1 /*-------------------------------------------------------------------------
2 * Vulkan Conformance Tests
3 * ------------------------
5 * Copyright (c) 2015 Google Inc.
6 * Copyright (c) 2016 The Khronos Group Inc.
8 * Licensed under the Apache License, Version 2.0 (the "License");
9 * you may not use this file except in compliance with the License.
10 * You may obtain a copy of the License at
12 * http://www.apache.org/licenses/LICENSE-2.0
14 * Unless required by applicable law or agreed to in writing, software
15 * distributed under the License is distributed on an "AS IS" BASIS,
16 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17 * See the License for the specific language governing permissions and
18 * limitations under the License.
22 * \brief SPIR-V Assembly Tests for Instructions (special opcode/operand)
23 *//*--------------------------------------------------------------------*/
25 #include "vktSpvAsmInstructionTests.hpp"
27 #include "tcuCommandLine.hpp"
28 #include "tcuFormatUtil.hpp"
29 #include "tcuFloat.hpp"
30 #include "tcuRGBA.hpp"
31 #include "tcuStringTemplate.hpp"
32 #include "tcuTestLog.hpp"
33 #include "tcuVectorUtil.hpp"
34 #include "tcuInterval.hpp"
37 #include "vkDeviceUtil.hpp"
38 #include "vkMemUtil.hpp"
39 #include "vkPlatform.hpp"
40 #include "vkPrograms.hpp"
41 #include "vkQueryUtil.hpp"
43 #include "vkRefUtil.hpp"
44 #include "vkStrUtil.hpp"
45 #include "vkTypeUtil.hpp"
47 #include "deStringUtil.hpp"
48 #include "deUniquePtr.hpp"
50 #include "tcuStringTemplate.hpp"
52 #include "vktSpvAsm16bitStorageTests.hpp"
53 #include "vktSpvAsmUboMatrixPaddingTests.hpp"
54 #include "vktSpvAsmConditionalBranchTests.hpp"
55 #include "vktSpvAsmIndexingTests.hpp"
56 #include "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 vector<deUint8> input1Bytes;
256 vector<deUint8> input2Bytes;
257 vector<deUint8> expectedBytes;
259 inputs[0]->getBytes(input1Bytes);
260 inputs[1]->getBytes(input2Bytes);
261 expectedOutputs[0]->getBytes(expectedBytes);
263 const deInt32* const expectedOutputAsInt = reinterpret_cast<const deInt32* const>(&expectedBytes.front());
264 const deInt32* const outputAsInt = static_cast<const deInt32* const>(outputAllocs[0]->getHostPtr());
265 const float* const input1AsFloat = reinterpret_cast<const float* const>(&input1Bytes.front());
266 const float* const input2AsFloat = reinterpret_cast<const float* const>(&input2Bytes.front());
267 bool returnValue = true;
269 for (size_t idx = 0; idx < expectedBytes.size() / sizeof(deInt32); ++idx)
271 if (outputAsInt[idx] != expectedOutputAsInt[idx])
273 log << TestLog::Message << "ERROR: Sub-case failed. inputs: " << input1AsFloat[idx] << "," << input2AsFloat[idx] << " output: " << outputAsInt[idx]<< " expected output: " << expectedOutputAsInt[idx] << TestLog::EndMessage;
280 typedef VkBool32 (*compareFuncType) (float, float);
286 compareFuncType compareFunc;
288 OpFUnordCase (const char* _name, const char* _opCode, compareFuncType _compareFunc)
291 , compareFunc (_compareFunc) {}
294 #define ADD_OPFUNORD_CASE(NAME, OPCODE, OPERATOR) \
296 struct compare_##NAME { static VkBool32 compare(float x, float y) { return (x OPERATOR y) ? VK_TRUE : VK_FALSE; } }; \
297 cases.push_back(OpFUnordCase(#NAME, OPCODE, compare_##NAME::compare)); \
298 } while (deGetFalse())
300 tcu::TestCaseGroup* createOpFUnordGroup (tcu::TestContext& testCtx)
302 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfunord", "Test the OpFUnord* opcodes"));
303 de::Random rnd (deStringHash(group->getName()));
304 const int numElements = 100;
305 vector<OpFUnordCase> cases;
307 const StringTemplate shaderTemplate (
309 string(getComputeAsmShaderPreamble()) +
311 "OpSource GLSL 430\n"
312 "OpName %main \"main\"\n"
313 "OpName %id \"gl_GlobalInvocationID\"\n"
315 "OpDecorate %id BuiltIn GlobalInvocationId\n"
317 "OpDecorate %buf BufferBlock\n"
318 "OpDecorate %buf2 BufferBlock\n"
319 "OpDecorate %indata1 DescriptorSet 0\n"
320 "OpDecorate %indata1 Binding 0\n"
321 "OpDecorate %indata2 DescriptorSet 0\n"
322 "OpDecorate %indata2 Binding 1\n"
323 "OpDecorate %outdata DescriptorSet 0\n"
324 "OpDecorate %outdata Binding 2\n"
325 "OpDecorate %f32arr ArrayStride 4\n"
326 "OpDecorate %i32arr ArrayStride 4\n"
327 "OpMemberDecorate %buf 0 Offset 0\n"
328 "OpMemberDecorate %buf2 0 Offset 0\n"
330 + string(getComputeAsmCommonTypes()) +
332 "%buf = OpTypeStruct %f32arr\n"
333 "%bufptr = OpTypePointer Uniform %buf\n"
334 "%indata1 = OpVariable %bufptr Uniform\n"
335 "%indata2 = OpVariable %bufptr Uniform\n"
337 "%buf2 = OpTypeStruct %i32arr\n"
338 "%buf2ptr = OpTypePointer Uniform %buf2\n"
339 "%outdata = OpVariable %buf2ptr Uniform\n"
341 "%id = OpVariable %uvec3ptr Input\n"
342 "%zero = OpConstant %i32 0\n"
343 "%consti1 = OpConstant %i32 1\n"
344 "%constf1 = OpConstant %f32 1.0\n"
346 "%main = OpFunction %void None %voidf\n"
348 "%idval = OpLoad %uvec3 %id\n"
349 "%x = OpCompositeExtract %u32 %idval 0\n"
351 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
352 "%inval1 = OpLoad %f32 %inloc1\n"
353 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
354 "%inval2 = OpLoad %f32 %inloc2\n"
355 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
357 "%result = ${OPCODE} %bool %inval1 %inval2\n"
358 "%int_res = OpSelect %i32 %result %consti1 %zero\n"
359 " OpStore %outloc %int_res\n"
364 ADD_OPFUNORD_CASE(equal, "OpFUnordEqual", ==);
365 ADD_OPFUNORD_CASE(less, "OpFUnordLessThan", <);
366 ADD_OPFUNORD_CASE(lessequal, "OpFUnordLessThanEqual", <=);
367 ADD_OPFUNORD_CASE(greater, "OpFUnordGreaterThan", >);
368 ADD_OPFUNORD_CASE(greaterequal, "OpFUnordGreaterThanEqual", >=);
369 ADD_OPFUNORD_CASE(notequal, "OpFUnordNotEqual", !=);
371 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
373 map<string, string> specializations;
374 ComputeShaderSpec spec;
375 const float NaN = std::numeric_limits<float>::quiet_NaN();
376 vector<float> inputFloats1 (numElements, 0);
377 vector<float> inputFloats2 (numElements, 0);
378 vector<deInt32> expectedInts (numElements, 0);
380 specializations["OPCODE"] = cases[caseNdx].opCode;
381 spec.assembly = shaderTemplate.specialize(specializations);
383 fillRandomScalars(rnd, 1.f, 100.f, &inputFloats1[0], numElements);
384 for (size_t ndx = 0; ndx < numElements; ++ndx)
388 case 0: inputFloats2[ndx] = inputFloats1[ndx] + 1.0f; break;
389 case 1: inputFloats2[ndx] = inputFloats1[ndx] - 1.0f; break;
390 case 2: inputFloats2[ndx] = inputFloats1[ndx]; break;
391 case 3: inputFloats2[ndx] = NaN; break;
392 case 4: inputFloats2[ndx] = inputFloats1[ndx]; inputFloats1[ndx] = NaN; break;
393 case 5: inputFloats2[ndx] = NaN; inputFloats1[ndx] = NaN; break;
395 expectedInts[ndx] = tcu::Float32(inputFloats1[ndx]).isNaN() || tcu::Float32(inputFloats2[ndx]).isNaN() || cases[caseNdx].compareFunc(inputFloats1[ndx], inputFloats2[ndx]);
398 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
399 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
400 spec.outputs.push_back(BufferSp(new Int32Buffer(expectedInts)));
401 spec.numWorkGroups = IVec3(numElements, 1, 1);
402 spec.verifyIO = &compareFUnord;
403 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
406 return group.release();
412 const char* assembly;
413 OpAtomicType opAtomic;
414 deInt32 numOutputElements;
416 OpAtomicCase (const char* _name, const char* _assembly, OpAtomicType _opAtomic, deInt32 _numOutputElements)
418 , assembly (_assembly)
419 , opAtomic (_opAtomic)
420 , numOutputElements (_numOutputElements) {}
423 tcu::TestCaseGroup* createOpAtomicGroup (tcu::TestContext& testCtx, bool useStorageBuffer)
425 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx,
426 useStorageBuffer ? "opatomic_storage_buffer" : "opatomic",
427 "Test the OpAtomic* opcodes"));
428 const int numElements = 65535;
429 vector<OpAtomicCase> cases;
431 const StringTemplate shaderTemplate (
433 string("OpCapability Shader\n") +
434 (useStorageBuffer ? "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" : "") +
435 "OpMemoryModel Logical GLSL450\n"
436 "OpEntryPoint GLCompute %main \"main\" %id\n"
437 "OpExecutionMode %main LocalSize 1 1 1\n" +
439 "OpSource GLSL 430\n"
440 "OpName %main \"main\"\n"
441 "OpName %id \"gl_GlobalInvocationID\"\n"
443 "OpDecorate %id BuiltIn GlobalInvocationId\n"
445 "OpDecorate %buf ${BLOCK_DECORATION}\n"
446 "OpDecorate %indata DescriptorSet 0\n"
447 "OpDecorate %indata Binding 0\n"
448 "OpDecorate %i32arr ArrayStride 4\n"
449 "OpMemberDecorate %buf 0 Offset 0\n"
451 "OpDecorate %sumbuf ${BLOCK_DECORATION}\n"
452 "OpDecorate %sum DescriptorSet 0\n"
453 "OpDecorate %sum Binding 1\n"
454 "OpMemberDecorate %sumbuf 0 Coherent\n"
455 "OpMemberDecorate %sumbuf 0 Offset 0\n"
457 + getComputeAsmCommonTypes("${BLOCK_POINTER_TYPE}") +
459 "%buf = OpTypeStruct %i32arr\n"
460 "%bufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %buf\n"
461 "%indata = OpVariable %bufptr ${BLOCK_POINTER_TYPE}\n"
463 "%sumbuf = OpTypeStruct %i32arr\n"
464 "%sumbufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %sumbuf\n"
465 "%sum = OpVariable %sumbufptr ${BLOCK_POINTER_TYPE}\n"
467 "%id = OpVariable %uvec3ptr Input\n"
468 "%minusone = OpConstant %i32 -1\n"
469 "%zero = OpConstant %i32 0\n"
470 "%one = OpConstant %u32 1\n"
471 "%two = OpConstant %i32 2\n"
473 "%main = OpFunction %void None %voidf\n"
475 "%idval = OpLoad %uvec3 %id\n"
476 "%x = OpCompositeExtract %u32 %idval 0\n"
478 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
479 "%inval = OpLoad %i32 %inloc\n"
481 "%outloc = OpAccessChain %i32ptr %sum %zero ${INDEX}\n"
487 #define ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, NUM_OUTPUT_ELEMENTS) \
489 DE_STATIC_ASSERT((NUM_OUTPUT_ELEMENTS) == 1 || (NUM_OUTPUT_ELEMENTS) == numElements); \
490 cases.push_back(OpAtomicCase(#NAME, ASSEMBLY, OPATOMIC, NUM_OUTPUT_ELEMENTS)); \
491 } while (deGetFalse())
492 #define ADD_OPATOMIC_CASE_1(NAME, ASSEMBLY, OPATOMIC) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, 1)
493 #define ADD_OPATOMIC_CASE_N(NAME, ASSEMBLY, OPATOMIC) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, OPATOMIC, numElements)
495 ADD_OPATOMIC_CASE_1(iadd, "%unused = OpAtomicIAdd %i32 %outloc %one %zero %inval\n", OPATOMIC_IADD );
496 ADD_OPATOMIC_CASE_1(isub, "%unused = OpAtomicISub %i32 %outloc %one %zero %inval\n", OPATOMIC_ISUB );
497 ADD_OPATOMIC_CASE_1(iinc, "%unused = OpAtomicIIncrement %i32 %outloc %one %zero\n", OPATOMIC_IINC );
498 ADD_OPATOMIC_CASE_1(idec, "%unused = OpAtomicIDecrement %i32 %outloc %one %zero\n", OPATOMIC_IDEC );
499 ADD_OPATOMIC_CASE_N(load, "%inval2 = OpAtomicLoad %i32 %inloc %zero %zero\n"
500 " OpStore %outloc %inval2\n", OPATOMIC_LOAD );
501 ADD_OPATOMIC_CASE_N(store, " OpAtomicStore %outloc %zero %zero %inval\n", OPATOMIC_STORE );
502 ADD_OPATOMIC_CASE_N(compex, "%even = OpSMod %i32 %inval %two\n"
503 " OpStore %outloc %even\n"
504 "%unused = OpAtomicCompareExchange %i32 %outloc %one %zero %zero %minusone %zero\n", OPATOMIC_COMPEX );
506 #undef ADD_OPATOMIC_CASE
507 #undef ADD_OPATOMIC_CASE_1
508 #undef ADD_OPATOMIC_CASE_N
510 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
512 map<string, string> specializations;
513 ComputeShaderSpec spec;
514 vector<deInt32> inputInts (numElements, 0);
515 vector<deInt32> expected (cases[caseNdx].numOutputElements, -1);
517 specializations["INDEX"] = (cases[caseNdx].numOutputElements == 1) ? "%zero" : "%x";
518 specializations["INSTRUCTION"] = cases[caseNdx].assembly;
519 specializations["BLOCK_DECORATION"] = useStorageBuffer ? "Block" : "BufferBlock";
520 specializations["BLOCK_POINTER_TYPE"] = useStorageBuffer ? "StorageBuffer" : "Uniform";
521 spec.assembly = shaderTemplate.specialize(specializations);
523 if (useStorageBuffer)
524 spec.extensions.push_back("VK_KHR_storage_buffer_storage_class");
526 spec.inputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_INPUT)));
527 spec.outputs.push_back(BufferSp(new OpAtomicBuffer(numElements, cases[caseNdx].numOutputElements, cases[caseNdx].opAtomic, BUFFERTYPE_EXPECTED)));
528 spec.numWorkGroups = IVec3(numElements, 1, 1);
529 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
532 return group.release();
535 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
537 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
538 ComputeShaderSpec spec;
539 de::Random rnd (deStringHash(group->getName()));
540 const int numElements = 100;
541 vector<float> positiveFloats (numElements, 0);
542 vector<float> negativeFloats (numElements, 0);
544 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
546 for (size_t ndx = 0; ndx < numElements; ++ndx)
547 negativeFloats[ndx] = -positiveFloats[ndx];
550 string(getComputeAsmShaderPreamble()) +
552 "%fname1 = OpString \"negateInputs.comp\"\n"
553 "%fname2 = OpString \"negateInputs\"\n"
555 "OpSource GLSL 430\n"
556 "OpName %main \"main\"\n"
557 "OpName %id \"gl_GlobalInvocationID\"\n"
559 "OpDecorate %id BuiltIn GlobalInvocationId\n"
561 + string(getComputeAsmInputOutputBufferTraits()) +
563 "OpLine %fname1 0 0\n" // At the earliest possible position
565 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
567 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
568 "OpLine %fname2 1 0\n" // Different filenames
569 "OpLine %fname1 1000 100000\n"
571 "%id = OpVariable %uvec3ptr Input\n"
572 "%zero = OpConstant %i32 0\n"
574 "OpLine %fname1 1 1\n" // Before a function
576 "%main = OpFunction %void None %voidf\n"
579 "OpLine %fname1 1 1\n" // In a function
581 "%idval = OpLoad %uvec3 %id\n"
582 "%x = OpCompositeExtract %u32 %idval 0\n"
583 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
584 "%inval = OpLoad %f32 %inloc\n"
585 "%neg = OpFNegate %f32 %inval\n"
586 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
587 " OpStore %outloc %neg\n"
590 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
591 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
592 spec.numWorkGroups = IVec3(numElements, 1, 1);
594 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
596 return group.release();
599 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
601 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
602 ComputeShaderSpec spec;
603 de::Random rnd (deStringHash(group->getName()));
604 const int numElements = 100;
605 vector<float> positiveFloats (numElements, 0);
606 vector<float> negativeFloats (numElements, 0);
608 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
610 for (size_t ndx = 0; ndx < numElements; ++ndx)
611 negativeFloats[ndx] = -positiveFloats[ndx];
614 string(getComputeAsmShaderPreamble()) +
616 "%fname = OpString \"negateInputs.comp\"\n"
618 "OpSource GLSL 430\n"
619 "OpName %main \"main\"\n"
620 "OpName %id \"gl_GlobalInvocationID\"\n"
622 "OpDecorate %id BuiltIn GlobalInvocationId\n"
624 + string(getComputeAsmInputOutputBufferTraits()) +
626 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
628 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
630 "OpLine %fname 0 1\n"
631 "OpNoLine\n" // Immediately following a preceding OpLine
633 "OpLine %fname 1000 1\n"
635 "%id = OpVariable %uvec3ptr Input\n"
636 "%zero = OpConstant %i32 0\n"
638 "OpNoLine\n" // Contents after the previous OpLine
640 "%main = OpFunction %void None %voidf\n"
642 "%idval = OpLoad %uvec3 %id\n"
643 "%x = OpCompositeExtract %u32 %idval 0\n"
645 "OpNoLine\n" // Multiple OpNoLine
649 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
650 "%inval = OpLoad %f32 %inloc\n"
651 "%neg = OpFNegate %f32 %inval\n"
652 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
653 " OpStore %outloc %neg\n"
656 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
657 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
658 spec.numWorkGroups = IVec3(numElements, 1, 1);
660 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
662 return group.release();
665 // Compare instruction for the contraction compute case.
666 // Returns true if the output is what is expected from the test case.
667 bool compareNoContractCase(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
669 if (outputAllocs.size() != 1)
672 // Only size is needed because we are not comparing the exact values.
673 size_t byteSize = expectedOutputs[0]->getByteSize();
675 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
677 for(size_t i = 0; i < byteSize / 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 vector<deUint8> expectedBytes;
783 expectedOutputs[0]->getBytes(expectedBytes);
785 const float* expectedOutputAsFloat = reinterpret_cast<const float*>(&expectedBytes.front());
786 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
788 for (size_t idx = 0; idx < expectedBytes.size() / sizeof(float); ++idx)
790 const float f0 = expectedOutputAsFloat[idx];
791 const float f1 = outputAsFloat[idx];
792 // \todo relative error needs to be fairly high because FRem may be implemented as
793 // (roughly) frac(a/b)*b, so LSB errors can be magnified. But this should be fine for now.
794 if (deFloatAbs((f1 - f0) / f0) > 0.02)
801 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
803 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
804 ComputeShaderSpec spec;
805 de::Random rnd (deStringHash(group->getName()));
806 const int numElements = 200;
807 vector<float> inputFloats1 (numElements, 0);
808 vector<float> inputFloats2 (numElements, 0);
809 vector<float> outputFloats (numElements, 0);
811 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
812 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
814 for (size_t ndx = 0; ndx < numElements; ++ndx)
816 // Guard against divisors near zero.
817 if (std::fabs(inputFloats2[ndx]) < 1e-3)
818 inputFloats2[ndx] = 8.f;
820 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
821 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
825 string(getComputeAsmShaderPreamble()) +
827 "OpName %main \"main\"\n"
828 "OpName %id \"gl_GlobalInvocationID\"\n"
830 "OpDecorate %id BuiltIn GlobalInvocationId\n"
832 "OpDecorate %buf BufferBlock\n"
833 "OpDecorate %indata1 DescriptorSet 0\n"
834 "OpDecorate %indata1 Binding 0\n"
835 "OpDecorate %indata2 DescriptorSet 0\n"
836 "OpDecorate %indata2 Binding 1\n"
837 "OpDecorate %outdata DescriptorSet 0\n"
838 "OpDecorate %outdata Binding 2\n"
839 "OpDecorate %f32arr ArrayStride 4\n"
840 "OpMemberDecorate %buf 0 Offset 0\n"
842 + string(getComputeAsmCommonTypes()) +
844 "%buf = OpTypeStruct %f32arr\n"
845 "%bufptr = OpTypePointer Uniform %buf\n"
846 "%indata1 = OpVariable %bufptr Uniform\n"
847 "%indata2 = OpVariable %bufptr Uniform\n"
848 "%outdata = OpVariable %bufptr Uniform\n"
850 "%id = OpVariable %uvec3ptr Input\n"
851 "%zero = OpConstant %i32 0\n"
853 "%main = OpFunction %void None %voidf\n"
855 "%idval = OpLoad %uvec3 %id\n"
856 "%x = OpCompositeExtract %u32 %idval 0\n"
857 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
858 "%inval1 = OpLoad %f32 %inloc1\n"
859 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
860 "%inval2 = OpLoad %f32 %inloc2\n"
861 "%rem = OpFRem %f32 %inval1 %inval2\n"
862 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
863 " OpStore %outloc %rem\n"
867 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
868 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
869 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
870 spec.numWorkGroups = IVec3(numElements, 1, 1);
871 spec.verifyIO = &compareFRem;
873 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
875 return group.release();
878 bool compareNMin (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
880 if (outputAllocs.size() != 1)
883 const BufferSp& expectedOutput (expectedOutputs[0]);
884 std::vector<deUint8> data;
885 expectedOutput->getBytes(data);
887 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
888 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
890 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float); ++idx)
892 const float f0 = expectedOutputAsFloat[idx];
893 const float f1 = outputAsFloat[idx];
895 // For NMin, we accept NaN as output if both inputs were NaN.
896 // Otherwise the NaN is the wrong choise, as on architectures that
897 // do not handle NaN, those are huge values.
898 if (!(tcu::Float32(f1).isNaN() && tcu::Float32(f0).isNaN()) && deFloatAbs(f1 - f0) > 0.00001f)
905 tcu::TestCaseGroup* createOpNMinGroup (tcu::TestContext& testCtx)
907 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnmin", "Test the OpNMin instruction"));
908 ComputeShaderSpec spec;
909 de::Random rnd (deStringHash(group->getName()));
910 const int numElements = 200;
911 vector<float> inputFloats1 (numElements, 0);
912 vector<float> inputFloats2 (numElements, 0);
913 vector<float> outputFloats (numElements, 0);
915 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
916 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
918 // Make the first case a full-NAN case.
919 inputFloats1[0] = TCU_NAN;
920 inputFloats2[0] = TCU_NAN;
922 for (size_t ndx = 0; ndx < numElements; ++ndx)
924 // By default, pick the smallest
925 outputFloats[ndx] = std::min(inputFloats1[ndx], inputFloats2[ndx]);
927 // Make half of the cases NaN cases
930 // Alternate between the NaN operand
933 outputFloats[ndx] = inputFloats2[ndx];
934 inputFloats1[ndx] = TCU_NAN;
938 outputFloats[ndx] = inputFloats1[ndx];
939 inputFloats2[ndx] = TCU_NAN;
945 "OpCapability Shader\n"
946 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
947 "OpMemoryModel Logical GLSL450\n"
948 "OpEntryPoint GLCompute %main \"main\" %id\n"
949 "OpExecutionMode %main LocalSize 1 1 1\n"
951 "OpName %main \"main\"\n"
952 "OpName %id \"gl_GlobalInvocationID\"\n"
954 "OpDecorate %id BuiltIn GlobalInvocationId\n"
956 "OpDecorate %buf BufferBlock\n"
957 "OpDecorate %indata1 DescriptorSet 0\n"
958 "OpDecorate %indata1 Binding 0\n"
959 "OpDecorate %indata2 DescriptorSet 0\n"
960 "OpDecorate %indata2 Binding 1\n"
961 "OpDecorate %outdata DescriptorSet 0\n"
962 "OpDecorate %outdata Binding 2\n"
963 "OpDecorate %f32arr ArrayStride 4\n"
964 "OpMemberDecorate %buf 0 Offset 0\n"
966 + string(getComputeAsmCommonTypes()) +
968 "%buf = OpTypeStruct %f32arr\n"
969 "%bufptr = OpTypePointer Uniform %buf\n"
970 "%indata1 = OpVariable %bufptr Uniform\n"
971 "%indata2 = OpVariable %bufptr Uniform\n"
972 "%outdata = OpVariable %bufptr Uniform\n"
974 "%id = OpVariable %uvec3ptr Input\n"
975 "%zero = OpConstant %i32 0\n"
977 "%main = OpFunction %void None %voidf\n"
979 "%idval = OpLoad %uvec3 %id\n"
980 "%x = OpCompositeExtract %u32 %idval 0\n"
981 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
982 "%inval1 = OpLoad %f32 %inloc1\n"
983 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
984 "%inval2 = OpLoad %f32 %inloc2\n"
985 "%rem = OpExtInst %f32 %std450 NMin %inval1 %inval2\n"
986 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
987 " OpStore %outloc %rem\n"
991 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
992 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
993 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
994 spec.numWorkGroups = IVec3(numElements, 1, 1);
995 spec.verifyIO = &compareNMin;
997 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
999 return group.release();
1002 bool compareNMax (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1004 if (outputAllocs.size() != 1)
1007 const BufferSp& expectedOutput = expectedOutputs[0];
1008 std::vector<deUint8> data;
1009 expectedOutput->getBytes(data);
1011 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1012 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1014 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float); ++idx)
1016 const float f0 = expectedOutputAsFloat[idx];
1017 const float f1 = outputAsFloat[idx];
1019 // For NMax, NaN is considered acceptable result, since in
1020 // architectures that do not handle NaNs, those are huge values.
1021 if (!tcu::Float32(f1).isNaN() && deFloatAbs(f1 - f0) > 0.00001f)
1028 tcu::TestCaseGroup* createOpNMaxGroup (tcu::TestContext& testCtx)
1030 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, "opnmax", "Test the OpNMax instruction"));
1031 ComputeShaderSpec spec;
1032 de::Random rnd (deStringHash(group->getName()));
1033 const int numElements = 200;
1034 vector<float> inputFloats1 (numElements, 0);
1035 vector<float> inputFloats2 (numElements, 0);
1036 vector<float> outputFloats (numElements, 0);
1038 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1039 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1041 // Make the first case a full-NAN case.
1042 inputFloats1[0] = TCU_NAN;
1043 inputFloats2[0] = TCU_NAN;
1045 for (size_t ndx = 0; ndx < numElements; ++ndx)
1047 // By default, pick the biggest
1048 outputFloats[ndx] = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1050 // Make half of the cases NaN cases
1053 // Alternate between the NaN operand
1056 outputFloats[ndx] = inputFloats2[ndx];
1057 inputFloats1[ndx] = TCU_NAN;
1061 outputFloats[ndx] = inputFloats1[ndx];
1062 inputFloats2[ndx] = TCU_NAN;
1068 "OpCapability Shader\n"
1069 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1070 "OpMemoryModel Logical GLSL450\n"
1071 "OpEntryPoint GLCompute %main \"main\" %id\n"
1072 "OpExecutionMode %main LocalSize 1 1 1\n"
1074 "OpName %main \"main\"\n"
1075 "OpName %id \"gl_GlobalInvocationID\"\n"
1077 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1079 "OpDecorate %buf BufferBlock\n"
1080 "OpDecorate %indata1 DescriptorSet 0\n"
1081 "OpDecorate %indata1 Binding 0\n"
1082 "OpDecorate %indata2 DescriptorSet 0\n"
1083 "OpDecorate %indata2 Binding 1\n"
1084 "OpDecorate %outdata DescriptorSet 0\n"
1085 "OpDecorate %outdata Binding 2\n"
1086 "OpDecorate %f32arr ArrayStride 4\n"
1087 "OpMemberDecorate %buf 0 Offset 0\n"
1089 + string(getComputeAsmCommonTypes()) +
1091 "%buf = OpTypeStruct %f32arr\n"
1092 "%bufptr = OpTypePointer Uniform %buf\n"
1093 "%indata1 = OpVariable %bufptr Uniform\n"
1094 "%indata2 = OpVariable %bufptr Uniform\n"
1095 "%outdata = OpVariable %bufptr Uniform\n"
1097 "%id = OpVariable %uvec3ptr Input\n"
1098 "%zero = OpConstant %i32 0\n"
1100 "%main = OpFunction %void None %voidf\n"
1101 "%label = OpLabel\n"
1102 "%idval = OpLoad %uvec3 %id\n"
1103 "%x = OpCompositeExtract %u32 %idval 0\n"
1104 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1105 "%inval1 = OpLoad %f32 %inloc1\n"
1106 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1107 "%inval2 = OpLoad %f32 %inloc2\n"
1108 "%rem = OpExtInst %f32 %std450 NMax %inval1 %inval2\n"
1109 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1110 " OpStore %outloc %rem\n"
1114 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1115 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1116 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1117 spec.numWorkGroups = IVec3(numElements, 1, 1);
1118 spec.verifyIO = &compareNMax;
1120 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1122 return group.release();
1125 bool compareNClamp (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1127 if (outputAllocs.size() != 1)
1130 const BufferSp& expectedOutput = expectedOutputs[0];
1131 std::vector<deUint8> data;
1132 expectedOutput->getBytes(data);
1134 const float* const expectedOutputAsFloat = reinterpret_cast<const float*>(&data.front());
1135 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
1137 for (size_t idx = 0; idx < expectedOutput->getByteSize() / sizeof(float) / 2; ++idx)
1139 const float e0 = expectedOutputAsFloat[idx * 2];
1140 const float e1 = expectedOutputAsFloat[idx * 2 + 1];
1141 const float res = outputAsFloat[idx];
1143 // For NClamp, we have two possible outcomes based on
1144 // whether NaNs are handled or not.
1145 // If either min or max value is NaN, the result is undefined,
1146 // so this test doesn't stress those. If the clamped value is
1147 // NaN, and NaNs are handled, the result is min; if NaNs are not
1148 // handled, they are big values that result in max.
1149 // If all three parameters are NaN, the result should be NaN.
1150 if (!((tcu::Float32(e0).isNaN() && tcu::Float32(res).isNaN()) ||
1151 (deFloatAbs(e0 - res) < 0.00001f) ||
1152 (deFloatAbs(e1 - res) < 0.00001f)))
1159 tcu::TestCaseGroup* createOpNClampGroup (tcu::TestContext& testCtx)
1161 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnclamp", "Test the OpNClamp instruction"));
1162 ComputeShaderSpec spec;
1163 de::Random rnd (deStringHash(group->getName()));
1164 const int numElements = 200;
1165 vector<float> inputFloats1 (numElements, 0);
1166 vector<float> inputFloats2 (numElements, 0);
1167 vector<float> inputFloats3 (numElements, 0);
1168 vector<float> outputFloats (numElements * 2, 0);
1170 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1171 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1172 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats3[0], numElements);
1174 for (size_t ndx = 0; ndx < numElements; ++ndx)
1176 // Results are only defined if max value is bigger than min value.
1177 if (inputFloats2[ndx] > inputFloats3[ndx])
1179 float t = inputFloats2[ndx];
1180 inputFloats2[ndx] = inputFloats3[ndx];
1181 inputFloats3[ndx] = t;
1184 // By default, do the clamp, setting both possible answers
1185 float defaultRes = std::min(std::max(inputFloats1[ndx], inputFloats2[ndx]), inputFloats3[ndx]);
1187 float maxResA = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1188 float maxResB = maxResA;
1190 // Alternate between the NaN cases
1193 inputFloats1[ndx] = TCU_NAN;
1194 // If NaN is handled, the result should be same as the clamp minimum.
1195 // If NaN is not handled, the result should clamp to the clamp maximum.
1196 maxResA = inputFloats2[ndx];
1197 maxResB = inputFloats3[ndx];
1201 // Not a NaN case - only one legal result.
1202 maxResA = defaultRes;
1203 maxResB = defaultRes;
1206 outputFloats[ndx * 2] = maxResA;
1207 outputFloats[ndx * 2 + 1] = maxResB;
1210 // Make the first case a full-NAN case.
1211 inputFloats1[0] = TCU_NAN;
1212 inputFloats2[0] = TCU_NAN;
1213 inputFloats3[0] = TCU_NAN;
1214 outputFloats[0] = TCU_NAN;
1215 outputFloats[1] = TCU_NAN;
1218 "OpCapability Shader\n"
1219 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1220 "OpMemoryModel Logical GLSL450\n"
1221 "OpEntryPoint GLCompute %main \"main\" %id\n"
1222 "OpExecutionMode %main LocalSize 1 1 1\n"
1224 "OpName %main \"main\"\n"
1225 "OpName %id \"gl_GlobalInvocationID\"\n"
1227 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1229 "OpDecorate %buf BufferBlock\n"
1230 "OpDecorate %indata1 DescriptorSet 0\n"
1231 "OpDecorate %indata1 Binding 0\n"
1232 "OpDecorate %indata2 DescriptorSet 0\n"
1233 "OpDecorate %indata2 Binding 1\n"
1234 "OpDecorate %indata3 DescriptorSet 0\n"
1235 "OpDecorate %indata3 Binding 2\n"
1236 "OpDecorate %outdata DescriptorSet 0\n"
1237 "OpDecorate %outdata Binding 3\n"
1238 "OpDecorate %f32arr ArrayStride 4\n"
1239 "OpMemberDecorate %buf 0 Offset 0\n"
1241 + string(getComputeAsmCommonTypes()) +
1243 "%buf = OpTypeStruct %f32arr\n"
1244 "%bufptr = OpTypePointer Uniform %buf\n"
1245 "%indata1 = OpVariable %bufptr Uniform\n"
1246 "%indata2 = OpVariable %bufptr Uniform\n"
1247 "%indata3 = OpVariable %bufptr Uniform\n"
1248 "%outdata = OpVariable %bufptr Uniform\n"
1250 "%id = OpVariable %uvec3ptr Input\n"
1251 "%zero = OpConstant %i32 0\n"
1253 "%main = OpFunction %void None %voidf\n"
1254 "%label = OpLabel\n"
1255 "%idval = OpLoad %uvec3 %id\n"
1256 "%x = OpCompositeExtract %u32 %idval 0\n"
1257 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1258 "%inval1 = OpLoad %f32 %inloc1\n"
1259 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1260 "%inval2 = OpLoad %f32 %inloc2\n"
1261 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1262 "%inval3 = OpLoad %f32 %inloc3\n"
1263 "%rem = OpExtInst %f32 %std450 NClamp %inval1 %inval2 %inval3\n"
1264 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1265 " OpStore %outloc %rem\n"
1269 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1270 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1271 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1272 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1273 spec.numWorkGroups = IVec3(numElements, 1, 1);
1274 spec.verifyIO = &compareNClamp;
1276 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1278 return group.release();
1281 tcu::TestCaseGroup* createOpSRemComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1283 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem", "Test the OpSRem instruction"));
1284 de::Random rnd (deStringHash(group->getName()));
1285 const int numElements = 200;
1287 const struct CaseParams
1290 const char* failMessage; // customized status message
1291 qpTestResult failResult; // override status on failure
1292 int op1Min, op1Max; // operand ranges
1296 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1297 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1299 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1301 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1303 const CaseParams& params = cases[caseNdx];
1304 ComputeShaderSpec spec;
1305 vector<deInt32> inputInts1 (numElements, 0);
1306 vector<deInt32> inputInts2 (numElements, 0);
1307 vector<deInt32> outputInts (numElements, 0);
1309 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1310 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1312 for (int ndx = 0; ndx < numElements; ++ndx)
1314 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1315 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1319 string(getComputeAsmShaderPreamble()) +
1321 "OpName %main \"main\"\n"
1322 "OpName %id \"gl_GlobalInvocationID\"\n"
1324 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1326 "OpDecorate %buf BufferBlock\n"
1327 "OpDecorate %indata1 DescriptorSet 0\n"
1328 "OpDecorate %indata1 Binding 0\n"
1329 "OpDecorate %indata2 DescriptorSet 0\n"
1330 "OpDecorate %indata2 Binding 1\n"
1331 "OpDecorate %outdata DescriptorSet 0\n"
1332 "OpDecorate %outdata Binding 2\n"
1333 "OpDecorate %i32arr ArrayStride 4\n"
1334 "OpMemberDecorate %buf 0 Offset 0\n"
1336 + string(getComputeAsmCommonTypes()) +
1338 "%buf = OpTypeStruct %i32arr\n"
1339 "%bufptr = OpTypePointer Uniform %buf\n"
1340 "%indata1 = OpVariable %bufptr Uniform\n"
1341 "%indata2 = OpVariable %bufptr Uniform\n"
1342 "%outdata = OpVariable %bufptr Uniform\n"
1344 "%id = OpVariable %uvec3ptr Input\n"
1345 "%zero = OpConstant %i32 0\n"
1347 "%main = OpFunction %void None %voidf\n"
1348 "%label = OpLabel\n"
1349 "%idval = OpLoad %uvec3 %id\n"
1350 "%x = OpCompositeExtract %u32 %idval 0\n"
1351 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1352 "%inval1 = OpLoad %i32 %inloc1\n"
1353 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1354 "%inval2 = OpLoad %i32 %inloc2\n"
1355 "%rem = OpSRem %i32 %inval1 %inval2\n"
1356 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1357 " OpStore %outloc %rem\n"
1361 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1362 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1363 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1364 spec.numWorkGroups = IVec3(numElements, 1, 1);
1365 spec.failResult = params.failResult;
1366 spec.failMessage = params.failMessage;
1368 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1371 return group.release();
1374 tcu::TestCaseGroup* createOpSRemComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1376 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem64", "Test the 64-bit OpSRem instruction"));
1377 de::Random rnd (deStringHash(group->getName()));
1378 const int numElements = 200;
1380 const struct CaseParams
1383 const char* failMessage; // customized status message
1384 qpTestResult failResult; // override status on failure
1388 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1389 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1391 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1393 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1395 const CaseParams& params = cases[caseNdx];
1396 ComputeShaderSpec spec;
1397 vector<deInt64> inputInts1 (numElements, 0);
1398 vector<deInt64> inputInts2 (numElements, 0);
1399 vector<deInt64> outputInts (numElements, 0);
1401 if (params.positive)
1403 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1404 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1408 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1409 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1412 for (int ndx = 0; ndx < numElements; ++ndx)
1414 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1415 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1419 "OpCapability Int64\n"
1421 + string(getComputeAsmShaderPreamble()) +
1423 "OpName %main \"main\"\n"
1424 "OpName %id \"gl_GlobalInvocationID\"\n"
1426 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1428 "OpDecorate %buf BufferBlock\n"
1429 "OpDecorate %indata1 DescriptorSet 0\n"
1430 "OpDecorate %indata1 Binding 0\n"
1431 "OpDecorate %indata2 DescriptorSet 0\n"
1432 "OpDecorate %indata2 Binding 1\n"
1433 "OpDecorate %outdata DescriptorSet 0\n"
1434 "OpDecorate %outdata Binding 2\n"
1435 "OpDecorate %i64arr ArrayStride 8\n"
1436 "OpMemberDecorate %buf 0 Offset 0\n"
1438 + string(getComputeAsmCommonTypes())
1439 + string(getComputeAsmCommonInt64Types()) +
1441 "%buf = OpTypeStruct %i64arr\n"
1442 "%bufptr = OpTypePointer Uniform %buf\n"
1443 "%indata1 = OpVariable %bufptr Uniform\n"
1444 "%indata2 = OpVariable %bufptr Uniform\n"
1445 "%outdata = OpVariable %bufptr Uniform\n"
1447 "%id = OpVariable %uvec3ptr Input\n"
1448 "%zero = OpConstant %i64 0\n"
1450 "%main = OpFunction %void None %voidf\n"
1451 "%label = OpLabel\n"
1452 "%idval = OpLoad %uvec3 %id\n"
1453 "%x = OpCompositeExtract %u32 %idval 0\n"
1454 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1455 "%inval1 = OpLoad %i64 %inloc1\n"
1456 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1457 "%inval2 = OpLoad %i64 %inloc2\n"
1458 "%rem = OpSRem %i64 %inval1 %inval2\n"
1459 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1460 " OpStore %outloc %rem\n"
1464 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1465 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1466 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1467 spec.numWorkGroups = IVec3(numElements, 1, 1);
1468 spec.failResult = params.failResult;
1469 spec.failMessage = params.failMessage;
1471 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1474 return group.release();
1477 tcu::TestCaseGroup* createOpSModComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1479 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod", "Test the OpSMod instruction"));
1480 de::Random rnd (deStringHash(group->getName()));
1481 const int numElements = 200;
1483 const struct CaseParams
1486 const char* failMessage; // customized status message
1487 qpTestResult failResult; // override status on failure
1488 int op1Min, op1Max; // operand ranges
1492 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1493 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1495 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1497 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1499 const CaseParams& params = cases[caseNdx];
1501 ComputeShaderSpec spec;
1502 vector<deInt32> inputInts1 (numElements, 0);
1503 vector<deInt32> inputInts2 (numElements, 0);
1504 vector<deInt32> outputInts (numElements, 0);
1506 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1507 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1509 for (int ndx = 0; ndx < numElements; ++ndx)
1511 deInt32 rem = inputInts1[ndx] % inputInts2[ndx];
1514 outputInts[ndx] = 0;
1516 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1518 // They have the same sign
1519 outputInts[ndx] = rem;
1523 // They have opposite sign. The remainder operation takes the
1524 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1525 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1526 // the result has the correct sign and that it is still
1527 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1529 // See also http://mathforum.org/library/drmath/view/52343.html
1530 outputInts[ndx] = rem + inputInts2[ndx];
1535 string(getComputeAsmShaderPreamble()) +
1537 "OpName %main \"main\"\n"
1538 "OpName %id \"gl_GlobalInvocationID\"\n"
1540 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1542 "OpDecorate %buf BufferBlock\n"
1543 "OpDecorate %indata1 DescriptorSet 0\n"
1544 "OpDecorate %indata1 Binding 0\n"
1545 "OpDecorate %indata2 DescriptorSet 0\n"
1546 "OpDecorate %indata2 Binding 1\n"
1547 "OpDecorate %outdata DescriptorSet 0\n"
1548 "OpDecorate %outdata Binding 2\n"
1549 "OpDecorate %i32arr ArrayStride 4\n"
1550 "OpMemberDecorate %buf 0 Offset 0\n"
1552 + string(getComputeAsmCommonTypes()) +
1554 "%buf = OpTypeStruct %i32arr\n"
1555 "%bufptr = OpTypePointer Uniform %buf\n"
1556 "%indata1 = OpVariable %bufptr Uniform\n"
1557 "%indata2 = OpVariable %bufptr Uniform\n"
1558 "%outdata = OpVariable %bufptr Uniform\n"
1560 "%id = OpVariable %uvec3ptr Input\n"
1561 "%zero = OpConstant %i32 0\n"
1563 "%main = OpFunction %void None %voidf\n"
1564 "%label = OpLabel\n"
1565 "%idval = OpLoad %uvec3 %id\n"
1566 "%x = OpCompositeExtract %u32 %idval 0\n"
1567 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1568 "%inval1 = OpLoad %i32 %inloc1\n"
1569 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1570 "%inval2 = OpLoad %i32 %inloc2\n"
1571 "%rem = OpSMod %i32 %inval1 %inval2\n"
1572 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1573 " OpStore %outloc %rem\n"
1577 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1578 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1579 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1580 spec.numWorkGroups = IVec3(numElements, 1, 1);
1581 spec.failResult = params.failResult;
1582 spec.failMessage = params.failMessage;
1584 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1587 return group.release();
1590 tcu::TestCaseGroup* createOpSModComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1592 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod64", "Test the OpSMod instruction"));
1593 de::Random rnd (deStringHash(group->getName()));
1594 const int numElements = 200;
1596 const struct CaseParams
1599 const char* failMessage; // customized status message
1600 qpTestResult failResult; // override status on failure
1604 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1605 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1607 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1609 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1611 const CaseParams& params = cases[caseNdx];
1613 ComputeShaderSpec spec;
1614 vector<deInt64> inputInts1 (numElements, 0);
1615 vector<deInt64> inputInts2 (numElements, 0);
1616 vector<deInt64> outputInts (numElements, 0);
1619 if (params.positive)
1621 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1622 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1626 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1627 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1630 for (int ndx = 0; ndx < numElements; ++ndx)
1632 deInt64 rem = inputInts1[ndx] % inputInts2[ndx];
1635 outputInts[ndx] = 0;
1637 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1639 // They have the same sign
1640 outputInts[ndx] = rem;
1644 // They have opposite sign. The remainder operation takes the
1645 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1646 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1647 // the result has the correct sign and that it is still
1648 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1650 // See also http://mathforum.org/library/drmath/view/52343.html
1651 outputInts[ndx] = rem + inputInts2[ndx];
1656 "OpCapability Int64\n"
1658 + string(getComputeAsmShaderPreamble()) +
1660 "OpName %main \"main\"\n"
1661 "OpName %id \"gl_GlobalInvocationID\"\n"
1663 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1665 "OpDecorate %buf BufferBlock\n"
1666 "OpDecorate %indata1 DescriptorSet 0\n"
1667 "OpDecorate %indata1 Binding 0\n"
1668 "OpDecorate %indata2 DescriptorSet 0\n"
1669 "OpDecorate %indata2 Binding 1\n"
1670 "OpDecorate %outdata DescriptorSet 0\n"
1671 "OpDecorate %outdata Binding 2\n"
1672 "OpDecorate %i64arr ArrayStride 8\n"
1673 "OpMemberDecorate %buf 0 Offset 0\n"
1675 + string(getComputeAsmCommonTypes())
1676 + string(getComputeAsmCommonInt64Types()) +
1678 "%buf = OpTypeStruct %i64arr\n"
1679 "%bufptr = OpTypePointer Uniform %buf\n"
1680 "%indata1 = OpVariable %bufptr Uniform\n"
1681 "%indata2 = OpVariable %bufptr Uniform\n"
1682 "%outdata = OpVariable %bufptr Uniform\n"
1684 "%id = OpVariable %uvec3ptr Input\n"
1685 "%zero = OpConstant %i64 0\n"
1687 "%main = OpFunction %void None %voidf\n"
1688 "%label = OpLabel\n"
1689 "%idval = OpLoad %uvec3 %id\n"
1690 "%x = OpCompositeExtract %u32 %idval 0\n"
1691 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1692 "%inval1 = OpLoad %i64 %inloc1\n"
1693 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1694 "%inval2 = OpLoad %i64 %inloc2\n"
1695 "%rem = OpSMod %i64 %inval1 %inval2\n"
1696 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1697 " OpStore %outloc %rem\n"
1701 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1702 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1703 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1704 spec.numWorkGroups = IVec3(numElements, 1, 1);
1705 spec.failResult = params.failResult;
1706 spec.failMessage = params.failMessage;
1708 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1711 return group.release();
1714 // Copy contents in the input buffer to the output buffer.
1715 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
1717 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
1718 de::Random rnd (deStringHash(group->getName()));
1719 const int numElements = 100;
1721 // 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.
1722 ComputeShaderSpec spec1;
1723 vector<Vec4> inputFloats1 (numElements);
1724 vector<Vec4> outputFloats1 (numElements);
1726 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
1728 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1729 floorAll(inputFloats1);
1731 for (size_t ndx = 0; ndx < numElements; ++ndx)
1732 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
1735 string(getComputeAsmShaderPreamble()) +
1737 "OpName %main \"main\"\n"
1738 "OpName %id \"gl_GlobalInvocationID\"\n"
1740 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1741 "OpDecorate %vec4arr ArrayStride 16\n"
1743 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1745 "%vec4 = OpTypeVector %f32 4\n"
1746 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
1747 "%vec4ptr_f = OpTypePointer Function %vec4\n"
1748 "%vec4arr = OpTypeRuntimeArray %vec4\n"
1749 "%buf = OpTypeStruct %vec4arr\n"
1750 "%bufptr = OpTypePointer Uniform %buf\n"
1751 "%indata = OpVariable %bufptr Uniform\n"
1752 "%outdata = OpVariable %bufptr Uniform\n"
1754 "%id = OpVariable %uvec3ptr Input\n"
1755 "%zero = OpConstant %i32 0\n"
1756 "%c_f_0 = OpConstant %f32 0.\n"
1757 "%c_f_0_5 = OpConstant %f32 0.5\n"
1758 "%c_f_1_5 = OpConstant %f32 1.5\n"
1759 "%c_f_2_5 = OpConstant %f32 2.5\n"
1760 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
1762 "%main = OpFunction %void None %voidf\n"
1763 "%label = OpLabel\n"
1764 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
1765 "%idval = OpLoad %uvec3 %id\n"
1766 "%x = OpCompositeExtract %u32 %idval 0\n"
1767 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
1768 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
1769 " OpCopyMemory %v_vec4 %inloc\n"
1770 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
1771 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
1772 " OpStore %outloc %add\n"
1776 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
1777 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
1778 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1780 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
1782 // The following case copies a float[100] variable from the input buffer to the output buffer.
1783 ComputeShaderSpec spec2;
1784 vector<float> inputFloats2 (numElements);
1785 vector<float> outputFloats2 (numElements);
1787 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
1789 for (size_t ndx = 0; ndx < numElements; ++ndx)
1790 outputFloats2[ndx] = inputFloats2[ndx];
1793 string(getComputeAsmShaderPreamble()) +
1795 "OpName %main \"main\"\n"
1796 "OpName %id \"gl_GlobalInvocationID\"\n"
1798 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1799 "OpDecorate %f32arr100 ArrayStride 4\n"
1801 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1803 "%hundred = OpConstant %u32 100\n"
1804 "%f32arr100 = OpTypeArray %f32 %hundred\n"
1805 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
1806 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
1807 "%buf = OpTypeStruct %f32arr100\n"
1808 "%bufptr = OpTypePointer Uniform %buf\n"
1809 "%indata = OpVariable %bufptr Uniform\n"
1810 "%outdata = OpVariable %bufptr Uniform\n"
1812 "%id = OpVariable %uvec3ptr Input\n"
1813 "%zero = OpConstant %i32 0\n"
1815 "%main = OpFunction %void None %voidf\n"
1816 "%label = OpLabel\n"
1817 "%var = OpVariable %f32arr100ptr_f Function\n"
1818 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
1819 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
1820 " OpCopyMemory %var %inarr\n"
1821 " OpCopyMemory %outarr %var\n"
1825 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1826 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1827 spec2.numWorkGroups = IVec3(1, 1, 1);
1829 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
1831 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
1832 ComputeShaderSpec spec3;
1833 vector<float> inputFloats3 (16);
1834 vector<float> outputFloats3 (16);
1836 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
1838 for (size_t ndx = 0; ndx < 16; ++ndx)
1839 outputFloats3[ndx] = inputFloats3[ndx];
1842 string(getComputeAsmShaderPreamble()) +
1844 "OpName %main \"main\"\n"
1845 "OpName %id \"gl_GlobalInvocationID\"\n"
1847 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1848 "OpMemberDecorate %buf 0 Offset 0\n"
1849 "OpMemberDecorate %buf 1 Offset 16\n"
1850 "OpMemberDecorate %buf 2 Offset 32\n"
1851 "OpMemberDecorate %buf 3 Offset 48\n"
1853 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1855 "%vec4 = OpTypeVector %f32 4\n"
1856 "%buf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
1857 "%bufptr = OpTypePointer Uniform %buf\n"
1858 "%indata = OpVariable %bufptr Uniform\n"
1859 "%outdata = OpVariable %bufptr Uniform\n"
1860 "%vec4stptr = OpTypePointer Function %buf\n"
1862 "%id = OpVariable %uvec3ptr Input\n"
1863 "%zero = OpConstant %i32 0\n"
1865 "%main = OpFunction %void None %voidf\n"
1866 "%label = OpLabel\n"
1867 "%var = OpVariable %vec4stptr Function\n"
1868 " OpCopyMemory %var %indata\n"
1869 " OpCopyMemory %outdata %var\n"
1873 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1874 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
1875 spec3.numWorkGroups = IVec3(1, 1, 1);
1877 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
1879 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
1880 ComputeShaderSpec spec4;
1881 vector<float> inputFloats4 (numElements);
1882 vector<float> outputFloats4 (numElements);
1884 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
1886 for (size_t ndx = 0; ndx < numElements; ++ndx)
1887 outputFloats4[ndx] = -inputFloats4[ndx];
1890 string(getComputeAsmShaderPreamble()) +
1892 "OpName %main \"main\"\n"
1893 "OpName %id \"gl_GlobalInvocationID\"\n"
1895 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1897 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
1899 "%f32ptr_f = OpTypePointer Function %f32\n"
1900 "%id = OpVariable %uvec3ptr Input\n"
1901 "%zero = OpConstant %i32 0\n"
1903 "%main = OpFunction %void None %voidf\n"
1904 "%label = OpLabel\n"
1905 "%var = OpVariable %f32ptr_f Function\n"
1906 "%idval = OpLoad %uvec3 %id\n"
1907 "%x = OpCompositeExtract %u32 %idval 0\n"
1908 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1909 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1910 " OpCopyMemory %var %inloc\n"
1911 "%val = OpLoad %f32 %var\n"
1912 "%neg = OpFNegate %f32 %val\n"
1913 " OpStore %outloc %neg\n"
1917 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
1918 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
1919 spec4.numWorkGroups = IVec3(numElements, 1, 1);
1921 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
1923 return group.release();
1926 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
1928 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
1929 ComputeShaderSpec spec;
1930 de::Random rnd (deStringHash(group->getName()));
1931 const int numElements = 100;
1932 vector<float> inputFloats (numElements, 0);
1933 vector<float> outputFloats (numElements, 0);
1935 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
1937 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1938 floorAll(inputFloats);
1940 for (size_t ndx = 0; ndx < numElements; ++ndx)
1941 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
1944 string(getComputeAsmShaderPreamble()) +
1946 "OpName %main \"main\"\n"
1947 "OpName %id \"gl_GlobalInvocationID\"\n"
1949 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1951 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1953 "%fmat = OpTypeMatrix %fvec3 3\n"
1954 "%three = OpConstant %u32 3\n"
1955 "%farr = OpTypeArray %f32 %three\n"
1956 "%fst = OpTypeStruct %f32 %f32\n"
1958 + string(getComputeAsmInputOutputBuffer()) +
1960 "%id = OpVariable %uvec3ptr Input\n"
1961 "%zero = OpConstant %i32 0\n"
1962 "%c_f = OpConstant %f32 1.5\n"
1963 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
1964 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
1965 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
1966 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
1968 "%main = OpFunction %void None %voidf\n"
1969 "%label = OpLabel\n"
1970 "%c_f_copy = OpCopyObject %f32 %c_f\n"
1971 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
1972 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
1973 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
1974 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
1975 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
1976 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
1977 "%farr_elem = OpCompositeExtract %f32 %c_farr_copy 2\n"
1978 "%fst_elem = OpCompositeExtract %f32 %c_fst_copy 1\n"
1979 // Add up. 1.5 * 5 = 7.5.
1980 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
1981 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
1982 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
1983 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
1985 "%idval = OpLoad %uvec3 %id\n"
1986 "%x = OpCompositeExtract %u32 %idval 0\n"
1987 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1988 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1989 "%inval = OpLoad %f32 %inloc\n"
1990 "%add = OpFAdd %f32 %add4 %inval\n"
1991 " OpStore %outloc %add\n"
1994 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1995 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1996 spec.numWorkGroups = IVec3(numElements, 1, 1);
1998 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
2000 return group.release();
2002 // Assembly code used for testing OpUnreachable is based on GLSL source code:
2006 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2007 // float elements[];
2009 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2010 // float elements[];
2013 // void not_called_func() {
2014 // // place OpUnreachable here
2017 // uint modulo4(uint val) {
2018 // switch (val % uint(4)) {
2019 // case 0: return 3;
2020 // case 1: return 2;
2021 // case 2: return 1;
2022 // case 3: return 0;
2023 // default: return 100; // place OpUnreachable here
2029 // // place OpUnreachable here
2033 // uint x = gl_GlobalInvocationID.x;
2034 // if (const5() > modulo4(1000)) {
2035 // output_data.elements[x] = -input_data.elements[x];
2037 // // place OpUnreachable here
2038 // output_data.elements[x] = input_data.elements[x];
2042 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
2044 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
2045 ComputeShaderSpec spec;
2046 de::Random rnd (deStringHash(group->getName()));
2047 const int numElements = 100;
2048 vector<float> positiveFloats (numElements, 0);
2049 vector<float> negativeFloats (numElements, 0);
2051 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2053 for (size_t ndx = 0; ndx < numElements; ++ndx)
2054 negativeFloats[ndx] = -positiveFloats[ndx];
2057 string(getComputeAsmShaderPreamble()) +
2059 "OpSource GLSL 430\n"
2060 "OpName %main \"main\"\n"
2061 "OpName %func_not_called_func \"not_called_func(\"\n"
2062 "OpName %func_modulo4 \"modulo4(u1;\"\n"
2063 "OpName %func_const5 \"const5(\"\n"
2064 "OpName %id \"gl_GlobalInvocationID\"\n"
2066 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2068 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2070 "%u32ptr = OpTypePointer Function %u32\n"
2071 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
2072 "%unitf = OpTypeFunction %u32\n"
2074 "%id = OpVariable %uvec3ptr Input\n"
2075 "%zero = OpConstant %u32 0\n"
2076 "%one = OpConstant %u32 1\n"
2077 "%two = OpConstant %u32 2\n"
2078 "%three = OpConstant %u32 3\n"
2079 "%four = OpConstant %u32 4\n"
2080 "%five = OpConstant %u32 5\n"
2081 "%hundred = OpConstant %u32 100\n"
2082 "%thousand = OpConstant %u32 1000\n"
2084 + string(getComputeAsmInputOutputBuffer()) +
2087 "%main = OpFunction %void None %voidf\n"
2088 "%main_entry = OpLabel\n"
2089 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
2090 "%idval = OpLoad %uvec3 %id\n"
2091 "%x = OpCompositeExtract %u32 %idval 0\n"
2092 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2093 "%inval = OpLoad %f32 %inloc\n"
2094 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2095 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
2096 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
2097 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
2098 " OpSelectionMerge %if_end None\n"
2099 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2100 "%if_true = OpLabel\n"
2101 "%negate = OpFNegate %f32 %inval\n"
2102 " OpStore %outloc %negate\n"
2103 " OpBranch %if_end\n"
2104 "%if_false = OpLabel\n"
2105 " OpUnreachable\n" // Unreachable else branch for if statement
2106 "%if_end = OpLabel\n"
2110 // not_called_function()
2111 "%func_not_called_func = OpFunction %void None %voidf\n"
2112 "%not_called_func_entry = OpLabel\n"
2113 " OpUnreachable\n" // Unreachable entry block in not called static function
2117 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
2118 "%valptr = OpFunctionParameter %u32ptr\n"
2119 "%modulo4_entry = OpLabel\n"
2120 "%val = OpLoad %u32 %valptr\n"
2121 "%modulo = OpUMod %u32 %val %four\n"
2122 " OpSelectionMerge %switch_merge None\n"
2123 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
2124 "%case0 = OpLabel\n"
2125 " OpReturnValue %three\n"
2126 "%case1 = OpLabel\n"
2127 " OpReturnValue %two\n"
2128 "%case2 = OpLabel\n"
2129 " OpReturnValue %one\n"
2130 "%case3 = OpLabel\n"
2131 " OpReturnValue %zero\n"
2132 "%default = OpLabel\n"
2133 " OpUnreachable\n" // Unreachable default case for switch statement
2134 "%switch_merge = OpLabel\n"
2135 " OpUnreachable\n" // Unreachable merge block for switch statement
2139 "%func_const5 = OpFunction %u32 None %unitf\n"
2140 "%const5_entry = OpLabel\n"
2141 " OpReturnValue %five\n"
2142 "%unreachable = OpLabel\n"
2143 " OpUnreachable\n" // Unreachable block in function
2145 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2146 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2147 spec.numWorkGroups = IVec3(numElements, 1, 1);
2149 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
2151 return group.release();
2154 // Assembly code used for testing decoration group is based on GLSL source code:
2158 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
2159 // float elements[];
2161 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
2162 // float elements[];
2164 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
2165 // float elements[];
2167 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
2168 // float elements[];
2170 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
2171 // float elements[];
2173 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
2174 // float elements[];
2178 // uint x = gl_GlobalInvocationID.x;
2179 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
2181 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
2183 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
2184 ComputeShaderSpec spec;
2185 de::Random rnd (deStringHash(group->getName()));
2186 const int numElements = 100;
2187 vector<float> inputFloats0 (numElements, 0);
2188 vector<float> inputFloats1 (numElements, 0);
2189 vector<float> inputFloats2 (numElements, 0);
2190 vector<float> inputFloats3 (numElements, 0);
2191 vector<float> inputFloats4 (numElements, 0);
2192 vector<float> outputFloats (numElements, 0);
2194 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
2195 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
2196 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
2197 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
2198 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
2200 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2201 floorAll(inputFloats0);
2202 floorAll(inputFloats1);
2203 floorAll(inputFloats2);
2204 floorAll(inputFloats3);
2205 floorAll(inputFloats4);
2207 for (size_t ndx = 0; ndx < numElements; ++ndx)
2208 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
2211 string(getComputeAsmShaderPreamble()) +
2213 "OpSource GLSL 430\n"
2214 "OpName %main \"main\"\n"
2215 "OpName %id \"gl_GlobalInvocationID\"\n"
2217 // Not using group decoration on variable.
2218 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2219 // Not using group decoration on type.
2220 "OpDecorate %f32arr ArrayStride 4\n"
2222 "OpDecorate %groups BufferBlock\n"
2223 "OpDecorate %groupm Offset 0\n"
2224 "%groups = OpDecorationGroup\n"
2225 "%groupm = OpDecorationGroup\n"
2227 // Group decoration on multiple structs.
2228 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
2229 // Group decoration on multiple struct members.
2230 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
2232 "OpDecorate %group1 DescriptorSet 0\n"
2233 "OpDecorate %group3 DescriptorSet 0\n"
2234 "OpDecorate %group3 NonWritable\n"
2235 "OpDecorate %group3 Restrict\n"
2236 "%group0 = OpDecorationGroup\n"
2237 "%group1 = OpDecorationGroup\n"
2238 "%group3 = OpDecorationGroup\n"
2240 // Applying the same decoration group multiple times.
2241 "OpGroupDecorate %group1 %outdata\n"
2242 "OpGroupDecorate %group1 %outdata\n"
2243 "OpGroupDecorate %group1 %outdata\n"
2244 "OpDecorate %outdata DescriptorSet 0\n"
2245 "OpDecorate %outdata Binding 5\n"
2246 // Applying decoration group containing nothing.
2247 "OpGroupDecorate %group0 %indata0\n"
2248 "OpDecorate %indata0 DescriptorSet 0\n"
2249 "OpDecorate %indata0 Binding 0\n"
2250 // Applying decoration group containing one decoration.
2251 "OpGroupDecorate %group1 %indata1\n"
2252 "OpDecorate %indata1 Binding 1\n"
2253 // Applying decoration group containing multiple decorations.
2254 "OpGroupDecorate %group3 %indata2 %indata3\n"
2255 "OpDecorate %indata2 Binding 2\n"
2256 "OpDecorate %indata3 Binding 3\n"
2257 // Applying multiple decoration groups (with overlapping).
2258 "OpGroupDecorate %group0 %indata4\n"
2259 "OpGroupDecorate %group1 %indata4\n"
2260 "OpGroupDecorate %group3 %indata4\n"
2261 "OpDecorate %indata4 Binding 4\n"
2263 + string(getComputeAsmCommonTypes()) +
2265 "%id = OpVariable %uvec3ptr Input\n"
2266 "%zero = OpConstant %i32 0\n"
2268 "%outbuf = OpTypeStruct %f32arr\n"
2269 "%outbufptr = OpTypePointer Uniform %outbuf\n"
2270 "%outdata = OpVariable %outbufptr Uniform\n"
2271 "%inbuf0 = OpTypeStruct %f32arr\n"
2272 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
2273 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
2274 "%inbuf1 = OpTypeStruct %f32arr\n"
2275 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
2276 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
2277 "%inbuf2 = OpTypeStruct %f32arr\n"
2278 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
2279 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
2280 "%inbuf3 = OpTypeStruct %f32arr\n"
2281 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
2282 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
2283 "%inbuf4 = OpTypeStruct %f32arr\n"
2284 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
2285 "%indata4 = OpVariable %inbufptr Uniform\n"
2287 "%main = OpFunction %void None %voidf\n"
2288 "%label = OpLabel\n"
2289 "%idval = OpLoad %uvec3 %id\n"
2290 "%x = OpCompositeExtract %u32 %idval 0\n"
2291 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
2292 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
2293 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
2294 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
2295 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
2296 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2297 "%inval0 = OpLoad %f32 %inloc0\n"
2298 "%inval1 = OpLoad %f32 %inloc1\n"
2299 "%inval2 = OpLoad %f32 %inloc2\n"
2300 "%inval3 = OpLoad %f32 %inloc3\n"
2301 "%inval4 = OpLoad %f32 %inloc4\n"
2302 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
2303 "%add1 = OpFAdd %f32 %add0 %inval2\n"
2304 "%add2 = OpFAdd %f32 %add1 %inval3\n"
2305 "%add = OpFAdd %f32 %add2 %inval4\n"
2306 " OpStore %outloc %add\n"
2309 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
2310 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
2311 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2312 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2313 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2314 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2315 spec.numWorkGroups = IVec3(numElements, 1, 1);
2317 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
2319 return group.release();
2322 struct SpecConstantTwoIntCase
2324 const char* caseName;
2325 const char* scDefinition0;
2326 const char* scDefinition1;
2327 const char* scResultType;
2328 const char* scOperation;
2329 deInt32 scActualValue0;
2330 deInt32 scActualValue1;
2331 const char* resultOperation;
2332 vector<deInt32> expectedOutput;
2334 SpecConstantTwoIntCase (const char* name,
2335 const char* definition0,
2336 const char* definition1,
2337 const char* resultType,
2338 const char* operation,
2341 const char* resultOp,
2342 const vector<deInt32>& output)
2344 , scDefinition0 (definition0)
2345 , scDefinition1 (definition1)
2346 , scResultType (resultType)
2347 , scOperation (operation)
2348 , scActualValue0 (value0)
2349 , scActualValue1 (value1)
2350 , resultOperation (resultOp)
2351 , expectedOutput (output) {}
2354 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
2356 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
2357 vector<SpecConstantTwoIntCase> cases;
2358 de::Random rnd (deStringHash(group->getName()));
2359 const int numElements = 100;
2360 vector<deInt32> inputInts (numElements, 0);
2361 vector<deInt32> outputInts1 (numElements, 0);
2362 vector<deInt32> outputInts2 (numElements, 0);
2363 vector<deInt32> outputInts3 (numElements, 0);
2364 vector<deInt32> outputInts4 (numElements, 0);
2365 const StringTemplate shaderTemplate (
2366 string(getComputeAsmShaderPreamble()) +
2368 "OpName %main \"main\"\n"
2369 "OpName %id \"gl_GlobalInvocationID\"\n"
2371 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2372 "OpDecorate %sc_0 SpecId 0\n"
2373 "OpDecorate %sc_1 SpecId 1\n"
2374 "OpDecorate %i32arr ArrayStride 4\n"
2376 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2378 "%buf = OpTypeStruct %i32arr\n"
2379 "%bufptr = OpTypePointer Uniform %buf\n"
2380 "%indata = OpVariable %bufptr Uniform\n"
2381 "%outdata = OpVariable %bufptr Uniform\n"
2383 "%id = OpVariable %uvec3ptr Input\n"
2384 "%zero = OpConstant %i32 0\n"
2386 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
2387 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
2388 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
2390 "%main = OpFunction %void None %voidf\n"
2391 "%label = OpLabel\n"
2392 "%idval = OpLoad %uvec3 %id\n"
2393 "%x = OpCompositeExtract %u32 %idval 0\n"
2394 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2395 "%inval = OpLoad %i32 %inloc\n"
2396 "%final = ${GEN_RESULT}\n"
2397 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2398 " OpStore %outloc %final\n"
2400 " OpFunctionEnd\n");
2402 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
2404 for (size_t ndx = 0; ndx < numElements; ++ndx)
2406 outputInts1[ndx] = inputInts[ndx] + 42;
2407 outputInts2[ndx] = inputInts[ndx];
2408 outputInts3[ndx] = inputInts[ndx] - 11200;
2409 outputInts4[ndx] = inputInts[ndx] + 1;
2412 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
2413 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
2414 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
2416 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
2417 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
2418 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
2419 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
2420 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
2421 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2422 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2423 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
2424 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
2425 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
2426 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
2427 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2428 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2429 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
2430 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
2431 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
2432 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2433 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
2434 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
2435 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
2436 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2437 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
2438 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
2439 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2440 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2441 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2442 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2443 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
2444 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
2445 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
2446 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
2447 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
2449 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2451 map<string, string> specializations;
2452 ComputeShaderSpec spec;
2454 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
2455 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
2456 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
2457 specializations["SC_OP"] = cases[caseNdx].scOperation;
2458 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
2460 spec.assembly = shaderTemplate.specialize(specializations);
2461 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2462 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
2463 spec.numWorkGroups = IVec3(numElements, 1, 1);
2464 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
2465 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
2467 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec));
2470 ComputeShaderSpec spec;
2473 string(getComputeAsmShaderPreamble()) +
2475 "OpName %main \"main\"\n"
2476 "OpName %id \"gl_GlobalInvocationID\"\n"
2478 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2479 "OpDecorate %sc_0 SpecId 0\n"
2480 "OpDecorate %sc_1 SpecId 1\n"
2481 "OpDecorate %sc_2 SpecId 2\n"
2482 "OpDecorate %i32arr ArrayStride 4\n"
2484 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2486 "%ivec3 = OpTypeVector %i32 3\n"
2487 "%buf = OpTypeStruct %i32arr\n"
2488 "%bufptr = OpTypePointer Uniform %buf\n"
2489 "%indata = OpVariable %bufptr Uniform\n"
2490 "%outdata = OpVariable %bufptr Uniform\n"
2492 "%id = OpVariable %uvec3ptr Input\n"
2493 "%zero = OpConstant %i32 0\n"
2494 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
2495 "%vec3_undef = OpUndef %ivec3\n"
2497 "%sc_0 = OpSpecConstant %i32 0\n"
2498 "%sc_1 = OpSpecConstant %i32 0\n"
2499 "%sc_2 = OpSpecConstant %i32 0\n"
2500 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
2501 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
2502 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
2503 "%sc_vec3_0_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
2504 "%sc_vec3_1_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
2505 "%sc_vec3_2_s = OpSpecConstantOp %ivec3 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
2506 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
2507 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
2508 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
2509 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
2510 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
2511 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
2512 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
2514 "%main = OpFunction %void None %voidf\n"
2515 "%label = OpLabel\n"
2516 "%idval = OpLoad %uvec3 %id\n"
2517 "%x = OpCompositeExtract %u32 %idval 0\n"
2518 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2519 "%inval = OpLoad %i32 %inloc\n"
2520 "%final = OpIAdd %i32 %inval %sc_final\n"
2521 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2522 " OpStore %outloc %final\n"
2525 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2526 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
2527 spec.numWorkGroups = IVec3(numElements, 1, 1);
2528 spec.specConstants.push_back(123);
2529 spec.specConstants.push_back(56);
2530 spec.specConstants.push_back(-77);
2532 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
2534 return group.release();
2537 string generateConstantDefinitions (int count)
2539 std::stringstream r;
2540 for (int i = 0; i < count; i++)
2541 r << "%cf" << (i * 10 + 5) << " = OpConstant %f32 " <<(i * 10 + 5) << ".0\n";
2542 return r.str() + string("\n");
2545 string generateSwitchCases (int count)
2547 std::stringstream r;
2548 for (int i = 0; i < count; i++)
2549 r << " " << i << " %case" << i;
2550 return r.str() + string("\n");
2553 string generateSwitchTargets (int count)
2555 std::stringstream r;
2556 for (int i = 0; i < count; i++)
2557 r << "%case" << i << " = OpLabel\n OpBranch %phi\n";
2558 return r.str() + string("\n");
2561 string generateOpPhiParams (int count)
2563 std::stringstream r;
2564 for (int i = 0; i < count; i++)
2565 r << " %cf" << (i * 10 + 5) << " %case" << i;
2566 return r.str() + string("\n");
2569 string generateIntWidth (int value)
2571 std::stringstream r;
2576 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
2578 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
2579 ComputeShaderSpec spec1;
2580 ComputeShaderSpec spec2;
2581 ComputeShaderSpec spec3;
2582 ComputeShaderSpec spec4;
2583 de::Random rnd (deStringHash(group->getName()));
2584 const int numElements = 100;
2585 vector<float> inputFloats (numElements, 0);
2586 vector<float> outputFloats1 (numElements, 0);
2587 vector<float> outputFloats2 (numElements, 0);
2588 vector<float> outputFloats3 (numElements, 0);
2589 vector<float> outputFloats4 (numElements, 0);
2590 const int test4Width = 1024;
2592 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
2594 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2595 floorAll(inputFloats);
2597 for (size_t ndx = 0; ndx < numElements; ++ndx)
2601 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
2602 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
2603 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
2606 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
2607 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
2609 int index4 = (int)deFloor(deAbs((float)ndx * inputFloats[ndx]));
2610 outputFloats4[ndx] = (float)(index4 % test4Width) * 10.0f + 5.0f;
2614 string(getComputeAsmShaderPreamble()) +
2616 "OpSource GLSL 430\n"
2617 "OpName %main \"main\"\n"
2618 "OpName %id \"gl_GlobalInvocationID\"\n"
2620 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2622 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2624 "%id = OpVariable %uvec3ptr Input\n"
2625 "%zero = OpConstant %i32 0\n"
2626 "%three = OpConstant %u32 3\n"
2627 "%constf5p5 = OpConstant %f32 5.5\n"
2628 "%constf20p5 = OpConstant %f32 20.5\n"
2629 "%constf1p75 = OpConstant %f32 1.75\n"
2630 "%constf8p5 = OpConstant %f32 8.5\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 "%selector = OpUMod %u32 %x %three\n"
2638 " OpSelectionMerge %phi None\n"
2639 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
2641 // Case 1 before OpPhi.
2642 "%case1 = OpLabel\n"
2645 "%default = OpLabel\n"
2649 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
2650 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2651 "%inval = OpLoad %f32 %inloc\n"
2652 "%add = OpFAdd %f32 %inval %operand\n"
2653 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2654 " OpStore %outloc %add\n"
2657 // Case 0 after OpPhi.
2658 "%case0 = OpLabel\n"
2662 // Case 2 after OpPhi.
2663 "%case2 = OpLabel\n"
2667 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2668 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
2669 spec1.numWorkGroups = IVec3(numElements, 1, 1);
2671 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
2674 string(getComputeAsmShaderPreamble()) +
2676 "OpName %main \"main\"\n"
2677 "OpName %id \"gl_GlobalInvocationID\"\n"
2679 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2681 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2683 "%id = OpVariable %uvec3ptr Input\n"
2684 "%zero = OpConstant %i32 0\n"
2685 "%one = OpConstant %i32 1\n"
2686 "%three = OpConstant %i32 3\n"
2687 "%constf6p5 = OpConstant %f32 6.5\n"
2689 "%main = OpFunction %void None %voidf\n"
2690 "%entry = OpLabel\n"
2691 "%idval = OpLoad %uvec3 %id\n"
2692 "%x = OpCompositeExtract %u32 %idval 0\n"
2693 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2694 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2695 "%inval = OpLoad %f32 %inloc\n"
2699 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
2700 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
2701 "%step_next = OpIAdd %i32 %step %one\n"
2702 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
2703 "%still_loop = OpSLessThan %bool %step %three\n"
2704 " OpLoopMerge %exit %phi None\n"
2705 " OpBranchConditional %still_loop %phi %exit\n"
2708 " OpStore %outloc %accum\n"
2711 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2712 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2713 spec2.numWorkGroups = IVec3(numElements, 1, 1);
2715 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
2718 string(getComputeAsmShaderPreamble()) +
2720 "OpName %main \"main\"\n"
2721 "OpName %id \"gl_GlobalInvocationID\"\n"
2723 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2725 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2727 "%f32ptr_f = OpTypePointer Function %f32\n"
2728 "%id = OpVariable %uvec3ptr Input\n"
2729 "%true = OpConstantTrue %bool\n"
2730 "%false = OpConstantFalse %bool\n"
2731 "%zero = OpConstant %i32 0\n"
2732 "%constf8p5 = OpConstant %f32 8.5\n"
2734 "%main = OpFunction %void None %voidf\n"
2735 "%entry = OpLabel\n"
2736 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
2737 "%idval = OpLoad %uvec3 %id\n"
2738 "%x = OpCompositeExtract %u32 %idval 0\n"
2739 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2740 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2741 "%a_init = OpLoad %f32 %inloc\n"
2742 "%b_init = OpLoad %f32 %b\n"
2746 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
2747 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
2748 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
2749 " OpLoopMerge %exit %phi None\n"
2750 " OpBranchConditional %still_loop %phi %exit\n"
2753 "%sub = OpFSub %f32 %a_next %b_next\n"
2754 " OpStore %outloc %sub\n"
2757 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2758 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
2759 spec3.numWorkGroups = IVec3(numElements, 1, 1);
2761 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
2764 "OpCapability Shader\n"
2765 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
2766 "OpMemoryModel Logical GLSL450\n"
2767 "OpEntryPoint GLCompute %main \"main\" %id\n"
2768 "OpExecutionMode %main LocalSize 1 1 1\n"
2770 "OpSource GLSL 430\n"
2771 "OpName %main \"main\"\n"
2772 "OpName %id \"gl_GlobalInvocationID\"\n"
2774 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2776 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2778 "%id = OpVariable %uvec3ptr Input\n"
2779 "%zero = OpConstant %i32 0\n"
2780 "%cimod = OpConstant %u32 " + generateIntWidth(test4Width) + "\n"
2782 + generateConstantDefinitions(test4Width) +
2784 "%main = OpFunction %void None %voidf\n"
2785 "%entry = OpLabel\n"
2786 "%idval = OpLoad %uvec3 %id\n"
2787 "%x = OpCompositeExtract %u32 %idval 0\n"
2788 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2789 "%inval = OpLoad %f32 %inloc\n"
2790 "%xf = OpConvertUToF %f32 %x\n"
2791 "%xm = OpFMul %f32 %xf %inval\n"
2792 "%xa = OpExtInst %f32 %ext FAbs %xm\n"
2793 "%xi = OpConvertFToU %u32 %xa\n"
2794 "%selector = OpUMod %u32 %xi %cimod\n"
2795 " OpSelectionMerge %phi None\n"
2796 " OpSwitch %selector %default "
2798 + generateSwitchCases(test4Width) +
2800 "%default = OpLabel\n"
2803 + generateSwitchTargets(test4Width) +
2806 "%result = OpPhi %f32"
2808 + generateOpPhiParams(test4Width) +
2810 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2811 " OpStore %outloc %result\n"
2815 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2816 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
2817 spec4.numWorkGroups = IVec3(numElements, 1, 1);
2819 group->addChild(new SpvAsmComputeShaderCase(testCtx, "wide", "OpPhi with a lot of parameters", spec4));
2821 return group.release();
2824 // Assembly code used for testing block order is based on GLSL source code:
2828 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2829 // float elements[];
2831 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2832 // float elements[];
2836 // uint x = gl_GlobalInvocationID.x;
2837 // output_data.elements[x] = input_data.elements[x];
2838 // if (x > uint(50)) {
2839 // switch (x % uint(3)) {
2840 // case 0: output_data.elements[x] += 1.5f; break;
2841 // case 1: output_data.elements[x] += 42.f; break;
2842 // case 2: output_data.elements[x] -= 27.f; break;
2846 // output_data.elements[x] = -input_data.elements[x];
2849 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
2851 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
2852 ComputeShaderSpec spec;
2853 de::Random rnd (deStringHash(group->getName()));
2854 const int numElements = 100;
2855 vector<float> inputFloats (numElements, 0);
2856 vector<float> outputFloats (numElements, 0);
2858 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2860 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2861 floorAll(inputFloats);
2863 for (size_t ndx = 0; ndx <= 50; ++ndx)
2864 outputFloats[ndx] = -inputFloats[ndx];
2866 for (size_t ndx = 51; ndx < numElements; ++ndx)
2870 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
2871 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
2872 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
2878 string(getComputeAsmShaderPreamble()) +
2880 "OpSource GLSL 430\n"
2881 "OpName %main \"main\"\n"
2882 "OpName %id \"gl_GlobalInvocationID\"\n"
2884 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2886 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2888 "%u32ptr = OpTypePointer Function %u32\n"
2889 "%u32ptr_input = OpTypePointer Input %u32\n"
2891 + string(getComputeAsmInputOutputBuffer()) +
2893 "%id = OpVariable %uvec3ptr Input\n"
2894 "%zero = OpConstant %i32 0\n"
2895 "%const3 = OpConstant %u32 3\n"
2896 "%const50 = OpConstant %u32 50\n"
2897 "%constf1p5 = OpConstant %f32 1.5\n"
2898 "%constf27 = OpConstant %f32 27.0\n"
2899 "%constf42 = OpConstant %f32 42.0\n"
2901 "%main = OpFunction %void None %voidf\n"
2904 "%entry = OpLabel\n"
2906 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
2907 "%xvar = OpVariable %u32ptr Function\n"
2908 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
2909 "%x = OpLoad %u32 %xptr\n"
2910 " OpStore %xvar %x\n"
2912 "%cmp = OpUGreaterThan %bool %x %const50\n"
2913 " OpSelectionMerge %if_merge None\n"
2914 " OpBranchConditional %cmp %if_true %if_false\n"
2916 // False branch for if-statement: placed in the middle of switch cases and before true branch.
2917 "%if_false = OpLabel\n"
2918 "%x_f = OpLoad %u32 %xvar\n"
2919 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
2920 "%inval_f = OpLoad %f32 %inloc_f\n"
2921 "%negate = OpFNegate %f32 %inval_f\n"
2922 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
2923 " OpStore %outloc_f %negate\n"
2924 " OpBranch %if_merge\n"
2926 // Merge block for if-statement: placed in the middle of true and false branch.
2927 "%if_merge = OpLabel\n"
2930 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
2931 "%if_true = OpLabel\n"
2932 "%xval_t = OpLoad %u32 %xvar\n"
2933 "%mod = OpUMod %u32 %xval_t %const3\n"
2934 " OpSelectionMerge %switch_merge None\n"
2935 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
2937 // Merge block for switch-statement: placed before the case
2938 // bodies. But it must follow OpSwitch which dominates it.
2939 "%switch_merge = OpLabel\n"
2940 " OpBranch %if_merge\n"
2942 // Case 1 for switch-statement: placed before case 0.
2943 // It must follow the OpSwitch that dominates it.
2944 "%case1 = OpLabel\n"
2945 "%x_1 = OpLoad %u32 %xvar\n"
2946 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
2947 "%inval_1 = OpLoad %f32 %inloc_1\n"
2948 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
2949 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
2950 " OpStore %outloc_1 %addf42\n"
2951 " OpBranch %switch_merge\n"
2953 // Case 2 for switch-statement.
2954 "%case2 = OpLabel\n"
2955 "%x_2 = OpLoad %u32 %xvar\n"
2956 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
2957 "%inval_2 = OpLoad %f32 %inloc_2\n"
2958 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
2959 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
2960 " OpStore %outloc_2 %subf27\n"
2961 " OpBranch %switch_merge\n"
2963 // Default case for switch-statement: placed in the middle of normal cases.
2964 "%default = OpLabel\n"
2965 " OpBranch %switch_merge\n"
2967 // Case 0 for switch-statement: out of order.
2968 "%case0 = OpLabel\n"
2969 "%x_0 = OpLoad %u32 %xvar\n"
2970 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
2971 "%inval_0 = OpLoad %f32 %inloc_0\n"
2972 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
2973 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
2974 " OpStore %outloc_0 %addf1p5\n"
2975 " OpBranch %switch_merge\n"
2978 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2979 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2980 spec.numWorkGroups = IVec3(numElements, 1, 1);
2982 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
2984 return group.release();
2987 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
2989 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
2990 ComputeShaderSpec spec1;
2991 ComputeShaderSpec spec2;
2992 de::Random rnd (deStringHash(group->getName()));
2993 const int numElements = 100;
2994 vector<float> inputFloats (numElements, 0);
2995 vector<float> outputFloats1 (numElements, 0);
2996 vector<float> outputFloats2 (numElements, 0);
2997 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
2999 for (size_t ndx = 0; ndx < numElements; ++ndx)
3001 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
3002 outputFloats2[ndx] = -inputFloats[ndx];
3005 const string assembly(
3006 "OpCapability Shader\n"
3007 "OpCapability ClipDistance\n"
3008 "OpMemoryModel Logical GLSL450\n"
3009 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
3010 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
3011 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
3012 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
3013 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
3014 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
3016 "OpName %comp_main1 \"entrypoint1\"\n"
3017 "OpName %comp_main2 \"entrypoint2\"\n"
3018 "OpName %vert_main \"entrypoint2\"\n"
3019 "OpName %id \"gl_GlobalInvocationID\"\n"
3020 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
3021 "OpName %vertexIndex \"gl_VertexIndex\"\n"
3022 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
3023 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
3024 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
3025 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
3027 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3028 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
3029 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
3030 "OpDecorate %vert_builtin_st Block\n"
3031 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
3032 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
3033 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
3035 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3037 "%zero = OpConstant %i32 0\n"
3038 "%one = OpConstant %u32 1\n"
3039 "%c_f32_1 = OpConstant %f32 1\n"
3041 "%i32inputptr = OpTypePointer Input %i32\n"
3042 "%vec4 = OpTypeVector %f32 4\n"
3043 "%vec4ptr = OpTypePointer Output %vec4\n"
3044 "%f32arr1 = OpTypeArray %f32 %one\n"
3045 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
3046 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
3047 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
3049 "%id = OpVariable %uvec3ptr Input\n"
3050 "%vertexIndex = OpVariable %i32inputptr Input\n"
3051 "%instanceIndex = OpVariable %i32inputptr Input\n"
3052 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
3054 // gl_Position = vec4(1.);
3055 "%vert_main = OpFunction %void None %voidf\n"
3056 "%vert_entry = OpLabel\n"
3057 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
3058 " OpStore %position %c_vec4_1\n"
3063 "%comp_main1 = OpFunction %void None %voidf\n"
3064 "%comp1_entry = OpLabel\n"
3065 "%idval1 = OpLoad %uvec3 %id\n"
3066 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
3067 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
3068 "%inval1 = OpLoad %f32 %inloc1\n"
3069 "%add = OpFAdd %f32 %inval1 %inval1\n"
3070 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
3071 " OpStore %outloc1 %add\n"
3076 "%comp_main2 = OpFunction %void None %voidf\n"
3077 "%comp2_entry = OpLabel\n"
3078 "%idval2 = OpLoad %uvec3 %id\n"
3079 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
3080 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
3081 "%inval2 = OpLoad %f32 %inloc2\n"
3082 "%neg = OpFNegate %f32 %inval2\n"
3083 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
3084 " OpStore %outloc2 %neg\n"
3086 " OpFunctionEnd\n");
3088 spec1.assembly = assembly;
3089 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3090 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
3091 spec1.numWorkGroups = IVec3(numElements, 1, 1);
3092 spec1.entryPoint = "entrypoint1";
3094 spec2.assembly = assembly;
3095 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3096 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
3097 spec2.numWorkGroups = IVec3(numElements, 1, 1);
3098 spec2.entryPoint = "entrypoint2";
3100 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
3101 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
3103 return group.release();
3106 inline std::string makeLongUTF8String (size_t num4ByteChars)
3108 // An example of a longest valid UTF-8 character. Be explicit about the
3109 // character type because Microsoft compilers can otherwise interpret the
3110 // character string as being over wide (16-bit) characters. Ideally, we
3111 // would just use a C++11 UTF-8 string literal, but we want to support older
3112 // Microsoft compilers.
3113 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
3114 std::string longString;
3115 longString.reserve(num4ByteChars * 4);
3116 for (size_t count = 0; count < num4ByteChars; count++)
3118 longString += earthAfrica;
3123 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
3125 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
3126 vector<CaseParameter> cases;
3127 de::Random rnd (deStringHash(group->getName()));
3128 const int numElements = 100;
3129 vector<float> positiveFloats (numElements, 0);
3130 vector<float> negativeFloats (numElements, 0);
3131 const StringTemplate shaderTemplate (
3132 "OpCapability Shader\n"
3133 "OpMemoryModel Logical GLSL450\n"
3135 "OpEntryPoint GLCompute %main \"main\" %id\n"
3136 "OpExecutionMode %main LocalSize 1 1 1\n"
3140 "OpName %main \"main\"\n"
3141 "OpName %id \"gl_GlobalInvocationID\"\n"
3143 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3145 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3147 "%id = OpVariable %uvec3ptr Input\n"
3148 "%zero = OpConstant %i32 0\n"
3150 "%main = OpFunction %void None %voidf\n"
3151 "%label = OpLabel\n"
3152 "%idval = OpLoad %uvec3 %id\n"
3153 "%x = OpCompositeExtract %u32 %idval 0\n"
3154 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3155 "%inval = OpLoad %f32 %inloc\n"
3156 "%neg = OpFNegate %f32 %inval\n"
3157 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3158 " OpStore %outloc %neg\n"
3160 " OpFunctionEnd\n");
3162 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
3163 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
3164 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
3165 "OpSource GLSL 430 %fname"));
3166 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
3167 "OpSource GLSL 430 %fname"));
3168 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
3169 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
3170 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
3171 "OpSource GLSL 430 %fname \"\""));
3172 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
3173 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
3174 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
3175 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
3176 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
3177 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
3178 "OpSourceContinued \"id main() {}\""));
3179 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
3180 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3181 "OpSourceContinued \"\""));
3182 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
3183 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3184 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
3185 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
3186 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3187 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
3188 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
3189 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
3190 "OpSourceContinued \"void\"\n"
3191 "OpSourceContinued \"main()\"\n"
3192 "OpSourceContinued \"{}\""));
3193 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
3194 "OpSource GLSL 430 %fname \"\"\n"
3195 "OpSourceContinued \"#version 430\nvoid main() {}\""));
3197 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3199 for (size_t ndx = 0; ndx < numElements; ++ndx)
3200 negativeFloats[ndx] = -positiveFloats[ndx];
3202 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3204 map<string, string> specializations;
3205 ComputeShaderSpec spec;
3207 specializations["SOURCE"] = cases[caseNdx].param;
3208 spec.assembly = shaderTemplate.specialize(specializations);
3209 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3210 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3211 spec.numWorkGroups = IVec3(numElements, 1, 1);
3213 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3216 return group.release();
3219 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
3221 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
3222 vector<CaseParameter> cases;
3223 de::Random rnd (deStringHash(group->getName()));
3224 const int numElements = 100;
3225 vector<float> inputFloats (numElements, 0);
3226 vector<float> outputFloats (numElements, 0);
3227 const StringTemplate shaderTemplate (
3228 string(getComputeAsmShaderPreamble()) +
3230 "OpSourceExtension \"${EXTENSION}\"\n"
3232 "OpName %main \"main\"\n"
3233 "OpName %id \"gl_GlobalInvocationID\"\n"
3235 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3237 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3239 "%id = OpVariable %uvec3ptr Input\n"
3240 "%zero = OpConstant %i32 0\n"
3242 "%main = OpFunction %void None %voidf\n"
3243 "%label = OpLabel\n"
3244 "%idval = OpLoad %uvec3 %id\n"
3245 "%x = OpCompositeExtract %u32 %idval 0\n"
3246 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3247 "%inval = OpLoad %f32 %inloc\n"
3248 "%neg = OpFNegate %f32 %inval\n"
3249 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3250 " OpStore %outloc %neg\n"
3252 " OpFunctionEnd\n");
3254 cases.push_back(CaseParameter("empty_extension", ""));
3255 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
3256 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
3257 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
3258 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
3260 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
3262 for (size_t ndx = 0; ndx < numElements; ++ndx)
3263 outputFloats[ndx] = -inputFloats[ndx];
3265 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3267 map<string, string> specializations;
3268 ComputeShaderSpec spec;
3270 specializations["EXTENSION"] = cases[caseNdx].param;
3271 spec.assembly = shaderTemplate.specialize(specializations);
3272 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3273 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3274 spec.numWorkGroups = IVec3(numElements, 1, 1);
3276 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3279 return group.release();
3282 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
3283 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
3285 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
3286 vector<CaseParameter> cases;
3287 de::Random rnd (deStringHash(group->getName()));
3288 const int numElements = 100;
3289 vector<float> positiveFloats (numElements, 0);
3290 vector<float> negativeFloats (numElements, 0);
3291 const StringTemplate shaderTemplate (
3292 string(getComputeAsmShaderPreamble()) +
3294 "OpSource GLSL 430\n"
3295 "OpName %main \"main\"\n"
3296 "OpName %id \"gl_GlobalInvocationID\"\n"
3298 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3300 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3301 "%uvec2 = OpTypeVector %u32 2\n"
3302 "%bvec3 = OpTypeVector %bool 3\n"
3303 "%fvec4 = OpTypeVector %f32 4\n"
3304 "%fmat33 = OpTypeMatrix %fvec3 3\n"
3305 "%const100 = OpConstant %u32 100\n"
3306 "%uarr100 = OpTypeArray %i32 %const100\n"
3307 "%struct = OpTypeStruct %f32 %i32 %u32\n"
3308 "%pointer = OpTypePointer Function %i32\n"
3309 + string(getComputeAsmInputOutputBuffer()) +
3311 "%null = OpConstantNull ${TYPE}\n"
3313 "%id = OpVariable %uvec3ptr Input\n"
3314 "%zero = OpConstant %i32 0\n"
3316 "%main = OpFunction %void None %voidf\n"
3317 "%label = OpLabel\n"
3318 "%idval = OpLoad %uvec3 %id\n"
3319 "%x = OpCompositeExtract %u32 %idval 0\n"
3320 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3321 "%inval = OpLoad %f32 %inloc\n"
3322 "%neg = OpFNegate %f32 %inval\n"
3323 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3324 " OpStore %outloc %neg\n"
3326 " OpFunctionEnd\n");
3328 cases.push_back(CaseParameter("bool", "%bool"));
3329 cases.push_back(CaseParameter("sint32", "%i32"));
3330 cases.push_back(CaseParameter("uint32", "%u32"));
3331 cases.push_back(CaseParameter("float32", "%f32"));
3332 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
3333 cases.push_back(CaseParameter("vec3bool", "%bvec3"));
3334 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
3335 cases.push_back(CaseParameter("matrix", "%fmat33"));
3336 cases.push_back(CaseParameter("array", "%uarr100"));
3337 cases.push_back(CaseParameter("struct", "%struct"));
3338 cases.push_back(CaseParameter("pointer", "%pointer"));
3340 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3342 for (size_t ndx = 0; ndx < numElements; ++ndx)
3343 negativeFloats[ndx] = -positiveFloats[ndx];
3345 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3347 map<string, string> specializations;
3348 ComputeShaderSpec spec;
3350 specializations["TYPE"] = cases[caseNdx].param;
3351 spec.assembly = shaderTemplate.specialize(specializations);
3352 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3353 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3354 spec.numWorkGroups = IVec3(numElements, 1, 1);
3356 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3359 return group.release();
3362 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
3363 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
3365 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
3366 vector<CaseParameter> cases;
3367 de::Random rnd (deStringHash(group->getName()));
3368 const int numElements = 100;
3369 vector<float> positiveFloats (numElements, 0);
3370 vector<float> negativeFloats (numElements, 0);
3371 const StringTemplate shaderTemplate (
3372 string(getComputeAsmShaderPreamble()) +
3374 "OpSource GLSL 430\n"
3375 "OpName %main \"main\"\n"
3376 "OpName %id \"gl_GlobalInvocationID\"\n"
3378 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3380 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3382 "%id = OpVariable %uvec3ptr Input\n"
3383 "%zero = OpConstant %i32 0\n"
3387 "%main = OpFunction %void None %voidf\n"
3388 "%label = OpLabel\n"
3389 "%idval = OpLoad %uvec3 %id\n"
3390 "%x = OpCompositeExtract %u32 %idval 0\n"
3391 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3392 "%inval = OpLoad %f32 %inloc\n"
3393 "%neg = OpFNegate %f32 %inval\n"
3394 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3395 " OpStore %outloc %neg\n"
3397 " OpFunctionEnd\n");
3399 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
3400 "%const = OpConstantComposite %uvec3 %five %zero %five"));
3401 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
3402 "%ten = OpConstant %f32 10.\n"
3403 "%fzero = OpConstant %f32 0.\n"
3404 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
3405 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
3406 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
3407 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
3408 "%fzero = OpConstant %f32 0.\n"
3409 "%one = OpConstant %f32 1.\n"
3410 "%point5 = OpConstant %f32 0.5\n"
3411 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
3412 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
3413 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
3414 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
3415 "%st2 = OpTypeStruct %i32 %i32\n"
3416 "%struct = OpTypeStruct %st1 %st2\n"
3417 "%point5 = OpConstant %f32 0.5\n"
3418 "%one = OpConstant %u32 1\n"
3419 "%ten = OpConstant %i32 10\n"
3420 "%st1val = OpConstantComposite %st1 %one %point5\n"
3421 "%st2val = OpConstantComposite %st2 %ten %ten\n"
3422 "%const = OpConstantComposite %struct %st1val %st2val"));
3424 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3426 for (size_t ndx = 0; ndx < numElements; ++ndx)
3427 negativeFloats[ndx] = -positiveFloats[ndx];
3429 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3431 map<string, string> specializations;
3432 ComputeShaderSpec spec;
3434 specializations["CONSTANT"] = cases[caseNdx].param;
3435 spec.assembly = shaderTemplate.specialize(specializations);
3436 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3437 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3438 spec.numWorkGroups = IVec3(numElements, 1, 1);
3440 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3443 return group.release();
3446 // Creates a floating point number with the given exponent, and significand
3447 // bits set. It can only create normalized numbers. Only the least significant
3448 // 24 bits of the significand will be examined. The final bit of the
3449 // significand will also be ignored. This allows alignment to be written
3450 // similarly to C99 hex-floats.
3451 // For example if you wanted to write 0x1.7f34p-12 you would call
3452 // constructNormalizedFloat(-12, 0x7f3400)
3453 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
3457 for (deInt32 idx = 0; idx < 23; ++idx)
3459 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
3463 return std::ldexp(f, exponent);
3466 // Compare instruction for the OpQuantizeF16 compute exact case.
3467 // Returns true if the output is what is expected from the test case.
3468 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
3470 if (outputAllocs.size() != 1)
3473 // Only size is needed because we cannot compare Nans.
3474 size_t byteSize = expectedOutputs[0]->getByteSize();
3476 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());
3478 if (byteSize != 4*sizeof(float)) {
3482 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
3483 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
3488 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
3489 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
3494 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
3495 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
3500 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
3501 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
3508 // Checks that every output from a test-case is a float NaN.
3509 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
3511 if (outputAllocs.size() != 1)
3514 // Only size is needed because we cannot compare Nans.
3515 size_t byteSize = expectedOutputs[0]->getByteSize();
3517 const float* const output_as_float = static_cast<const float* const>(outputAllocs[0]->getHostPtr());
3519 for (size_t idx = 0; idx < byteSize / sizeof(float); ++idx)
3521 if (!deFloatIsNaN(output_as_float[idx]))
3530 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
3531 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
3533 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
3535 const std::string shader (
3536 string(getComputeAsmShaderPreamble()) +
3538 "OpSource GLSL 430\n"
3539 "OpName %main \"main\"\n"
3540 "OpName %id \"gl_GlobalInvocationID\"\n"
3542 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3544 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3546 "%id = OpVariable %uvec3ptr Input\n"
3547 "%zero = OpConstant %i32 0\n"
3549 "%main = OpFunction %void None %voidf\n"
3550 "%label = OpLabel\n"
3551 "%idval = OpLoad %uvec3 %id\n"
3552 "%x = OpCompositeExtract %u32 %idval 0\n"
3553 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3554 "%inval = OpLoad %f32 %inloc\n"
3555 "%quant = OpQuantizeToF16 %f32 %inval\n"
3556 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3557 " OpStore %outloc %quant\n"
3559 " OpFunctionEnd\n");
3562 ComputeShaderSpec spec;
3563 const deUint32 numElements = 100;
3564 vector<float> infinities;
3565 vector<float> results;
3567 infinities.reserve(numElements);
3568 results.reserve(numElements);
3570 for (size_t idx = 0; idx < numElements; ++idx)
3575 infinities.push_back(std::numeric_limits<float>::infinity());
3576 results.push_back(std::numeric_limits<float>::infinity());
3579 infinities.push_back(-std::numeric_limits<float>::infinity());
3580 results.push_back(-std::numeric_limits<float>::infinity());
3583 infinities.push_back(std::ldexp(1.0f, 16));
3584 results.push_back(std::numeric_limits<float>::infinity());
3587 infinities.push_back(std::ldexp(-1.0f, 32));
3588 results.push_back(-std::numeric_limits<float>::infinity());
3593 spec.assembly = shader;
3594 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
3595 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
3596 spec.numWorkGroups = IVec3(numElements, 1, 1);
3598 group->addChild(new SpvAsmComputeShaderCase(
3599 testCtx, "infinities", "Check that infinities propagated and created", spec));
3603 ComputeShaderSpec spec;
3605 const deUint32 numElements = 100;
3607 nans.reserve(numElements);
3609 for (size_t idx = 0; idx < numElements; ++idx)
3613 nans.push_back(std::numeric_limits<float>::quiet_NaN());
3617 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
3621 spec.assembly = shader;
3622 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
3623 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
3624 spec.numWorkGroups = IVec3(numElements, 1, 1);
3625 spec.verifyIO = &compareNan;
3627 group->addChild(new SpvAsmComputeShaderCase(
3628 testCtx, "propagated_nans", "Check that nans are propagated", spec));
3632 ComputeShaderSpec spec;
3633 vector<float> small;
3634 vector<float> zeros;
3635 const deUint32 numElements = 100;
3637 small.reserve(numElements);
3638 zeros.reserve(numElements);
3640 for (size_t idx = 0; idx < numElements; ++idx)
3645 small.push_back(0.f);
3646 zeros.push_back(0.f);
3649 small.push_back(-0.f);
3650 zeros.push_back(-0.f);
3653 small.push_back(std::ldexp(1.0f, -16));
3654 zeros.push_back(0.f);
3657 small.push_back(std::ldexp(-1.0f, -32));
3658 zeros.push_back(-0.f);
3661 small.push_back(std::ldexp(1.0f, -127));
3662 zeros.push_back(0.f);
3665 small.push_back(-std::ldexp(1.0f, -128));
3666 zeros.push_back(-0.f);
3671 spec.assembly = shader;
3672 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
3673 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
3674 spec.numWorkGroups = IVec3(numElements, 1, 1);
3676 group->addChild(new SpvAsmComputeShaderCase(
3677 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
3681 ComputeShaderSpec spec;
3682 vector<float> exact;
3683 const deUint32 numElements = 200;
3685 exact.reserve(numElements);
3687 for (size_t idx = 0; idx < numElements; ++idx)
3688 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
3690 spec.assembly = shader;
3691 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
3692 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
3693 spec.numWorkGroups = IVec3(numElements, 1, 1);
3695 group->addChild(new SpvAsmComputeShaderCase(
3696 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
3700 ComputeShaderSpec spec;
3701 vector<float> inputs;
3702 const deUint32 numElements = 4;
3704 inputs.push_back(constructNormalizedFloat(8, 0x300300));
3705 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
3706 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
3707 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
3709 spec.assembly = shader;
3710 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
3711 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3712 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
3713 spec.numWorkGroups = IVec3(numElements, 1, 1);
3715 group->addChild(new SpvAsmComputeShaderCase(
3716 testCtx, "rounded", "Check that are rounded when needed", spec));
3719 return group.release();
3722 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
3724 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
3726 const std::string shader (
3727 string(getComputeAsmShaderPreamble()) +
3729 "OpName %main \"main\"\n"
3730 "OpName %id \"gl_GlobalInvocationID\"\n"
3732 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3734 "OpDecorate %sc_0 SpecId 0\n"
3735 "OpDecorate %sc_1 SpecId 1\n"
3736 "OpDecorate %sc_2 SpecId 2\n"
3737 "OpDecorate %sc_3 SpecId 3\n"
3738 "OpDecorate %sc_4 SpecId 4\n"
3739 "OpDecorate %sc_5 SpecId 5\n"
3741 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3743 "%id = OpVariable %uvec3ptr Input\n"
3744 "%zero = OpConstant %i32 0\n"
3745 "%c_u32_6 = OpConstant %u32 6\n"
3747 "%sc_0 = OpSpecConstant %f32 0.\n"
3748 "%sc_1 = OpSpecConstant %f32 0.\n"
3749 "%sc_2 = OpSpecConstant %f32 0.\n"
3750 "%sc_3 = OpSpecConstant %f32 0.\n"
3751 "%sc_4 = OpSpecConstant %f32 0.\n"
3752 "%sc_5 = OpSpecConstant %f32 0.\n"
3754 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
3755 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
3756 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
3757 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
3758 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
3759 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
3761 "%main = OpFunction %void None %voidf\n"
3762 "%label = OpLabel\n"
3763 "%idval = OpLoad %uvec3 %id\n"
3764 "%x = OpCompositeExtract %u32 %idval 0\n"
3765 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3766 "%selector = OpUMod %u32 %x %c_u32_6\n"
3767 " OpSelectionMerge %exit None\n"
3768 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
3770 "%case0 = OpLabel\n"
3771 " OpStore %outloc %sc_0_quant\n"
3774 "%case1 = OpLabel\n"
3775 " OpStore %outloc %sc_1_quant\n"
3778 "%case2 = OpLabel\n"
3779 " OpStore %outloc %sc_2_quant\n"
3782 "%case3 = OpLabel\n"
3783 " OpStore %outloc %sc_3_quant\n"
3786 "%case4 = OpLabel\n"
3787 " OpStore %outloc %sc_4_quant\n"
3790 "%case5 = OpLabel\n"
3791 " OpStore %outloc %sc_5_quant\n"
3797 " OpFunctionEnd\n");
3800 ComputeShaderSpec spec;
3801 const deUint8 numCases = 4;
3802 vector<float> inputs (numCases, 0.f);
3803 vector<float> outputs;
3805 spec.assembly = shader;
3806 spec.numWorkGroups = IVec3(numCases, 1, 1);
3808 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
3809 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
3810 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
3811 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
3813 outputs.push_back(std::numeric_limits<float>::infinity());
3814 outputs.push_back(-std::numeric_limits<float>::infinity());
3815 outputs.push_back(std::numeric_limits<float>::infinity());
3816 outputs.push_back(-std::numeric_limits<float>::infinity());
3818 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3819 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3821 group->addChild(new SpvAsmComputeShaderCase(
3822 testCtx, "infinities", "Check that infinities propagated and created", spec));
3826 ComputeShaderSpec spec;
3827 const deUint8 numCases = 2;
3828 vector<float> inputs (numCases, 0.f);
3829 vector<float> outputs;
3831 spec.assembly = shader;
3832 spec.numWorkGroups = IVec3(numCases, 1, 1);
3833 spec.verifyIO = &compareNan;
3835 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
3836 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
3838 for (deUint8 idx = 0; idx < numCases; ++idx)
3839 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
3841 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3842 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3844 group->addChild(new SpvAsmComputeShaderCase(
3845 testCtx, "propagated_nans", "Check that nans are propagated", spec));
3849 ComputeShaderSpec spec;
3850 const deUint8 numCases = 6;
3851 vector<float> inputs (numCases, 0.f);
3852 vector<float> outputs;
3854 spec.assembly = shader;
3855 spec.numWorkGroups = IVec3(numCases, 1, 1);
3857 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
3858 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
3859 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
3860 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
3861 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
3862 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
3864 outputs.push_back(0.f);
3865 outputs.push_back(-0.f);
3866 outputs.push_back(0.f);
3867 outputs.push_back(-0.f);
3868 outputs.push_back(0.f);
3869 outputs.push_back(-0.f);
3871 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3872 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3874 group->addChild(new SpvAsmComputeShaderCase(
3875 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
3879 ComputeShaderSpec spec;
3880 const deUint8 numCases = 6;
3881 vector<float> inputs (numCases, 0.f);
3882 vector<float> outputs;
3884 spec.assembly = shader;
3885 spec.numWorkGroups = IVec3(numCases, 1, 1);
3887 for (deUint8 idx = 0; idx < 6; ++idx)
3889 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
3890 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
3891 outputs.push_back(f);
3894 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3895 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3897 group->addChild(new SpvAsmComputeShaderCase(
3898 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
3902 ComputeShaderSpec spec;
3903 const deUint8 numCases = 4;
3904 vector<float> inputs (numCases, 0.f);
3905 vector<float> outputs;
3907 spec.assembly = shader;
3908 spec.numWorkGroups = IVec3(numCases, 1, 1);
3909 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
3911 outputs.push_back(constructNormalizedFloat(8, 0x300300));
3912 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
3913 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
3914 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
3916 for (deUint8 idx = 0; idx < numCases; ++idx)
3917 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
3919 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3920 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3922 group->addChild(new SpvAsmComputeShaderCase(
3923 testCtx, "rounded", "Check that are rounded when needed", spec));
3926 return group.release();
3929 // Checks that constant null/composite values can be used in computation.
3930 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
3932 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
3933 ComputeShaderSpec spec;
3934 de::Random rnd (deStringHash(group->getName()));
3935 const int numElements = 100;
3936 vector<float> positiveFloats (numElements, 0);
3937 vector<float> negativeFloats (numElements, 0);
3939 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3941 for (size_t ndx = 0; ndx < numElements; ++ndx)
3942 negativeFloats[ndx] = -positiveFloats[ndx];
3945 "OpCapability Shader\n"
3946 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
3947 "OpMemoryModel Logical GLSL450\n"
3948 "OpEntryPoint GLCompute %main \"main\" %id\n"
3949 "OpExecutionMode %main LocalSize 1 1 1\n"
3951 "OpSource GLSL 430\n"
3952 "OpName %main \"main\"\n"
3953 "OpName %id \"gl_GlobalInvocationID\"\n"
3955 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3957 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3959 "%fmat = OpTypeMatrix %fvec3 3\n"
3960 "%ten = OpConstant %u32 10\n"
3961 "%f32arr10 = OpTypeArray %f32 %ten\n"
3962 "%fst = OpTypeStruct %f32 %f32\n"
3964 + string(getComputeAsmInputOutputBuffer()) +
3966 "%id = OpVariable %uvec3ptr Input\n"
3967 "%zero = OpConstant %i32 0\n"
3969 // Create a bunch of null values
3970 "%unull = OpConstantNull %u32\n"
3971 "%fnull = OpConstantNull %f32\n"
3972 "%vnull = OpConstantNull %fvec3\n"
3973 "%mnull = OpConstantNull %fmat\n"
3974 "%anull = OpConstantNull %f32arr10\n"
3975 "%snull = OpConstantComposite %fst %fnull %fnull\n"
3977 "%main = OpFunction %void None %voidf\n"
3978 "%label = OpLabel\n"
3979 "%idval = OpLoad %uvec3 %id\n"
3980 "%x = OpCompositeExtract %u32 %idval 0\n"
3981 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3982 "%inval = OpLoad %f32 %inloc\n"
3983 "%neg = OpFNegate %f32 %inval\n"
3985 // Get the abs() of (a certain element of) those null values
3986 "%unull_cov = OpConvertUToF %f32 %unull\n"
3987 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
3988 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
3989 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
3990 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
3991 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
3992 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
3993 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
3994 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
3995 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
3996 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
3999 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
4000 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
4001 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
4002 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
4003 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
4004 "%final = OpFAdd %f32 %add5 %snull_abs\n"
4006 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4007 " OpStore %outloc %final\n" // write to output
4010 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4011 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4012 spec.numWorkGroups = IVec3(numElements, 1, 1);
4014 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
4016 return group.release();
4019 // Assembly code used for testing loop control is based on GLSL source code:
4022 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4023 // float elements[];
4025 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4026 // float elements[];
4030 // uint x = gl_GlobalInvocationID.x;
4031 // output_data.elements[x] = input_data.elements[x];
4032 // for (uint i = 0; i < 4; ++i)
4033 // output_data.elements[x] += 1.f;
4035 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
4037 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
4038 vector<CaseParameter> cases;
4039 de::Random rnd (deStringHash(group->getName()));
4040 const int numElements = 100;
4041 vector<float> inputFloats (numElements, 0);
4042 vector<float> outputFloats (numElements, 0);
4043 const StringTemplate shaderTemplate (
4044 string(getComputeAsmShaderPreamble()) +
4046 "OpSource GLSL 430\n"
4047 "OpName %main \"main\"\n"
4048 "OpName %id \"gl_GlobalInvocationID\"\n"
4050 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4052 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4054 "%u32ptr = OpTypePointer Function %u32\n"
4056 "%id = OpVariable %uvec3ptr Input\n"
4057 "%zero = OpConstant %i32 0\n"
4058 "%uzero = OpConstant %u32 0\n"
4059 "%one = OpConstant %i32 1\n"
4060 "%constf1 = OpConstant %f32 1.0\n"
4061 "%four = OpConstant %u32 4\n"
4063 "%main = OpFunction %void None %voidf\n"
4064 "%entry = OpLabel\n"
4065 "%i = OpVariable %u32ptr Function\n"
4066 " OpStore %i %uzero\n"
4068 "%idval = OpLoad %uvec3 %id\n"
4069 "%x = OpCompositeExtract %u32 %idval 0\n"
4070 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4071 "%inval = OpLoad %f32 %inloc\n"
4072 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4073 " OpStore %outloc %inval\n"
4074 " OpBranch %loop_entry\n"
4076 "%loop_entry = OpLabel\n"
4077 "%i_val = OpLoad %u32 %i\n"
4078 "%cmp_lt = OpULessThan %bool %i_val %four\n"
4079 " OpLoopMerge %loop_merge %loop_body ${CONTROL}\n"
4080 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
4081 "%loop_body = OpLabel\n"
4082 "%outval = OpLoad %f32 %outloc\n"
4083 "%addf1 = OpFAdd %f32 %outval %constf1\n"
4084 " OpStore %outloc %addf1\n"
4085 "%new_i = OpIAdd %u32 %i_val %one\n"
4086 " OpStore %i %new_i\n"
4087 " OpBranch %loop_entry\n"
4088 "%loop_merge = OpLabel\n"
4090 " OpFunctionEnd\n");
4092 cases.push_back(CaseParameter("none", "None"));
4093 cases.push_back(CaseParameter("unroll", "Unroll"));
4094 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
4095 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
4097 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4099 for (size_t ndx = 0; ndx < numElements; ++ndx)
4100 outputFloats[ndx] = inputFloats[ndx] + 4.f;
4102 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4104 map<string, string> specializations;
4105 ComputeShaderSpec spec;
4107 specializations["CONTROL"] = cases[caseNdx].param;
4108 spec.assembly = shaderTemplate.specialize(specializations);
4109 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4110 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4111 spec.numWorkGroups = IVec3(numElements, 1, 1);
4113 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4116 return group.release();
4119 // Assembly code used for testing selection control is based on GLSL source code:
4122 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4123 // float elements[];
4125 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4126 // float elements[];
4130 // uint x = gl_GlobalInvocationID.x;
4131 // float val = input_data.elements[x];
4133 // output_data.elements[x] = val + 1.f;
4135 // output_data.elements[x] = val - 1.f;
4137 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
4139 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
4140 vector<CaseParameter> cases;
4141 de::Random rnd (deStringHash(group->getName()));
4142 const int numElements = 100;
4143 vector<float> inputFloats (numElements, 0);
4144 vector<float> outputFloats (numElements, 0);
4145 const StringTemplate shaderTemplate (
4146 string(getComputeAsmShaderPreamble()) +
4148 "OpSource GLSL 430\n"
4149 "OpName %main \"main\"\n"
4150 "OpName %id \"gl_GlobalInvocationID\"\n"
4152 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4154 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4156 "%id = OpVariable %uvec3ptr Input\n"
4157 "%zero = OpConstant %i32 0\n"
4158 "%constf1 = OpConstant %f32 1.0\n"
4159 "%constf10 = OpConstant %f32 10.0\n"
4161 "%main = OpFunction %void None %voidf\n"
4162 "%entry = OpLabel\n"
4163 "%idval = OpLoad %uvec3 %id\n"
4164 "%x = OpCompositeExtract %u32 %idval 0\n"
4165 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4166 "%inval = OpLoad %f32 %inloc\n"
4167 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4168 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
4170 " OpSelectionMerge %if_end ${CONTROL}\n"
4171 " OpBranchConditional %cmp_gt %if_true %if_false\n"
4172 "%if_true = OpLabel\n"
4173 "%addf1 = OpFAdd %f32 %inval %constf1\n"
4174 " OpStore %outloc %addf1\n"
4175 " OpBranch %if_end\n"
4176 "%if_false = OpLabel\n"
4177 "%subf1 = OpFSub %f32 %inval %constf1\n"
4178 " OpStore %outloc %subf1\n"
4179 " OpBranch %if_end\n"
4180 "%if_end = OpLabel\n"
4182 " OpFunctionEnd\n");
4184 cases.push_back(CaseParameter("none", "None"));
4185 cases.push_back(CaseParameter("flatten", "Flatten"));
4186 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
4187 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
4189 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4191 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4192 floorAll(inputFloats);
4194 for (size_t ndx = 0; ndx < numElements; ++ndx)
4195 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
4197 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4199 map<string, string> specializations;
4200 ComputeShaderSpec spec;
4202 specializations["CONTROL"] = cases[caseNdx].param;
4203 spec.assembly = shaderTemplate.specialize(specializations);
4204 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4205 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4206 spec.numWorkGroups = IVec3(numElements, 1, 1);
4208 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4211 return group.release();
4214 // Assembly code used for testing function control is based on GLSL source code:
4218 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4219 // float elements[];
4221 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4222 // float elements[];
4225 // float const10() { return 10.f; }
4228 // uint x = gl_GlobalInvocationID.x;
4229 // output_data.elements[x] = input_data.elements[x] + const10();
4231 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
4233 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
4234 vector<CaseParameter> cases;
4235 de::Random rnd (deStringHash(group->getName()));
4236 const int numElements = 100;
4237 vector<float> inputFloats (numElements, 0);
4238 vector<float> outputFloats (numElements, 0);
4239 const StringTemplate shaderTemplate (
4240 string(getComputeAsmShaderPreamble()) +
4242 "OpSource GLSL 430\n"
4243 "OpName %main \"main\"\n"
4244 "OpName %func_const10 \"const10(\"\n"
4245 "OpName %id \"gl_GlobalInvocationID\"\n"
4247 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4249 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4251 "%f32f = OpTypeFunction %f32\n"
4252 "%id = OpVariable %uvec3ptr Input\n"
4253 "%zero = OpConstant %i32 0\n"
4254 "%constf10 = OpConstant %f32 10.0\n"
4256 "%main = OpFunction %void None %voidf\n"
4257 "%entry = OpLabel\n"
4258 "%idval = OpLoad %uvec3 %id\n"
4259 "%x = OpCompositeExtract %u32 %idval 0\n"
4260 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4261 "%inval = OpLoad %f32 %inloc\n"
4262 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
4263 "%fadd = OpFAdd %f32 %inval %ret_10\n"
4264 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4265 " OpStore %outloc %fadd\n"
4269 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
4270 "%label = OpLabel\n"
4271 " OpReturnValue %constf10\n"
4272 " OpFunctionEnd\n");
4274 cases.push_back(CaseParameter("none", "None"));
4275 cases.push_back(CaseParameter("inline", "Inline"));
4276 cases.push_back(CaseParameter("dont_inline", "DontInline"));
4277 cases.push_back(CaseParameter("pure", "Pure"));
4278 cases.push_back(CaseParameter("const", "Const"));
4279 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
4280 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
4281 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
4282 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
4284 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4286 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4287 floorAll(inputFloats);
4289 for (size_t ndx = 0; ndx < numElements; ++ndx)
4290 outputFloats[ndx] = inputFloats[ndx] + 10.f;
4292 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4294 map<string, string> specializations;
4295 ComputeShaderSpec spec;
4297 specializations["CONTROL"] = cases[caseNdx].param;
4298 spec.assembly = shaderTemplate.specialize(specializations);
4299 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4300 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4301 spec.numWorkGroups = IVec3(numElements, 1, 1);
4303 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4306 return group.release();
4309 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
4311 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
4312 vector<CaseParameter> cases;
4313 de::Random rnd (deStringHash(group->getName()));
4314 const int numElements = 100;
4315 vector<float> inputFloats (numElements, 0);
4316 vector<float> outputFloats (numElements, 0);
4317 const StringTemplate shaderTemplate (
4318 string(getComputeAsmShaderPreamble()) +
4320 "OpSource GLSL 430\n"
4321 "OpName %main \"main\"\n"
4322 "OpName %id \"gl_GlobalInvocationID\"\n"
4324 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4326 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4328 "%f32ptr_f = OpTypePointer Function %f32\n"
4330 "%id = OpVariable %uvec3ptr Input\n"
4331 "%zero = OpConstant %i32 0\n"
4332 "%four = OpConstant %i32 4\n"
4334 "%main = OpFunction %void None %voidf\n"
4335 "%label = OpLabel\n"
4336 "%copy = OpVariable %f32ptr_f Function\n"
4337 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
4338 "%x = OpCompositeExtract %u32 %idval 0\n"
4339 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4340 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4341 " OpCopyMemory %copy %inloc ${ACCESS}\n"
4342 "%val1 = OpLoad %f32 %copy\n"
4343 "%val2 = OpLoad %f32 %inloc\n"
4344 "%add = OpFAdd %f32 %val1 %val2\n"
4345 " OpStore %outloc %add ${ACCESS}\n"
4347 " OpFunctionEnd\n");
4349 cases.push_back(CaseParameter("null", ""));
4350 cases.push_back(CaseParameter("none", "None"));
4351 cases.push_back(CaseParameter("volatile", "Volatile"));
4352 cases.push_back(CaseParameter("aligned", "Aligned 4"));
4353 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
4354 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
4355 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
4357 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4359 for (size_t ndx = 0; ndx < numElements; ++ndx)
4360 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
4362 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4364 map<string, string> specializations;
4365 ComputeShaderSpec spec;
4367 specializations["ACCESS"] = cases[caseNdx].param;
4368 spec.assembly = shaderTemplate.specialize(specializations);
4369 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4370 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4371 spec.numWorkGroups = IVec3(numElements, 1, 1);
4373 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4376 return group.release();
4379 // Checks that we can get undefined values for various types, without exercising a computation with it.
4380 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
4382 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
4383 vector<CaseParameter> cases;
4384 de::Random rnd (deStringHash(group->getName()));
4385 const int numElements = 100;
4386 vector<float> positiveFloats (numElements, 0);
4387 vector<float> negativeFloats (numElements, 0);
4388 const StringTemplate shaderTemplate (
4389 string(getComputeAsmShaderPreamble()) +
4391 "OpSource GLSL 430\n"
4392 "OpName %main \"main\"\n"
4393 "OpName %id \"gl_GlobalInvocationID\"\n"
4395 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4397 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4398 "%uvec2 = OpTypeVector %u32 2\n"
4399 "%fvec4 = OpTypeVector %f32 4\n"
4400 "%fmat33 = OpTypeMatrix %fvec3 3\n"
4401 "%image = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
4402 "%sampler = OpTypeSampler\n"
4403 "%simage = OpTypeSampledImage %image\n"
4404 "%const100 = OpConstant %u32 100\n"
4405 "%uarr100 = OpTypeArray %i32 %const100\n"
4406 "%struct = OpTypeStruct %f32 %i32 %u32\n"
4407 "%pointer = OpTypePointer Function %i32\n"
4408 + string(getComputeAsmInputOutputBuffer()) +
4410 "%id = OpVariable %uvec3ptr Input\n"
4411 "%zero = OpConstant %i32 0\n"
4413 "%main = OpFunction %void None %voidf\n"
4414 "%label = OpLabel\n"
4416 "%undef = OpUndef ${TYPE}\n"
4418 "%idval = OpLoad %uvec3 %id\n"
4419 "%x = OpCompositeExtract %u32 %idval 0\n"
4421 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4422 "%inval = OpLoad %f32 %inloc\n"
4423 "%neg = OpFNegate %f32 %inval\n"
4424 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4425 " OpStore %outloc %neg\n"
4427 " OpFunctionEnd\n");
4429 cases.push_back(CaseParameter("bool", "%bool"));
4430 cases.push_back(CaseParameter("sint32", "%i32"));
4431 cases.push_back(CaseParameter("uint32", "%u32"));
4432 cases.push_back(CaseParameter("float32", "%f32"));
4433 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
4434 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
4435 cases.push_back(CaseParameter("matrix", "%fmat33"));
4436 cases.push_back(CaseParameter("image", "%image"));
4437 cases.push_back(CaseParameter("sampler", "%sampler"));
4438 cases.push_back(CaseParameter("sampledimage", "%simage"));
4439 cases.push_back(CaseParameter("array", "%uarr100"));
4440 cases.push_back(CaseParameter("runtimearray", "%f32arr"));
4441 cases.push_back(CaseParameter("struct", "%struct"));
4442 cases.push_back(CaseParameter("pointer", "%pointer"));
4444 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4446 for (size_t ndx = 0; ndx < numElements; ++ndx)
4447 negativeFloats[ndx] = -positiveFloats[ndx];
4449 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4451 map<string, string> specializations;
4452 ComputeShaderSpec spec;
4454 specializations["TYPE"] = cases[caseNdx].param;
4455 spec.assembly = shaderTemplate.specialize(specializations);
4456 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4457 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4458 spec.numWorkGroups = IVec3(numElements, 1, 1);
4460 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4463 return group.release();
4468 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
4470 struct NameCodePair { string name, code; };
4471 RGBA defaultColors[4];
4472 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
4473 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
4474 map<string, string> fragments = passthruFragments();
4475 const NameCodePair tests[] =
4477 {"unknown", "OpSource Unknown 321"},
4478 {"essl", "OpSource ESSL 310"},
4479 {"glsl", "OpSource GLSL 450"},
4480 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
4481 {"opencl_c", "OpSource OpenCL_C 120"},
4482 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
4483 {"file", opsourceGLSLWithFile},
4484 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
4485 // Longest possible source string: SPIR-V limits instructions to 65535
4486 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
4487 // contain 65530 UTF8 characters (one word each) plus one last word
4488 // containing 3 ASCII characters and \0.
4489 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
4492 getDefaultColors(defaultColors);
4493 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
4495 fragments["debug"] = tests[testNdx].code;
4496 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
4499 return opSourceTests.release();
4502 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
4504 struct NameCodePair { string name, code; };
4505 RGBA defaultColors[4];
4506 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
4507 map<string, string> fragments = passthruFragments();
4508 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
4509 const NameCodePair tests[] =
4511 {"empty", opsource + "OpSourceContinued \"\""},
4512 {"short", opsource + "OpSourceContinued \"abcde\""},
4513 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
4514 // Longest possible source string: SPIR-V limits instructions to 65535
4515 // words, of which the first one is OpSourceContinued/length; the rest
4516 // will contain 65533 UTF8 characters (one word each) plus one last word
4517 // containing 3 ASCII characters and \0.
4518 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
4521 getDefaultColors(defaultColors);
4522 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
4524 fragments["debug"] = tests[testNdx].code;
4525 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
4528 return opSourceTests.release();
4531 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
4533 RGBA defaultColors[4];
4534 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
4535 map<string, string> fragments;
4536 getDefaultColors(defaultColors);
4537 fragments["debug"] =
4538 "%name = OpString \"name\"\n";
4540 fragments["pre_main"] =
4543 "OpLine %name 1 1\n"
4545 "OpLine %name 1 1\n"
4546 "OpLine %name 1 1\n"
4547 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
4549 "OpLine %name 1 1\n"
4551 "OpLine %name 1 1\n"
4552 "OpLine %name 1 1\n"
4553 "%second_param1 = OpFunctionParameter %v4f32\n"
4556 "%label_secondfunction = OpLabel\n"
4558 "OpReturnValue %second_param1\n"
4563 fragments["testfun"] =
4564 // A %test_code function that returns its argument unchanged.
4567 "OpLine %name 1 1\n"
4568 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4570 "%param1 = OpFunctionParameter %v4f32\n"
4573 "%label_testfun = OpLabel\n"
4575 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
4576 "OpReturnValue %val1\n"
4578 "OpLine %name 1 1\n"
4581 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
4583 return opLineTests.release();
4587 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
4589 RGBA defaultColors[4];
4590 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
4591 map<string, string> fragments;
4592 std::vector<std::pair<std::string, std::string> > problemStrings;
4594 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
4595 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
4596 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
4597 getDefaultColors(defaultColors);
4599 fragments["debug"] =
4600 "%other_name = OpString \"other_name\"\n";
4602 fragments["pre_main"] =
4603 "OpLine %file_name 32 0\n"
4604 "OpLine %file_name 32 32\n"
4605 "OpLine %file_name 32 40\n"
4606 "OpLine %other_name 32 40\n"
4607 "OpLine %other_name 0 100\n"
4608 "OpLine %other_name 0 4294967295\n"
4609 "OpLine %other_name 4294967295 0\n"
4610 "OpLine %other_name 32 40\n"
4611 "OpLine %file_name 0 0\n"
4612 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
4613 "OpLine %file_name 1 0\n"
4614 "%second_param1 = OpFunctionParameter %v4f32\n"
4615 "OpLine %file_name 1 3\n"
4616 "OpLine %file_name 1 2\n"
4617 "%label_secondfunction = OpLabel\n"
4618 "OpLine %file_name 0 2\n"
4619 "OpReturnValue %second_param1\n"
4621 "OpLine %file_name 0 2\n"
4622 "OpLine %file_name 0 2\n";
4624 fragments["testfun"] =
4625 // A %test_code function that returns its argument unchanged.
4626 "OpLine %file_name 1 0\n"
4627 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4628 "OpLine %file_name 16 330\n"
4629 "%param1 = OpFunctionParameter %v4f32\n"
4630 "OpLine %file_name 14 442\n"
4631 "%label_testfun = OpLabel\n"
4632 "OpLine %file_name 11 1024\n"
4633 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
4634 "OpLine %file_name 2 97\n"
4635 "OpReturnValue %val1\n"
4637 "OpLine %file_name 5 32\n";
4639 for (size_t i = 0; i < problemStrings.size(); ++i)
4641 map<string, string> testFragments = fragments;
4642 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
4643 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
4646 return opLineTests.release();
4649 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
4651 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
4655 const char functionStart[] =
4656 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4657 "%param1 = OpFunctionParameter %v4f32\n"
4660 const char functionEnd[] =
4661 "OpReturnValue %transformed_param\n"
4664 struct NameConstantsCode
4671 NameConstantsCode tests[] =
4675 "%cnull = OpConstantNull %v4f32\n",
4676 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
4680 "%cnull = OpConstantNull %f32\n",
4681 "%vp = OpVariable %fp_v4f32 Function\n"
4682 "%v = OpLoad %v4f32 %vp\n"
4683 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
4684 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
4685 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
4686 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
4687 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
4691 "%cnull = OpConstantNull %bool\n",
4692 "%v = OpVariable %fp_v4f32 Function\n"
4693 " OpStore %v %param1\n"
4694 " OpSelectionMerge %false_label None\n"
4695 " OpBranchConditional %cnull %true_label %false_label\n"
4696 "%true_label = OpLabel\n"
4697 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
4698 " OpBranch %false_label\n"
4699 "%false_label = OpLabel\n"
4700 "%transformed_param = OpLoad %v4f32 %v\n"
4704 "%cnull = OpConstantNull %i32\n",
4705 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
4706 "%b = OpIEqual %bool %cnull %c_i32_0\n"
4707 " OpSelectionMerge %false_label None\n"
4708 " OpBranchConditional %b %true_label %false_label\n"
4709 "%true_label = OpLabel\n"
4710 " OpStore %v %param1\n"
4711 " OpBranch %false_label\n"
4712 "%false_label = OpLabel\n"
4713 "%transformed_param = OpLoad %v4f32 %v\n"
4717 "%stype = OpTypeStruct %f32 %v4f32\n"
4718 "%fp_stype = OpTypePointer Function %stype\n"
4719 "%cnull = OpConstantNull %stype\n",
4720 "%v = OpVariable %fp_stype Function %cnull\n"
4721 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
4722 "%f_val = OpLoad %v4f32 %f\n"
4723 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
4727 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
4728 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
4729 "%cnull = OpConstantNull %a4_v4f32\n",
4730 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
4731 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
4732 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
4733 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
4734 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
4735 "%f_val = OpLoad %v4f32 %f\n"
4736 "%f1_val = OpLoad %v4f32 %f1\n"
4737 "%f2_val = OpLoad %v4f32 %f2\n"
4738 "%f3_val = OpLoad %v4f32 %f3\n"
4739 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
4740 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
4741 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
4742 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
4746 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
4747 "%cnull = OpConstantNull %mat4x4_f32\n",
4748 // Our null matrix * any vector should result in a zero vector.
4749 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
4750 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
4754 getHalfColorsFullAlpha(colors);
4756 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
4758 map<string, string> fragments;
4759 fragments["pre_main"] = tests[testNdx].constants;
4760 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
4761 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
4763 return opConstantNullTests.release();
4765 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
4767 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
4768 RGBA inputColors[4];
4769 RGBA outputColors[4];
4772 const char functionStart[] =
4773 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4774 "%param1 = OpFunctionParameter %v4f32\n"
4777 const char functionEnd[] =
4778 "OpReturnValue %transformed_param\n"
4781 struct NameConstantsCode
4788 NameConstantsCode tests[] =
4793 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
4794 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
4799 "%stype = OpTypeStruct %v4f32 %f32\n"
4800 "%fp_stype = OpTypePointer Function %stype\n"
4801 "%f32_n_1 = OpConstant %f32 -1.0\n"
4802 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
4803 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
4804 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
4806 "%v = OpVariable %fp_stype Function %cval\n"
4807 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
4808 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
4809 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
4810 "%f32_val = OpLoad %f32 %f32_ptr\n"
4811 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
4812 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
4813 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
4816 // [1|0|0|0.5] [x] = x + 0.5
4817 // [0|1|0|0.5] [y] = y + 0.5
4818 // [0|0|1|0.5] [z] = z + 0.5
4819 // [0|0|0|1 ] [1] = 1
4822 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
4823 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
4824 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
4825 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
4826 "%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"
4827 "%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",
4829 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
4834 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4835 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
4836 "%f32_n_1 = OpConstant %f32 -1.0\n"
4837 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
4838 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
4840 "%v = OpVariable %fp_a4f32 Function %carr\n"
4841 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
4842 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
4843 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
4844 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
4845 "%f_val = OpLoad %f32 %f\n"
4846 "%f1_val = OpLoad %f32 %f1\n"
4847 "%f2_val = OpLoad %f32 %f2\n"
4848 "%f3_val = OpLoad %f32 %f3\n"
4849 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
4850 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
4851 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
4852 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
4853 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
4860 // [ 1.0, 1.0, 1.0, 1.0]
4864 // [ 0.0, 0.5, 0.0, 0.0]
4868 // [ 1.0, 1.0, 1.0, 1.0]
4871 "array_of_struct_of_array",
4873 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4874 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
4875 "%stype = OpTypeStruct %f32 %a4f32\n"
4876 "%a3stype = OpTypeArray %stype %c_u32_3\n"
4877 "%fp_a3stype = OpTypePointer Function %a3stype\n"
4878 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
4879 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
4880 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
4881 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
4882 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
4884 "%v = OpVariable %fp_a3stype Function %carr\n"
4885 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
4886 "%f_l = OpLoad %f32 %f\n"
4887 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
4888 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
4892 getHalfColorsFullAlpha(inputColors);
4893 outputColors[0] = RGBA(255, 255, 255, 255);
4894 outputColors[1] = RGBA(255, 127, 127, 255);
4895 outputColors[2] = RGBA(127, 255, 127, 255);
4896 outputColors[3] = RGBA(127, 127, 255, 255);
4898 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
4900 map<string, string> fragments;
4901 fragments["pre_main"] = tests[testNdx].constants;
4902 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
4903 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
4905 return opConstantCompositeTests.release();
4908 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
4910 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
4911 RGBA inputColors[4];
4912 RGBA outputColors[4];
4913 map<string, string> fragments;
4915 // vec4 test_code(vec4 param) {
4916 // vec4 result = param;
4917 // for (int i = 0; i < 4; ++i) {
4918 // if (i == 0) result[i] = 0.;
4919 // else result[i] = 1. - result[i];
4923 const char function[] =
4924 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4925 "%param1 = OpFunctionParameter %v4f32\n"
4927 "%iptr = OpVariable %fp_i32 Function\n"
4928 "%result = OpVariable %fp_v4f32 Function\n"
4929 " OpStore %iptr %c_i32_0\n"
4930 " OpStore %result %param1\n"
4933 // Loop entry block.
4935 "%ival = OpLoad %i32 %iptr\n"
4936 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
4937 " OpLoopMerge %exit %if_entry None\n"
4938 " OpBranchConditional %lt_4 %if_entry %exit\n"
4940 // Merge block for loop.
4942 "%ret = OpLoad %v4f32 %result\n"
4943 " OpReturnValue %ret\n"
4945 // If-statement entry block.
4946 "%if_entry = OpLabel\n"
4947 "%loc = OpAccessChain %fp_f32 %result %ival\n"
4948 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
4949 " OpSelectionMerge %if_exit None\n"
4950 " OpBranchConditional %eq_0 %if_true %if_false\n"
4952 // False branch for if-statement.
4953 "%if_false = OpLabel\n"
4954 "%val = OpLoad %f32 %loc\n"
4955 "%sub = OpFSub %f32 %c_f32_1 %val\n"
4956 " OpStore %loc %sub\n"
4957 " OpBranch %if_exit\n"
4959 // Merge block for if-statement.
4960 "%if_exit = OpLabel\n"
4961 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
4962 " OpStore %iptr %ival_next\n"
4965 // True branch for if-statement.
4966 "%if_true = OpLabel\n"
4967 " OpStore %loc %c_f32_0\n"
4968 " OpBranch %if_exit\n"
4972 fragments["testfun"] = function;
4974 inputColors[0] = RGBA(127, 127, 127, 0);
4975 inputColors[1] = RGBA(127, 0, 0, 0);
4976 inputColors[2] = RGBA(0, 127, 0, 0);
4977 inputColors[3] = RGBA(0, 0, 127, 0);
4979 outputColors[0] = RGBA(0, 128, 128, 255);
4980 outputColors[1] = RGBA(0, 255, 255, 255);
4981 outputColors[2] = RGBA(0, 128, 255, 255);
4982 outputColors[3] = RGBA(0, 255, 128, 255);
4984 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
4986 return group.release();
4989 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
4991 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
4992 RGBA inputColors[4];
4993 RGBA outputColors[4];
4994 map<string, string> fragments;
4996 const char typesAndConstants[] =
4997 "%c_f32_p2 = OpConstant %f32 0.2\n"
4998 "%c_f32_p4 = OpConstant %f32 0.4\n"
4999 "%c_f32_p6 = OpConstant %f32 0.6\n"
5000 "%c_f32_p8 = OpConstant %f32 0.8\n";
5002 // vec4 test_code(vec4 param) {
5003 // vec4 result = param;
5004 // for (int i = 0; i < 4; ++i) {
5006 // case 0: result[i] += .2; break;
5007 // case 1: result[i] += .6; break;
5008 // case 2: result[i] += .4; break;
5009 // case 3: result[i] += .8; break;
5010 // default: break; // unreachable
5015 const char function[] =
5016 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5017 "%param1 = OpFunctionParameter %v4f32\n"
5019 "%iptr = OpVariable %fp_i32 Function\n"
5020 "%result = OpVariable %fp_v4f32 Function\n"
5021 " OpStore %iptr %c_i32_0\n"
5022 " OpStore %result %param1\n"
5025 // Loop entry block.
5027 "%ival = OpLoad %i32 %iptr\n"
5028 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5029 " OpLoopMerge %exit %switch_exit None\n"
5030 " OpBranchConditional %lt_4 %switch_entry %exit\n"
5032 // Merge block for loop.
5034 "%ret = OpLoad %v4f32 %result\n"
5035 " OpReturnValue %ret\n"
5037 // Switch-statement entry block.
5038 "%switch_entry = OpLabel\n"
5039 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5040 "%val = OpLoad %f32 %loc\n"
5041 " OpSelectionMerge %switch_exit None\n"
5042 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5044 "%case2 = OpLabel\n"
5045 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
5046 " OpStore %loc %addp4\n"
5047 " OpBranch %switch_exit\n"
5049 "%switch_default = OpLabel\n"
5052 "%case3 = OpLabel\n"
5053 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
5054 " OpStore %loc %addp8\n"
5055 " OpBranch %switch_exit\n"
5057 "%case0 = OpLabel\n"
5058 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
5059 " OpStore %loc %addp2\n"
5060 " OpBranch %switch_exit\n"
5062 // Merge block for switch-statement.
5063 "%switch_exit = OpLabel\n"
5064 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5065 " OpStore %iptr %ival_next\n"
5068 "%case1 = OpLabel\n"
5069 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
5070 " OpStore %loc %addp6\n"
5071 " OpBranch %switch_exit\n"
5075 fragments["pre_main"] = typesAndConstants;
5076 fragments["testfun"] = function;
5078 inputColors[0] = RGBA(127, 27, 127, 51);
5079 inputColors[1] = RGBA(127, 0, 0, 51);
5080 inputColors[2] = RGBA(0, 27, 0, 51);
5081 inputColors[3] = RGBA(0, 0, 127, 51);
5083 outputColors[0] = RGBA(178, 180, 229, 255);
5084 outputColors[1] = RGBA(178, 153, 102, 255);
5085 outputColors[2] = RGBA(51, 180, 102, 255);
5086 outputColors[3] = RGBA(51, 153, 229, 255);
5088 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5090 return group.release();
5093 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
5095 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
5096 RGBA inputColors[4];
5097 RGBA outputColors[4];
5098 map<string, string> fragments;
5100 const char decorations[] =
5101 "OpDecorate %array_group ArrayStride 4\n"
5102 "OpDecorate %struct_member_group Offset 0\n"
5103 "%array_group = OpDecorationGroup\n"
5104 "%struct_member_group = OpDecorationGroup\n"
5106 "OpDecorate %group1 RelaxedPrecision\n"
5107 "OpDecorate %group3 RelaxedPrecision\n"
5108 "OpDecorate %group3 Invariant\n"
5109 "OpDecorate %group3 Restrict\n"
5110 "%group0 = OpDecorationGroup\n"
5111 "%group1 = OpDecorationGroup\n"
5112 "%group3 = OpDecorationGroup\n";
5114 const char typesAndConstants[] =
5115 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
5116 "%struct1 = OpTypeStruct %a3f32\n"
5117 "%struct2 = OpTypeStruct %a3f32\n"
5118 "%fp_struct1 = OpTypePointer Function %struct1\n"
5119 "%fp_struct2 = OpTypePointer Function %struct2\n"
5120 "%c_f32_2 = OpConstant %f32 2.\n"
5121 "%c_f32_n2 = OpConstant %f32 -2.\n"
5123 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
5124 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
5125 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
5126 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
5128 const char function[] =
5129 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5130 "%param = OpFunctionParameter %v4f32\n"
5131 "%entry = OpLabel\n"
5132 "%result = OpVariable %fp_v4f32 Function\n"
5133 "%v_struct1 = OpVariable %fp_struct1 Function\n"
5134 "%v_struct2 = OpVariable %fp_struct2 Function\n"
5135 " OpStore %result %param\n"
5136 " OpStore %v_struct1 %c_struct1\n"
5137 " OpStore %v_struct2 %c_struct2\n"
5138 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
5139 "%val1 = OpLoad %f32 %ptr1\n"
5140 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
5141 "%val2 = OpLoad %f32 %ptr2\n"
5142 "%addvalues = OpFAdd %f32 %val1 %val2\n"
5143 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
5144 "%val = OpLoad %f32 %ptr\n"
5145 "%addresult = OpFAdd %f32 %addvalues %val\n"
5146 " OpStore %ptr %addresult\n"
5147 "%ret = OpLoad %v4f32 %result\n"
5148 " OpReturnValue %ret\n"
5151 struct CaseNameDecoration
5157 CaseNameDecoration tests[] =
5160 "same_decoration_group_on_multiple_types",
5161 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
5164 "empty_decoration_group",
5165 "OpGroupDecorate %group0 %a3f32\n"
5166 "OpGroupDecorate %group0 %result\n"
5169 "one_element_decoration_group",
5170 "OpGroupDecorate %array_group %a3f32\n"
5173 "multiple_elements_decoration_group",
5174 "OpGroupDecorate %group3 %v_struct1\n"
5177 "multiple_decoration_groups_on_same_variable",
5178 "OpGroupDecorate %group0 %v_struct2\n"
5179 "OpGroupDecorate %group1 %v_struct2\n"
5180 "OpGroupDecorate %group3 %v_struct2\n"
5183 "same_decoration_group_multiple_times",
5184 "OpGroupDecorate %group1 %addvalues\n"
5185 "OpGroupDecorate %group1 %addvalues\n"
5186 "OpGroupDecorate %group1 %addvalues\n"
5191 getHalfColorsFullAlpha(inputColors);
5192 getHalfColorsFullAlpha(outputColors);
5194 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
5196 fragments["decoration"] = decorations + tests[idx].decoration;
5197 fragments["pre_main"] = typesAndConstants;
5198 fragments["testfun"] = function;
5200 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
5203 return group.release();
5206 struct SpecConstantTwoIntGraphicsCase
5208 const char* caseName;
5209 const char* scDefinition0;
5210 const char* scDefinition1;
5211 const char* scResultType;
5212 const char* scOperation;
5213 deInt32 scActualValue0;
5214 deInt32 scActualValue1;
5215 const char* resultOperation;
5216 RGBA expectedColors[4];
5218 SpecConstantTwoIntGraphicsCase (const char* name,
5219 const char* definition0,
5220 const char* definition1,
5221 const char* resultType,
5222 const char* operation,
5225 const char* resultOp,
5226 const RGBA (&output)[4])
5228 , scDefinition0 (definition0)
5229 , scDefinition1 (definition1)
5230 , scResultType (resultType)
5231 , scOperation (operation)
5232 , scActualValue0 (value0)
5233 , scActualValue1 (value1)
5234 , resultOperation (resultOp)
5236 expectedColors[0] = output[0];
5237 expectedColors[1] = output[1];
5238 expectedColors[2] = output[2];
5239 expectedColors[3] = output[3];
5243 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
5245 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
5246 vector<SpecConstantTwoIntGraphicsCase> cases;
5247 RGBA inputColors[4];
5248 RGBA outputColors0[4];
5249 RGBA outputColors1[4];
5250 RGBA outputColors2[4];
5252 const char decorations1[] =
5253 "OpDecorate %sc_0 SpecId 0\n"
5254 "OpDecorate %sc_1 SpecId 1\n";
5256 const char typesAndConstants1[] =
5257 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
5258 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
5259 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
5261 const char function1[] =
5262 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5263 "%param = OpFunctionParameter %v4f32\n"
5264 "%label = OpLabel\n"
5265 "%result = OpVariable %fp_v4f32 Function\n"
5266 " OpStore %result %param\n"
5267 "%gen = ${GEN_RESULT}\n"
5268 "%index = OpIAdd %i32 %gen %c_i32_1\n"
5269 "%loc = OpAccessChain %fp_f32 %result %index\n"
5270 "%val = OpLoad %f32 %loc\n"
5271 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
5272 " OpStore %loc %add\n"
5273 "%ret = OpLoad %v4f32 %result\n"
5274 " OpReturnValue %ret\n"
5277 inputColors[0] = RGBA(127, 127, 127, 255);
5278 inputColors[1] = RGBA(127, 0, 0, 255);
5279 inputColors[2] = RGBA(0, 127, 0, 255);
5280 inputColors[3] = RGBA(0, 0, 127, 255);
5282 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
5283 outputColors0[0] = RGBA(255, 127, 127, 255);
5284 outputColors0[1] = RGBA(255, 0, 0, 255);
5285 outputColors0[2] = RGBA(128, 127, 0, 255);
5286 outputColors0[3] = RGBA(128, 0, 127, 255);
5288 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
5289 outputColors1[0] = RGBA(127, 255, 127, 255);
5290 outputColors1[1] = RGBA(127, 128, 0, 255);
5291 outputColors1[2] = RGBA(0, 255, 0, 255);
5292 outputColors1[3] = RGBA(0, 128, 127, 255);
5294 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
5295 outputColors2[0] = RGBA(127, 127, 255, 255);
5296 outputColors2[1] = RGBA(127, 0, 128, 255);
5297 outputColors2[2] = RGBA(0, 127, 128, 255);
5298 outputColors2[3] = RGBA(0, 0, 255, 255);
5300 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
5301 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
5302 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
5304 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
5305 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
5306 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
5307 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
5308 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
5309 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
5310 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
5311 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
5312 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
5313 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
5314 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
5315 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
5316 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
5317 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
5318 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
5319 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
5320 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
5321 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
5322 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
5323 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
5324 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
5325 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
5326 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
5327 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
5328 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
5329 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
5330 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
5331 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
5332 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
5333 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
5334 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
5335 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
5336 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
5338 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5340 map<string, string> specializations;
5341 map<string, string> fragments;
5342 vector<deInt32> specConstants;
5344 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
5345 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
5346 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
5347 specializations["SC_OP"] = cases[caseNdx].scOperation;
5348 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
5350 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
5351 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
5352 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
5354 specConstants.push_back(cases[caseNdx].scActualValue0);
5355 specConstants.push_back(cases[caseNdx].scActualValue1);
5357 createTestsForAllStages(cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants, group.get());
5360 const char decorations2[] =
5361 "OpDecorate %sc_0 SpecId 0\n"
5362 "OpDecorate %sc_1 SpecId 1\n"
5363 "OpDecorate %sc_2 SpecId 2\n";
5365 const char typesAndConstants2[] =
5366 "%v3i32 = OpTypeVector %i32 3\n"
5367 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
5368 "%vec3_undef = OpUndef %v3i32\n"
5370 "%sc_0 = OpSpecConstant %i32 0\n"
5371 "%sc_1 = OpSpecConstant %i32 0\n"
5372 "%sc_2 = OpSpecConstant %i32 0\n"
5373 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
5374 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
5375 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
5376 "%sc_vec3_0_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
5377 "%sc_vec3_1_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
5378 "%sc_vec3_2_s = OpSpecConstantOp %v3i32 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
5379 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
5380 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
5381 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
5382 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
5383 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
5384 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
5385 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
5387 const char function2[] =
5388 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5389 "%param = OpFunctionParameter %v4f32\n"
5390 "%label = OpLabel\n"
5391 "%result = OpVariable %fp_v4f32 Function\n"
5392 " OpStore %result %param\n"
5393 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
5394 "%val = OpLoad %f32 %loc\n"
5395 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
5396 " OpStore %loc %add\n"
5397 "%ret = OpLoad %v4f32 %result\n"
5398 " OpReturnValue %ret\n"
5401 map<string, string> fragments;
5402 vector<deInt32> specConstants;
5404 fragments["decoration"] = decorations2;
5405 fragments["pre_main"] = typesAndConstants2;
5406 fragments["testfun"] = function2;
5408 specConstants.push_back(56789);
5409 specConstants.push_back(-2);
5410 specConstants.push_back(56788);
5412 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
5414 return group.release();
5417 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
5419 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
5420 RGBA inputColors[4];
5421 RGBA outputColors1[4];
5422 RGBA outputColors2[4];
5423 RGBA outputColors3[4];
5424 map<string, string> fragments1;
5425 map<string, string> fragments2;
5426 map<string, string> fragments3;
5428 const char typesAndConstants1[] =
5429 "%c_f32_p2 = OpConstant %f32 0.2\n"
5430 "%c_f32_p4 = OpConstant %f32 0.4\n"
5431 "%c_f32_p5 = OpConstant %f32 0.5\n"
5432 "%c_f32_p8 = OpConstant %f32 0.8\n";
5434 // vec4 test_code(vec4 param) {
5435 // vec4 result = param;
5436 // for (int i = 0; i < 4; ++i) {
5439 // case 0: operand = .2; break;
5440 // case 1: operand = .5; break;
5441 // case 2: operand = .4; break;
5442 // case 3: operand = .0; break;
5443 // default: break; // unreachable
5445 // result[i] += operand;
5449 const char function1[] =
5450 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5451 "%param1 = OpFunctionParameter %v4f32\n"
5453 "%iptr = OpVariable %fp_i32 Function\n"
5454 "%result = OpVariable %fp_v4f32 Function\n"
5455 " OpStore %iptr %c_i32_0\n"
5456 " OpStore %result %param1\n"
5460 "%ival = OpLoad %i32 %iptr\n"
5461 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5462 " OpLoopMerge %exit %phi None\n"
5463 " OpBranchConditional %lt_4 %entry %exit\n"
5465 "%entry = OpLabel\n"
5466 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5467 "%val = OpLoad %f32 %loc\n"
5468 " OpSelectionMerge %phi None\n"
5469 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5471 "%case0 = OpLabel\n"
5473 "%case1 = OpLabel\n"
5475 "%case2 = OpLabel\n"
5477 "%case3 = OpLabel\n"
5480 "%default = OpLabel\n"
5484 "%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
5485 "%add = OpFAdd %f32 %val %operand\n"
5486 " OpStore %loc %add\n"
5487 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5488 " OpStore %iptr %ival_next\n"
5492 "%ret = OpLoad %v4f32 %result\n"
5493 " OpReturnValue %ret\n"
5497 fragments1["pre_main"] = typesAndConstants1;
5498 fragments1["testfun"] = function1;
5500 getHalfColorsFullAlpha(inputColors);
5502 outputColors1[0] = RGBA(178, 255, 229, 255);
5503 outputColors1[1] = RGBA(178, 127, 102, 255);
5504 outputColors1[2] = RGBA(51, 255, 102, 255);
5505 outputColors1[3] = RGBA(51, 127, 229, 255);
5507 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
5509 const char typesAndConstants2[] =
5510 "%c_f32_p2 = OpConstant %f32 0.2\n";
5512 // Add .4 to the second element of the given parameter.
5513 const char function2[] =
5514 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5515 "%param = OpFunctionParameter %v4f32\n"
5516 "%entry = OpLabel\n"
5517 "%result = OpVariable %fp_v4f32 Function\n"
5518 " OpStore %result %param\n"
5519 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
5520 "%val = OpLoad %f32 %loc\n"
5524 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
5525 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
5526 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
5527 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
5528 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
5529 " OpLoopMerge %exit %phi None\n"
5530 " OpBranchConditional %still_loop %phi %exit\n"
5533 " OpStore %loc %accum\n"
5534 "%ret = OpLoad %v4f32 %result\n"
5535 " OpReturnValue %ret\n"
5539 fragments2["pre_main"] = typesAndConstants2;
5540 fragments2["testfun"] = function2;
5542 outputColors2[0] = RGBA(127, 229, 127, 255);
5543 outputColors2[1] = RGBA(127, 102, 0, 255);
5544 outputColors2[2] = RGBA(0, 229, 0, 255);
5545 outputColors2[3] = RGBA(0, 102, 127, 255);
5547 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
5549 const char typesAndConstants3[] =
5550 "%true = OpConstantTrue %bool\n"
5551 "%false = OpConstantFalse %bool\n"
5552 "%c_f32_p2 = OpConstant %f32 0.2\n";
5554 // Swap the second and the third element of the given parameter.
5555 const char function3[] =
5556 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5557 "%param = OpFunctionParameter %v4f32\n"
5558 "%entry = OpLabel\n"
5559 "%result = OpVariable %fp_v4f32 Function\n"
5560 " OpStore %result %param\n"
5561 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
5562 "%a_init = OpLoad %f32 %a_loc\n"
5563 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
5564 "%b_init = OpLoad %f32 %b_loc\n"
5568 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
5569 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
5570 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
5571 " OpLoopMerge %exit %phi None\n"
5572 " OpBranchConditional %still_loop %phi %exit\n"
5575 " OpStore %a_loc %a_next\n"
5576 " OpStore %b_loc %b_next\n"
5577 "%ret = OpLoad %v4f32 %result\n"
5578 " OpReturnValue %ret\n"
5582 fragments3["pre_main"] = typesAndConstants3;
5583 fragments3["testfun"] = function3;
5585 outputColors3[0] = RGBA(127, 127, 127, 255);
5586 outputColors3[1] = RGBA(127, 0, 0, 255);
5587 outputColors3[2] = RGBA(0, 0, 127, 255);
5588 outputColors3[3] = RGBA(0, 127, 0, 255);
5590 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
5592 return group.release();
5595 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
5597 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
5598 RGBA inputColors[4];
5599 RGBA outputColors[4];
5601 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
5602 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
5603 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
5604 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
5605 const char constantsAndTypes[] =
5606 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
5607 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5608 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
5609 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
5610 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n"
5613 const char function[] =
5614 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5615 "%param = OpFunctionParameter %v4f32\n"
5616 "%label = OpLabel\n"
5617 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
5618 "%var2 = OpVariable %fp_f32 Function\n"
5619 "%red = OpCompositeExtract %f32 %param 0\n"
5620 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
5621 " OpStore %var2 %plus_red\n"
5622 "%val1 = OpLoad %f32 %var1\n"
5623 "%val2 = OpLoad %f32 %var2\n"
5624 "%mul = OpFMul %f32 %val1 %val2\n"
5625 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
5626 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
5627 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
5628 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
5629 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
5630 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
5631 " OpReturnValue %ret\n"
5634 struct CaseNameDecoration
5641 CaseNameDecoration tests[] = {
5642 {"multiplication", "OpDecorate %mul NoContraction"},
5643 {"addition", "OpDecorate %add NoContraction"},
5644 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
5647 getHalfColorsFullAlpha(inputColors);
5649 for (deUint8 idx = 0; idx < 4; ++idx)
5651 inputColors[idx].setRed(0);
5652 outputColors[idx] = RGBA(0, 0, 0, 255);
5655 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
5657 map<string, string> fragments;
5659 fragments["decoration"] = tests[testNdx].decoration;
5660 fragments["pre_main"] = constantsAndTypes;
5661 fragments["testfun"] = function;
5663 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
5666 return group.release();
5669 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
5671 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
5674 const char constantsAndTypes[] =
5675 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
5676 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
5677 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
5678 "%fp_stype = OpTypePointer Function %stype\n";
5680 const char function[] =
5681 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5682 "%param1 = OpFunctionParameter %v4f32\n"
5684 "%v1 = OpVariable %fp_v4f32 Function\n"
5685 "%v2 = OpVariable %fp_a2f32 Function\n"
5686 "%v3 = OpVariable %fp_f32 Function\n"
5687 "%v = OpVariable %fp_stype Function\n"
5688 "%vv = OpVariable %fp_stype Function\n"
5689 "%vvv = OpVariable %fp_f32 Function\n"
5691 " OpStore %v1 %c_v4f32_1_1_1_1\n"
5692 " OpStore %v2 %c_a2f32_1\n"
5693 " OpStore %v3 %c_f32_1\n"
5695 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5696 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
5697 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5698 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
5699 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
5700 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
5702 " OpStore %p_v4f32 %v1_v ${access_type}\n"
5703 " OpStore %p_a2f32 %v2_v ${access_type}\n"
5704 " OpStore %p_f32 %v3_v ${access_type}\n"
5706 " OpCopyMemory %vv %v ${access_type}\n"
5707 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
5709 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
5710 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
5711 "%v_f32_3 = OpLoad %f32 %vvv\n"
5713 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
5714 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
5715 " OpReturnValue %ret2\n"
5718 struct NameMemoryAccess
5725 NameMemoryAccess tests[] =
5728 { "volatile", "Volatile" },
5729 { "aligned", "Aligned 1" },
5730 { "volatile_aligned", "Volatile|Aligned 1" },
5731 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
5732 { "volatile_nontemporal", "Volatile|Nontemporal" },
5733 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
5736 getHalfColorsFullAlpha(colors);
5738 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
5740 map<string, string> fragments;
5741 map<string, string> memoryAccess;
5742 memoryAccess["access_type"] = tests[testNdx].accessType;
5744 fragments["pre_main"] = constantsAndTypes;
5745 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
5746 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
5748 return memoryAccessTests.release();
5750 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
5752 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
5753 RGBA defaultColors[4];
5754 map<string, string> fragments;
5755 getDefaultColors(defaultColors);
5757 // First, simple cases that don't do anything with the OpUndef result.
5758 struct NameCodePair { string name, decl, type; };
5759 const NameCodePair tests[] =
5761 {"bool", "", "%bool"},
5762 {"vec2uint32", "%type = OpTypeVector %u32 2", "%type"},
5763 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown", "%type"},
5764 {"sampler", "%type = OpTypeSampler", "%type"},
5765 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img", "%type"},
5766 {"pointer", "", "%fp_i32"},
5767 {"runtimearray", "%type = OpTypeRuntimeArray %f32", "%type"},
5768 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100", "%type"},
5769 {"struct", "%type = OpTypeStruct %f32 %i32 %u32", "%type"}};
5770 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5772 fragments["undef_type"] = tests[testNdx].type;
5773 fragments["testfun"] = StringTemplate(
5774 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5775 "%param1 = OpFunctionParameter %v4f32\n"
5776 "%label_testfun = OpLabel\n"
5777 "%undef = OpUndef ${undef_type}\n"
5778 "OpReturnValue %param1\n"
5779 "OpFunctionEnd\n").specialize(fragments);
5780 fragments["pre_main"] = tests[testNdx].decl;
5781 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
5785 fragments["testfun"] =
5786 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5787 "%param1 = OpFunctionParameter %v4f32\n"
5788 "%label_testfun = OpLabel\n"
5789 "%undef = OpUndef %f32\n"
5790 "%zero = OpFMul %f32 %undef %c_f32_0\n"
5791 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
5792 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
5793 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5794 "%b = OpFAdd %f32 %a %actually_zero\n"
5795 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
5796 "OpReturnValue %ret\n"
5799 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
5801 fragments["testfun"] =
5802 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5803 "%param1 = OpFunctionParameter %v4f32\n"
5804 "%label_testfun = OpLabel\n"
5805 "%undef = OpUndef %i32\n"
5806 "%zero = OpIMul %i32 %undef %c_i32_0\n"
5807 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
5808 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
5809 "OpReturnValue %ret\n"
5812 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
5814 fragments["testfun"] =
5815 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5816 "%param1 = OpFunctionParameter %v4f32\n"
5817 "%label_testfun = OpLabel\n"
5818 "%undef = OpUndef %u32\n"
5819 "%zero = OpIMul %u32 %undef %c_i32_0\n"
5820 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
5821 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
5822 "OpReturnValue %ret\n"
5825 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
5827 fragments["testfun"] =
5828 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5829 "%param1 = OpFunctionParameter %v4f32\n"
5830 "%label_testfun = OpLabel\n"
5831 "%undef = OpUndef %v4f32\n"
5832 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
5833 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
5834 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
5835 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
5836 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
5837 "%is_nan_0 = OpIsNan %bool %zero_0\n"
5838 "%is_nan_1 = OpIsNan %bool %zero_1\n"
5839 "%is_nan_2 = OpIsNan %bool %zero_2\n"
5840 "%is_nan_3 = OpIsNan %bool %zero_3\n"
5841 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
5842 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
5843 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
5844 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
5845 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5846 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
5847 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
5848 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
5849 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
5850 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
5851 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
5852 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
5853 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
5854 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
5855 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
5856 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
5857 "OpReturnValue %ret\n"
5860 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
5862 fragments["pre_main"] =
5863 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
5864 fragments["testfun"] =
5865 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5866 "%param1 = OpFunctionParameter %v4f32\n"
5867 "%label_testfun = OpLabel\n"
5868 "%undef = OpUndef %m2x2f32\n"
5869 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
5870 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
5871 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
5872 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
5873 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
5874 "%is_nan_0 = OpIsNan %bool %zero_0\n"
5875 "%is_nan_1 = OpIsNan %bool %zero_1\n"
5876 "%is_nan_2 = OpIsNan %bool %zero_2\n"
5877 "%is_nan_3 = OpIsNan %bool %zero_3\n"
5878 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
5879 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
5880 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
5881 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
5882 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5883 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
5884 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
5885 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
5886 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
5887 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
5888 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
5889 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
5890 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
5891 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
5892 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
5893 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
5894 "OpReturnValue %ret\n"
5897 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
5899 return opUndefTests.release();
5902 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
5904 const RGBA inputColors[4] =
5907 RGBA(0, 0, 255, 255),
5908 RGBA(0, 255, 0, 255),
5909 RGBA(0, 255, 255, 255)
5912 const RGBA expectedColors[4] =
5914 RGBA(255, 0, 0, 255),
5915 RGBA(255, 0, 0, 255),
5916 RGBA(255, 0, 0, 255),
5917 RGBA(255, 0, 0, 255)
5920 const struct SingleFP16Possibility
5923 const char* constant; // Value to assign to %test_constant.
5925 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
5931 -constructNormalizedFloat(1, 0x300000),
5932 "%cond = OpFOrdEqual %bool %c %test_constant\n"
5937 constructNormalizedFloat(7, 0x000000),
5938 "%cond = OpFOrdEqual %bool %c %test_constant\n"
5940 // SPIR-V requires that OpQuantizeToF16 flushes
5941 // any numbers that would end up denormalized in F16 to zero.
5945 std::ldexp(1.5f, -140),
5946 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5951 -std::ldexp(1.5f, -140),
5952 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5957 std::ldexp(1.0f, -16),
5958 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5959 }, // too small positive
5961 "negative_too_small",
5963 -std::ldexp(1.0f, -32),
5964 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5965 }, // too small negative
5969 -std::ldexp(1.0f, 128),
5971 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
5972 "%inf = OpIsInf %bool %c\n"
5973 "%cond = OpLogicalAnd %bool %gz %inf\n"
5978 std::ldexp(1.0f, 128),
5980 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
5981 "%inf = OpIsInf %bool %c\n"
5982 "%cond = OpLogicalAnd %bool %gz %inf\n"
5985 "round_to_negative_inf",
5987 -std::ldexp(1.0f, 32),
5989 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
5990 "%inf = OpIsInf %bool %c\n"
5991 "%cond = OpLogicalAnd %bool %gz %inf\n"
5996 std::ldexp(1.0f, 16),
5998 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
5999 "%inf = OpIsInf %bool %c\n"
6000 "%cond = OpLogicalAnd %bool %gz %inf\n"
6005 std::numeric_limits<float>::quiet_NaN(),
6007 // Test for any NaN value, as NaNs are not preserved
6008 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6009 "%cond = OpIsNan %bool %direct_quant\n"
6014 std::numeric_limits<float>::quiet_NaN(),
6016 // Test for any NaN value, as NaNs are not preserved
6017 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6018 "%cond = OpIsNan %bool %direct_quant\n"
6021 const char* constants =
6022 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
6024 StringTemplate function (
6025 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6026 "%param1 = OpFunctionParameter %v4f32\n"
6027 "%label_testfun = OpLabel\n"
6028 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6029 "%b = OpFAdd %f32 %test_constant %a\n"
6030 "%c = OpQuantizeToF16 %f32 %b\n"
6032 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6033 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
6034 " OpReturnValue %retval\n"
6038 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
6039 const char* specConstants =
6040 "%test_constant = OpSpecConstant %f32 0.\n"
6041 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
6043 StringTemplate specConstantFunction(
6044 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6045 "%param1 = OpFunctionParameter %v4f32\n"
6046 "%label_testfun = OpLabel\n"
6048 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6049 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
6050 " OpReturnValue %retval\n"
6054 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6056 map<string, string> codeSpecialization;
6057 map<string, string> fragments;
6058 codeSpecialization["condition"] = tests[idx].condition;
6059 fragments["testfun"] = function.specialize(codeSpecialization);
6060 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
6061 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6064 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6066 map<string, string> codeSpecialization;
6067 map<string, string> fragments;
6068 vector<deInt32> passConstants;
6069 deInt32 specConstant;
6071 codeSpecialization["condition"] = tests[idx].condition;
6072 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
6073 fragments["decoration"] = specDecorations;
6074 fragments["pre_main"] = specConstants;
6076 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
6077 passConstants.push_back(specConstant);
6079 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6083 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
6085 RGBA inputColors[4] = {
6087 RGBA(0, 0, 255, 255),
6088 RGBA(0, 255, 0, 255),
6089 RGBA(0, 255, 255, 255)
6092 RGBA expectedColors[4] =
6094 RGBA(255, 0, 0, 255),
6095 RGBA(255, 0, 0, 255),
6096 RGBA(255, 0, 0, 255),
6097 RGBA(255, 0, 0, 255)
6100 struct DualFP16Possibility
6105 const char* possibleOutput1;
6106 const char* possibleOutput2;
6109 "positive_round_up_or_round_down",
6111 constructNormalizedFloat(8, 0x300300),
6116 "negative_round_up_or_round_down",
6118 -constructNormalizedFloat(-7, 0x600800),
6125 constructNormalizedFloat(2, 0x01e000),
6130 "carry_to_exponent",
6132 constructNormalizedFloat(1, 0xffe000),
6137 StringTemplate constants (
6138 "%input_const = OpConstant %f32 ${input}\n"
6139 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6140 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6143 StringTemplate specConstants (
6144 "%input_const = OpSpecConstant %f32 0.\n"
6145 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6146 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6149 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
6151 const char* function =
6152 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6153 "%param1 = OpFunctionParameter %v4f32\n"
6154 "%label_testfun = OpLabel\n"
6155 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6156 // For the purposes of this test we assume that 0.f will always get
6157 // faithfully passed through the pipeline stages.
6158 "%b = OpFAdd %f32 %input_const %a\n"
6159 "%c = OpQuantizeToF16 %f32 %b\n"
6160 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
6161 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
6162 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
6163 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6164 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
6165 " OpReturnValue %retval\n"
6168 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6169 map<string, string> fragments;
6170 map<string, string> constantSpecialization;
6172 constantSpecialization["input"] = tests[idx].input;
6173 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6174 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6175 fragments["testfun"] = function;
6176 fragments["pre_main"] = constants.specialize(constantSpecialization);
6177 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6180 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6181 map<string, string> fragments;
6182 map<string, string> constantSpecialization;
6183 vector<deInt32> passConstants;
6184 deInt32 specConstant;
6186 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6187 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6188 fragments["testfun"] = function;
6189 fragments["decoration"] = specDecorations;
6190 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
6192 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
6193 passConstants.push_back(specConstant);
6195 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6199 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
6201 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
6202 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
6203 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
6204 return opQuantizeTests.release();
6207 struct ShaderPermutation
6209 deUint8 vertexPermutation;
6210 deUint8 geometryPermutation;
6211 deUint8 tesscPermutation;
6212 deUint8 tessePermutation;
6213 deUint8 fragmentPermutation;
6216 ShaderPermutation getShaderPermutation(deUint8 inputValue)
6218 ShaderPermutation permutation =
6220 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
6221 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
6222 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
6223 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
6224 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
6229 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
6231 RGBA defaultColors[4];
6232 RGBA invertedColors[4];
6233 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
6235 const ShaderElement combinedPipeline[] =
6237 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
6238 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
6239 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
6240 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
6241 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
6244 getDefaultColors(defaultColors);
6245 getInvertedDefaultColors(invertedColors);
6246 addFunctionCaseWithPrograms<InstanceContext>(
6247 moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline,
6248 createInstanceContext(combinedPipeline, map<string, string>()));
6250 const char* numbers[] =
6255 for (deInt8 idx = 0; idx < 32; ++idx)
6257 ShaderPermutation permutation = getShaderPermutation(idx);
6258 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
6259 const ShaderElement pipeline[] =
6261 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
6262 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
6263 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
6264 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
6265 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
6268 // If there are an even number of swaps, then it should be no-op.
6269 // If there are an odd number, the color should be flipped.
6270 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
6272 addFunctionCaseWithPrograms<InstanceContext>(
6273 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
6274 createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
6278 addFunctionCaseWithPrograms<InstanceContext>(
6279 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
6280 createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
6283 return moduleTests.release();
6286 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
6288 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
6289 RGBA defaultColors[4];
6290 getDefaultColors(defaultColors);
6291 map<string, string> fragments;
6292 fragments["pre_main"] =
6293 "%c_f32_5 = OpConstant %f32 5.\n";
6295 // A loop with a single block. The Continue Target is the loop block
6296 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
6297 // -- the "continue construct" forms the entire loop.
6298 fragments["testfun"] =
6299 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6300 "%param1 = OpFunctionParameter %v4f32\n"
6302 "%entry = OpLabel\n"
6303 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6306 ";adds and subtracts 1.0 to %val in alternate iterations\n"
6308 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
6309 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
6310 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
6311 "%val = OpFAdd %f32 %val1 %delta\n"
6312 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
6313 "%count__ = OpISub %i32 %count %c_i32_1\n"
6314 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6315 "OpLoopMerge %exit %loop None\n"
6316 "OpBranchConditional %again %loop %exit\n"
6319 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
6320 "OpReturnValue %result\n"
6324 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
6326 // Body comprised of multiple basic blocks.
6327 const StringTemplate multiBlock(
6328 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6329 "%param1 = OpFunctionParameter %v4f32\n"
6331 "%entry = OpLabel\n"
6332 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6335 ";adds and subtracts 1.0 to %val in alternate iterations\n"
6337 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
6338 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
6339 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
6340 // There are several possibilities for the Continue Target below. Each
6341 // will be specialized into a separate test case.
6342 "OpLoopMerge %exit ${continue_target} None\n"
6346 ";delta_next = (delta > 0) ? -1 : 1;\n"
6347 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
6348 "OpSelectionMerge %gather DontFlatten\n"
6349 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
6352 "OpBranch %gather\n"
6355 "OpBranch %gather\n"
6357 "%gather = OpLabel\n"
6358 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
6359 "%val = OpFAdd %f32 %val1 %delta\n"
6360 "%count__ = OpISub %i32 %count %c_i32_1\n"
6361 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6362 "OpBranchConditional %again %loop %exit\n"
6365 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
6366 "OpReturnValue %result\n"
6370 map<string, string> continue_target;
6372 // The Continue Target is the loop block itself.
6373 continue_target["continue_target"] = "%loop";
6374 fragments["testfun"] = multiBlock.specialize(continue_target);
6375 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
6377 // The Continue Target is at the end of the loop.
6378 continue_target["continue_target"] = "%gather";
6379 fragments["testfun"] = multiBlock.specialize(continue_target);
6380 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
6382 // A loop with continue statement.
6383 fragments["testfun"] =
6384 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6385 "%param1 = OpFunctionParameter %v4f32\n"
6387 "%entry = OpLabel\n"
6388 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6391 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
6393 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6394 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
6395 "OpLoopMerge %exit %continue None\n"
6399 ";skip if %count==2\n"
6400 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
6401 "OpSelectionMerge %continue DontFlatten\n"
6402 "OpBranchConditional %eq2 %continue %body\n"
6405 "%fcount = OpConvertSToF %f32 %count\n"
6406 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6407 "OpBranch %continue\n"
6409 "%continue = OpLabel\n"
6410 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
6411 "%count__ = OpISub %i32 %count %c_i32_1\n"
6412 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6413 "OpBranchConditional %again %loop %exit\n"
6416 "%same = OpFSub %f32 %val %c_f32_8\n"
6417 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6418 "OpReturnValue %result\n"
6420 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
6422 // A loop with break.
6423 fragments["testfun"] =
6424 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6425 "%param1 = OpFunctionParameter %v4f32\n"
6427 "%entry = OpLabel\n"
6428 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6429 "%dot = OpDot %f32 %param1 %param1\n"
6430 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6431 "%zero = OpConvertFToU %u32 %div\n"
6432 "%two = OpIAdd %i32 %zero %c_i32_2\n"
6433 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6436 ";adds 4 and 3 to %val0 (exits early)\n"
6438 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6439 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
6440 "OpLoopMerge %exit %continue None\n"
6444 ";end loop if %count==%two\n"
6445 "%above2 = OpSGreaterThan %bool %count %two\n"
6446 "OpSelectionMerge %continue DontFlatten\n"
6447 "OpBranchConditional %above2 %body %exit\n"
6450 "%fcount = OpConvertSToF %f32 %count\n"
6451 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6452 "OpBranch %continue\n"
6454 "%continue = OpLabel\n"
6455 "%count__ = OpISub %i32 %count %c_i32_1\n"
6456 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6457 "OpBranchConditional %again %loop %exit\n"
6460 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
6461 "%same = OpFSub %f32 %val_post %c_f32_7\n"
6462 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6463 "OpReturnValue %result\n"
6465 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
6467 // A loop with return.
6468 fragments["testfun"] =
6469 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6470 "%param1 = OpFunctionParameter %v4f32\n"
6472 "%entry = OpLabel\n"
6473 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6474 "%dot = OpDot %f32 %param1 %param1\n"
6475 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6476 "%zero = OpConvertFToU %u32 %div\n"
6477 "%two = OpIAdd %i32 %zero %c_i32_2\n"
6478 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6481 ";returns early without modifying %param1\n"
6483 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6484 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
6485 "OpLoopMerge %exit %continue None\n"
6489 ";return if %count==%two\n"
6490 "%above2 = OpSGreaterThan %bool %count %two\n"
6491 "OpSelectionMerge %continue DontFlatten\n"
6492 "OpBranchConditional %above2 %body %early_exit\n"
6494 "%early_exit = OpLabel\n"
6495 "OpReturnValue %param1\n"
6498 "%fcount = OpConvertSToF %f32 %count\n"
6499 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6500 "OpBranch %continue\n"
6502 "%continue = OpLabel\n"
6503 "%count__ = OpISub %i32 %count %c_i32_1\n"
6504 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6505 "OpBranchConditional %again %loop %exit\n"
6508 ";should never get here, so return an incorrect result\n"
6509 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
6510 "OpReturnValue %result\n"
6512 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
6514 return testGroup.release();
6517 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
6518 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
6520 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
6521 map<string, string> fragments;
6523 // A barrier inside a function body.
6524 fragments["pre_main"] =
6525 "%Workgroup = OpConstant %i32 2\n"
6526 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
6527 fragments["testfun"] =
6528 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6529 "%param1 = OpFunctionParameter %v4f32\n"
6530 "%label_testfun = OpLabel\n"
6531 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6532 "OpReturnValue %param1\n"
6534 addTessCtrlTest(testGroup.get(), "in_function", fragments);
6536 // Common setup code for the following tests.
6537 fragments["pre_main"] =
6538 "%Workgroup = OpConstant %i32 2\n"
6539 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
6540 "%c_f32_5 = OpConstant %f32 5.\n";
6541 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
6542 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6543 "%param1 = OpFunctionParameter %v4f32\n"
6544 "%entry = OpLabel\n"
6545 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6546 "%dot = OpDot %f32 %param1 %param1\n"
6547 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6548 "%zero = OpConvertFToU %u32 %div\n";
6550 // Barriers inside OpSwitch branches.
6551 fragments["testfun"] =
6553 "OpSelectionMerge %switch_exit None\n"
6554 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
6556 "%case1 = OpLabel\n"
6557 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6558 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6559 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6560 "OpBranch %switch_exit\n"
6562 "%switch_default = OpLabel\n"
6563 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6564 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6565 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6566 "OpBranch %switch_exit\n"
6568 "%case0 = OpLabel\n"
6569 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6570 "OpBranch %switch_exit\n"
6572 "%switch_exit = OpLabel\n"
6573 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
6574 "OpReturnValue %ret\n"
6576 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
6578 // Barriers inside if-then-else.
6579 fragments["testfun"] =
6581 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
6582 "OpSelectionMerge %exit DontFlatten\n"
6583 "OpBranchConditional %eq0 %then %else\n"
6586 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6587 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6588 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6592 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6596 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
6597 "OpReturnValue %ret\n"
6599 addTessCtrlTest(testGroup.get(), "in_if", fragments);
6601 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
6602 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
6603 fragments["testfun"] =
6605 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
6606 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
6607 "OpSelectionMerge %exit DontFlatten\n"
6608 "OpBranchConditional %thread0 %then %else\n"
6611 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6615 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
6619 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
6620 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6621 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
6622 "OpReturnValue %ret\n"
6624 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
6626 // A barrier inside a loop.
6627 fragments["pre_main"] =
6628 "%Workgroup = OpConstant %i32 2\n"
6629 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
6630 "%c_f32_10 = OpConstant %f32 10.\n";
6631 fragments["testfun"] =
6632 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6633 "%param1 = OpFunctionParameter %v4f32\n"
6634 "%entry = OpLabel\n"
6635 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6638 ";adds 4, 3, 2, and 1 to %val0\n"
6640 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
6641 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
6642 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6643 "%fcount = OpConvertSToF %f32 %count\n"
6644 "%val = OpFAdd %f32 %val1 %fcount\n"
6645 "%count__ = OpISub %i32 %count %c_i32_1\n"
6646 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6647 "OpLoopMerge %exit %loop None\n"
6648 "OpBranchConditional %again %loop %exit\n"
6651 "%same = OpFSub %f32 %val %c_f32_10\n"
6652 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6653 "OpReturnValue %ret\n"
6655 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
6657 return testGroup.release();
6660 // Test for the OpFRem instruction.
6661 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
6663 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
6664 map<string, string> fragments;
6665 RGBA inputColors[4];
6666 RGBA outputColors[4];
6668 fragments["pre_main"] =
6669 "%c_f32_3 = OpConstant %f32 3.0\n"
6670 "%c_f32_n3 = OpConstant %f32 -3.0\n"
6671 "%c_f32_4 = OpConstant %f32 4.0\n"
6672 "%c_f32_p75 = OpConstant %f32 0.75\n"
6673 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
6674 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
6675 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
6677 // The test does the following.
6678 // vec4 result = (param1 * 8.0) - 4.0;
6679 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
6680 fragments["testfun"] =
6681 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6682 "%param1 = OpFunctionParameter %v4f32\n"
6683 "%label_testfun = OpLabel\n"
6684 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
6685 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
6686 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
6687 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
6688 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
6689 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
6690 "OpReturnValue %xy_0_1\n"
6694 inputColors[0] = RGBA(16, 16, 0, 255);
6695 inputColors[1] = RGBA(232, 232, 0, 255);
6696 inputColors[2] = RGBA(232, 16, 0, 255);
6697 inputColors[3] = RGBA(16, 232, 0, 255);
6699 outputColors[0] = RGBA(64, 64, 0, 255);
6700 outputColors[1] = RGBA(255, 255, 0, 255);
6701 outputColors[2] = RGBA(255, 64, 0, 255);
6702 outputColors[3] = RGBA(64, 255, 0, 255);
6704 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
6705 return testGroup.release();
6708 // Test for the OpSRem instruction.
6709 tcu::TestCaseGroup* createOpSRemGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
6711 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "srem", "OpSRem"));
6712 map<string, string> fragments;
6714 fragments["pre_main"] =
6715 "%c_f32_255 = OpConstant %f32 255.0\n"
6716 "%c_i32_128 = OpConstant %i32 128\n"
6717 "%c_i32_255 = OpConstant %i32 255\n"
6718 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
6719 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
6720 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
6722 // The test does the following.
6723 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
6724 // ivec4 result = ivec4(srem(ints.x, ints.y), srem(ints.y, ints.z), srem(ints.z, ints.x), 255);
6725 // return float(result + 128) / 255.0;
6726 fragments["testfun"] =
6727 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6728 "%param1 = OpFunctionParameter %v4f32\n"
6729 "%label_testfun = OpLabel\n"
6730 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
6731 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
6732 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
6733 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
6734 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
6735 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
6736 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
6737 "%x_out = OpSRem %i32 %x_in %y_in\n"
6738 "%y_out = OpSRem %i32 %y_in %z_in\n"
6739 "%z_out = OpSRem %i32 %z_in %x_in\n"
6740 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
6741 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
6742 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
6743 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
6744 "OpReturnValue %float_out\n"
6747 const struct CaseParams
6750 const char* failMessageTemplate; // customized status message
6751 qpTestResult failResult; // override status on failure
6752 int operands[4][3]; // four (x, y, z) vectors of operands
6753 int results[4][3]; // four (x, y, z) vectors of results
6759 QP_TEST_RESULT_FAIL,
6760 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
6761 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
6765 "Inconsistent results, but within specification: ${reason}",
6766 negFailResult, // negative operands, not required by the spec
6767 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
6768 { { 5, 12, -2 }, { 0, -5, 2 }, { 3, 8, -6 }, { 25, -60, 0 } }, // results
6771 // If either operand is negative the result is undefined. Some implementations may still return correct values.
6773 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
6775 const CaseParams& params = cases[caseNdx];
6776 RGBA inputColors[4];
6777 RGBA outputColors[4];
6779 for (int i = 0; i < 4; ++i)
6781 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
6782 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
6785 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
6788 return testGroup.release();
6791 // Test for the OpSMod instruction.
6792 tcu::TestCaseGroup* createOpSModGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
6794 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "smod", "OpSMod"));
6795 map<string, string> fragments;
6797 fragments["pre_main"] =
6798 "%c_f32_255 = OpConstant %f32 255.0\n"
6799 "%c_i32_128 = OpConstant %i32 128\n"
6800 "%c_i32_255 = OpConstant %i32 255\n"
6801 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
6802 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
6803 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
6805 // The test does the following.
6806 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
6807 // ivec4 result = ivec4(smod(ints.x, ints.y), smod(ints.y, ints.z), smod(ints.z, ints.x), 255);
6808 // return float(result + 128) / 255.0;
6809 fragments["testfun"] =
6810 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6811 "%param1 = OpFunctionParameter %v4f32\n"
6812 "%label_testfun = OpLabel\n"
6813 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
6814 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
6815 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
6816 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
6817 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
6818 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
6819 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
6820 "%x_out = OpSMod %i32 %x_in %y_in\n"
6821 "%y_out = OpSMod %i32 %y_in %z_in\n"
6822 "%z_out = OpSMod %i32 %z_in %x_in\n"
6823 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
6824 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
6825 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
6826 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
6827 "OpReturnValue %float_out\n"
6830 const struct CaseParams
6833 const char* failMessageTemplate; // customized status message
6834 qpTestResult failResult; // override status on failure
6835 int operands[4][3]; // four (x, y, z) vectors of operands
6836 int results[4][3]; // four (x, y, z) vectors of results
6842 QP_TEST_RESULT_FAIL,
6843 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
6844 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
6848 "Inconsistent results, but within specification: ${reason}",
6849 negFailResult, // negative operands, not required by the spec
6850 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
6851 { { 5, -5, 3 }, { 0, 2, -3 }, { 3, -73, 69 }, { -35, 40, 0 } }, // results
6854 // If either operand is negative the result is undefined. Some implementations may still return correct values.
6856 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
6858 const CaseParams& params = cases[caseNdx];
6859 RGBA inputColors[4];
6860 RGBA outputColors[4];
6862 for (int i = 0; i < 4; ++i)
6864 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
6865 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
6868 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
6871 return testGroup.release();
6876 INTEGER_TYPE_SIGNED_16,
6877 INTEGER_TYPE_SIGNED_32,
6878 INTEGER_TYPE_SIGNED_64,
6880 INTEGER_TYPE_UNSIGNED_16,
6881 INTEGER_TYPE_UNSIGNED_32,
6882 INTEGER_TYPE_UNSIGNED_64,
6885 const string getBitWidthStr (IntegerType type)
6889 case INTEGER_TYPE_SIGNED_16:
6890 case INTEGER_TYPE_UNSIGNED_16: return "16";
6892 case INTEGER_TYPE_SIGNED_32:
6893 case INTEGER_TYPE_UNSIGNED_32: return "32";
6895 case INTEGER_TYPE_SIGNED_64:
6896 case INTEGER_TYPE_UNSIGNED_64: return "64";
6898 default: DE_ASSERT(false);
6903 const string getByteWidthStr (IntegerType type)
6907 case INTEGER_TYPE_SIGNED_16:
6908 case INTEGER_TYPE_UNSIGNED_16: return "2";
6910 case INTEGER_TYPE_SIGNED_32:
6911 case INTEGER_TYPE_UNSIGNED_32: return "4";
6913 case INTEGER_TYPE_SIGNED_64:
6914 case INTEGER_TYPE_UNSIGNED_64: return "8";
6916 default: DE_ASSERT(false);
6921 bool isSigned (IntegerType type)
6923 return (type <= INTEGER_TYPE_SIGNED_64);
6926 const string getTypeName (IntegerType type)
6928 string prefix = isSigned(type) ? "" : "u";
6929 return prefix + "int" + getBitWidthStr(type);
6932 const string getTestName (IntegerType from, IntegerType to)
6934 return getTypeName(from) + "_to_" + getTypeName(to);
6937 const string getAsmTypeDeclaration (IntegerType type)
6939 string sign = isSigned(type) ? " 1" : " 0";
6940 return "OpTypeInt " + getBitWidthStr(type) + sign;
6943 template<typename T>
6944 BufferSp getSpecializedBuffer (deInt64 number)
6946 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
6949 BufferSp getBuffer (IntegerType type, deInt64 number)
6953 case INTEGER_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
6954 case INTEGER_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
6955 case INTEGER_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
6957 case INTEGER_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
6958 case INTEGER_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
6959 case INTEGER_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
6961 default: DE_ASSERT(false);
6962 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
6966 bool usesInt16 (IntegerType from, IntegerType to)
6968 return (from == INTEGER_TYPE_SIGNED_16 || from == INTEGER_TYPE_UNSIGNED_16
6969 || to == INTEGER_TYPE_SIGNED_16 || to == INTEGER_TYPE_UNSIGNED_16);
6972 bool usesInt64 (IntegerType from, IntegerType to)
6974 return (from == INTEGER_TYPE_SIGNED_64 || from == INTEGER_TYPE_UNSIGNED_64
6975 || to == INTEGER_TYPE_SIGNED_64 || to == INTEGER_TYPE_UNSIGNED_64);
6978 ComputeTestFeatures getConversionUsedFeatures (IntegerType from, IntegerType to)
6980 if (usesInt16(from, to))
6982 if (usesInt64(from, to))
6984 return COMPUTE_TEST_USES_INT16_INT64;
6988 return COMPUTE_TEST_USES_INT16;
6993 return COMPUTE_TEST_USES_INT64;
6999 ConvertCase (IntegerType from, IntegerType to, deInt64 number)
7002 , m_features (getConversionUsedFeatures(from, to))
7003 , m_name (getTestName(from, to))
7004 , m_inputBuffer (getBuffer(from, number))
7005 , m_outputBuffer (getBuffer(to, number))
7007 m_asmTypes["inputType"] = getAsmTypeDeclaration(from);
7008 m_asmTypes["outputType"] = getAsmTypeDeclaration(to);
7010 if (m_features == COMPUTE_TEST_USES_INT16)
7012 m_asmTypes["int_capabilities"] = "OpCapability Int16\n";
7014 else if (m_features == COMPUTE_TEST_USES_INT64)
7016 m_asmTypes["int_capabilities"] = "OpCapability Int64\n";
7018 else if (m_features == COMPUTE_TEST_USES_INT16_INT64)
7020 m_asmTypes["int_capabilities"] = string("OpCapability Int16\n") +
7021 "OpCapability Int64\n";
7029 IntegerType m_fromType;
7030 IntegerType m_toType;
7031 ComputeTestFeatures m_features;
7033 map<string, string> m_asmTypes;
7034 BufferSp m_inputBuffer;
7035 BufferSp m_outputBuffer;
7038 const string getConvertCaseShaderStr (const string& instruction, const ConvertCase& convertCase)
7040 map<string, string> params = convertCase.m_asmTypes;
7042 params["instruction"] = instruction;
7044 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
7045 params["outDecorator"] = getByteWidthStr(convertCase.m_toType);
7047 const StringTemplate shader (
7048 "OpCapability Shader\n"
7049 "${int_capabilities}"
7050 "OpMemoryModel Logical GLSL450\n"
7051 "OpEntryPoint GLCompute %main \"main\" %id\n"
7052 "OpExecutionMode %main LocalSize 1 1 1\n"
7053 "OpSource GLSL 430\n"
7054 "OpName %main \"main\"\n"
7055 "OpName %id \"gl_GlobalInvocationID\"\n"
7057 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7058 "OpDecorate %indata DescriptorSet 0\n"
7059 "OpDecorate %indata Binding 0\n"
7060 "OpDecorate %outdata DescriptorSet 0\n"
7061 "OpDecorate %outdata Binding 1\n"
7062 "OpDecorate %in_arr ArrayStride ${inDecorator}\n"
7063 "OpDecorate %out_arr ArrayStride ${outDecorator}\n"
7064 "OpDecorate %in_buf BufferBlock\n"
7065 "OpDecorate %out_buf BufferBlock\n"
7066 "OpMemberDecorate %in_buf 0 Offset 0\n"
7067 "OpMemberDecorate %out_buf 0 Offset 0\n"
7069 "%void = OpTypeVoid\n"
7070 "%voidf = OpTypeFunction %void\n"
7071 "%u32 = OpTypeInt 32 0\n"
7072 "%i32 = OpTypeInt 32 1\n"
7073 "%uvec3 = OpTypeVector %u32 3\n"
7074 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7076 "%in_type = ${inputType}\n"
7077 "%out_type = ${outputType}\n"
7079 "%in_ptr = OpTypePointer Uniform %in_type\n"
7080 "%out_ptr = OpTypePointer Uniform %out_type\n"
7081 "%in_arr = OpTypeRuntimeArray %in_type\n"
7082 "%out_arr = OpTypeRuntimeArray %out_type\n"
7083 "%in_buf = OpTypeStruct %in_arr\n"
7084 "%out_buf = OpTypeStruct %out_arr\n"
7085 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
7086 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
7087 "%indata = OpVariable %in_bufptr Uniform\n"
7088 "%outdata = OpVariable %out_bufptr Uniform\n"
7089 "%inputptr = OpTypePointer Input %in_type\n"
7090 "%id = OpVariable %uvec3ptr Input\n"
7092 "%zero = OpConstant %i32 0\n"
7094 "%main = OpFunction %void None %voidf\n"
7095 "%label = OpLabel\n"
7096 "%idval = OpLoad %uvec3 %id\n"
7097 "%x = OpCompositeExtract %u32 %idval 0\n"
7098 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
7099 "%outloc = OpAccessChain %out_ptr %outdata %zero %x\n"
7100 "%inval = OpLoad %in_type %inloc\n"
7101 "%conv = ${instruction} %out_type %inval\n"
7102 " OpStore %outloc %conv\n"
7107 return shader.specialize(params);
7110 void createSConvertCases (vector<ConvertCase>& testCases)
7112 // Convert int to int
7113 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_32, 14669));
7114 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_64, 3341));
7116 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_SIGNED_64, 973610259));
7118 // Convert int to unsigned int
7119 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_32, 9288));
7120 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_64, 15460));
7122 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_UNSIGNED_64, 346213461));
7125 // Test for the OpSConvert instruction.
7126 tcu::TestCaseGroup* createSConvertTests (tcu::TestContext& testCtx)
7128 const string instruction ("OpSConvert");
7129 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "sconvert", "OpSConvert"));
7130 vector<ConvertCase> testCases;
7131 createSConvertCases(testCases);
7133 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7135 ComputeShaderSpec spec;
7137 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7138 spec.inputs.push_back(test->m_inputBuffer);
7139 spec.outputs.push_back(test->m_outputBuffer);
7140 spec.numWorkGroups = IVec3(1, 1, 1);
7142 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpSConvert.", spec, test->m_features));
7145 return group.release();
7148 void createUConvertCases (vector<ConvertCase>& testCases)
7150 // Convert unsigned int to unsigned int
7151 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_32, 60653));
7152 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_64, 17991));
7154 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_UNSIGNED_64, 904256275));
7156 // Convert unsigned int to int
7157 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_32, 38002));
7158 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_64, 64921));
7160 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_SIGNED_64, 4294956295ll));
7163 // Test for the OpUConvert instruction.
7164 tcu::TestCaseGroup* createUConvertTests (tcu::TestContext& testCtx)
7166 const string instruction ("OpUConvert");
7167 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "uconvert", "OpUConvert"));
7168 vector<ConvertCase> testCases;
7169 createUConvertCases(testCases);
7171 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7173 ComputeShaderSpec spec;
7175 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7176 spec.inputs.push_back(test->m_inputBuffer);
7177 spec.outputs.push_back(test->m_outputBuffer);
7178 spec.numWorkGroups = IVec3(1, 1, 1);
7180 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpUConvert.", spec, test->m_features));
7182 return group.release();
7185 const string getNumberTypeName (const NumberType type)
7187 if (type == NUMBERTYPE_INT32)
7191 else if (type == NUMBERTYPE_UINT32)
7195 else if (type == NUMBERTYPE_FLOAT32)
7206 deInt32 getInt(de::Random& rnd)
7208 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
7211 const string repeatString (const string& str, int times)
7214 for (int i = 0; i < times; ++i)
7221 const string getRandomConstantString (const NumberType type, de::Random& rnd)
7223 if (type == NUMBERTYPE_INT32)
7225 return numberToString<deInt32>(getInt(rnd));
7227 else if (type == NUMBERTYPE_UINT32)
7229 return numberToString<deUint32>(rnd.getUint32());
7231 else if (type == NUMBERTYPE_FLOAT32)
7233 return numberToString<float>(rnd.getFloat());
7242 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7244 map<string, string> params;
7247 for (int width = 2; width <= 4; ++width)
7249 string randomConst = numberToString(getInt(rnd));
7250 string widthStr = numberToString(width);
7251 int index = rnd.getInt(0, width-1);
7253 params["type"] = "vec";
7254 params["name"] = params["type"] + "_" + widthStr;
7255 params["compositeType"] = "%composite = OpTypeVector %custom " + widthStr +"\n";
7256 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7257 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7258 params["indexes"] = numberToString(index);
7259 testCases.push_back(params);
7263 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7265 const int limit = 10;
7266 map<string, string> params;
7268 for (int width = 2; width <= limit; ++width)
7270 string randomConst = numberToString(getInt(rnd));
7271 string widthStr = numberToString(width);
7272 int index = rnd.getInt(0, width-1);
7274 params["type"] = "array";
7275 params["name"] = params["type"] + "_" + widthStr;
7276 params["compositeType"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
7277 + "%composite = OpTypeArray %custom %arraywidth\n";
7279 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7280 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7281 params["indexes"] = numberToString(index);
7282 testCases.push_back(params);
7286 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7288 const int limit = 10;
7289 map<string, string> params;
7291 for (int width = 2; width <= limit; ++width)
7293 string randomConst = numberToString(getInt(rnd));
7294 int index = rnd.getInt(0, width-1);
7296 params["type"] = "struct";
7297 params["name"] = params["type"] + "_" + numberToString(width);
7298 params["compositeType"] = "%composite = OpTypeStruct" + repeatString(" %custom", width) + "\n";
7299 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7300 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7301 params["indexes"] = numberToString(index);
7302 testCases.push_back(params);
7306 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7308 map<string, string> params;
7311 for (int width = 2; width <= 4; ++width)
7313 string widthStr = numberToString(width);
7315 for (int column = 2 ; column <= 4; ++column)
7317 int index_0 = rnd.getInt(0, column-1);
7318 int index_1 = rnd.getInt(0, width-1);
7319 string columnStr = numberToString(column);
7321 params["type"] = "matrix";
7322 params["name"] = params["type"] + "_" + widthStr + "x" + columnStr;
7323 params["compositeType"] = string("%vectype = OpTypeVector %custom " + widthStr + "\n")
7324 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
7326 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n"
7327 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
7329 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
7330 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
7331 testCases.push_back(params);
7336 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7338 createVectorCompositeCases(testCases, rnd, type);
7339 createArrayCompositeCases(testCases, rnd, type);
7340 createStructCompositeCases(testCases, rnd, type);
7341 // Matrix only supports float types
7342 if (type == NUMBERTYPE_FLOAT32)
7344 createMatrixCompositeCases(testCases, rnd, type);
7348 const string getAssemblyTypeDeclaration (const NumberType type)
7352 case NUMBERTYPE_INT32: return "OpTypeInt 32 1";
7353 case NUMBERTYPE_UINT32: return "OpTypeInt 32 0";
7354 case NUMBERTYPE_FLOAT32: return "OpTypeFloat 32";
7355 default: DE_ASSERT(false); return "";
7359 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
7361 map<string, string> parameters(params);
7363 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7365 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
7367 return StringTemplate (
7368 "OpCapability Shader\n"
7369 "OpCapability Matrix\n"
7370 "OpMemoryModel Logical GLSL450\n"
7371 "OpEntryPoint GLCompute %main \"main\" %id\n"
7372 "OpExecutionMode %main LocalSize 1 1 1\n"
7374 "OpSource GLSL 430\n"
7375 "OpName %main \"main\"\n"
7376 "OpName %id \"gl_GlobalInvocationID\"\n"
7379 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7380 "OpDecorate %buf BufferBlock\n"
7381 "OpDecorate %indata DescriptorSet 0\n"
7382 "OpDecorate %indata Binding 0\n"
7383 "OpDecorate %outdata DescriptorSet 0\n"
7384 "OpDecorate %outdata Binding 1\n"
7385 "OpDecorate %customarr ArrayStride 4\n"
7386 "${compositeDecorator}"
7387 "OpMemberDecorate %buf 0 Offset 0\n"
7390 "%void = OpTypeVoid\n"
7391 "%voidf = OpTypeFunction %void\n"
7392 "%u32 = OpTypeInt 32 0\n"
7393 "%i32 = OpTypeInt 32 1\n"
7394 "%uvec3 = OpTypeVector %u32 3\n"
7395 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7398 "%custom = ${typeDeclaration}\n"
7404 // Inherited from custom
7405 "%customptr = OpTypePointer Uniform %custom\n"
7406 "%customarr = OpTypeRuntimeArray %custom\n"
7407 "%buf = OpTypeStruct %customarr\n"
7408 "%bufptr = OpTypePointer Uniform %buf\n"
7410 "%indata = OpVariable %bufptr Uniform\n"
7411 "%outdata = OpVariable %bufptr Uniform\n"
7413 "%id = OpVariable %uvec3ptr Input\n"
7414 "%zero = OpConstant %i32 0\n"
7416 "%main = OpFunction %void None %voidf\n"
7417 "%label = OpLabel\n"
7418 "%idval = OpLoad %uvec3 %id\n"
7419 "%x = OpCompositeExtract %u32 %idval 0\n"
7421 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
7422 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
7423 // Read the input value
7424 "%inval = OpLoad %custom %inloc\n"
7425 // Create the composite and fill it
7426 "${compositeConstruct}"
7427 // Insert the input value to a place
7428 "%instance2 = OpCompositeInsert %composite %inval %instance ${indexes}\n"
7429 // Read back the value from the position
7430 "%out_val = OpCompositeExtract %custom %instance2 ${indexes}\n"
7431 // Store it in the output position
7432 " OpStore %outloc %out_val\n"
7435 ).specialize(parameters);
7438 template<typename T>
7439 BufferSp createCompositeBuffer(T number)
7441 return BufferSp(new Buffer<T>(vector<T>(1, number)));
7444 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
7446 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
7447 de::Random rnd (deStringHash(group->getName()));
7449 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7451 NumberType numberType = NumberType(type);
7452 const string typeName = getNumberTypeName(numberType);
7453 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
7454 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7455 vector<map<string, string> > testCases;
7457 createCompositeCases(testCases, rnd, numberType);
7459 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7461 ComputeShaderSpec spec;
7463 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
7467 case NUMBERTYPE_INT32:
7469 deInt32 number = getInt(rnd);
7470 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7471 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7474 case NUMBERTYPE_UINT32:
7476 deUint32 number = rnd.getUint32();
7477 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7478 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7481 case NUMBERTYPE_FLOAT32:
7483 float number = rnd.getFloat();
7484 spec.inputs.push_back(createCompositeBuffer<float>(number));
7485 spec.outputs.push_back(createCompositeBuffer<float>(number));
7492 spec.numWorkGroups = IVec3(1, 1, 1);
7493 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
7495 group->addChild(subGroup.release());
7497 return group.release();
7500 struct AssemblyStructInfo
7502 AssemblyStructInfo (const deUint32 comp, const deUint32 idx)
7507 deUint32 components;
7511 const string specializeInBoundsShaderTemplate (const NumberType type, const AssemblyStructInfo& structInfo, const map<string, string>& params)
7513 // Create the full index string
7514 string fullIndex = numberToString(structInfo.index) + " " + params.at("indexes");
7515 // Convert it to list of indexes
7516 vector<string> indexes = de::splitString(fullIndex, ' ');
7518 map<string, string> parameters (params);
7519 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7520 parameters["structType"] = repeatString(" %composite", structInfo.components);
7521 parameters["structConstruct"] = repeatString(" %instance", structInfo.components);
7522 parameters["insertIndexes"] = fullIndex;
7524 // In matrix cases the last two index is the CompositeExtract indexes
7525 const deUint32 extractIndexes = (parameters["type"] == "matrix") ? 2 : 1;
7527 // Construct the extractIndex
7528 for (vector<string>::const_iterator index = indexes.end() - extractIndexes; index != indexes.end(); ++index)
7530 parameters["extractIndexes"] += " " + *index;
7533 // Remove the last 1 or 2 element depends on matrix case or not
7534 indexes.erase(indexes.end() - extractIndexes, indexes.end());
7537 // Generate AccessChain index expressions (except for the last one, because we use ptr to the composite)
7538 for (vector<string>::const_iterator index = indexes.begin(); index != indexes.end(); ++index)
7540 string indexId = "%index_" + numberToString(id++);
7541 parameters["accessChainConstDeclaration"] += indexId + " = OpConstant %u32 " + *index + "\n";
7542 parameters["accessChainIndexes"] += " " + indexId;
7545 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
7547 return StringTemplate (
7548 "OpCapability Shader\n"
7549 "OpCapability Matrix\n"
7550 "OpMemoryModel Logical GLSL450\n"
7551 "OpEntryPoint GLCompute %main \"main\" %id\n"
7552 "OpExecutionMode %main LocalSize 1 1 1\n"
7554 "OpSource GLSL 430\n"
7555 "OpName %main \"main\"\n"
7556 "OpName %id \"gl_GlobalInvocationID\"\n"
7558 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7559 "OpDecorate %buf BufferBlock\n"
7560 "OpDecorate %indata DescriptorSet 0\n"
7561 "OpDecorate %indata Binding 0\n"
7562 "OpDecorate %outdata DescriptorSet 0\n"
7563 "OpDecorate %outdata Binding 1\n"
7564 "OpDecorate %customarr ArrayStride 4\n"
7565 "${compositeDecorator}"
7566 "OpMemberDecorate %buf 0 Offset 0\n"
7568 "%void = OpTypeVoid\n"
7569 "%voidf = OpTypeFunction %void\n"
7570 "%u32 = OpTypeInt 32 0\n"
7571 "%uvec3 = OpTypeVector %u32 3\n"
7572 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7574 "%custom = ${typeDeclaration}\n"
7577 // Inherited from composite
7578 "%composite_p = OpTypePointer Function %composite\n"
7579 "%struct_t = OpTypeStruct${structType}\n"
7580 "%struct_p = OpTypePointer Function %struct_t\n"
7583 "${accessChainConstDeclaration}"
7584 // Inherited from custom
7585 "%customptr = OpTypePointer Uniform %custom\n"
7586 "%customarr = OpTypeRuntimeArray %custom\n"
7587 "%buf = OpTypeStruct %customarr\n"
7588 "%bufptr = OpTypePointer Uniform %buf\n"
7589 "%indata = OpVariable %bufptr Uniform\n"
7590 "%outdata = OpVariable %bufptr Uniform\n"
7592 "%id = OpVariable %uvec3ptr Input\n"
7593 "%zero = OpConstant %u32 0\n"
7594 "%main = OpFunction %void None %voidf\n"
7595 "%label = OpLabel\n"
7596 "%struct_v = OpVariable %struct_p Function\n"
7597 "%idval = OpLoad %uvec3 %id\n"
7598 "%x = OpCompositeExtract %u32 %idval 0\n"
7599 // Create the input/output type
7600 "%inloc = OpInBoundsAccessChain %customptr %indata %zero %x\n"
7601 "%outloc = OpInBoundsAccessChain %customptr %outdata %zero %x\n"
7602 // Read the input value
7603 "%inval = OpLoad %custom %inloc\n"
7604 // Create the composite and fill it
7605 "${compositeConstruct}"
7606 // Create the struct and fill it with the composite
7607 "%struct = OpCompositeConstruct %struct_t${structConstruct}\n"
7609 "%comp_obj = OpCompositeInsert %struct_t %inval %struct ${insertIndexes}\n"
7611 " OpStore %struct_v %comp_obj\n"
7612 // Get deepest possible composite pointer
7613 "%inner_ptr = OpInBoundsAccessChain %composite_p %struct_v${accessChainIndexes}\n"
7614 "%read_obj = OpLoad %composite %inner_ptr\n"
7615 // Read back the stored value
7616 "%read_val = OpCompositeExtract %custom %read_obj${extractIndexes}\n"
7617 " OpStore %outloc %read_val\n"
7619 " OpFunctionEnd\n").specialize(parameters);
7622 tcu::TestCaseGroup* createOpInBoundsAccessChainGroup (tcu::TestContext& testCtx)
7624 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opinboundsaccesschain", "Test the OpInBoundsAccessChain instruction"));
7625 de::Random rnd (deStringHash(group->getName()));
7627 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7629 NumberType numberType = NumberType(type);
7630 const string typeName = getNumberTypeName(numberType);
7631 const string description = "Test the OpInBoundsAccessChain instruction with " + typeName + "s";
7632 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7634 vector<map<string, string> > testCases;
7635 createCompositeCases(testCases, rnd, numberType);
7637 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7639 ComputeShaderSpec spec;
7641 // Number of components inside of a struct
7642 deUint32 structComponents = rnd.getInt(2, 8);
7643 // Component index value
7644 deUint32 structIndex = rnd.getInt(0, structComponents - 1);
7645 AssemblyStructInfo structInfo(structComponents, structIndex);
7647 spec.assembly = specializeInBoundsShaderTemplate(numberType, structInfo, *test);
7651 case NUMBERTYPE_INT32:
7653 deInt32 number = getInt(rnd);
7654 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7655 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7658 case NUMBERTYPE_UINT32:
7660 deUint32 number = rnd.getUint32();
7661 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7662 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7665 case NUMBERTYPE_FLOAT32:
7667 float number = rnd.getFloat();
7668 spec.inputs.push_back(createCompositeBuffer<float>(number));
7669 spec.outputs.push_back(createCompositeBuffer<float>(number));
7675 spec.numWorkGroups = IVec3(1, 1, 1);
7676 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpInBoundsAccessChain test", spec));
7678 group->addChild(subGroup.release());
7680 return group.release();
7683 // If the params missing, uninitialized case
7684 const string specializeDefaultOutputShaderTemplate (const NumberType type, const map<string, string>& params = map<string, string>())
7686 map<string, string> parameters(params);
7688 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7690 // Declare the const value, and use it in the initializer
7691 if (params.find("constValue") != params.end())
7693 parameters["constDeclaration"] = "%const = OpConstant %in_type " + params.at("constValue") + "\n";
7694 parameters["variableInitializer"] = "%const";
7696 // Uninitialized case
7699 parameters["constDeclaration"] = "";
7700 parameters["variableInitializer"] = "";
7703 return StringTemplate(
7704 "OpCapability Shader\n"
7705 "OpMemoryModel Logical GLSL450\n"
7706 "OpEntryPoint GLCompute %main \"main\" %id\n"
7707 "OpExecutionMode %main LocalSize 1 1 1\n"
7708 "OpSource GLSL 430\n"
7709 "OpName %main \"main\"\n"
7710 "OpName %id \"gl_GlobalInvocationID\"\n"
7712 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7713 "OpDecorate %indata DescriptorSet 0\n"
7714 "OpDecorate %indata Binding 0\n"
7715 "OpDecorate %outdata DescriptorSet 0\n"
7716 "OpDecorate %outdata Binding 1\n"
7717 "OpDecorate %in_arr ArrayStride 4\n"
7718 "OpDecorate %in_buf BufferBlock\n"
7719 "OpMemberDecorate %in_buf 0 Offset 0\n"
7721 "%void = OpTypeVoid\n"
7722 "%voidf = OpTypeFunction %void\n"
7723 "%u32 = OpTypeInt 32 0\n"
7724 "%i32 = OpTypeInt 32 1\n"
7725 "%uvec3 = OpTypeVector %u32 3\n"
7726 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7728 "%in_type = ${typeDeclaration}\n"
7729 // "%const = OpConstant %in_type ${constValue}\n"
7730 "${constDeclaration}\n"
7732 "%in_ptr = OpTypePointer Uniform %in_type\n"
7733 "%in_arr = OpTypeRuntimeArray %in_type\n"
7734 "%in_buf = OpTypeStruct %in_arr\n"
7735 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
7736 "%indata = OpVariable %in_bufptr Uniform\n"
7737 "%outdata = OpVariable %in_bufptr Uniform\n"
7738 "%id = OpVariable %uvec3ptr Input\n"
7739 "%var_ptr = OpTypePointer Function %in_type\n"
7741 "%zero = OpConstant %i32 0\n"
7743 "%main = OpFunction %void None %voidf\n"
7744 "%label = OpLabel\n"
7745 "%out_var = OpVariable %var_ptr Function ${variableInitializer}\n"
7746 "%idval = OpLoad %uvec3 %id\n"
7747 "%x = OpCompositeExtract %u32 %idval 0\n"
7748 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
7749 "%outloc = OpAccessChain %in_ptr %outdata %zero %x\n"
7751 "%outval = OpLoad %in_type %out_var\n"
7752 " OpStore %outloc %outval\n"
7755 ).specialize(parameters);
7758 bool compareFloats (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
7760 DE_ASSERT(outputAllocs.size() != 0);
7761 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
7763 // Use custom epsilon because of the float->string conversion
7764 const float epsilon = 0.00001f;
7766 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
7768 vector<deUint8> expectedBytes;
7772 expectedOutputs[outputNdx]->getBytes(expectedBytes);
7773 memcpy(&expected, &expectedBytes.front(), expectedBytes.size());
7774 memcpy(&actual, outputAllocs[outputNdx]->getHostPtr(), expectedBytes.size());
7776 // Test with epsilon
7777 if (fabs(expected - actual) > epsilon)
7779 log << TestLog::Message << "Error: The actual and expected values not matching."
7780 << " Expected: " << expected << " Actual: " << actual << " Epsilon: " << epsilon << TestLog::EndMessage;
7787 // Checks if the driver crash with uninitialized cases
7788 bool passthruVerify (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
7790 DE_ASSERT(outputAllocs.size() != 0);
7791 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
7793 // Copy and discard the result.
7794 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
7796 vector<deUint8> expectedBytes;
7797 expectedOutputs[outputNdx]->getBytes(expectedBytes);
7799 const size_t width = expectedBytes.size();
7800 vector<char> data (width);
7802 memcpy(&data[0], outputAllocs[outputNdx]->getHostPtr(), width);
7807 tcu::TestCaseGroup* createShaderDefaultOutputGroup (tcu::TestContext& testCtx)
7809 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "shader_default_output", "Test shader default output."));
7810 de::Random rnd (deStringHash(group->getName()));
7812 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7814 NumberType numberType = NumberType(type);
7815 const string typeName = getNumberTypeName(numberType);
7816 const string description = "Test the OpVariable initializer with " + typeName + ".";
7817 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7819 // 2 similar subcases (initialized and uninitialized)
7820 for (int subCase = 0; subCase < 2; ++subCase)
7822 ComputeShaderSpec spec;
7823 spec.numWorkGroups = IVec3(1, 1, 1);
7825 map<string, string> params;
7829 case NUMBERTYPE_INT32:
7831 deInt32 number = getInt(rnd);
7832 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7833 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7834 params["constValue"] = numberToString(number);
7837 case NUMBERTYPE_UINT32:
7839 deUint32 number = rnd.getUint32();
7840 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7841 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7842 params["constValue"] = numberToString(number);
7845 case NUMBERTYPE_FLOAT32:
7847 float number = rnd.getFloat();
7848 spec.inputs.push_back(createCompositeBuffer<float>(number));
7849 spec.outputs.push_back(createCompositeBuffer<float>(number));
7850 spec.verifyIO = &compareFloats;
7851 params["constValue"] = numberToString(number);
7858 // Initialized subcase
7861 spec.assembly = specializeDefaultOutputShaderTemplate(numberType, params);
7862 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "initialized", "OpVariable initializer tests.", spec));
7864 // Uninitialized subcase
7867 spec.assembly = specializeDefaultOutputShaderTemplate(numberType);
7868 spec.verifyIO = &passthruVerify;
7869 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "uninitialized", "OpVariable initializer tests.", spec));
7872 group->addChild(subGroup.release());
7874 return group.release();
7877 tcu::TestCaseGroup* createOpNopTests (tcu::TestContext& testCtx)
7879 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
7880 RGBA defaultColors[4];
7881 map<string, string> opNopFragments;
7883 getDefaultColors(defaultColors);
7885 opNopFragments["testfun"] =
7886 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7887 "%param1 = OpFunctionParameter %v4f32\n"
7888 "%label_testfun = OpLabel\n"
7897 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7898 "%b = OpFAdd %f32 %a %a\n"
7900 "%c = OpFSub %f32 %b %a\n"
7901 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
7904 "OpReturnValue %ret\n"
7907 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, testGroup.get());
7909 return testGroup.release();
7912 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
7914 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
7915 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
7916 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
7918 computeTests->addChild(createOpNopGroup(testCtx));
7919 computeTests->addChild(createOpFUnordGroup(testCtx));
7920 computeTests->addChild(createOpAtomicGroup(testCtx, false));
7921 computeTests->addChild(createOpAtomicGroup(testCtx, true)); // Using new StorageBuffer decoration
7922 computeTests->addChild(createOpLineGroup(testCtx));
7923 computeTests->addChild(createOpNoLineGroup(testCtx));
7924 computeTests->addChild(createOpConstantNullGroup(testCtx));
7925 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
7926 computeTests->addChild(createOpConstantUsageGroup(testCtx));
7927 computeTests->addChild(createSpecConstantGroup(testCtx));
7928 computeTests->addChild(createOpSourceGroup(testCtx));
7929 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
7930 computeTests->addChild(createDecorationGroupGroup(testCtx));
7931 computeTests->addChild(createOpPhiGroup(testCtx));
7932 computeTests->addChild(createLoopControlGroup(testCtx));
7933 computeTests->addChild(createFunctionControlGroup(testCtx));
7934 computeTests->addChild(createSelectionControlGroup(testCtx));
7935 computeTests->addChild(createBlockOrderGroup(testCtx));
7936 computeTests->addChild(createMultipleShaderGroup(testCtx));
7937 computeTests->addChild(createMemoryAccessGroup(testCtx));
7938 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
7939 computeTests->addChild(createOpCopyObjectGroup(testCtx));
7940 computeTests->addChild(createNoContractionGroup(testCtx));
7941 computeTests->addChild(createOpUndefGroup(testCtx));
7942 computeTests->addChild(createOpUnreachableGroup(testCtx));
7943 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
7944 computeTests ->addChild(createOpFRemGroup(testCtx));
7945 computeTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_PASS));
7946 computeTests->addChild(createOpSRemComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
7947 computeTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_PASS));
7948 computeTests->addChild(createOpSModComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
7949 computeTests->addChild(createSConvertTests(testCtx));
7950 computeTests->addChild(createUConvertTests(testCtx));
7951 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
7952 computeTests->addChild(createOpInBoundsAccessChainGroup(testCtx));
7953 computeTests->addChild(createShaderDefaultOutputGroup(testCtx));
7954 computeTests->addChild(createOpNMinGroup(testCtx));
7955 computeTests->addChild(createOpNMaxGroup(testCtx));
7956 computeTests->addChild(createOpNClampGroup(testCtx));
7958 de::MovePtr<tcu::TestCaseGroup> computeAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
7960 computeAndroidTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7961 computeAndroidTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7963 computeTests->addChild(computeAndroidTests.release());
7966 computeTests->addChild(create16BitStorageComputeGroup(testCtx));
7967 computeTests->addChild(createUboMatrixPaddingComputeGroup(testCtx));
7968 computeTests->addChild(createConditionalBranchComputeGroup(testCtx));
7969 computeTests->addChild(createIndexingComputeGroup(testCtx));
7970 computeTests->addChild(createVariablePointersComputeGroup(testCtx));
7971 graphicsTests->addChild(createOpNopTests(testCtx));
7972 graphicsTests->addChild(createOpSourceTests(testCtx));
7973 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
7974 graphicsTests->addChild(createOpLineTests(testCtx));
7975 graphicsTests->addChild(createOpNoLineTests(testCtx));
7976 graphicsTests->addChild(createOpConstantNullTests(testCtx));
7977 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
7978 graphicsTests->addChild(createMemoryAccessTests(testCtx));
7979 graphicsTests->addChild(createOpUndefTests(testCtx));
7980 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
7981 graphicsTests->addChild(createModuleTests(testCtx));
7982 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
7983 graphicsTests->addChild(createOpPhiTests(testCtx));
7984 graphicsTests->addChild(createNoContractionTests(testCtx));
7985 graphicsTests->addChild(createOpQuantizeTests(testCtx));
7986 graphicsTests->addChild(createLoopTests(testCtx));
7987 graphicsTests->addChild(createSpecConstantTests(testCtx));
7988 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
7989 graphicsTests->addChild(createBarrierTests(testCtx));
7990 graphicsTests->addChild(createDecorationGroupTests(testCtx));
7991 graphicsTests->addChild(createFRemTests(testCtx));
7992 graphicsTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
7993 graphicsTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
7996 de::MovePtr<tcu::TestCaseGroup> graphicsAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
7998 graphicsAndroidTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7999 graphicsAndroidTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
8001 graphicsTests->addChild(graphicsAndroidTests.release());
8004 graphicsTests->addChild(create16BitStorageGraphicsGroup(testCtx));
8005 graphicsTests->addChild(createUboMatrixPaddingGraphicsGroup(testCtx));
8006 graphicsTests->addChild(createConditionalBranchGraphicsGroup(testCtx));
8007 graphicsTests->addChild(createIndexingGraphicsGroup(testCtx));
8008 graphicsTests->addChild(createVariablePointersGraphicsGroup(testCtx));
8010 instructionTests->addChild(computeTests.release());
8011 instructionTests->addChild(graphicsTests.release());
8013 return instructionTests.release();