_testfun = OpLabel\n"
4459 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
4460 "OpReturnValue %val1\n"
4462 "OpLine %name 1 1\n"
4465 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
4467 return opLineTests.release();
4471 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
4473 RGBA defaultColors[4];
4474 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
4475 map<string, string> fragments;
4476 std::vector<std::pair<std::string, std::string> > problemStrings;
4478 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
4479 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
4480 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
4481 getDefaultColors(defaultColors);
4483 fragments["debug"] =
4484 "%other_name = OpString \"other_name\"\n";
4486 fragments["pre_main"] =
4487 "OpLine %file_name 32 0\n"
4488 "OpLine %file_name 32 32\n"
4489 "OpLine %file_name 32 40\n"
4490 "OpLine %other_name 32 40\n"
4491 "OpLine %other_name 0 100\n"
4492 "OpLine %other_name 0 4294967295\n"
4493 "OpLine %other_name 4294967295 0\n"
4494 "OpLine %other_name 32 40\n"
4495 "OpLine %file_name 0 0\n"
4496 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
4497 "OpLine %file_name 1 0\n"
4498 "%second_param1 = OpFunctionParameter %v4f32\n"
4499 "OpLine %file_name 1 3\n"
4500 "OpLine %file_name 1 2\n"
4501 "%label_secondfunction = OpLabel\n"
4502 "OpLine %file_name 0 2\n"
4503 "OpReturnValue %second_param1\n"
4505 "OpLine %file_name 0 2\n"
4506 "OpLine %file_name 0 2\n";
4508 fragments["testfun"] =
4509 // A %test_code function that returns its argument unchanged.
4510 "OpLine %file_name 1 0\n"
4511 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4512 "OpLine %file_name 16 330\n"
4513 "%param1 = OpFunctionParameter %v4f32\n"
4514 "OpLine %file_name 14 442\n"
4515 "%label_testfun = OpLabel\n"
4516 "OpLine %file_name 11 1024\n"
4517 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
4518 "OpLine %file_name 2 97\n"
4519 "OpReturnValue %val1\n"
4521 "OpLine %file_name 5 32\n";
4523 for (size_t i = 0; i < problemStrings.size(); ++i)
4525 map<string, string> testFragments = fragments;
4526 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
4527 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
4530 return opLineTests.release();
4533 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
4535 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
4539 const char functionStart[] =
4540 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4541 "%param1 = OpFunctionParameter %v4f32\n"
4544 const char functionEnd[] =
4545 "OpReturnValue %transformed_param\n"
4548 struct NameConstantsCode
4555 NameConstantsCode tests[] =
4559 "%cnull = OpConstantNull %v4f32\n",
4560 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
4564 "%cnull = OpConstantNull %f32\n",
4565 "%vp = OpVariable %fp_v4f32 Function\n"
4566 "%v = OpLoad %v4f32 %vp\n"
4567 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
4568 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
4569 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
4570 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
4571 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
4575 "%cnull = OpConstantNull %bool\n",
4576 "%v = OpVariable %fp_v4f32 Function\n"
4577 " OpStore %v %param1\n"
4578 " OpSelectionMerge %false_label None\n"
4579 " OpBranchConditional %cnull %true_label %false_label\n"
4580 "%true_label = OpLabel\n"
4581 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
4582 " OpBranch %false_label\n"
4583 "%false_label = OpLabel\n"
4584 "%transformed_param = OpLoad %v4f32 %v\n"
4588 "%cnull = OpConstantNull %i32\n",
4589 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
4590 "%b = OpIEqual %bool %cnull %c_i32_0\n"
4591 " OpSelectionMerge %false_label None\n"
4592 " OpBranchConditional %b %true_label %false_label\n"
4593 "%true_label = OpLabel\n"
4594 " OpStore %v %param1\n"
4595 " OpBranch %false_label\n"
4596 "%false_label = OpLabel\n"
4597 "%transformed_param = OpLoad %v4f32 %v\n"
4601 "%stype = OpTypeStruct %f32 %v4f32\n"
4602 "%fp_stype = OpTypePointer Function %stype\n"
4603 "%cnull = OpConstantNull %stype\n",
4604 "%v = OpVariable %fp_stype Function %cnull\n"
4605 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
4606 "%f_val = OpLoad %v4f32 %f\n"
4607 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
4611 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
4612 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
4613 "%cnull = OpConstantNull %a4_v4f32\n",
4614 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
4615 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
4616 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
4617 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
4618 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
4619 "%f_val = OpLoad %v4f32 %f\n"
4620 "%f1_val = OpLoad %v4f32 %f1\n"
4621 "%f2_val = OpLoad %v4f32 %f2\n"
4622 "%f3_val = OpLoad %v4f32 %f3\n"
4623 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
4624 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
4625 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
4626 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
4630 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
4631 "%cnull = OpConstantNull %mat4x4_f32\n",
4632 // Our null matrix * any vector should result in a zero vector.
4633 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
4634 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
4638 getHalfColorsFullAlpha(colors);
4640 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
4642 map<string, string> fragments;
4643 fragments["pre_main"] = tests[testNdx].constants;
4644 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
4645 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
4647 return opConstantNullTests.release();
4649 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
4651 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
4652 RGBA inputColors[4];
4653 RGBA outputColors[4];
4656 const char functionStart[] =
4657 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4658 "%param1 = OpFunctionParameter %v4f32\n"
4661 const char functionEnd[] =
4662 "OpReturnValue %transformed_param\n"
4665 struct NameConstantsCode
4672 NameConstantsCode tests[] =
4677 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
4678 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
4683 "%stype = OpTypeStruct %v4f32 %f32\n"
4684 "%fp_stype = OpTypePointer Function %stype\n"
4685 "%f32_n_1 = OpConstant %f32 -1.0\n"
4686 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
4687 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
4688 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
4690 "%v = OpVariable %fp_stype Function %cval\n"
4691 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
4692 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
4693 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
4694 "%f32_val = OpLoad %f32 %f32_ptr\n"
4695 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
4696 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
4697 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
4700 // [1|0|0|0.5] [x] = x + 0.5
4701 // [0|1|0|0.5] [y] = y + 0.5
4702 // [0|0|1|0.5] [z] = z + 0.5
4703 // [0|0|0|1 ] [1] = 1
4706 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
4707 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
4708 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
4709 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
4710 "%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"
4711 "%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",
4713 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
4718 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4719 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
4720 "%f32_n_1 = OpConstant %f32 -1.0\n"
4721 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
4722 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
4724 "%v = OpVariable %fp_a4f32 Function %carr\n"
4725 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
4726 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
4727 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
4728 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
4729 "%f_val = OpLoad %f32 %f\n"
4730 "%f1_val = OpLoad %f32 %f1\n"
4731 "%f2_val = OpLoad %f32 %f2\n"
4732 "%f3_val = OpLoad %f32 %f3\n"
4733 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
4734 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
4735 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
4736 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
4737 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
4744 // [ 1.0, 1.0, 1.0, 1.0]
4748 // [ 0.0, 0.5, 0.0, 0.0]
4752 // [ 1.0, 1.0, 1.0, 1.0]
4755 "array_of_struct_of_array",
4757 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4758 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
4759 "%stype = OpTypeStruct %f32 %a4f32\n"
4760 "%a3stype = OpTypeArray %stype %c_u32_3\n"
4761 "%fp_a3stype = OpTypePointer Function %a3stype\n"
4762 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
4763 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
4764 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
4765 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
4766 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
4768 "%v = OpVariable %fp_a3stype Function %carr\n"
4769 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
4770 "%f_l = OpLoad %f32 %f\n"
4771 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
4772 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
4776 getHalfColorsFullAlpha(inputColors);
4777 outputColors[0] = RGBA(255, 255, 255, 255);
4778 outputColors[1] = RGBA(255, 127, 127, 255);
4779 outputColors[2] = RGBA(127, 255, 127, 255);
4780 outputColors[3] = RGBA(127, 127, 255, 255);
4782 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
4784 map<string, string> fragments;
4785 fragments["pre_main"] = tests[testNdx].constants;
4786 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
4787 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
4789 return opConstantCompositeTests.release();
4792 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
4794 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
4795 RGBA inputColors[4];
4796 RGBA outputColors[4];
4797 map<string, string> fragments;
4799 // vec4 test_code(vec4 param) {
4800 // vec4 result = param;
4801 // for (int i = 0; i < 4; ++i) {
4802 // if (i == 0) result[i] = 0.;
4803 // else result[i] = 1. - result[i];
4807 const char function[] =
4808 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4809 "%param1 = OpFunctionParameter %v4f32\n"
4811 "%iptr = OpVariable %fp_i32 Function\n"
4812 "%result = OpVariable %fp_v4f32 Function\n"
4813 " OpStore %iptr %c_i32_0\n"
4814 " OpStore %result %param1\n"
4817 // Loop entry block.
4819 "%ival = OpLoad %i32 %iptr\n"
4820 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
4821 " OpLoopMerge %exit %if_entry None\n"
4822 " OpBranchConditional %lt_4 %if_entry %exit\n"
4824 // Merge block for loop.
4826 "%ret = OpLoad %v4f32 %result\n"
4827 " OpReturnValue %ret\n"
4829 // If-statement entry block.
4830 "%if_entry = OpLabel\n"
4831 "%loc = OpAccessChain %fp_f32 %result %ival\n"
4832 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
4833 " OpSelectionMerge %if_exit None\n"
4834 " OpBranchConditional %eq_0 %if_true %if_false\n"
4836 // False branch for if-statement.
4837 "%if_false = OpLabel\n"
4838 "%val = OpLoad %f32 %loc\n"
4839 "%sub = OpFSub %f32 %c_f32_1 %val\n"
4840 " OpStore %loc %sub\n"
4841 " OpBranch %if_exit\n"
4843 // Merge block for if-statement.
4844 "%if_exit = OpLabel\n"
4845 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
4846 " OpStore %iptr %ival_next\n"
4849 // True branch for if-statement.
4850 "%if_true = OpLabel\n"
4851 " OpStore %loc %c_f32_0\n"
4852 " OpBranch %if_exit\n"
4856 fragments["testfun"] = function;
4858 inputColors[0] = RGBA(127, 127, 127, 0);
4859 inputColors[1] = RGBA(127, 0, 0, 0);
4860 inputColors[2] = RGBA(0, 127, 0, 0);
4861 inputColors[3] = RGBA(0, 0, 127, 0);
4863 outputColors[0] = RGBA(0, 128, 128, 255);
4864 outputColors[1] = RGBA(0, 255, 255, 255);
4865 outputColors[2] = RGBA(0, 128, 255, 255);
4866 outputColors[3] = RGBA(0, 255, 128, 255);
4868 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
4870 return group.release();
4873 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
4875 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
4876 RGBA inputColors[4];
4877 RGBA outputColors[4];
4878 map<string, string> fragments;
4880 const char typesAndConstants[] =
4881 "%c_f32_p2 = OpConstant %f32 0.2\n"
4882 "%c_f32_p4 = OpConstant %f32 0.4\n"
4883 "%c_f32_p6 = OpConstant %f32 0.6\n"
4884 "%c_f32_p8 = OpConstant %f32 0.8\n";
4886 // vec4 test_code(vec4 param) {
4887 // vec4 result = param;
4888 // for (int i = 0; i < 4; ++i) {
4890 // case 0: result[i] += .2; break;
4891 // case 1: result[i] += .6; break;
4892 // case 2: result[i] += .4; break;
4893 // case 3: result[i] += .8; break;
4894 // default: break; // unreachable
4899 const char function[] =
4900 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4901 "%param1 = OpFunctionParameter %v4f32\n"
4903 "%iptr = OpVariable %fp_i32 Function\n"
4904 "%result = OpVariable %fp_v4f32 Function\n"
4905 " OpStore %iptr %c_i32_0\n"
4906 " OpStore %result %param1\n"
4909 // Loop entry block.
4911 "%ival = OpLoad %i32 %iptr\n"
4912 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
4913 " OpLoopMerge %exit %switch_exit None\n"
4914 " OpBranchConditional %lt_4 %switch_entry %exit\n"
4916 // Merge block for loop.
4918 "%ret = OpLoad %v4f32 %result\n"
4919 " OpReturnValue %ret\n"
4921 // Switch-statement entry block.
4922 "%switch_entry = OpLabel\n"
4923 "%loc = OpAccessChain %fp_f32 %result %ival\n"
4924 "%val = OpLoad %f32 %loc\n"
4925 " OpSelectionMerge %switch_exit None\n"
4926 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
4928 "%case2 = OpLabel\n"
4929 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
4930 " OpStore %loc %addp4\n"
4931 " OpBranch %switch_exit\n"
4933 "%switch_default = OpLabel\n"
4936 "%case3 = OpLabel\n"
4937 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
4938 " OpStore %loc %addp8\n"
4939 " OpBranch %switch_exit\n"
4941 "%case0 = OpLabel\n"
4942 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
4943 " OpStore %loc %addp2\n"
4944 " OpBranch %switch_exit\n"
4946 // Merge block for switch-statement.
4947 "%switch_exit = OpLabel\n"
4948 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
4949 " OpStore %iptr %ival_next\n"
4952 "%case1 = OpLabel\n"
4953 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
4954 " OpStore %loc %addp6\n"
4955 " OpBranch %switch_exit\n"
4959 fragments["pre_main"] = typesAndConstants;
4960 fragments["testfun"] = function;
4962 inputColors[0] = RGBA(127, 27, 127, 51);
4963 inputColors[1] = RGBA(127, 0, 0, 51);
4964 inputColors[2] = RGBA(0, 27, 0, 51);
4965 inputColors[3] = RGBA(0, 0, 127, 51);
4967 outputColors[0] = RGBA(178, 180, 229, 255);
4968 outputColors[1] = RGBA(178, 153, 102, 255);
4969 outputColors[2] = RGBA(51, 180, 102, 255);
4970 outputColors[3] = RGBA(51, 153, 229, 255);
4972 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
4974 return group.release();
4977 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
4979 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
4980 RGBA inputColors[4];
4981 RGBA outputColors[4];
4982 map<string, string> fragments;
4984 const char decorations[] =
4985 "OpDecorate %array_group ArrayStride 4\n"
4986 "OpDecorate %struct_member_group Offset 0\n"
4987 "%array_group = OpDecorationGroup\n"
4988 "%struct_member_group = OpDecorationGroup\n"
4990 "OpDecorate %group1 RelaxedPrecision\n"
4991 "OpDecorate %group3 RelaxedPrecision\n"
4992 "OpDecorate %group3 Invariant\n"
4993 "OpDecorate %group3 Restrict\n"
4994 "%group0 = OpDecorationGroup\n"
4995 "%group1 = OpDecorationGroup\n"
4996 "%group3 = OpDecorationGroup\n";
4998 const char typesAndConstants[] =
4999 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
5000 "%struct1 = OpTypeStruct %a3f32\n"
5001 "%struct2 = OpTypeStruct %a3f32\n"
5002 "%fp_struct1 = OpTypePointer Function %struct1\n"
5003 "%fp_struct2 = OpTypePointer Function %struct2\n"
5004 "%c_f32_2 = OpConstant %f32 2.\n"
5005 "%c_f32_n2 = OpConstant %f32 -2.\n"
5007 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
5008 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
5009 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
5010 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
5012 const char function[] =
5013 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5014 "%param = OpFunctionParameter %v4f32\n"
5015 "%entry = OpLabel\n"
5016 "%result = OpVariable %fp_v4f32 Function\n"
5017 "%v_struct1 = OpVariable %fp_struct1 Function\n"
5018 "%v_struct2 = OpVariable %fp_struct2 Function\n"
5019 " OpStore %result %param\n"
5020 " OpStore %v_struct1 %c_struct1\n"
5021 " OpStore %v_struct2 %c_struct2\n"
5022 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
5023 "%val1 = OpLoad %f32 %ptr1\n"
5024 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
5025 "%val2 = OpLoad %f32 %ptr2\n"
5026 "%addvalues = OpFAdd %f32 %val1 %val2\n"
5027 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
5028 "%val = OpLoad %f32 %ptr\n"
5029 "%addresult = OpFAdd %f32 %addvalues %val\n"
5030 " OpStore %ptr %addresult\n"
5031 "%ret = OpLoad %v4f32 %result\n"
5032 " OpReturnValue %ret\n"
5035 struct CaseNameDecoration
5041 CaseNameDecoration tests[] =
5044 "same_decoration_group_on_multiple_types",
5045 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
5048 "empty_decoration_group",
5049 "OpGroupDecorate %group0 %a3f32\n"
5050 "OpGroupDecorate %group0 %result\n"
5053 "one_element_decoration_group",
5054 "OpGroupDecorate %array_group %a3f32\n"
5057 "multiple_elements_decoration_group",
5058 "OpGroupDecorate %group3 %v_struct1\n"
5061 "multiple_decoration_groups_on_same_variable",
5062 "OpGroupDecorate %group0 %v_struct2\n"
5063 "OpGroupDecorate %group1 %v_struct2\n"
5064 "OpGroupDecorate %group3 %v_struct2\n"
5067 "same_decoration_group_multiple_times",
5068 "OpGroupDecorate %group1 %addvalues\n"
5069 "OpGroupDecorate %group1 %addvalues\n"
5070 "OpGroupDecorate %group1 %addvalues\n"
5075 getHalfColorsFullAlpha(inputColors);
5076 getHalfColorsFullAlpha(outputColors);
5078 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
5080 fragments["decoration"] = decorations + tests[idx].decoration;
5081 fragments["pre_main"] = typesAndConstants;
5082 fragments["testfun"] = function;
5084 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
5087 return group.release();
5090 struct SpecConstantTwoIntGraphicsCase
5092 const char* caseName;
5093 const char* scDefinition0;
5094 const char* scDefinition1;
5095 const char* scResultType;
5096 const char* scOperation;
5097 deInt32 scActualValue0;
5098 deInt32 scActualValue1;
5099 const char* resultOperation;
5100 RGBA expectedColors[4];
5102 SpecConstantTwoIntGraphicsCase (const char* name,
5103 const char* definition0,
5104 const char* definition1,
5105 const char* resultType,
5106 const char* operation,
5109 const char* resultOp,
5110 const RGBA (&output)[4])
5112 , scDefinition0 (definition0)
5113 , scDefinition1 (definition1)
5114 , scResultType (resultType)
5115 , scOperation (operation)
5116 , scActualValue0 (value0)
5117 , scActualValue1 (value1)
5118 , resultOperation (resultOp)
5120 expectedColors[0] = output[0];
5121 expectedColors[1] = output[1];
5122 expectedColors[2] = output[2];
5123 expectedColors[3] = output[3];
5127 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
5129 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
5130 vector<SpecConstantTwoIntGraphicsCase> cases;
5131 RGBA inputColors[4];
5132 RGBA outputColors0[4];
5133 RGBA outputColors1[4];
5134 RGBA outputColors2[4];
5136 const char decorations1[] =
5137 "OpDecorate %sc_0 SpecId 0\n"
5138 "OpDecorate %sc_1 SpecId 1\n";
5140 const char typesAndConstants1[] =
5141 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
5142 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
5143 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
5145 const char function1[] =
5146 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5147 "%param = OpFunctionParameter %v4f32\n"
5148 "%label = OpLabel\n"
5149 "%result = OpVariable %fp_v4f32 Function\n"
5150 " OpStore %result %param\n"
5151 "%gen = ${GEN_RESULT}\n"
5152 "%index = OpIAdd %i32 %gen %c_i32_1\n"
5153 "%loc = OpAccessChain %fp_f32 %result %index\n"
5154 "%val = OpLoad %f32 %loc\n"
5155 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
5156 " OpStore %loc %add\n"
5157 "%ret = OpLoad %v4f32 %result\n"
5158 " OpReturnValue %ret\n"
5161 inputColors[0] = RGBA(127, 127, 127, 255);
5162 inputColors[1] = RGBA(127, 0, 0, 255);
5163 inputColors[2] = RGBA(0, 127, 0, 255);
5164 inputColors[3] = RGBA(0, 0, 127, 255);
5166 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
5167 outputColors0[0] = RGBA(255, 127, 127, 255);
5168 outputColors0[1] = RGBA(255, 0, 0, 255);
5169 outputColors0[2] = RGBA(128, 127, 0, 255);
5170 outputColors0[3] = RGBA(128, 0, 127, 255);
5172 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
5173 outputColors1[0] = RGBA(127, 255, 127, 255);
5174 outputColors1[1] = RGBA(127, 128, 0, 255);
5175 outputColors1[2] = RGBA(0, 255, 0, 255);
5176 outputColors1[3] = RGBA(0, 128, 127, 255);
5178 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
5179 outputColors2[0] = RGBA(127, 127, 255, 255);
5180 outputColors2[1] = RGBA(127, 0, 128, 255);
5181 outputColors2[2] = RGBA(0, 127, 128, 255);
5182 outputColors2[3] = RGBA(0, 0, 255, 255);
5184 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
5185 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
5186 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
5188 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
5189 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
5190 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
5191 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
5192 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
5193 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
5194 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
5195 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
5196 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
5197 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
5198 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
5199 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
5200 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
5201 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
5202 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
5203 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
5204 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
5205 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
5206 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
5207 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
5208 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
5209 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
5210 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
5211 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
5212 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
5213 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
5214 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
5215 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
5216 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
5217 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
5218 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
5219 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
5220 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
5222 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5224 map<string, string> specializations;
5225 map<string, string> fragments;
5226 vector<deInt32> specConstants;
5228 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
5229 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
5230 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
5231 specializations["SC_OP"] = cases[caseNdx].scOperation;
5232 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
5234 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
5235 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
5236 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
5238 specConstants.push_back(cases[caseNdx].scActualValue0);
5239 specConstants.push_back(cases[caseNdx].scActualValue1);
5241 createTestsForAllStages(cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants, group.get());
5244 const char decorations2[] =
5245 "OpDecorate %sc_0 SpecId 0\n"
5246 "OpDecorate %sc_1 SpecId 1\n"
5247 "OpDecorate %sc_2 SpecId 2\n";
5249 const char typesAndConstants2[] =
5250 "%v3i32 = OpTypeVector %i32 3\n"
5251 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
5252 "%vec3_undef = OpUndef %v3i32\n"
5254 "%sc_0 = OpSpecConstant %i32 0\n"
5255 "%sc_1 = OpSpecConstant %i32 0\n"
5256 "%sc_2 = OpSpecConstant %i32 0\n"
5257 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
5258 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
5259 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
5260 "%sc_vec3_0_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
5261 "%sc_vec3_1_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
5262 "%sc_vec3_2_s = OpSpecConstantOp %v3i32 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
5263 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
5264 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
5265 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
5266 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
5267 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
5268 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
5269 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
5271 const char function2[] =
5272 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5273 "%param = OpFunctionParameter %v4f32\n"
5274 "%label = OpLabel\n"
5275 "%result = OpVariable %fp_v4f32 Function\n"
5276 " OpStore %result %param\n"
5277 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
5278 "%val = OpLoad %f32 %loc\n"
5279 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
5280 " OpStore %loc %add\n"
5281 "%ret = OpLoad %v4f32 %result\n"
5282 " OpReturnValue %ret\n"
5285 map<string, string> fragments;
5286 vector<deInt32> specConstants;
5288 fragments["decoration"] = decorations2;
5289 fragments["pre_main"] = typesAndConstants2;
5290 fragments["testfun"] = function2;
5292 specConstants.push_back(56789);
5293 specConstants.push_back(-2);
5294 specConstants.push_back(56788);
5296 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
5298 return group.release();
5301 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
5303 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
5304 RGBA inputColors[4];
5305 RGBA outputColors1[4];
5306 RGBA outputColors2[4];
5307 RGBA outputColors3[4];
5308 map<string, string> fragments1;
5309 map<string, string> fragments2;
5310 map<string, string> fragments3;
5312 const char typesAndConstants1[] =
5313 "%c_f32_p2 = OpConstant %f32 0.2\n"
5314 "%c_f32_p4 = OpConstant %f32 0.4\n"
5315 "%c_f32_p5 = OpConstant %f32 0.5\n"
5316 "%c_f32_p8 = OpConstant %f32 0.8\n";
5318 // vec4 test_code(vec4 param) {
5319 // vec4 result = param;
5320 // for (int i = 0; i < 4; ++i) {
5323 // case 0: operand = .2; break;
5324 // case 1: operand = .5; break;
5325 // case 2: operand = .4; break;
5326 // case 3: operand = .0; break;
5327 // default: break; // unreachable
5329 // result[i] += operand;
5333 const char function1[] =
5334 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5335 "%param1 = OpFunctionParameter %v4f32\n"
5337 "%iptr = OpVariable %fp_i32 Function\n"
5338 "%result = OpVariable %fp_v4f32 Function\n"
5339 " OpStore %iptr %c_i32_0\n"
5340 " OpStore %result %param1\n"
5344 "%ival = OpLoad %i32 %iptr\n"
5345 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5346 " OpLoopMerge %exit %phi None\n"
5347 " OpBranchConditional %lt_4 %entry %exit\n"
5349 "%entry = OpLabel\n"
5350 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5351 "%val = OpLoad %f32 %loc\n"
5352 " OpSelectionMerge %phi None\n"
5353 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5355 "%case0 = OpLabel\n"
5357 "%case1 = OpLabel\n"
5359 "%case2 = OpLabel\n"
5361 "%case3 = OpLabel\n"
5364 "%default = OpLabel\n"
5368 "%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
5369 "%add = OpFAdd %f32 %val %operand\n"
5370 " OpStore %loc %add\n"
5371 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5372 " OpStore %iptr %ival_next\n"
5376 "%ret = OpLoad %v4f32 %result\n"
5377 " OpReturnValue %ret\n"
5381 fragments1["pre_main"] = typesAndConstants1;
5382 fragments1["testfun"] = function1;
5384 getHalfColorsFullAlpha(inputColors);
5386 outputColors1[0] = RGBA(178, 255, 229, 255);
5387 outputColors1[1] = RGBA(178, 127, 102, 255);
5388 outputColors1[2] = RGBA(51, 255, 102, 255);
5389 outputColors1[3] = RGBA(51, 127, 229, 255);
5391 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
5393 const char typesAndConstants2[] =
5394 "%c_f32_p2 = OpConstant %f32 0.2\n";
5396 // Add .4 to the second element of the given parameter.
5397 const char function2[] =
5398 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5399 "%param = OpFunctionParameter %v4f32\n"
5400 "%entry = OpLabel\n"
5401 "%result = OpVariable %fp_v4f32 Function\n"
5402 " OpStore %result %param\n"
5403 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
5404 "%val = OpLoad %f32 %loc\n"
5408 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
5409 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
5410 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
5411 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
5412 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
5413 " OpLoopMerge %exit %phi None\n"
5414 " OpBranchConditional %still_loop %phi %exit\n"
5417 " OpStore %loc %accum\n"
5418 "%ret = OpLoad %v4f32 %result\n"
5419 " OpReturnValue %ret\n"
5423 fragments2["pre_main"] = typesAndConstants2;
5424 fragments2["testfun"] = function2;
5426 outputColors2[0] = RGBA(127, 229, 127, 255);
5427 outputColors2[1] = RGBA(127, 102, 0, 255);
5428 outputColors2[2] = RGBA(0, 229, 0, 255);
5429 outputColors2[3] = RGBA(0, 102, 127, 255);
5431 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
5433 const char typesAndConstants3[] =
5434 "%true = OpConstantTrue %bool\n"
5435 "%false = OpConstantFalse %bool\n"
5436 "%c_f32_p2 = OpConstant %f32 0.2\n";
5438 // Swap the second and the third element of the given parameter.
5439 const char function3[] =
5440 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5441 "%param = OpFunctionParameter %v4f32\n"
5442 "%entry = OpLabel\n"
5443 "%result = OpVariable %fp_v4f32 Function\n"
5444 " OpStore %result %param\n"
5445 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
5446 "%a_init = OpLoad %f32 %a_loc\n"
5447 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
5448 "%b_init = OpLoad %f32 %b_loc\n"
5452 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
5453 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
5454 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
5455 " OpLoopMerge %exit %phi None\n"
5456 " OpBranchConditional %still_loop %phi %exit\n"
5459 " OpStore %a_loc %a_next\n"
5460 " OpStore %b_loc %b_next\n"
5461 "%ret = OpLoad %v4f32 %result\n"
5462 " OpReturnValue %ret\n"
5466 fragments3["pre_main"] = typesAndConstants3;
5467 fragments3["testfun"] = function3;
5469 outputColors3[0] = RGBA(127, 127, 127, 255);
5470 outputColors3[1] = RGBA(127, 0, 0, 255);
5471 outputColors3[2] = RGBA(0, 0, 127, 255);
5472 outputColors3[3] = RGBA(0, 127, 0, 255);
5474 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
5476 return group.release();
5479 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
5481 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
5482 RGBA inputColors[4];
5483 RGBA outputColors[4];
5485 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
5486 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
5487 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
5488 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
5489 const char constantsAndTypes[] =
5490 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
5491 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5492 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
5493 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
5494 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n"
5497 const char function[] =
5498 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5499 "%param = OpFunctionParameter %v4f32\n"
5500 "%label = OpLabel\n"
5501 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
5502 "%var2 = OpVariable %fp_f32 Function\n"
5503 "%red = OpCompositeExtract %f32 %param 0\n"
5504 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
5505 " OpStore %var2 %plus_red\n"
5506 "%val1 = OpLoad %f32 %var1\n"
5507 "%val2 = OpLoad %f32 %var2\n"
5508 "%mul = OpFMul %f32 %val1 %val2\n"
5509 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
5510 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
5511 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
5512 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
5513 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
5514 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
5515 " OpReturnValue %ret\n"
5518 struct CaseNameDecoration
5525 CaseNameDecoration tests[] = {
5526 {"multiplication", "OpDecorate %mul NoContraction"},
5527 {"addition", "OpDecorate %add NoContraction"},
5528 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
5531 getHalfColorsFullAlpha(inputColors);
5533 for (deUint8 idx = 0; idx < 4; ++idx)
5535 inputColors[idx].setRed(0);
5536 outputColors[idx] = RGBA(0, 0, 0, 255);
5539 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
5541 map<string, string> fragments;
5543 fragments["decoration"] = tests[testNdx].decoration;
5544 fragments["pre_main"] = constantsAndTypes;
5545 fragments["testfun"] = function;
5547 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
5550 return group.release();
5553 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
5555 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
5558 const char constantsAndTypes[] =
5559 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
5560 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
5561 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
5562 "%fp_stype = OpTypePointer Function %stype\n";
5564 const char function[] =
5565 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5566 "%param1 = OpFunctionParameter %v4f32\n"
5568 "%v1 = OpVariable %fp_v4f32 Function\n"
5569 "%v2 = OpVariable %fp_a2f32 Function\n"
5570 "%v3 = OpVariable %fp_f32 Function\n"
5571 "%v = OpVariable %fp_stype Function\n"
5572 "%vv = OpVariable %fp_stype Function\n"
5573 "%vvv = OpVariable %fp_f32 Function\n"
5575 " OpStore %v1 %c_v4f32_1_1_1_1\n"
5576 " OpStore %v2 %c_a2f32_1\n"
5577 " OpStore %v3 %c_f32_1\n"
5579 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5580 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
5581 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5582 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
5583 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
5584 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
5586 " OpStore %p_v4f32 %v1_v ${access_type}\n"
5587 " OpStore %p_a2f32 %v2_v ${access_type}\n"
5588 " OpStore %p_f32 %v3_v ${access_type}\n"
5590 " OpCopyMemory %vv %v ${access_type}\n"
5591 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
5593 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
5594 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
5595 "%v_f32_3 = OpLoad %f32 %vvv\n"
5597 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
5598 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
5599 " OpReturnValue %ret2\n"
5602 struct NameMemoryAccess
5609 NameMemoryAccess tests[] =
5612 { "volatile", "Volatile" },
5613 { "aligned", "Aligned 1" },
5614 { "volatile_aligned", "Volatile|Aligned 1" },
5615 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
5616 { "volatile_nontemporal", "Volatile|Nontemporal" },
5617 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
5620 getHalfColorsFullAlpha(colors);
5622 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
5624 map<string, string> fragments;
5625 map<string, string> memoryAccess;
5626 memoryAccess["access_type"] = tests[testNdx].accessType;
5628 fragments["pre_main"] = constantsAndTypes;
5629 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
5630 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
5632 return memoryAccessTests.release();
5634 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
5636 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
5637 RGBA defaultColors[4];
5638 map<string, string> fragments;
5639 getDefaultColors(defaultColors);
5641 // First, simple cases that don't do anything with the OpUndef result.
5642 struct NameCodePair { string name, decl, type; };
5643 const NameCodePair tests[] =
5645 {"bool", "", "%bool"},
5646 {"vec2uint32", "%type = OpTypeVector %u32 2", "%type"},
5647 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown", "%type"},
5648 {"sampler", "%type = OpTypeSampler", "%type"},
5649 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img", "%type"},
5650 {"pointer", "", "%fp_i32"},
5651 {"runtimearray", "%type = OpTypeRuntimeArray %f32", "%type"},
5652 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100", "%type"},
5653 {"struct", "%type = OpTypeStruct %f32 %i32 %u32", "%type"}};
5654 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5656 fragments["undef_type"] = tests[testNdx].type;
5657 fragments["testfun"] = StringTemplate(
5658 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5659 "%param1 = OpFunctionParameter %v4f32\n"
5660 "%label_testfun = OpLabel\n"
5661 "%undef = OpUndef ${undef_type}\n"
5662 "OpReturnValue %param1\n"
5663 "OpFunctionEnd\n").specialize(fragments);
5664 fragments["pre_main"] = tests[testNdx].decl;
5665 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
5669 fragments["testfun"] =
5670 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5671 "%param1 = OpFunctionParameter %v4f32\n"
5672 "%label_testfun = OpLabel\n"
5673 "%undef = OpUndef %f32\n"
5674 "%zero = OpFMul %f32 %undef %c_f32_0\n"
5675 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
5676 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
5677 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5678 "%b = OpFAdd %f32 %a %actually_zero\n"
5679 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
5680 "OpReturnValue %ret\n"
5683 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
5685 fragments["testfun"] =
5686 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5687 "%param1 = OpFunctionParameter %v4f32\n"
5688 "%label_testfun = OpLabel\n"
5689 "%undef = OpUndef %i32\n"
5690 "%zero = OpIMul %i32 %undef %c_i32_0\n"
5691 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
5692 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
5693 "OpReturnValue %ret\n"
5696 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
5698 fragments["testfun"] =
5699 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5700 "%param1 = OpFunctionParameter %v4f32\n"
5701 "%label_testfun = OpLabel\n"
5702 "%undef = OpUndef %u32\n"
5703 "%zero = OpIMul %u32 %undef %c_i32_0\n"
5704 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
5705 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
5706 "OpReturnValue %ret\n"
5709 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
5711 fragments["testfun"] =
5712 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5713 "%param1 = OpFunctionParameter %v4f32\n"
5714 "%label_testfun = OpLabel\n"
5715 "%undef = OpUndef %v4f32\n"
5716 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
5717 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
5718 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
5719 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
5720 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
5721 "%is_nan_0 = OpIsNan %bool %zero_0\n"
5722 "%is_nan_1 = OpIsNan %bool %zero_1\n"
5723 "%is_nan_2 = OpIsNan %bool %zero_2\n"
5724 "%is_nan_3 = OpIsNan %bool %zero_3\n"
5725 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
5726 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
5727 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
5728 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
5729 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5730 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
5731 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
5732 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
5733 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
5734 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
5735 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
5736 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
5737 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
5738 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
5739 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
5740 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
5741 "OpReturnValue %ret\n"
5744 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
5746 fragments["pre_main"] =
5747 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
5748 fragments["testfun"] =
5749 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5750 "%param1 = OpFunctionParameter %v4f32\n"
5751 "%label_testfun = OpLabel\n"
5752 "%undef = OpUndef %m2x2f32\n"
5753 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
5754 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
5755 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
5756 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
5757 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
5758 "%is_nan_0 = OpIsNan %bool %zero_0\n"
5759 "%is_nan_1 = OpIsNan %bool %zero_1\n"
5760 "%is_nan_2 = OpIsNan %bool %zero_2\n"
5761 "%is_nan_3 = OpIsNan %bool %zero_3\n"
5762 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
5763 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
5764 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
5765 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
5766 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5767 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
5768 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
5769 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
5770 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
5771 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
5772 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
5773 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
5774 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
5775 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
5776 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
5777 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
5778 "OpReturnValue %ret\n"
5781 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
5783 return opUndefTests.release();
5786 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
5788 const RGBA inputColors[4] =
5791 RGBA(0, 0, 255, 255),
5792 RGBA(0, 255, 0, 255),
5793 RGBA(0, 255, 255, 255)
5796 const RGBA expectedColors[4] =
5798 RGBA(255, 0, 0, 255),
5799 RGBA(255, 0, 0, 255),
5800 RGBA(255, 0, 0, 255),
5801 RGBA(255, 0, 0, 255)
5804 const struct SingleFP16Possibility
5807 const char* constant; // Value to assign to %test_constant.
5809 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
5815 -constructNormalizedFloat(1, 0x300000),
5816 "%cond = OpFOrdEqual %bool %c %test_constant\n"
5821 constructNormalizedFloat(7, 0x000000),
5822 "%cond = OpFOrdEqual %bool %c %test_constant\n"
5824 // SPIR-V requires that OpQuantizeToF16 flushes
5825 // any numbers that would end up denormalized in F16 to zero.
5829 std::ldexp(1.5f, -140),
5830 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5835 -std::ldexp(1.5f, -140),
5836 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5841 std::ldexp(1.0f, -16),
5842 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5843 }, // too small positive
5845 "negative_too_small",
5847 -std::ldexp(1.0f, -32),
5848 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5849 }, // too small negative
5853 -std::ldexp(1.0f, 128),
5855 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
5856 "%inf = OpIsInf %bool %c\n"
5857 "%cond = OpLogicalAnd %bool %gz %inf\n"
5862 std::ldexp(1.0f, 128),
5864 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
5865 "%inf = OpIsInf %bool %c\n"
5866 "%cond = OpLogicalAnd %bool %gz %inf\n"
5869 "round_to_negative_inf",
5871 -std::ldexp(1.0f, 32),
5873 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
5874 "%inf = OpIsInf %bool %c\n"
5875 "%cond = OpLogicalAnd %bool %gz %inf\n"
5880 std::ldexp(1.0f, 16),
5882 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
5883 "%inf = OpIsInf %bool %c\n"
5884 "%cond = OpLogicalAnd %bool %gz %inf\n"
5889 std::numeric_limits<float>::quiet_NaN(),
5891 // Test for any NaN value, as NaNs are not preserved
5892 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
5893 "%cond = OpIsNan %bool %direct_quant\n"
5898 std::numeric_limits<float>::quiet_NaN(),
5900 // Test for any NaN value, as NaNs are not preserved
5901 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
5902 "%cond = OpIsNan %bool %direct_quant\n"
5905 const char* constants =
5906 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
5908 StringTemplate function (
5909 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5910 "%param1 = OpFunctionParameter %v4f32\n"
5911 "%label_testfun = OpLabel\n"
5912 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5913 "%b = OpFAdd %f32 %test_constant %a\n"
5914 "%c = OpQuantizeToF16 %f32 %b\n"
5916 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
5917 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
5918 " OpReturnValue %retval\n"
5922 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
5923 const char* specConstants =
5924 "%test_constant = OpSpecConstant %f32 0.\n"
5925 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
5927 StringTemplate specConstantFunction(
5928 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5929 "%param1 = OpFunctionParameter %v4f32\n"
5930 "%label_testfun = OpLabel\n"
5932 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
5933 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
5934 " OpReturnValue %retval\n"
5938 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
5940 map<string, string> codeSpecialization;
5941 map<string, string> fragments;
5942 codeSpecialization["condition"] = tests[idx].condition;
5943 fragments["testfun"] = function.specialize(codeSpecialization);
5944 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
5945 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
5948 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
5950 map<string, string> codeSpecialization;
5951 map<string, string> fragments;
5952 vector<deInt32> passConstants;
5953 deInt32 specConstant;
5955 codeSpecialization["condition"] = tests[idx].condition;
5956 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
5957 fragments["decoration"] = specDecorations;
5958 fragments["pre_main"] = specConstants;
5960 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
5961 passConstants.push_back(specConstant);
5963 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
5967 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
5969 RGBA inputColors[4] = {
5971 RGBA(0, 0, 255, 255),
5972 RGBA(0, 255, 0, 255),
5973 RGBA(0, 255, 255, 255)
5976 RGBA expectedColors[4] =
5978 RGBA(255, 0, 0, 255),
5979 RGBA(255, 0, 0, 255),
5980 RGBA(255, 0, 0, 255),
5981 RGBA(255, 0, 0, 255)
5984 struct DualFP16Possibility
5989 const char* possibleOutput1;
5990 const char* possibleOutput2;
5993 "positive_round_up_or_round_down",
5995 constructNormalizedFloat(8, 0x300300),
6000 "negative_round_up_or_round_down",
6002 -constructNormalizedFloat(-7, 0x600800),
6009 constructNormalizedFloat(2, 0x01e000),
6014 "carry_to_exponent",
6016 constructNormalizedFloat(1, 0xffe000),
6021 StringTemplate constants (
6022 "%input_const = OpConstant %f32 ${input}\n"
6023 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6024 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6027 StringTemplate specConstants (
6028 "%input_const = OpSpecConstant %f32 0.\n"
6029 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6030 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6033 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
6035 const char* function =
6036 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6037 "%param1 = OpFunctionParameter %v4f32\n"
6038 "%label_testfun = OpLabel\n"
6039 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6040 // For the purposes of this test we assume that 0.f will always get
6041 // faithfully passed through the pipeline stages.
6042 "%b = OpFAdd %f32 %input_const %a\n"
6043 "%c = OpQuantizeToF16 %f32 %b\n"
6044 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
6045 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
6046 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
6047 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6048 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
6049 " OpReturnValue %retval\n"
6052 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6053 map<string, string> fragments;
6054 map<string, string> constantSpecialization;
6056 constantSpecialization["input"] = tests[idx].input;
6057 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6058 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6059 fragments["testfun"] = function;
6060 fragments["pre_main"] = constants.specialize(constantSpecialization);
6061 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6064 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6065 map<string, string> fragments;
6066 map<string, string> constantSpecialization;
6067 vector<deInt32> passConstants;
6068 deInt32 specConstant;
6070 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6071 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6072 fragments["testfun"] = function;
6073 fragments["decoration"] = specDecorations;
6074 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
6076 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
6077 passConstants.push_back(specConstant);
6079 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6083 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
6085 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
6086 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
6087 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
6088 return opQuantizeTests.release();
6091 struct ShaderPermutation
6093 deUint8 vertexPermutation;
6094 deUint8 geometryPermutation;
6095 deUint8 tesscPermutation;
6096 deUint8 tessePermutation;
6097 deUint8 fragmentPermutation;
6100 ShaderPermutation getShaderPermutation(deUint8 inputValue)
6102 ShaderPermutation permutation =
6104 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
6105 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
6106 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
6107 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
6108 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
6113 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
6115 RGBA defaultColors[4];
6116 RGBA invertedColors[4];
6117 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
6119 const ShaderElement combinedPipeline[] =
6121 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
6122 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
6123 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
6124 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
6125 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
6128 getDefaultColors(defaultColors);
6129 getInvertedDefaultColors(invertedColors);
6130 addFunctionCaseWithPrograms<InstanceContext>(
6131 moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline,
6132 createInstanceContext(combinedPipeline, map<string, string>()));
6134 const char* numbers[] =
6139 for (deInt8 idx = 0; idx < 32; ++idx)
6141 ShaderPermutation permutation = getShaderPermutation(idx);
6142 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
6143 const ShaderElement pipeline[] =
6145 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
6146 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
6147 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
6148 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
6149 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
6152 // If there are an even number of swaps, then it should be no-op.
6153 // If there are an odd number, the color should be flipped.
6154 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
6156 addFunctionCaseWithPrograms<InstanceContext>(
6157 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
6158 createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
6162 addFunctionCaseWithPrograms<InstanceContext>(
6163 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
6164 createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
6167 return moduleTests.release();
6170 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
6172 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
6173 RGBA defaultColors[4];
6174 getDefaultColors(defaultColors);
6175 map<string, string> fragments;
6176 fragments["pre_main"] =
6177 "%c_f32_5 = OpConstant %f32 5.\n";
6179 // A loop with a single block. The Continue Target is the loop block
6180 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
6181 // -- the "continue construct" forms the entire loop.
6182 fragments["testfun"] =
6183 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6184 "%param1 = OpFunctionParameter %v4f32\n"
6186 "%entry = OpLabel\n"
6187 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6190 ";adds and subtracts 1.0 to %val in alternate iterations\n"
6192 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
6193 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
6194 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
6195 "%val = OpFAdd %f32 %val1 %delta\n"
6196 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
6197 "%count__ = OpISub %i32 %count %c_i32_1\n"
6198 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6199 "OpLoopMerge %exit %loop None\n"
6200 "OpBranchConditional %again %loop %exit\n"
6203 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
6204 "OpReturnValue %result\n"
6208 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
6210 // Body comprised of multiple basic blocks.
6211 const StringTemplate multiBlock(
6212 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6213 "%param1 = OpFunctionParameter %v4f32\n"
6215 "%entry = OpLabel\n"
6216 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6219 ";adds and subtracts 1.0 to %val in alternate iterations\n"
6221 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
6222 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
6223 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
6224 // There are several possibilities for the Continue Target below. Each
6225 // will be specialized into a separate test case.
6226 "OpLoopMerge %exit ${continue_target} None\n"
6230 ";delta_next = (delta > 0) ? -1 : 1;\n"
6231 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
6232 "OpSelectionMerge %gather DontFlatten\n"
6233 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
6236 "OpBranch %gather\n"
6239 "OpBranch %gather\n"
6241 "%gather = OpLabel\n"
6242 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
6243 "%val = OpFAdd %f32 %val1 %delta\n"
6244 "%count__ = OpISub %i32 %count %c_i32_1\n"
6245 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6246 "OpBranchConditional %again %loop %exit\n"
6249 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
6250 "OpReturnValue %result\n"
6254 map<string, string> continue_target;
6256 // The Continue Target is the loop block itself.
6257 continue_target["continue_target"] = "%loop";
6258 fragments["testfun"] = multiBlock.specialize(continue_target);
6259 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
6261 // The Continue Target is at the end of the loop.
6262 continue_target["continue_target"] = "%gather";
6263 fragments["testfun"] = multiBlock.specialize(continue_target);
6264 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
6266 // A loop with continue statement.
6267 fragments["testfun"] =
6268 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6269 "%param1 = OpFunctionParameter %v4f32\n"
6271 "%entry = OpLabel\n"
6272 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6275 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
6277 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6278 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
6279 "OpLoopMerge %exit %continue None\n"
6283 ";skip if %count==2\n"
6284 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
6285 "OpSelectionMerge %continue DontFlatten\n"
6286 "OpBranchConditional %eq2 %continue %body\n"
6289 "%fcount = OpConvertSToF %f32 %count\n"
6290 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6291 "OpBranch %continue\n"
6293 "%continue = OpLabel\n"
6294 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
6295 "%count__ = OpISub %i32 %count %c_i32_1\n"
6296 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6297 "OpBranchConditional %again %loop %exit\n"
6300 "%same = OpFSub %f32 %val %c_f32_8\n"
6301 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6302 "OpReturnValue %result\n"
6304 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
6306 // A loop with break.
6307 fragments["testfun"] =
6308 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6309 "%param1 = OpFunctionParameter %v4f32\n"
6311 "%entry = OpLabel\n"
6312 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6313 "%dot = OpDot %f32 %param1 %param1\n"
6314 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6315 "%zero = OpConvertFToU %u32 %div\n"
6316 "%two = OpIAdd %i32 %zero %c_i32_2\n"
6317 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6320 ";adds 4 and 3 to %val0 (exits early)\n"
6322 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6323 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
6324 "OpLoopMerge %exit %continue None\n"
6328 ";end loop if %count==%two\n"
6329 "%above2 = OpSGreaterThan %bool %count %two\n"
6330 "OpSelectionMerge %continue DontFlatten\n"
6331 "OpBranchConditional %above2 %body %exit\n"
6334 "%fcount = OpConvertSToF %f32 %count\n"
6335 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6336 "OpBranch %continue\n"
6338 "%continue = OpLabel\n"
6339 "%count__ = OpISub %i32 %count %c_i32_1\n"
6340 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6341 "OpBranchConditional %again %loop %exit\n"
6344 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
6345 "%same = OpFSub %f32 %val_post %c_f32_7\n"
6346 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6347 "OpReturnValue %result\n"
6349 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
6351 // A loop with return.
6352 fragments["testfun"] =
6353 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6354 "%param1 = OpFunctionParameter %v4f32\n"
6356 "%entry = OpLabel\n"
6357 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6358 "%dot = OpDot %f32 %param1 %param1\n"
6359 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6360 "%zero = OpConvertFToU %u32 %div\n"
6361 "%two = OpIAdd %i32 %zero %c_i32_2\n"
6362 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6365 ";returns early without modifying %param1\n"
6367 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6368 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
6369 "OpLoopMerge %exit %continue None\n"
6373 ";return if %count==%two\n"
6374 "%above2 = OpSGreaterThan %bool %count %two\n"
6375 "OpSelectionMerge %continue DontFlatten\n"
6376 "OpBranchConditional %above2 %body %early_exit\n"
6378 "%early_exit = OpLabel\n"
6379 "OpReturnValue %param1\n"
6382 "%fcount = OpConvertSToF %f32 %count\n"
6383 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6384 "OpBranch %continue\n"
6386 "%continue = OpLabel\n"
6387 "%count__ = OpISub %i32 %count %c_i32_1\n"
6388 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6389 "OpBranchConditional %again %loop %exit\n"
6392 ";should never get here, so return an incorrect result\n"
6393 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
6394 "OpReturnValue %result\n"
6396 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
6398 return testGroup.release();
6401 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
6402 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
6404 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
6405 map<string, string> fragments;
6407 // A barrier inside a function body.
6408 fragments["pre_main"] =
6409 "%Workgroup = OpConstant %i32 2\n"
6410 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
6411 fragments["testfun"] =
6412 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6413 "%param1 = OpFunctionParameter %v4f32\n"
6414 "%label_testfun = OpLabel\n"
6415 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6416 "OpReturnValue %param1\n"
6418 addTessCtrlTest(testGroup.get(), "in_function", fragments);
6420 // Common setup code for the following tests.
6421 fragments["pre_main"] =
6422 "%Workgroup = OpConstant %i32 2\n"
6423 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
6424 "%c_f32_5 = OpConstant %f32 5.\n";
6425 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
6426 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6427 "%param1 = OpFunctionParameter %v4f32\n"
6428 "%entry = OpLabel\n"
6429 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6430 "%dot = OpDot %f32 %param1 %param1\n"
6431 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6432 "%zero = OpConvertFToU %u32 %div\n";
6434 // Barriers inside OpSwitch branches.
6435 fragments["testfun"] =
6437 "OpSelectionMerge %switch_exit None\n"
6438 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
6440 "%case1 = OpLabel\n"
6441 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6442 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6443 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6444 "OpBranch %switch_exit\n"
6446 "%switch_default = OpLabel\n"
6447 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6448 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6449 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6450 "OpBranch %switch_exit\n"
6452 "%case0 = OpLabel\n"
6453 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6454 "OpBranch %switch_exit\n"
6456 "%switch_exit = OpLabel\n"
6457 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
6458 "OpReturnValue %ret\n"
6460 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
6462 // Barriers inside if-then-else.
6463 fragments["testfun"] =
6465 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
6466 "OpSelectionMerge %exit DontFlatten\n"
6467 "OpBranchConditional %eq0 %then %else\n"
6470 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6471 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6472 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6476 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6480 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
6481 "OpReturnValue %ret\n"
6483 addTessCtrlTest(testGroup.get(), "in_if", fragments);
6485 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
6486 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
6487 fragments["testfun"] =
6489 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
6490 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
6491 "OpSelectionMerge %exit DontFlatten\n"
6492 "OpBranchConditional %thread0 %then %else\n"
6495 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6499 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
6503 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
6504 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6505 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
6506 "OpReturnValue %ret\n"
6508 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
6510 // A barrier inside a loop.
6511 fragments["pre_main"] =
6512 "%Workgroup = OpConstant %i32 2\n"
6513 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
6514 "%c_f32_10 = OpConstant %f32 10.\n";
6515 fragments["testfun"] =
6516 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6517 "%param1 = OpFunctionParameter %v4f32\n"
6518 "%entry = OpLabel\n"
6519 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6522 ";adds 4, 3, 2, and 1 to %val0\n"
6524 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
6525 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
6526 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6527 "%fcount = OpConvertSToF %f32 %count\n"
6528 "%val = OpFAdd %f32 %val1 %fcount\n"
6529 "%count__ = OpISub %i32 %count %c_i32_1\n"
6530 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6531 "OpLoopMerge %exit %loop None\n"
6532 "OpBranchConditional %again %loop %exit\n"
6535 "%same = OpFSub %f32 %val %c_f32_10\n"
6536 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6537 "OpReturnValue %ret\n"
6539 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
6541 return testGroup.release();
6544 // Test for the OpFRem instruction.
6545 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
6547 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
6548 map<string, string> fragments;
6549 RGBA inputColors[4];
6550 RGBA outputColors[4];
6552 fragments["pre_main"] =
6553 "%c_f32_3 = OpConstant %f32 3.0\n"
6554 "%c_f32_n3 = OpConstant %f32 -3.0\n"
6555 "%c_f32_4 = OpConstant %f32 4.0\n"
6556 "%c_f32_p75 = OpConstant %f32 0.75\n"
6557 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
6558 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
6559 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
6561 // The test does the following.
6562 // vec4 result = (param1 * 8.0) - 4.0;
6563 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
6564 fragments["testfun"] =
6565 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6566 "%param1 = OpFunctionParameter %v4f32\n"
6567 "%label_testfun = OpLabel\n"
6568 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
6569 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
6570 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
6571 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
6572 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
6573 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
6574 "OpReturnValue %xy_0_1\n"
6578 inputColors[0] = RGBA(16, 16, 0, 255);
6579 inputColors[1] = RGBA(232, 232, 0, 255);
6580 inputColors[2] = RGBA(232, 16, 0, 255);
6581 inputColors[3] = RGBA(16, 232, 0, 255);
6583 outputColors[0] = RGBA(64, 64, 0, 255);
6584 outputColors[1] = RGBA(255, 255, 0, 255);
6585 outputColors[2] = RGBA(255, 64, 0, 255);
6586 outputColors[3] = RGBA(64, 255, 0, 255);
6588 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
6589 return testGroup.release();
6592 // Test for the OpSRem instruction.
6593 tcu::TestCaseGroup* createOpSRemGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
6595 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "srem", "OpSRem"));
6596 map<string, string> fragments;
6598 fragments["pre_main"] =
6599 "%c_f32_255 = OpConstant %f32 255.0\n"
6600 "%c_i32_128 = OpConstant %i32 128\n"
6601 "%c_i32_255 = OpConstant %i32 255\n"
6602 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
6603 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
6604 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
6606 // The test does the following.
6607 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
6608 // ivec4 result = ivec4(srem(ints.x, ints.y), srem(ints.y, ints.z), srem(ints.z, ints.x), 255);
6609 // return float(result + 128) / 255.0;
6610 fragments["testfun"] =
6611 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6612 "%param1 = OpFunctionParameter %v4f32\n"
6613 "%label_testfun = OpLabel\n"
6614 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
6615 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
6616 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
6617 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
6618 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
6619 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
6620 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
6621 "%x_out = OpSRem %i32 %x_in %y_in\n"
6622 "%y_out = OpSRem %i32 %y_in %z_in\n"
6623 "%z_out = OpSRem %i32 %z_in %x_in\n"
6624 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
6625 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
6626 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
6627 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
6628 "OpReturnValue %float_out\n"
6631 const struct CaseParams
6634 const char* failMessageTemplate; // customized status message
6635 qpTestResult failResult; // override status on failure
6636 int operands[4][3]; // four (x, y, z) vectors of operands
6637 int results[4][3]; // four (x, y, z) vectors of results
6643 QP_TEST_RESULT_FAIL,
6644 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
6645 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
6649 "Inconsistent results, but within specification: ${reason}",
6650 negFailResult, // negative operands, not required by the spec
6651 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
6652 { { 5, 12, -2 }, { 0, -5, 2 }, { 3, 8, -6 }, { 25, -60, 0 } }, // results
6655 // If either operand is negative the result is undefined. Some implementations may still return correct values.
6657 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
6659 const CaseParams& params = cases[caseNdx];
6660 RGBA inputColors[4];
6661 RGBA outputColors[4];
6663 for (int i = 0; i < 4; ++i)
6665 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
6666 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
6669 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
6672 return testGroup.release();
6675 // Test for the OpSMod instruction.
6676 tcu::TestCaseGroup* createOpSModGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
6678 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "smod", "OpSMod"));
6679 map<string, string> fragments;
6681 fragments["pre_main"] =
6682 "%c_f32_255 = OpConstant %f32 255.0\n"
6683 "%c_i32_128 = OpConstant %i32 128\n"
6684 "%c_i32_255 = OpConstant %i32 255\n"
6685 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
6686 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
6687 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
6689 // The test does the following.
6690 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
6691 // ivec4 result = ivec4(smod(ints.x, ints.y), smod(ints.y, ints.z), smod(ints.z, ints.x), 255);
6692 // return float(result + 128) / 255.0;
6693 fragments["testfun"] =
6694 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6695 "%param1 = OpFunctionParameter %v4f32\n"
6696 "%label_testfun = OpLabel\n"
6697 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
6698 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
6699 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
6700 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
6701 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
6702 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
6703 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
6704 "%x_out = OpSMod %i32 %x_in %y_in\n"
6705 "%y_out = OpSMod %i32 %y_in %z_in\n"
6706 "%z_out = OpSMod %i32 %z_in %x_in\n"
6707 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
6708 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
6709 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
6710 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
6711 "OpReturnValue %float_out\n"
6714 const struct CaseParams
6717 const char* failMessageTemplate; // customized status message
6718 qpTestResult failResult; // override status on failure
6719 int operands[4][3]; // four (x, y, z) vectors of operands
6720 int results[4][3]; // four (x, y, z) vectors of results
6726 QP_TEST_RESULT_FAIL,
6727 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
6728 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
6732 "Inconsistent results, but within specification: ${reason}",
6733 negFailResult, // negative operands, not required by the spec
6734 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
6735 { { 5, -5, 3 }, { 0, 2, -3 }, { 3, -73, 69 }, { -35, 40, 0 } }, // results
6738 // If either operand is negative the result is undefined. Some implementations may still return correct values.
6740 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
6742 const CaseParams& params = cases[caseNdx];
6743 RGBA inputColors[4];
6744 RGBA outputColors[4];
6746 for (int i = 0; i < 4; ++i)
6748 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
6749 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
6752 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
6755 return testGroup.release();
6760 INTEGER_TYPE_SIGNED_16,
6761 INTEGER_TYPE_SIGNED_32,
6762 INTEGER_TYPE_SIGNED_64,
6764 INTEGER_TYPE_UNSIGNED_16,
6765 INTEGER_TYPE_UNSIGNED_32,
6766 INTEGER_TYPE_UNSIGNED_64,
6769 const string getBitWidthStr (IntegerType type)
6773 case INTEGER_TYPE_SIGNED_16:
6774 case INTEGER_TYPE_UNSIGNED_16: return "16";
6776 case INTEGER_TYPE_SIGNED_32:
6777 case INTEGER_TYPE_UNSIGNED_32: return "32";
6779 case INTEGER_TYPE_SIGNED_64:
6780 case INTEGER_TYPE_UNSIGNED_64: return "64";
6782 default: DE_ASSERT(false);
6787 const string getByteWidthStr (IntegerType type)
6791 case INTEGER_TYPE_SIGNED_16:
6792 case INTEGER_TYPE_UNSIGNED_16: return "2";
6794 case INTEGER_TYPE_SIGNED_32:
6795 case INTEGER_TYPE_UNSIGNED_32: return "4";
6797 case INTEGER_TYPE_SIGNED_64:
6798 case INTEGER_TYPE_UNSIGNED_64: return "8";
6800 default: DE_ASSERT(false);
6805 bool isSigned (IntegerType type)
6807 return (type <= INTEGER_TYPE_SIGNED_64);
6810 const string getTypeName (IntegerType type)
6812 string prefix = isSigned(type) ? "" : "u";
6813 return prefix + "int" + getBitWidthStr(type);
6816 const string getTestName (IntegerType from, IntegerType to)
6818 return getTypeName(from) + "_to_" + getTypeName(to);
6821 const string getAsmTypeDeclaration (IntegerType type)
6823 string sign = isSigned(type) ? " 1" : " 0";
6824 return "OpTypeInt " + getBitWidthStr(type) + sign;
6827 template<typename T>
6828 BufferSp getSpecializedBuffer (deInt64 number)
6830 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
6833 BufferSp getBuffer (IntegerType type, deInt64 number)
6837 case INTEGER_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
6838 case INTEGER_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
6839 case INTEGER_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
6841 case INTEGER_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
6842 case INTEGER_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
6843 case INTEGER_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
6845 default: DE_ASSERT(false);
6846 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
6850 bool usesInt16 (IntegerType from, IntegerType to)
6852 return (from == INTEGER_TYPE_SIGNED_16 || from == INTEGER_TYPE_UNSIGNED_16
6853 || to == INTEGER_TYPE_SIGNED_16 || to == INTEGER_TYPE_UNSIGNED_16);
6856 bool usesInt64 (IntegerType from, IntegerType to)
6858 return (from == INTEGER_TYPE_SIGNED_64 || from == INTEGER_TYPE_UNSIGNED_64
6859 || to == INTEGER_TYPE_SIGNED_64 || to == INTEGER_TYPE_UNSIGNED_64);
6862 ComputeTestFeatures getConversionUsedFeatures (IntegerType from, IntegerType to)
6864 if (usesInt16(from, to))
6866 if (usesInt64(from, to))
6868 return COMPUTE_TEST_USES_INT16_INT64;
6872 return COMPUTE_TEST_USES_INT16;
6877 return COMPUTE_TEST_USES_INT64;
6883 ConvertCase (IntegerType from, IntegerType to, deInt64 number)
6886 , m_features (getConversionUsedFeatures(from, to))
6887 , m_name (getTestName(from, to))
6888 , m_inputBuffer (getBuffer(from, number))
6889 , m_outputBuffer (getBuffer(to, number))
6891 m_asmTypes["inputType"] = getAsmTypeDeclaration(from);
6892 m_asmTypes["outputType"] = getAsmTypeDeclaration(to);
6894 if (m_features == COMPUTE_TEST_USES_INT16)
6896 m_asmTypes["int_capabilities"] = "OpCapability Int16\n";
6898 else if (m_features == COMPUTE_TEST_USES_INT64)
6900 m_asmTypes["int_capabilities"] = "OpCapability Int64\n";
6902 else if (m_features == COMPUTE_TEST_USES_INT16_INT64)
6904 m_asmTypes["int_capabilities"] = string("OpCapability Int16\n") +
6905 "OpCapability Int64\n";
6913 IntegerType m_fromType;
6914 IntegerType m_toType;
6915 ComputeTestFeatures m_features;
6917 map<string, string> m_asmTypes;
6918 BufferSp m_inputBuffer;
6919 BufferSp m_outputBuffer;
6922 const string getConvertCaseShaderStr (const string& instruction, const ConvertCase& convertCase)
6924 map<string, string> params = convertCase.m_asmTypes;
6926 params["instruction"] = instruction;
6928 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
6929 params["outDecorator"] = getByteWidthStr(convertCase.m_toType);
6931 const StringTemplate shader (
6932 "OpCapability Shader\n"
6933 "${int_capabilities}"
6934 "OpMemoryModel Logical GLSL450\n"
6935 "OpEntryPoint GLCompute %main \"main\" %id\n"
6936 "OpExecutionMode %main LocalSize 1 1 1\n"
6937 "OpSource GLSL 430\n"
6938 "OpName %main \"main\"\n"
6939 "OpName %id \"gl_GlobalInvocationID\"\n"
6941 "OpDecorate %id BuiltIn GlobalInvocationId\n"
6942 "OpDecorate %indata DescriptorSet 0\n"
6943 "OpDecorate %indata Binding 0\n"
6944 "OpDecorate %outdata DescriptorSet 0\n"
6945 "OpDecorate %outdata Binding 1\n"
6946 "OpDecorate %in_arr ArrayStride ${inDecorator}\n"
6947 "OpDecorate %out_arr ArrayStride ${outDecorator}\n"
6948 "OpDecorate %in_buf BufferBlock\n"
6949 "OpDecorate %out_buf BufferBlock\n"
6950 "OpMemberDecorate %in_buf 0 Offset 0\n"
6951 "OpMemberDecorate %out_buf 0 Offset 0\n"
6953 "%void = OpTypeVoid\n"
6954 "%voidf = OpTypeFunction %void\n"
6955 "%u32 = OpTypeInt 32 0\n"
6956 "%i32 = OpTypeInt 32 1\n"
6957 "%uvec3 = OpTypeVector %u32 3\n"
6958 "%uvec3ptr = OpTypePointer Input %uvec3\n"
6960 "%in_type = ${inputType}\n"
6961 "%out_type = ${outputType}\n"
6963 "%in_ptr = OpTypePointer Uniform %in_type\n"
6964 "%out_ptr = OpTypePointer Uniform %out_type\n"
6965 "%in_arr = OpTypeRuntimeArray %in_type\n"
6966 "%out_arr = OpTypeRuntimeArray %out_type\n"
6967 "%in_buf = OpTypeStruct %in_arr\n"
6968 "%out_buf = OpTypeStruct %out_arr\n"
6969 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
6970 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
6971 "%indata = OpVariable %in_bufptr Uniform\n"
6972 "%outdata = OpVariable %out_bufptr Uniform\n"
6973 "%inputptr = OpTypePointer Input %in_type\n"
6974 "%id = OpVariable %uvec3ptr Input\n"
6976 "%zero = OpConstant %i32 0\n"
6978 "%main = OpFunction %void None %voidf\n"
6979 "%label = OpLabel\n"
6980 "%idval = OpLoad %uvec3 %id\n"
6981 "%x = OpCompositeExtract %u32 %idval 0\n"
6982 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
6983 "%outloc = OpAccessChain %out_ptr %outdata %zero %x\n"
6984 "%inval = OpLoad %in_type %inloc\n"
6985 "%conv = ${instruction} %out_type %inval\n"
6986 " OpStore %outloc %conv\n"
6991 return shader.specialize(params);
6994 void createSConvertCases (vector<ConvertCase>& testCases)
6996 // Convert int to int
6997 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_32, 14669));
6998 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_64, 3341));
7000 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_SIGNED_64, 973610259));
7002 // Convert int to unsigned int
7003 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_32, 9288));
7004 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_64, 15460));
7006 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_UNSIGNED_64, 346213461));
7009 // Test for the OpSConvert instruction.
7010 tcu::TestCaseGroup* createSConvertTests (tcu::TestContext& testCtx)
7012 const string instruction ("OpSConvert");
7013 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "sconvert", "OpSConvert"));
7014 vector<ConvertCase> testCases;
7015 createSConvertCases(testCases);
7017 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7019 ComputeShaderSpec spec;
7021 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7022 spec.inputs.push_back(test->m_inputBuffer);
7023 spec.outputs.push_back(test->m_outputBuffer);
7024 spec.numWorkGroups = IVec3(1, 1, 1);
7026 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpSConvert.", spec, test->m_features));
7029 return group.release();
7032 void createUConvertCases (vector<ConvertCase>& testCases)
7034 // Convert unsigned int to unsigned int
7035 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_32, 60653));
7036 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_64, 17991));
7038 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_UNSIGNED_64, 904256275));
7040 // Convert unsigned int to int
7041 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_32, 38002));
7042 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_64, 64921));
7044 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_SIGNED_64, 4294956295ll));
7047 // Test for the OpUConvert instruction.
7048 tcu::TestCaseGroup* createUConvertTests (tcu::TestContext& testCtx)
7050 const string instruction ("OpUConvert");
7051 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "uconvert", "OpUConvert"));
7052 vector<ConvertCase> testCases;
7053 createUConvertCases(testCases);
7055 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7057 ComputeShaderSpec spec;
7059 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7060 spec.inputs.push_back(test->m_inputBuffer);
7061 spec.outputs.push_back(test->m_outputBuffer);
7062 spec.numWorkGroups = IVec3(1, 1, 1);
7064 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpUConvert.", spec, test->m_features));
7066 return group.release();
7069 const string getNumberTypeName (const NumberType type)
7071 if (type == NUMBERTYPE_INT32)
7075 else if (type == NUMBERTYPE_UINT32)
7079 else if (type == NUMBERTYPE_FLOAT32)
7090 deInt32 getInt(de::Random& rnd)
7092 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
7095 const string repeatString (const string& str, int times)
7098 for (int i = 0; i < times; ++i)
7105 const string getRandomConstantString (const NumberType type, de::Random& rnd)
7107 if (type == NUMBERTYPE_INT32)
7109 return numberToString<deInt32>(getInt(rnd));
7111 else if (type == NUMBERTYPE_UINT32)
7113 return numberToString<deUint32>(rnd.getUint32());
7115 else if (type == NUMBERTYPE_FLOAT32)
7117 return numberToString<float>(rnd.getFloat());
7126 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7128 map<string, string> params;
7131 for (int width = 2; width <= 4; ++width)
7133 string randomConst = numberToString(getInt(rnd));
7134 string widthStr = numberToString(width);
7135 int index = rnd.getInt(0, width-1);
7137 params["type"] = "vec";
7138 params["name"] = params["type"] + "_" + widthStr;
7139 params["compositeType"] = "%composite = OpTypeVector %custom " + widthStr +"\n";
7140 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7141 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7142 params["indexes"] = numberToString(index);
7143 testCases.push_back(params);
7147 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7149 const int limit = 10;
7150 map<string, string> params;
7152 for (int width = 2; width <= limit; ++width)
7154 string randomConst = numberToString(getInt(rnd));
7155 string widthStr = numberToString(width);
7156 int index = rnd.getInt(0, width-1);
7158 params["type"] = "array";
7159 params["name"] = params["type"] + "_" + widthStr;
7160 params["compositeType"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
7161 + "%composite = OpTypeArray %custom %arraywidth\n";
7163 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7164 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7165 params["indexes"] = numberToString(index);
7166 testCases.push_back(params);
7170 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7172 const int limit = 10;
7173 map<string, string> params;
7175 for (int width = 2; width <= limit; ++width)
7177 string randomConst = numberToString(getInt(rnd));
7178 int index = rnd.getInt(0, width-1);
7180 params["type"] = "struct";
7181 params["name"] = params["type"] + "_" + numberToString(width);
7182 params["compositeType"] = "%composite = OpTypeStruct" + repeatString(" %custom", width) + "\n";
7183 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7184 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7185 params["indexes"] = numberToString(index);
7186 testCases.push_back(params);
7190 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7192 map<string, string> params;
7195 for (int width = 2; width <= 4; ++width)
7197 string widthStr = numberToString(width);
7199 for (int column = 2 ; column <= 4; ++column)
7201 int index_0 = rnd.getInt(0, column-1);
7202 int index_1 = rnd.getInt(0, width-1);
7203 string columnStr = numberToString(column);
7205 params["type"] = "matrix";
7206 params["name"] = params["type"] + "_" + widthStr + "x" + columnStr;
7207 params["compositeType"] = string("%vectype = OpTypeVector %custom " + widthStr + "\n")
7208 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
7210 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n"
7211 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
7213 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
7214 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
7215 testCases.push_back(params);
7220 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7222 createVectorCompositeCases(testCases, rnd, type);
7223 createArrayCompositeCases(testCases, rnd, type);
7224 createStructCompositeCases(testCases, rnd, type);
7225 // Matrix only supports float types
7226 if (type == NUMBERTYPE_FLOAT32)
7228 createMatrixCompositeCases(testCases, rnd, type);
7232 const string getAssemblyTypeDeclaration (const NumberType type)
7236 case NUMBERTYPE_INT32: return "OpTypeInt 32 1";
7237 case NUMBERTYPE_UINT32: return "OpTypeInt 32 0";
7238 case NUMBERTYPE_FLOAT32: return "OpTypeFloat 32";
7239 default: DE_ASSERT(false); return "";
7243 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
7245 map<string, string> parameters(params);
7247 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7249 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
7251 return StringTemplate (
7252 "OpCapability Shader\n"
7253 "OpCapability Matrix\n"
7254 "OpMemoryModel Logical GLSL450\n"
7255 "OpEntryPoint GLCompute %main \"main\" %id\n"
7256 "OpExecutionMode %main LocalSize 1 1 1\n"
7258 "OpSource GLSL 430\n"
7259 "OpName %main \"main\"\n"
7260 "OpName %id \"gl_GlobalInvocationID\"\n"
7263 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7264 "OpDecorate %buf BufferBlock\n"
7265 "OpDecorate %indata DescriptorSet 0\n"
7266 "OpDecorate %indata Binding 0\n"
7267 "OpDecorate %outdata DescriptorSet 0\n"
7268 "OpDecorate %outdata Binding 1\n"
7269 "OpDecorate %customarr ArrayStride 4\n"
7270 "${compositeDecorator}"
7271 "OpMemberDecorate %buf 0 Offset 0\n"
7274 "%void = OpTypeVoid\n"
7275 "%voidf = OpTypeFunction %void\n"
7276 "%u32 = OpTypeInt 32 0\n"
7277 "%i32 = OpTypeInt 32 1\n"
7278 "%uvec3 = OpTypeVector %u32 3\n"
7279 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7282 "%custom = ${typeDeclaration}\n"
7288 // Inherited from custom
7289 "%customptr = OpTypePointer Uniform %custom\n"
7290 "%customarr = OpTypeRuntimeArray %custom\n"
7291 "%buf = OpTypeStruct %customarr\n"
7292 "%bufptr = OpTypePointer Uniform %buf\n"
7294 "%indata = OpVariable %bufptr Uniform\n"
7295 "%outdata = OpVariable %bufptr Uniform\n"
7297 "%id = OpVariable %uvec3ptr Input\n"
7298 "%zero = OpConstant %i32 0\n"
7300 "%main = OpFunction %void None %voidf\n"
7301 "%label = OpLabel\n"
7302 "%idval = OpLoad %uvec3 %id\n"
7303 "%x = OpCompositeExtract %u32 %idval 0\n"
7305 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
7306 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
7307 // Read the input value
7308 "%inval = OpLoad %custom %inloc\n"
7309 // Create the composite and fill it
7310 "${compositeConstruct}"
7311 // Insert the input value to a place
7312 "%instance2 = OpCompositeInsert %composite %inval %instance ${indexes}\n"
7313 // Read back the value from the position
7314 "%out_val = OpCompositeExtract %custom %instance2 ${indexes}\n"
7315 // Store it in the output position
7316 " OpStore %outloc %out_val\n"
7319 ).specialize(parameters);
7322 template<typename T>
7323 BufferSp createCompositeBuffer(T number)
7325 return BufferSp(new Buffer<T>(vector<T>(1, number)));
7328 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
7330 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
7331 de::Random rnd (deStringHash(group->getName()));
7333 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7335 NumberType numberType = NumberType(type);
7336 const string typeName = getNumberTypeName(numberType);
7337 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
7338 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7339 vector<map<string, string> > testCases;
7341 createCompositeCases(testCases, rnd, numberType);
7343 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7345 ComputeShaderSpec spec;
7347 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
7351 case NUMBERTYPE_INT32:
7353 deInt32 number = getInt(rnd);
7354 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7355 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7358 case NUMBERTYPE_UINT32:
7360 deUint32 number = rnd.getUint32();
7361 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7362 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7365 case NUMBERTYPE_FLOAT32:
7367 float number = rnd.getFloat();
7368 spec.inputs.push_back(createCompositeBuffer<float>(number));
7369 spec.outputs.push_back(createCompositeBuffer<float>(number));
7376 spec.numWorkGroups = IVec3(1, 1, 1);
7377 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
7379 group->addChild(subGroup.release());
7381 return group.release();
7384 struct AssemblyStructInfo
7386 AssemblyStructInfo (const deUint32 comp, const deUint32 idx)
7391 deUint32 components;
7395 const string specializeInBoundsShaderTemplate (const NumberType type, const AssemblyStructInfo& structInfo, const map<string, string>& params)
7397 // Create the full index string
7398 string fullIndex = numberToString(structInfo.index) + " " + params.at("indexes");
7399 // Convert it to list of indexes
7400 vector<string> indexes = de::splitString(fullIndex, ' ');
7402 map<string, string> parameters (params);
7403 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7404 parameters["structType"] = repeatString(" %composite", structInfo.components);
7405 parameters["structConstruct"] = repeatString(" %instance", structInfo.components);
7406 parameters["insertIndexes"] = fullIndex;
7408 // In matrix cases the last two index is the CompositeExtract indexes
7409 const deUint32 extractIndexes = (parameters["type"] == "matrix") ? 2 : 1;
7411 // Construct the extractIndex
7412 for (vector<string>::const_iterator index = indexes.end() - extractIndexes; index != indexes.end(); ++index)
7414 parameters["extractIndexes"] += " " + *index;
7417 // Remove the last 1 or 2 element depends on matrix case or not
7418 indexes.erase(indexes.end() - extractIndexes, indexes.end());
7421 // Generate AccessChain index expressions (except for the last one, because we use ptr to the composite)
7422 for (vector<string>::const_iterator index = indexes.begin(); index != indexes.end(); ++index)
7424 string indexId = "%index_" + numberToString(id++);
7425 parameters["accessChainConstDeclaration"] += indexId + " = OpConstant %u32 " + *index + "\n";
7426 parameters["accessChainIndexes"] += " " + indexId;
7429 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
7431 return StringTemplate (
7432 "OpCapability Shader\n"
7433 "OpCapability Matrix\n"
7434 "OpMemoryModel Logical GLSL450\n"
7435 "OpEntryPoint GLCompute %main \"main\" %id\n"
7436 "OpExecutionMode %main LocalSize 1 1 1\n"
7438 "OpSource GLSL 430\n"
7439 "OpName %main \"main\"\n"
7440 "OpName %id \"gl_GlobalInvocationID\"\n"
7442 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7443 "OpDecorate %buf BufferBlock\n"
7444 "OpDecorate %indata DescriptorSet 0\n"
7445 "OpDecorate %indata Binding 0\n"
7446 "OpDecorate %outdata DescriptorSet 0\n"
7447 "OpDecorate %outdata Binding 1\n"
7448 "OpDecorate %customarr ArrayStride 4\n"
7449 "${compositeDecorator}"
7450 "OpMemberDecorate %buf 0 Offset 0\n"
7452 "%void = OpTypeVoid\n"
7453 "%voidf = OpTypeFunction %void\n"
7454 "%u32 = OpTypeInt 32 0\n"
7455 "%uvec3 = OpTypeVector %u32 3\n"
7456 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7458 "%custom = ${typeDeclaration}\n"
7461 // Inherited from composite
7462 "%composite_p = OpTypePointer Function %composite\n"
7463 "%struct_t = OpTypeStruct${structType}\n"
7464 "%struct_p = OpTypePointer Function %struct_t\n"
7467 "${accessChainConstDeclaration}"
7468 // Inherited from custom
7469 "%customptr = OpTypePointer Uniform %custom\n"
7470 "%customarr = OpTypeRuntimeArray %custom\n"
7471 "%buf = OpTypeStruct %customarr\n"
7472 "%bufptr = OpTypePointer Uniform %buf\n"
7473 "%indata = OpVariable %bufptr Uniform\n"
7474 "%outdata = OpVariable %bufptr Uniform\n"
7476 "%id = OpVariable %uvec3ptr Input\n"
7477 "%zero = OpConstant %u32 0\n"
7478 "%main = OpFunction %void None %voidf\n"
7479 "%label = OpLabel\n"
7480 "%struct_v = OpVariable %struct_p Function\n"
7481 "%idval = OpLoad %uvec3 %id\n"
7482 "%x = OpCompositeExtract %u32 %idval 0\n"
7483 // Create the input/output type
7484 "%inloc = OpInBoundsAccessChain %customptr %indata %zero %x\n"
7485 "%outloc = OpInBoundsAccessChain %customptr %outdata %zero %x\n"
7486 // Read the input value
7487 "%inval = OpLoad %custom %inloc\n"
7488 // Create the composite and fill it
7489 "${compositeConstruct}"
7490 // Create the struct and fill it with the composite
7491 "%struct = OpCompositeConstruct %struct_t${structConstruct}\n"
7493 "%comp_obj = OpCompositeInsert %struct_t %inval %struct ${insertIndexes}\n"
7495 " OpStore %struct_v %comp_obj\n"
7496 // Get deepest possible composite pointer
7497 "%inner_ptr = OpInBoundsAccessChain %composite_p %struct_v${accessChainIndexes}\n"
7498 "%read_obj = OpLoad %composite %inner_ptr\n"
7499 // Read back the stored value
7500 "%read_val = OpCompositeExtract %custom %read_obj${extractIndexes}\n"
7501 " OpStore %outloc %read_val\n"
7503 " OpFunctionEnd\n").specialize(parameters);
7506 tcu::TestCaseGroup* createOpInBoundsAccessChainGroup (tcu::TestContext& testCtx)
7508 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opinboundsaccesschain", "Test the OpInBoundsAccessChain instruction"));
7509 de::Random rnd (deStringHash(group->getName()));
7511 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7513 NumberType numberType = NumberType(type);
7514 const string typeName = getNumberTypeName(numberType);
7515 const string description = "Test the OpInBoundsAccessChain instruction with " + typeName + "s";
7516 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7518 vector<map<string, string> > testCases;
7519 createCompositeCases(testCases, rnd, numberType);
7521 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7523 ComputeShaderSpec spec;
7525 // Number of components inside of a struct
7526 deUint32 structComponents = rnd.getInt(2, 8);
7527 // Component index value
7528 deUint32 structIndex = rnd.getInt(0, structComponents - 1);
7529 AssemblyStructInfo structInfo(structComponents, structIndex);
7531 spec.assembly = specializeInBoundsShaderTemplate(numberType, structInfo, *test);
7535 case NUMBERTYPE_INT32:
7537 deInt32 number = getInt(rnd);
7538 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7539 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7542 case NUMBERTYPE_UINT32:
7544 deUint32 number = rnd.getUint32();
7545 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7546 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7549 case NUMBERTYPE_FLOAT32:
7551 float number = rnd.getFloat();
7552 spec.inputs.push_back(createCompositeBuffer<float>(number));
7553 spec.outputs.push_back(createCompositeBuffer<float>(number));
7559 spec.numWorkGroups = IVec3(1, 1, 1);
7560 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpInBoundsAccessChain test", spec));
7562 group->addChild(subGroup.release());
7564 return group.release();
7567 // If the params missing, uninitialized case
7568 const string specializeDefaultOutputShaderTemplate (const NumberType type, const map<string, string>& params = map<string, string>())
7570 map<string, string> parameters(params);
7572 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7574 // Declare the const value, and use it in the initializer
7575 if (params.find("constValue") != params.end())
7577 parameters["constDeclaration"] = "%const = OpConstant %in_type " + params.at("constValue") + "\n";
7578 parameters["variableInitializer"] = "%const";
7580 // Uninitialized case
7583 parameters["constDeclaration"] = "";
7584 parameters["variableInitializer"] = "";
7587 return StringTemplate(
7588 "OpCapability Shader\n"
7589 "OpMemoryModel Logical GLSL450\n"
7590 "OpEntryPoint GLCompute %main \"main\" %id\n"
7591 "OpExecutionMode %main LocalSize 1 1 1\n"
7592 "OpSource GLSL 430\n"
7593 "OpName %main \"main\"\n"
7594 "OpName %id \"gl_GlobalInvocationID\"\n"
7596 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7597 "OpDecorate %indata DescriptorSet 0\n"
7598 "OpDecorate %indata Binding 0\n"
7599 "OpDecorate %outdata DescriptorSet 0\n"
7600 "OpDecorate %outdata Binding 1\n"
7601 "OpDecorate %in_arr ArrayStride 4\n"
7602 "OpDecorate %in_buf BufferBlock\n"
7603 "OpMemberDecorate %in_buf 0 Offset 0\n"
7605 "%void = OpTypeVoid\n"
7606 "%voidf = OpTypeFunction %void\n"
7607 "%u32 = OpTypeInt 32 0\n"
7608 "%i32 = OpTypeInt 32 1\n"
7609 "%uvec3 = OpTypeVector %u32 3\n"
7610 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7612 "%in_type = ${typeDeclaration}\n"
7613 // "%const = OpConstant %in_type ${constValue}\n"
7614 "${constDeclaration}\n"
7616 "%in_ptr = OpTypePointer Uniform %in_type\n"
7617 "%in_arr = OpTypeRuntimeArray %in_type\n"
7618 "%in_buf = OpTypeStruct %in_arr\n"
7619 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
7620 "%indata = OpVariable %in_bufptr Uniform\n"
7621 "%outdata = OpVariable %in_bufptr Uniform\n"
7622 "%id = OpVariable %uvec3ptr Input\n"
7623 "%var_ptr = OpTypePointer Function %in_type\n"
7625 "%zero = OpConstant %i32 0\n"
7627 "%main = OpFunction %void None %voidf\n"
7628 "%label = OpLabel\n"
7629 "%out_var = OpVariable %var_ptr Function ${variableInitializer}\n"
7630 "%idval = OpLoad %uvec3 %id\n"
7631 "%x = OpCompositeExtract %u32 %idval 0\n"
7632 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
7633 "%outloc = OpAccessChain %in_ptr %outdata %zero %x\n"
7635 "%outval = OpLoad %in_type %out_var\n"
7636 " OpStore %outloc %outval\n"
7639 ).specialize(parameters);
7642 bool compareFloats (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
7644 DE_ASSERT(outputAllocs.size() != 0);
7645 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
7647 // Use custom epsilon because of the float->string conversion
7648 const float epsilon = 0.00001f;
7650 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
7653 memcpy(&expected, expectedOutputs[outputNdx]->data(), expectedOutputs[outputNdx]->getNumBytes());
7656 memcpy(&actual, outputAllocs[outputNdx]->getHostPtr(), expectedOutputs[outputNdx]->getNumBytes());
7658 // Test with epsilon
7659 if (fabs(expected - actual) > epsilon)
7661 log << TestLog::Message << "Error: The actual and expected values not matching."
7662 << " Expected: " << expected << " Actual: " << actual << " Epsilon: " << epsilon << TestLog::EndMessage;
7669 // Checks if the driver crash with uninitialized cases
7670 bool passthruVerify (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
7672 DE_ASSERT(outputAllocs.size() != 0);
7673 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
7675 // Copy and discard the result.
7676 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
7678 size_t width = expectedOutputs[outputNdx]->getNumBytes();
7680 vector<char> data(width);
7681 memcpy(&data[0], outputAllocs[outputNdx]->getHostPtr(), width);
7686 tcu::TestCaseGroup* createShaderDefaultOutputGroup (tcu::TestContext& testCtx)
7688 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "shader_default_output", "Test shader default output."));
7689 de::Random rnd (deStringHash(group->getName()));
7691 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7693 NumberType numberType = NumberType(type);
7694 const string typeName = getNumberTypeName(numberType);
7695 const string description = "Test the OpVariable initializer with " + typeName + ".";
7696 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7698 // 2 similar subcases (initialized and uninitialized)
7699 for (int subCase = 0; subCase < 2; ++subCase)
7701 ComputeShaderSpec spec;
7702 spec.numWorkGroups = IVec3(1, 1, 1);
7704 map<string, string> params;
7708 case NUMBERTYPE_INT32:
7710 deInt32 number = getInt(rnd);
7711 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7712 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7713 params["constValue"] = numberToString(number);
7716 case NUMBERTYPE_UINT32:
7718 deUint32 number = rnd.getUint32();
7719 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7720 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7721 params["constValue"] = numberToString(number);
7724 case NUMBERTYPE_FLOAT32:
7726 float number = rnd.getFloat();
7727 spec.inputs.push_back(createCompositeBuffer<float>(number));
7728 spec.outputs.push_back(createCompositeBuffer<float>(number));
7729 spec.verifyIO = &compareFloats;
7730 params["constValue"] = numberToString(number);
7737 // Initialized subcase
7740 spec.assembly = specializeDefaultOutputShaderTemplate(numberType, params);
7741 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "initialized", "OpVariable initializer tests.", spec));
7743 // Uninitialized subcase
7746 spec.assembly = specializeDefaultOutputShaderTemplate(numberType);
7747 spec.verifyIO = &passthruVerify;
7748 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "uninitialized", "OpVariable initializer tests.", spec));
7751 group->addChild(subGroup.release());
7753 return group.release();
7756 tcu::TestCaseGroup* createOpNopTests (tcu::TestContext& testCtx)
7758 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
7759 RGBA defaultColors[4];
7760 map<string, string> opNopFragments;
7762 getDefaultColors(defaultColors);
7764 opNopFragments["testfun"] =
7765 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7766 "%param1 = OpFunctionParameter %v4f32\n"
7767 "%label_testfun = OpLabel\n"
7776 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7777 "%b = OpFAdd %f32 %a %a\n"
7779 "%c = OpFSub %f32 %b %a\n"
7780 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
7783 "OpReturnValue %ret\n"
7786 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, testGroup.get());
7788 return testGroup.release();
7791 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
7793 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
7794 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
7795 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
7797 computeTests->addChild(createOpNopGroup(testCtx));
7798 computeTests->addChild(createOpFUnordGroup(testCtx));
7799 computeTests->addChild(createOpAtomicGroup(testCtx, false));
7800 computeTests->addChild(createOpAtomicGroup(testCtx, true)); // Using new StorageBuffer decoration
7801 computeTests->addChild(createOpLineGroup(testCtx));
7802 computeTests->addChild(createOpNoLineGroup(testCtx));
7803 computeTests->addChild(createOpConstantNullGroup(testCtx));
7804 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
7805 computeTests->addChild(createOpConstantUsageGroup(testCtx));
7806 computeTests->addChild(createSpecConstantGroup(testCtx));
7807 computeTests->addChild(createOpSourceGroup(testCtx));
7808 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
7809 computeTests->addChild(createDecorationGroupGroup(testCtx));
7810 computeTests->addChild(createOpPhiGroup(testCtx));
7811 computeTests->addChild(createLoopControlGroup(testCtx));
7812 computeTests->addChild(createFunctionControlGroup(testCtx));
7813 computeTests->addChild(createSelectionControlGroup(testCtx));
7814 computeTests->addChild(createBlockOrderGroup(testCtx));
7815 computeTests->addChild(createMultipleShaderGroup(testCtx));
7816 computeTests->addChild(createMemoryAccessGroup(testCtx));
7817 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
7818 computeTests->addChild(createOpCopyObjectGroup(testCtx));
7819 computeTests->addChild(createNoContractionGroup(testCtx));
7820 computeTests->addChild(createOpUndefGroup(testCtx));
7821 computeTests->addChild(createOpUnreachableGroup(testCtx));
7822 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
7823 computeTests ->addChild(createOpFRemGroup(testCtx));
7824 computeTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_PASS));
7825 computeTests->addChild(createOpSRemComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
7826 computeTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_PASS));
7827 computeTests->addChild(createOpSModComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
7828 computeTests->addChild(createSConvertTests(testCtx));
7829 computeTests->addChild(createUConvertTests(testCtx));
7830 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
7831 computeTests->addChild(createOpInBoundsAccessChainGroup(testCtx));
7832 computeTests->addChild(createShaderDefaultOutputGroup(testCtx));
7833 computeTests->addChild(createOpNMinGroup(testCtx));
7834 computeTests->addChild(createOpNMaxGroup(testCtx));
7835 computeTests->addChild(createOpNClampGroup(testCtx));
7837 de::MovePtr<tcu::TestCaseGroup> computeAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
7839 computeAndroidTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7840 computeAndroidTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7842 computeTests->addChild(computeAndroidTests.release());
7845 computeTests->addChild(create16BitStorageComputeGroup(testCtx));
7846 computeTests->addChild(createUboMatrixPaddingComputeGroup(testCtx));
7847 computeTests->addChild(createConditionalBranchComputeGroup(testCtx));
7848 computeTests->addChild(createVariablePointersComputeGroup(testCtx));
7849 graphicsTests->addChild(createOpNopTests(testCtx));
7850 graphicsTests->addChild(createOpSourceTests(testCtx));
7851 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
7852 graphicsTests->addChild(createOpLineTests(testCtx));
7853 graphicsTests->addChild(createOpNoLineTests(testCtx));
7854 graphicsTests->addChild(createOpConstantNullTests(testCtx));
7855 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
7856 graphicsTests->addChild(createMemoryAccessTests(testCtx));
7857 graphicsTests->addChild(createOpUndefTests(testCtx));
7858 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
7859 graphicsTests->addChild(createModuleTests(testCtx));
7860 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
7861 graphicsTests->addChild(createOpPhiTests(testCtx));
7862 graphicsTests->addChild(createNoContractionTests(testCtx));
7863 graphicsTests->addChild(createOpQuantizeTests(testCtx));
7864 graphicsTests->addChild(createLoopTests(testCtx));
7865 graphicsTests->addChild(createSpecConstantTests(testCtx));
7866 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
7867 graphicsTests->addChild(createBarrierTests(testCtx));
7868 graphicsTests->addChild(createDecorationGroupTests(testCtx));
7869 graphicsTests->addChild(createFRemTests(testCtx));
7870 graphicsTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
7871 graphicsTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
7874 de::MovePtr<tcu::TestCaseGroup> graphicsAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
7876 graphicsAndroidTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7877 graphicsAndroidTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7879 graphicsTests->addChild(graphicsAndroidTests.release());
7882 graphicsTests->addChild(create16BitStorageGraphicsGroup(testCtx));
7883 graphicsTests->addChild(createUboMatrixPaddingGraphicsGroup(testCtx));
7884 graphicsTests->addChild(createConditionalBranchGraphicsGroup(testCtx));
7885 graphicsTests->addChild(createVariablePointersGraphicsGroup(testCtx));
7887 instructionTests->addChild(computeTests.release());
7888 instructionTests->addChild(graphicsTests.release());
7890 return instructionTests.release();