1 /*-------------------------------------------------------------------------
2 * Vulkan Conformance Tests
3 * ------------------------
5 * Copyright (c) 2015 Google Inc.
6 * Copyright (c) 2016 The Khronos Group Inc.
8 * Licensed under the Apache License, Version 2.0 (the "License");
9 * you may not use this file except in compliance with the License.
10 * You may obtain a copy of the License at
12 * http://www.apache.org/licenses/LICENSE-2.0
14 * Unless required by applicable law or agreed to in writing, software
15 * distributed under the License is distributed on an "AS IS" BASIS,
16 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17 * See the License for the specific language governing permissions and
18 * limitations under the License.
22 * \brief SPIR-V Assembly Tests for Instructions (special opcode/operand)
23 *//*--------------------------------------------------------------------*/
25 #include "vktSpvAsmInstructionTests.hpp"
27 #include "tcuCommandLine.hpp"
28 #include "tcuFormatUtil.hpp"
29 #include "tcuFloat.hpp"
30 #include "tcuRGBA.hpp"
31 #include "tcuStringTemplate.hpp"
32 #include "tcuTestLog.hpp"
33 #include "tcuVectorUtil.hpp"
34 #include "tcuInterval.hpp"
37 #include "vkDeviceUtil.hpp"
38 #include "vkMemUtil.hpp"
39 #include "vkPlatform.hpp"
40 #include "vkPrograms.hpp"
41 #include "vkQueryUtil.hpp"
43 #include "vkRefUtil.hpp"
44 #include "vkStrUtil.hpp"
45 #include "vkTypeUtil.hpp"
47 #include "deRandom.hpp"
48 #include "deStringUtil.hpp"
49 #include "deUniquePtr.hpp"
51 #include "tcuStringTemplate.hpp"
53 #include "vktSpvAsm16bitStorageTests.hpp"
54 #include "vktSpvAsmUboMatrixPaddingTests.hpp"
55 #include "vktSpvAsmComputeShaderCase.hpp"
56 #include "vktSpvAsmComputeShaderTestUtil.hpp"
57 #include "vktSpvAsmGraphicsShaderTestUtil.hpp"
58 #include "vktSpvAsmVariablePointersTests.hpp"
59 #include "vktTestCaseUtil.hpp"
70 namespace SpirVAssembly
84 using tcu::TestStatus;
87 using tcu::StringTemplate;
91 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0)
93 T* const typedPtr = (T*)dst;
94 for (int ndx = 0; ndx < numValues; ndx++)
95 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue);
98 // Filter is a function that returns true if a value should pass, false otherwise.
99 template<typename T, typename FilterT>
100 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, FilterT filter, int offset = 0)
102 T* const typedPtr = (T*)dst;
104 for (int ndx = 0; ndx < numValues; ndx++)
107 value = randomScalar<T>(rnd, minValue, maxValue);
108 while (!filter(value));
110 typedPtr[offset + ndx] = value;
114 // Gets a 64-bit integer with a more logarithmic distribution
115 deInt64 randomInt64LogDistributed (de::Random& rnd)
117 deInt64 val = rnd.getUint64();
118 val &= (1ull << rnd.getInt(1, 63)) - 1;
124 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues)
126 for (int ndx = 0; ndx < numValues; ndx++)
127 dst[ndx] = randomInt64LogDistributed(rnd);
130 template<typename FilterT>
131 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues, FilterT filter)
133 for (int ndx = 0; ndx < numValues; ndx++)
137 value = randomInt64LogDistributed(rnd);
138 } while (!filter(value));
143 inline bool filterNonNegative (const deInt64 value)
148 inline bool filterPositive (const deInt64 value)
153 inline bool filterNotZero (const deInt64 value)
158 static void floorAll (vector<float>& values)
160 for (size_t i = 0; i < values.size(); i++)
161 values[i] = deFloatFloor(values[i]);
164 static void floorAll (vector<Vec4>& values)
166 for (size_t i = 0; i < values.size(); i++)
167 values[i] = floor(values[i]);
175 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {}
178 // Assembly code used for testing OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code:
182 // layout(std140, set = 0, binding = 0) readonly buffer Input {
185 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
189 // layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
192 // uint x = gl_GlobalInvocationID.x;
193 // output_data.elements[x] = -input_data.elements[x];
196 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
198 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
199 ComputeShaderSpec spec;
200 de::Random rnd (deStringHash(group->getName()));
201 const int numElements = 100;
202 vector<float> positiveFloats (numElements, 0);
203 vector<float> negativeFloats (numElements, 0);
205 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
207 for (size_t ndx = 0; ndx < numElements; ++ndx)
208 negativeFloats[ndx] = -positiveFloats[ndx];
211 string(getComputeAsmShaderPreamble()) +
213 "OpSource GLSL 430\n"
214 "OpName %main \"main\"\n"
215 "OpName %id \"gl_GlobalInvocationID\"\n"
217 "OpDecorate %id BuiltIn GlobalInvocationId\n"
219 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes())
221 + string(getComputeAsmInputOutputBuffer()) +
223 "%id = OpVariable %uvec3ptr Input\n"
224 "%zero = OpConstant %i32 0\n"
226 "%main = OpFunction %void None %voidf\n"
228 "%idval = OpLoad %uvec3 %id\n"
229 "%x = OpCompositeExtract %u32 %idval 0\n"
231 " OpNop\n" // Inside a function body
233 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
234 "%inval = OpLoad %f32 %inloc\n"
235 "%neg = OpFNegate %f32 %inval\n"
236 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
237 " OpStore %outloc %neg\n"
240 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
241 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
242 spec.numWorkGroups = IVec3(numElements, 1, 1);
244 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
246 return group.release();
249 bool compareFUnord (const std::vector<BufferSp>& inputs, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
251 if (outputAllocs.size() != 1)
254 const BufferSp& expectedOutput = expectedOutputs[0];
255 const deInt32* expectedOutputAsInt = static_cast<const deInt32*>(expectedOutputs[0]->data());
256 const deInt32* outputAsInt = static_cast<const deInt32*>(outputAllocs[0]->getHostPtr());
257 const float* input1AsFloat = static_cast<const float*>(inputs[0]->data());
258 const float* input2AsFloat = static_cast<const float*>(inputs[1]->data());
259 bool returnValue = true;
261 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(deInt32); ++idx)
263 if (outputAsInt[idx] != expectedOutputAsInt[idx])
265 log << TestLog::Message << "ERROR: Sub-case failed. inputs: " << input1AsFloat[idx] << "," << input2AsFloat[idx] << " output: " << outputAsInt[idx]<< " expected output: " << expectedOutputAsInt[idx] << TestLog::EndMessage;
272 typedef VkBool32 (*compareFuncType) (float, float);
278 compareFuncType compareFunc;
280 OpFUnordCase (const char* _name, const char* _opCode, compareFuncType _compareFunc)
283 , compareFunc (_compareFunc) {}
286 #define ADD_OPFUNORD_CASE(NAME, OPCODE, OPERATOR) \
288 struct compare_##NAME { static VkBool32 compare(float x, float y) { return (x OPERATOR y) ? VK_TRUE : VK_FALSE; } }; \
289 cases.push_back(OpFUnordCase(#NAME, OPCODE, compare_##NAME::compare)); \
290 } while (deGetFalse())
292 tcu::TestCaseGroup* createOpFUnordGroup (tcu::TestContext& testCtx)
294 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfunord", "Test the OpFUnord* opcodes"));
295 de::Random rnd (deStringHash(group->getName()));
296 const int numElements = 100;
297 vector<OpFUnordCase> cases;
299 const StringTemplate shaderTemplate (
301 string(getComputeAsmShaderPreamble()) +
303 "OpSource GLSL 430\n"
304 "OpName %main \"main\"\n"
305 "OpName %id \"gl_GlobalInvocationID\"\n"
307 "OpDecorate %id BuiltIn GlobalInvocationId\n"
309 "OpDecorate %buf BufferBlock\n"
310 "OpDecorate %buf2 BufferBlock\n"
311 "OpDecorate %indata1 DescriptorSet 0\n"
312 "OpDecorate %indata1 Binding 0\n"
313 "OpDecorate %indata2 DescriptorSet 0\n"
314 "OpDecorate %indata2 Binding 1\n"
315 "OpDecorate %outdata DescriptorSet 0\n"
316 "OpDecorate %outdata Binding 2\n"
317 "OpDecorate %f32arr ArrayStride 4\n"
318 "OpDecorate %i32arr ArrayStride 4\n"
319 "OpMemberDecorate %buf 0 Offset 0\n"
320 "OpMemberDecorate %buf2 0 Offset 0\n"
322 + string(getComputeAsmCommonTypes()) +
324 "%buf = OpTypeStruct %f32arr\n"
325 "%bufptr = OpTypePointer Uniform %buf\n"
326 "%indata1 = OpVariable %bufptr Uniform\n"
327 "%indata2 = OpVariable %bufptr Uniform\n"
329 "%buf2 = OpTypeStruct %i32arr\n"
330 "%buf2ptr = OpTypePointer Uniform %buf2\n"
331 "%outdata = OpVariable %buf2ptr Uniform\n"
333 "%id = OpVariable %uvec3ptr Input\n"
334 "%zero = OpConstant %i32 0\n"
335 "%consti1 = OpConstant %i32 1\n"
336 "%constf1 = OpConstant %f32 1.0\n"
338 "%main = OpFunction %void None %voidf\n"
340 "%idval = OpLoad %uvec3 %id\n"
341 "%x = OpCompositeExtract %u32 %idval 0\n"
343 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
344 "%inval1 = OpLoad %f32 %inloc1\n"
345 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
346 "%inval2 = OpLoad %f32 %inloc2\n"
347 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
349 "%result = ${OPCODE} %bool %inval1 %inval2\n"
350 "%int_res = OpSelect %i32 %result %consti1 %zero\n"
351 " OpStore %outloc %int_res\n"
356 ADD_OPFUNORD_CASE(equal, "OpFUnordEqual", ==);
357 ADD_OPFUNORD_CASE(less, "OpFUnordLessThan", <);
358 ADD_OPFUNORD_CASE(lessequal, "OpFUnordLessThanEqual", <=);
359 ADD_OPFUNORD_CASE(greater, "OpFUnordGreaterThan", >);
360 ADD_OPFUNORD_CASE(greaterequal, "OpFUnordGreaterThanEqual", >=);
361 ADD_OPFUNORD_CASE(notequal, "OpFUnordNotEqual", !=);
363 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
365 map<string, string> specializations;
366 ComputeShaderSpec spec;
367 const float NaN = std::numeric_limits<float>::quiet_NaN();
368 vector<float> inputFloats1 (numElements, 0);
369 vector<float> inputFloats2 (numElements, 0);
370 vector<deInt32> expectedInts (numElements, 0);
372 specializations["OPCODE"] = cases[caseNdx].opCode;
373 spec.assembly = shaderTemplate.specialize(specializations);
375 fillRandomScalars(rnd, 1.f, 100.f, &inputFloats1[0], numElements);
376 for (size_t ndx = 0; ndx < numElements; ++ndx)
380 case 0: inputFloats2[ndx] = inputFloats1[ndx] + 1.0f; break;
381 case 1: inputFloats2[ndx] = inputFloats1[ndx] - 1.0f; break;
382 case 2: inputFloats2[ndx] = inputFloats1[ndx]; break;
383 case 3: inputFloats2[ndx] = NaN; break;
384 case 4: inputFloats2[ndx] = inputFloats1[ndx]; inputFloats1[ndx] = NaN; break;
385 case 5: inputFloats2[ndx] = NaN; inputFloats1[ndx] = NaN; break;
387 expectedInts[ndx] = tcu::Float32(inputFloats1[ndx]).isNaN() || tcu::Float32(inputFloats2[ndx]).isNaN() || cases[caseNdx].compareFunc(inputFloats1[ndx], inputFloats2[ndx]);
390 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
391 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
392 spec.outputs.push_back(BufferSp(new Int32Buffer(expectedInts)));
393 spec.numWorkGroups = IVec3(numElements, 1, 1);
394 spec.verifyIO = &compareFUnord;
395 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
398 return group.release();
404 const char* assembly;
405 void (*calculateExpected)(deInt32&, deInt32);
406 deInt32 numOutputElements;
408 OpAtomicCase (const char* _name, const char* _assembly, void (*_calculateExpected)(deInt32&, deInt32), deInt32 _numOutputElements)
410 , assembly (_assembly)
411 , calculateExpected (_calculateExpected)
412 , numOutputElements (_numOutputElements) {}
415 tcu::TestCaseGroup* createOpAtomicGroup (tcu::TestContext& testCtx, bool useStorageBuffer)
417 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx,
418 useStorageBuffer ? "opatomic_storage_buffer" : "opatomic",
419 "Test the OpAtomic* opcodes"));
420 de::Random rnd (deStringHash(group->getName()));
421 const int numElements = 65535;
422 vector<OpAtomicCase> cases;
424 const StringTemplate shaderTemplate (
426 string("OpCapability Shader\n") +
427 (useStorageBuffer ? "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" : "") +
428 "OpMemoryModel Logical GLSL450\n"
429 "OpEntryPoint GLCompute %main \"main\" %id\n"
430 "OpExecutionMode %main LocalSize 1 1 1\n" +
432 "OpSource GLSL 430\n"
433 "OpName %main \"main\"\n"
434 "OpName %id \"gl_GlobalInvocationID\"\n"
436 "OpDecorate %id BuiltIn GlobalInvocationId\n"
438 "OpDecorate %buf ${BLOCK_DECORATION}\n"
439 "OpDecorate %indata DescriptorSet 0\n"
440 "OpDecorate %indata Binding 0\n"
441 "OpDecorate %i32arr ArrayStride 4\n"
442 "OpMemberDecorate %buf 0 Offset 0\n"
444 "OpDecorate %sumbuf ${BLOCK_DECORATION}\n"
445 "OpDecorate %sum DescriptorSet 0\n"
446 "OpDecorate %sum Binding 1\n"
447 "OpMemberDecorate %sumbuf 0 Coherent\n"
448 "OpMemberDecorate %sumbuf 0 Offset 0\n"
450 + getComputeAsmCommonTypes("${BLOCK_POINTER_TYPE}") +
452 "%buf = OpTypeStruct %i32arr\n"
453 "%bufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %buf\n"
454 "%indata = OpVariable %bufptr ${BLOCK_POINTER_TYPE}\n"
456 "%sumbuf = OpTypeStruct %i32arr\n"
457 "%sumbufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %sumbuf\n"
458 "%sum = OpVariable %sumbufptr ${BLOCK_POINTER_TYPE}\n"
460 "%id = OpVariable %uvec3ptr Input\n"
461 "%minusone = OpConstant %i32 -1\n"
462 "%zero = OpConstant %i32 0\n"
463 "%one = OpConstant %u32 1\n"
464 "%two = OpConstant %i32 2\n"
466 "%main = OpFunction %void None %voidf\n"
468 "%idval = OpLoad %uvec3 %id\n"
469 "%x = OpCompositeExtract %u32 %idval 0\n"
471 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
472 "%inval = OpLoad %i32 %inloc\n"
474 "%outloc = OpAccessChain %i32ptr %sum %zero ${INDEX}\n"
480 #define ADD_OPATOMIC_CASE(NAME, ASSEMBLY, CALCULATE_EXPECTED, NUM_OUTPUT_ELEMENTS) \
482 DE_STATIC_ASSERT((NUM_OUTPUT_ELEMENTS) == 1 || (NUM_OUTPUT_ELEMENTS) == numElements); \
483 struct calculateExpected_##NAME { static void calculateExpected(deInt32& expected, deInt32 input) CALCULATE_EXPECTED }; /* NOLINT(CALCULATE_EXPECTED) */ \
484 cases.push_back(OpAtomicCase(#NAME, ASSEMBLY, calculateExpected_##NAME::calculateExpected, NUM_OUTPUT_ELEMENTS)); \
485 } while (deGetFalse())
486 #define ADD_OPATOMIC_CASE_1(NAME, ASSEMBLY, CALCULATE_EXPECTED) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, CALCULATE_EXPECTED, 1)
487 #define ADD_OPATOMIC_CASE_N(NAME, ASSEMBLY, CALCULATE_EXPECTED) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, CALCULATE_EXPECTED, numElements)
489 ADD_OPATOMIC_CASE_1(iadd, "%unused = OpAtomicIAdd %i32 %outloc %one %zero %inval\n", { expected += input; } );
490 ADD_OPATOMIC_CASE_1(isub, "%unused = OpAtomicISub %i32 %outloc %one %zero %inval\n", { expected -= input; } );
491 ADD_OPATOMIC_CASE_1(iinc, "%unused = OpAtomicIIncrement %i32 %outloc %one %zero\n", { ++expected; (void)input;} );
492 ADD_OPATOMIC_CASE_1(idec, "%unused = OpAtomicIDecrement %i32 %outloc %one %zero\n", { --expected; (void)input;} );
493 ADD_OPATOMIC_CASE_N(load, "%inval2 = OpAtomicLoad %i32 %inloc %zero %zero\n"
494 " OpStore %outloc %inval2\n", { expected = input;} );
495 ADD_OPATOMIC_CASE_N(store, " OpAtomicStore %outloc %zero %zero %inval\n", { expected = input;} );
496 ADD_OPATOMIC_CASE_N(compex, "%even = OpSMod %i32 %inval %two\n"
497 " OpStore %outloc %even\n"
498 "%unused = OpAtomicCompareExchange %i32 %outloc %one %zero %zero %minusone %zero\n", { expected = (input % 2) == 0 ? -1 : 1;} );
500 #undef ADD_OPATOMIC_CASE
501 #undef ADD_OPATOMIC_CASE_1
502 #undef ADD_OPATOMIC_CASE_N
504 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
506 map<string, string> specializations;
507 ComputeShaderSpec spec;
508 vector<deInt32> inputInts (numElements, 0);
509 vector<deInt32> expected (cases[caseNdx].numOutputElements, -1);
511 specializations["INDEX"] = (cases[caseNdx].numOutputElements == 1) ? "%zero" : "%x";
512 specializations["INSTRUCTION"] = cases[caseNdx].assembly;
513 specializations["BLOCK_DECORATION"] = useStorageBuffer ? "Block" : "BufferBlock";
514 specializations["BLOCK_POINTER_TYPE"] = useStorageBuffer ? "StorageBuffer" : "Uniform";
515 spec.assembly = shaderTemplate.specialize(specializations);
517 if (useStorageBuffer)
518 spec.extensions.push_back("VK_KHR_storage_buffer_storage_class");
520 fillRandomScalars(rnd, 1, 100, &inputInts[0], numElements);
521 for (size_t ndx = 0; ndx < numElements; ++ndx)
523 cases[caseNdx].calculateExpected((cases[caseNdx].numOutputElements == 1) ? expected[0] : expected[ndx], inputInts[ndx]);
526 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
527 spec.outputs.push_back(BufferSp(new Int32Buffer(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 // We really just need this for size because we are not comparing the exact values.
673 const BufferSp& expectedOutput = expectedOutputs[0];
674 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
676 for(size_t i = 0; i < expectedOutput->getNumBytes() / sizeof(float); ++i) {
677 if (outputAsFloat[i] != 0.f &&
678 outputAsFloat[i] != -ldexp(1, -24)) {
686 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
688 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
689 vector<CaseParameter> cases;
690 const int numElements = 100;
691 vector<float> inputFloats1 (numElements, 0);
692 vector<float> inputFloats2 (numElements, 0);
693 vector<float> outputFloats (numElements, 0);
694 const StringTemplate shaderTemplate (
695 string(getComputeAsmShaderPreamble()) +
697 "OpName %main \"main\"\n"
698 "OpName %id \"gl_GlobalInvocationID\"\n"
700 "OpDecorate %id BuiltIn GlobalInvocationId\n"
704 "OpDecorate %buf BufferBlock\n"
705 "OpDecorate %indata1 DescriptorSet 0\n"
706 "OpDecorate %indata1 Binding 0\n"
707 "OpDecorate %indata2 DescriptorSet 0\n"
708 "OpDecorate %indata2 Binding 1\n"
709 "OpDecorate %outdata DescriptorSet 0\n"
710 "OpDecorate %outdata Binding 2\n"
711 "OpDecorate %f32arr ArrayStride 4\n"
712 "OpMemberDecorate %buf 0 Offset 0\n"
714 + string(getComputeAsmCommonTypes()) +
716 "%buf = OpTypeStruct %f32arr\n"
717 "%bufptr = OpTypePointer Uniform %buf\n"
718 "%indata1 = OpVariable %bufptr Uniform\n"
719 "%indata2 = OpVariable %bufptr Uniform\n"
720 "%outdata = OpVariable %bufptr Uniform\n"
722 "%id = OpVariable %uvec3ptr Input\n"
723 "%zero = OpConstant %i32 0\n"
724 "%c_f_m1 = OpConstant %f32 -1.\n"
726 "%main = OpFunction %void None %voidf\n"
728 "%idval = OpLoad %uvec3 %id\n"
729 "%x = OpCompositeExtract %u32 %idval 0\n"
730 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
731 "%inval1 = OpLoad %f32 %inloc1\n"
732 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
733 "%inval2 = OpLoad %f32 %inloc2\n"
734 "%mul = OpFMul %f32 %inval1 %inval2\n"
735 "%add = OpFAdd %f32 %mul %c_f_m1\n"
736 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
737 " OpStore %outloc %add\n"
741 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
742 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
743 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
745 for (size_t ndx = 0; ndx < numElements; ++ndx)
747 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
748 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
749 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
750 // conducted separately and the result is rounded to 1, or 0x1.fffffcp-1
751 // So the final result will be 0.f or 0x1p-24.
752 // If the operation is combined into a precise fused multiply-add, then the result would be
753 // 2^-46 (0xa8800000).
754 outputFloats[ndx] = 0.f;
757 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
759 map<string, string> specializations;
760 ComputeShaderSpec spec;
762 specializations["DECORATION"] = cases[caseNdx].param;
763 spec.assembly = shaderTemplate.specialize(specializations);
764 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
765 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
766 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
767 spec.numWorkGroups = IVec3(numElements, 1, 1);
768 // Check against the two possible answers based on rounding mode.
769 spec.verifyIO = &compareNoContractCase;
771 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
773 return group.release();
776 bool compareFRem(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
778 if (outputAllocs.size() != 1)
781 const BufferSp& expectedOutput = expectedOutputs[0];
782 const float *expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
783 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
785 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
787 const float f0 = expectedOutputAsFloat[idx];
788 const float f1 = outputAsFloat[idx];
789 // \todo relative error needs to be fairly high because FRem may be implemented as
790 // (roughly) frac(a/b)*b, so LSB errors can be magnified. But this should be fine for now.
791 if (deFloatAbs((f1 - f0) / f0) > 0.02)
798 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
800 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
801 ComputeShaderSpec spec;
802 de::Random rnd (deStringHash(group->getName()));
803 const int numElements = 200;
804 vector<float> inputFloats1 (numElements, 0);
805 vector<float> inputFloats2 (numElements, 0);
806 vector<float> outputFloats (numElements, 0);
808 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
809 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
811 for (size_t ndx = 0; ndx < numElements; ++ndx)
813 // Guard against divisors near zero.
814 if (std::fabs(inputFloats2[ndx]) < 1e-3)
815 inputFloats2[ndx] = 8.f;
817 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
818 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
822 string(getComputeAsmShaderPreamble()) +
824 "OpName %main \"main\"\n"
825 "OpName %id \"gl_GlobalInvocationID\"\n"
827 "OpDecorate %id BuiltIn GlobalInvocationId\n"
829 "OpDecorate %buf BufferBlock\n"
830 "OpDecorate %indata1 DescriptorSet 0\n"
831 "OpDecorate %indata1 Binding 0\n"
832 "OpDecorate %indata2 DescriptorSet 0\n"
833 "OpDecorate %indata2 Binding 1\n"
834 "OpDecorate %outdata DescriptorSet 0\n"
835 "OpDecorate %outdata Binding 2\n"
836 "OpDecorate %f32arr ArrayStride 4\n"
837 "OpMemberDecorate %buf 0 Offset 0\n"
839 + string(getComputeAsmCommonTypes()) +
841 "%buf = OpTypeStruct %f32arr\n"
842 "%bufptr = OpTypePointer Uniform %buf\n"
843 "%indata1 = OpVariable %bufptr Uniform\n"
844 "%indata2 = OpVariable %bufptr Uniform\n"
845 "%outdata = OpVariable %bufptr Uniform\n"
847 "%id = OpVariable %uvec3ptr Input\n"
848 "%zero = OpConstant %i32 0\n"
850 "%main = OpFunction %void None %voidf\n"
852 "%idval = OpLoad %uvec3 %id\n"
853 "%x = OpCompositeExtract %u32 %idval 0\n"
854 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
855 "%inval1 = OpLoad %f32 %inloc1\n"
856 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
857 "%inval2 = OpLoad %f32 %inloc2\n"
858 "%rem = OpFRem %f32 %inval1 %inval2\n"
859 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
860 " OpStore %outloc %rem\n"
864 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
865 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
866 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
867 spec.numWorkGroups = IVec3(numElements, 1, 1);
868 spec.verifyIO = &compareFRem;
870 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
872 return group.release();
875 bool compareNMin (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
877 if (outputAllocs.size() != 1)
880 const BufferSp& expectedOutput = expectedOutputs[0];
881 const float* const expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
882 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
884 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
886 const float f0 = expectedOutputAsFloat[idx];
887 const float f1 = outputAsFloat[idx];
889 // For NMin, we accept NaN as output if both inputs were NaN.
890 // Otherwise the NaN is the wrong choise, as on architectures that
891 // do not handle NaN, those are huge values.
892 if (!(tcu::Float32(f1).isNaN() && tcu::Float32(f0).isNaN()) && deFloatAbs(f1 - f0) > 0.00001f)
899 tcu::TestCaseGroup* createOpNMinGroup (tcu::TestContext& testCtx)
901 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnmin", "Test the OpNMin instruction"));
902 ComputeShaderSpec spec;
903 de::Random rnd (deStringHash(group->getName()));
904 const int numElements = 200;
905 vector<float> inputFloats1 (numElements, 0);
906 vector<float> inputFloats2 (numElements, 0);
907 vector<float> outputFloats (numElements, 0);
909 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
910 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
912 // Make the first case a full-NAN case.
913 inputFloats1[0] = TCU_NAN;
914 inputFloats2[0] = TCU_NAN;
916 for (size_t ndx = 0; ndx < numElements; ++ndx)
918 // By default, pick the smallest
919 outputFloats[ndx] = std::min(inputFloats1[ndx], inputFloats2[ndx]);
921 // Make half of the cases NaN cases
924 // Alternate between the NaN operand
927 outputFloats[ndx] = inputFloats2[ndx];
928 inputFloats1[ndx] = TCU_NAN;
932 outputFloats[ndx] = inputFloats1[ndx];
933 inputFloats2[ndx] = TCU_NAN;
939 "OpCapability Shader\n"
940 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
941 "OpMemoryModel Logical GLSL450\n"
942 "OpEntryPoint GLCompute %main \"main\" %id\n"
943 "OpExecutionMode %main LocalSize 1 1 1\n"
945 "OpName %main \"main\"\n"
946 "OpName %id \"gl_GlobalInvocationID\"\n"
948 "OpDecorate %id BuiltIn GlobalInvocationId\n"
950 "OpDecorate %buf BufferBlock\n"
951 "OpDecorate %indata1 DescriptorSet 0\n"
952 "OpDecorate %indata1 Binding 0\n"
953 "OpDecorate %indata2 DescriptorSet 0\n"
954 "OpDecorate %indata2 Binding 1\n"
955 "OpDecorate %outdata DescriptorSet 0\n"
956 "OpDecorate %outdata Binding 2\n"
957 "OpDecorate %f32arr ArrayStride 4\n"
958 "OpMemberDecorate %buf 0 Offset 0\n"
960 + string(getComputeAsmCommonTypes()) +
962 "%buf = OpTypeStruct %f32arr\n"
963 "%bufptr = OpTypePointer Uniform %buf\n"
964 "%indata1 = OpVariable %bufptr Uniform\n"
965 "%indata2 = OpVariable %bufptr Uniform\n"
966 "%outdata = OpVariable %bufptr Uniform\n"
968 "%id = OpVariable %uvec3ptr Input\n"
969 "%zero = OpConstant %i32 0\n"
971 "%main = OpFunction %void None %voidf\n"
973 "%idval = OpLoad %uvec3 %id\n"
974 "%x = OpCompositeExtract %u32 %idval 0\n"
975 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
976 "%inval1 = OpLoad %f32 %inloc1\n"
977 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
978 "%inval2 = OpLoad %f32 %inloc2\n"
979 "%rem = OpExtInst %f32 %std450 NMin %inval1 %inval2\n"
980 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
981 " OpStore %outloc %rem\n"
985 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
986 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
987 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
988 spec.numWorkGroups = IVec3(numElements, 1, 1);
989 spec.verifyIO = &compareNMin;
991 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
993 return group.release();
996 bool compareNMax (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
998 if (outputAllocs.size() != 1)
1001 const BufferSp& expectedOutput = expectedOutputs[0];
1002 const float* const expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
1003 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
1005 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
1007 const float f0 = expectedOutputAsFloat[idx];
1008 const float f1 = outputAsFloat[idx];
1010 // For NMax, NaN is considered acceptable result, since in
1011 // architectures that do not handle NaNs, those are huge values.
1012 if (!tcu::Float32(f1).isNaN() && deFloatAbs(f1 - f0) > 0.00001f)
1019 tcu::TestCaseGroup* createOpNMaxGroup (tcu::TestContext& testCtx)
1021 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, "opnmax", "Test the OpNMax instruction"));
1022 ComputeShaderSpec spec;
1023 de::Random rnd (deStringHash(group->getName()));
1024 const int numElements = 200;
1025 vector<float> inputFloats1 (numElements, 0);
1026 vector<float> inputFloats2 (numElements, 0);
1027 vector<float> outputFloats (numElements, 0);
1029 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1030 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1032 // Make the first case a full-NAN case.
1033 inputFloats1[0] = TCU_NAN;
1034 inputFloats2[0] = TCU_NAN;
1036 for (size_t ndx = 0; ndx < numElements; ++ndx)
1038 // By default, pick the biggest
1039 outputFloats[ndx] = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1041 // Make half of the cases NaN cases
1044 // Alternate between the NaN operand
1047 outputFloats[ndx] = inputFloats2[ndx];
1048 inputFloats1[ndx] = TCU_NAN;
1052 outputFloats[ndx] = inputFloats1[ndx];
1053 inputFloats2[ndx] = TCU_NAN;
1059 "OpCapability Shader\n"
1060 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1061 "OpMemoryModel Logical GLSL450\n"
1062 "OpEntryPoint GLCompute %main \"main\" %id\n"
1063 "OpExecutionMode %main LocalSize 1 1 1\n"
1065 "OpName %main \"main\"\n"
1066 "OpName %id \"gl_GlobalInvocationID\"\n"
1068 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1070 "OpDecorate %buf BufferBlock\n"
1071 "OpDecorate %indata1 DescriptorSet 0\n"
1072 "OpDecorate %indata1 Binding 0\n"
1073 "OpDecorate %indata2 DescriptorSet 0\n"
1074 "OpDecorate %indata2 Binding 1\n"
1075 "OpDecorate %outdata DescriptorSet 0\n"
1076 "OpDecorate %outdata Binding 2\n"
1077 "OpDecorate %f32arr ArrayStride 4\n"
1078 "OpMemberDecorate %buf 0 Offset 0\n"
1080 + string(getComputeAsmCommonTypes()) +
1082 "%buf = OpTypeStruct %f32arr\n"
1083 "%bufptr = OpTypePointer Uniform %buf\n"
1084 "%indata1 = OpVariable %bufptr Uniform\n"
1085 "%indata2 = OpVariable %bufptr Uniform\n"
1086 "%outdata = OpVariable %bufptr Uniform\n"
1088 "%id = OpVariable %uvec3ptr Input\n"
1089 "%zero = OpConstant %i32 0\n"
1091 "%main = OpFunction %void None %voidf\n"
1092 "%label = OpLabel\n"
1093 "%idval = OpLoad %uvec3 %id\n"
1094 "%x = OpCompositeExtract %u32 %idval 0\n"
1095 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1096 "%inval1 = OpLoad %f32 %inloc1\n"
1097 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1098 "%inval2 = OpLoad %f32 %inloc2\n"
1099 "%rem = OpExtInst %f32 %std450 NMax %inval1 %inval2\n"
1100 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1101 " OpStore %outloc %rem\n"
1105 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1106 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1107 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1108 spec.numWorkGroups = IVec3(numElements, 1, 1);
1109 spec.verifyIO = &compareNMax;
1111 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1113 return group.release();
1116 bool compareNClamp (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1118 if (outputAllocs.size() != 1)
1121 const BufferSp& expectedOutput = expectedOutputs[0];
1122 const float* const expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
1123 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
1125 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float) / 2; ++idx)
1127 const float e0 = expectedOutputAsFloat[idx * 2];
1128 const float e1 = expectedOutputAsFloat[idx * 2 + 1];
1129 const float res = outputAsFloat[idx];
1131 // For NClamp, we have two possible outcomes based on
1132 // whether NaNs are handled or not.
1133 // If either min or max value is NaN, the result is undefined,
1134 // so this test doesn't stress those. If the clamped value is
1135 // NaN, and NaNs are handled, the result is min; if NaNs are not
1136 // handled, they are big values that result in max.
1137 // If all three parameters are NaN, the result should be NaN.
1138 if (!((tcu::Float32(e0).isNaN() && tcu::Float32(res).isNaN()) ||
1139 (deFloatAbs(e0 - res) < 0.00001f) ||
1140 (deFloatAbs(e1 - res) < 0.00001f)))
1147 tcu::TestCaseGroup* createOpNClampGroup (tcu::TestContext& testCtx)
1149 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnclamp", "Test the OpNClamp instruction"));
1150 ComputeShaderSpec spec;
1151 de::Random rnd (deStringHash(group->getName()));
1152 const int numElements = 200;
1153 vector<float> inputFloats1 (numElements, 0);
1154 vector<float> inputFloats2 (numElements, 0);
1155 vector<float> inputFloats3 (numElements, 0);
1156 vector<float> outputFloats (numElements * 2, 0);
1158 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1159 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1160 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats3[0], numElements);
1162 for (size_t ndx = 0; ndx < numElements; ++ndx)
1164 // Results are only defined if max value is bigger than min value.
1165 if (inputFloats2[ndx] > inputFloats3[ndx])
1167 float t = inputFloats2[ndx];
1168 inputFloats2[ndx] = inputFloats3[ndx];
1169 inputFloats3[ndx] = t;
1172 // By default, do the clamp, setting both possible answers
1173 float defaultRes = std::min(std::max(inputFloats1[ndx], inputFloats2[ndx]), inputFloats3[ndx]);
1175 float maxResA = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1176 float maxResB = maxResA;
1178 // Alternate between the NaN cases
1181 inputFloats1[ndx] = TCU_NAN;
1182 // If NaN is handled, the result should be same as the clamp minimum.
1183 // If NaN is not handled, the result should clamp to the clamp maximum.
1184 maxResA = inputFloats2[ndx];
1185 maxResB = inputFloats3[ndx];
1189 // Not a NaN case - only one legal result.
1190 maxResA = defaultRes;
1191 maxResB = defaultRes;
1194 outputFloats[ndx * 2] = maxResA;
1195 outputFloats[ndx * 2 + 1] = maxResB;
1198 // Make the first case a full-NAN case.
1199 inputFloats1[0] = TCU_NAN;
1200 inputFloats2[0] = TCU_NAN;
1201 inputFloats3[0] = TCU_NAN;
1202 outputFloats[0] = TCU_NAN;
1203 outputFloats[1] = TCU_NAN;
1206 "OpCapability Shader\n"
1207 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1208 "OpMemoryModel Logical GLSL450\n"
1209 "OpEntryPoint GLCompute %main \"main\" %id\n"
1210 "OpExecutionMode %main LocalSize 1 1 1\n"
1212 "OpName %main \"main\"\n"
1213 "OpName %id \"gl_GlobalInvocationID\"\n"
1215 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1217 "OpDecorate %buf BufferBlock\n"
1218 "OpDecorate %indata1 DescriptorSet 0\n"
1219 "OpDecorate %indata1 Binding 0\n"
1220 "OpDecorate %indata2 DescriptorSet 0\n"
1221 "OpDecorate %indata2 Binding 1\n"
1222 "OpDecorate %indata3 DescriptorSet 0\n"
1223 "OpDecorate %indata3 Binding 2\n"
1224 "OpDecorate %outdata DescriptorSet 0\n"
1225 "OpDecorate %outdata Binding 3\n"
1226 "OpDecorate %f32arr ArrayStride 4\n"
1227 "OpMemberDecorate %buf 0 Offset 0\n"
1229 + string(getComputeAsmCommonTypes()) +
1231 "%buf = OpTypeStruct %f32arr\n"
1232 "%bufptr = OpTypePointer Uniform %buf\n"
1233 "%indata1 = OpVariable %bufptr Uniform\n"
1234 "%indata2 = OpVariable %bufptr Uniform\n"
1235 "%indata3 = OpVariable %bufptr Uniform\n"
1236 "%outdata = OpVariable %bufptr Uniform\n"
1238 "%id = OpVariable %uvec3ptr Input\n"
1239 "%zero = OpConstant %i32 0\n"
1241 "%main = OpFunction %void None %voidf\n"
1242 "%label = OpLabel\n"
1243 "%idval = OpLoad %uvec3 %id\n"
1244 "%x = OpCompositeExtract %u32 %idval 0\n"
1245 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1246 "%inval1 = OpLoad %f32 %inloc1\n"
1247 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1248 "%inval2 = OpLoad %f32 %inloc2\n"
1249 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1250 "%inval3 = OpLoad %f32 %inloc3\n"
1251 "%rem = OpExtInst %f32 %std450 NClamp %inval1 %inval2 %inval3\n"
1252 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1253 " OpStore %outloc %rem\n"
1257 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1258 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1259 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1260 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1261 spec.numWorkGroups = IVec3(numElements, 1, 1);
1262 spec.verifyIO = &compareNClamp;
1264 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1266 return group.release();
1269 tcu::TestCaseGroup* createOpSRemComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1271 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem", "Test the OpSRem instruction"));
1272 de::Random rnd (deStringHash(group->getName()));
1273 const int numElements = 200;
1275 const struct CaseParams
1278 const char* failMessage; // customized status message
1279 qpTestResult failResult; // override status on failure
1280 int op1Min, op1Max; // operand ranges
1284 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1285 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1287 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1289 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1291 const CaseParams& params = cases[caseNdx];
1292 ComputeShaderSpec spec;
1293 vector<deInt32> inputInts1 (numElements, 0);
1294 vector<deInt32> inputInts2 (numElements, 0);
1295 vector<deInt32> outputInts (numElements, 0);
1297 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1298 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1300 for (int ndx = 0; ndx < numElements; ++ndx)
1302 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1303 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1307 string(getComputeAsmShaderPreamble()) +
1309 "OpName %main \"main\"\n"
1310 "OpName %id \"gl_GlobalInvocationID\"\n"
1312 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1314 "OpDecorate %buf BufferBlock\n"
1315 "OpDecorate %indata1 DescriptorSet 0\n"
1316 "OpDecorate %indata1 Binding 0\n"
1317 "OpDecorate %indata2 DescriptorSet 0\n"
1318 "OpDecorate %indata2 Binding 1\n"
1319 "OpDecorate %outdata DescriptorSet 0\n"
1320 "OpDecorate %outdata Binding 2\n"
1321 "OpDecorate %i32arr ArrayStride 4\n"
1322 "OpMemberDecorate %buf 0 Offset 0\n"
1324 + string(getComputeAsmCommonTypes()) +
1326 "%buf = OpTypeStruct %i32arr\n"
1327 "%bufptr = OpTypePointer Uniform %buf\n"
1328 "%indata1 = OpVariable %bufptr Uniform\n"
1329 "%indata2 = OpVariable %bufptr Uniform\n"
1330 "%outdata = OpVariable %bufptr Uniform\n"
1332 "%id = OpVariable %uvec3ptr Input\n"
1333 "%zero = OpConstant %i32 0\n"
1335 "%main = OpFunction %void None %voidf\n"
1336 "%label = OpLabel\n"
1337 "%idval = OpLoad %uvec3 %id\n"
1338 "%x = OpCompositeExtract %u32 %idval 0\n"
1339 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1340 "%inval1 = OpLoad %i32 %inloc1\n"
1341 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1342 "%inval2 = OpLoad %i32 %inloc2\n"
1343 "%rem = OpSRem %i32 %inval1 %inval2\n"
1344 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1345 " OpStore %outloc %rem\n"
1349 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1350 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1351 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1352 spec.numWorkGroups = IVec3(numElements, 1, 1);
1353 spec.failResult = params.failResult;
1354 spec.failMessage = params.failMessage;
1356 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1359 return group.release();
1362 tcu::TestCaseGroup* createOpSRemComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1364 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem64", "Test the 64-bit OpSRem instruction"));
1365 de::Random rnd (deStringHash(group->getName()));
1366 const int numElements = 200;
1368 const struct CaseParams
1371 const char* failMessage; // customized status message
1372 qpTestResult failResult; // override status on failure
1376 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1377 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1379 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1381 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1383 const CaseParams& params = cases[caseNdx];
1384 ComputeShaderSpec spec;
1385 vector<deInt64> inputInts1 (numElements, 0);
1386 vector<deInt64> inputInts2 (numElements, 0);
1387 vector<deInt64> outputInts (numElements, 0);
1389 if (params.positive)
1391 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1392 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1396 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1397 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1400 for (int ndx = 0; ndx < numElements; ++ndx)
1402 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1403 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1407 "OpCapability Int64\n"
1409 + string(getComputeAsmShaderPreamble()) +
1411 "OpName %main \"main\"\n"
1412 "OpName %id \"gl_GlobalInvocationID\"\n"
1414 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1416 "OpDecorate %buf BufferBlock\n"
1417 "OpDecorate %indata1 DescriptorSet 0\n"
1418 "OpDecorate %indata1 Binding 0\n"
1419 "OpDecorate %indata2 DescriptorSet 0\n"
1420 "OpDecorate %indata2 Binding 1\n"
1421 "OpDecorate %outdata DescriptorSet 0\n"
1422 "OpDecorate %outdata Binding 2\n"
1423 "OpDecorate %i64arr ArrayStride 8\n"
1424 "OpMemberDecorate %buf 0 Offset 0\n"
1426 + string(getComputeAsmCommonTypes())
1427 + string(getComputeAsmCommonInt64Types()) +
1429 "%buf = OpTypeStruct %i64arr\n"
1430 "%bufptr = OpTypePointer Uniform %buf\n"
1431 "%indata1 = OpVariable %bufptr Uniform\n"
1432 "%indata2 = OpVariable %bufptr Uniform\n"
1433 "%outdata = OpVariable %bufptr Uniform\n"
1435 "%id = OpVariable %uvec3ptr Input\n"
1436 "%zero = OpConstant %i64 0\n"
1438 "%main = OpFunction %void None %voidf\n"
1439 "%label = OpLabel\n"
1440 "%idval = OpLoad %uvec3 %id\n"
1441 "%x = OpCompositeExtract %u32 %idval 0\n"
1442 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1443 "%inval1 = OpLoad %i64 %inloc1\n"
1444 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1445 "%inval2 = OpLoad %i64 %inloc2\n"
1446 "%rem = OpSRem %i64 %inval1 %inval2\n"
1447 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1448 " OpStore %outloc %rem\n"
1452 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1453 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1454 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1455 spec.numWorkGroups = IVec3(numElements, 1, 1);
1456 spec.failResult = params.failResult;
1457 spec.failMessage = params.failMessage;
1459 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1462 return group.release();
1465 tcu::TestCaseGroup* createOpSModComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1467 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod", "Test the OpSMod instruction"));
1468 de::Random rnd (deStringHash(group->getName()));
1469 const int numElements = 200;
1471 const struct CaseParams
1474 const char* failMessage; // customized status message
1475 qpTestResult failResult; // override status on failure
1476 int op1Min, op1Max; // operand ranges
1480 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1481 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1483 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1485 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1487 const CaseParams& params = cases[caseNdx];
1489 ComputeShaderSpec spec;
1490 vector<deInt32> inputInts1 (numElements, 0);
1491 vector<deInt32> inputInts2 (numElements, 0);
1492 vector<deInt32> outputInts (numElements, 0);
1494 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1495 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1497 for (int ndx = 0; ndx < numElements; ++ndx)
1499 deInt32 rem = inputInts1[ndx] % inputInts2[ndx];
1502 outputInts[ndx] = 0;
1504 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1506 // They have the same sign
1507 outputInts[ndx] = rem;
1511 // They have opposite sign. The remainder operation takes the
1512 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1513 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1514 // the result has the correct sign and that it is still
1515 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1517 // See also http://mathforum.org/library/drmath/view/52343.html
1518 outputInts[ndx] = rem + inputInts2[ndx];
1523 string(getComputeAsmShaderPreamble()) +
1525 "OpName %main \"main\"\n"
1526 "OpName %id \"gl_GlobalInvocationID\"\n"
1528 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1530 "OpDecorate %buf BufferBlock\n"
1531 "OpDecorate %indata1 DescriptorSet 0\n"
1532 "OpDecorate %indata1 Binding 0\n"
1533 "OpDecorate %indata2 DescriptorSet 0\n"
1534 "OpDecorate %indata2 Binding 1\n"
1535 "OpDecorate %outdata DescriptorSet 0\n"
1536 "OpDecorate %outdata Binding 2\n"
1537 "OpDecorate %i32arr ArrayStride 4\n"
1538 "OpMemberDecorate %buf 0 Offset 0\n"
1540 + string(getComputeAsmCommonTypes()) +
1542 "%buf = OpTypeStruct %i32arr\n"
1543 "%bufptr = OpTypePointer Uniform %buf\n"
1544 "%indata1 = OpVariable %bufptr Uniform\n"
1545 "%indata2 = OpVariable %bufptr Uniform\n"
1546 "%outdata = OpVariable %bufptr Uniform\n"
1548 "%id = OpVariable %uvec3ptr Input\n"
1549 "%zero = OpConstant %i32 0\n"
1551 "%main = OpFunction %void None %voidf\n"
1552 "%label = OpLabel\n"
1553 "%idval = OpLoad %uvec3 %id\n"
1554 "%x = OpCompositeExtract %u32 %idval 0\n"
1555 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1556 "%inval1 = OpLoad %i32 %inloc1\n"
1557 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1558 "%inval2 = OpLoad %i32 %inloc2\n"
1559 "%rem = OpSMod %i32 %inval1 %inval2\n"
1560 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1561 " OpStore %outloc %rem\n"
1565 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1566 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1567 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1568 spec.numWorkGroups = IVec3(numElements, 1, 1);
1569 spec.failResult = params.failResult;
1570 spec.failMessage = params.failMessage;
1572 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1575 return group.release();
1578 tcu::TestCaseGroup* createOpSModComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1580 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod64", "Test the OpSMod instruction"));
1581 de::Random rnd (deStringHash(group->getName()));
1582 const int numElements = 200;
1584 const struct CaseParams
1587 const char* failMessage; // customized status message
1588 qpTestResult failResult; // override status on failure
1592 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1593 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1595 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1597 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1599 const CaseParams& params = cases[caseNdx];
1601 ComputeShaderSpec spec;
1602 vector<deInt64> inputInts1 (numElements, 0);
1603 vector<deInt64> inputInts2 (numElements, 0);
1604 vector<deInt64> outputInts (numElements, 0);
1607 if (params.positive)
1609 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1610 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1614 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1615 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1618 for (int ndx = 0; ndx < numElements; ++ndx)
1620 deInt64 rem = inputInts1[ndx] % inputInts2[ndx];
1623 outputInts[ndx] = 0;
1625 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1627 // They have the same sign
1628 outputInts[ndx] = rem;
1632 // They have opposite sign. The remainder operation takes the
1633 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1634 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1635 // the result has the correct sign and that it is still
1636 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1638 // See also http://mathforum.org/library/drmath/view/52343.html
1639 outputInts[ndx] = rem + inputInts2[ndx];
1644 "OpCapability Int64\n"
1646 + string(getComputeAsmShaderPreamble()) +
1648 "OpName %main \"main\"\n"
1649 "OpName %id \"gl_GlobalInvocationID\"\n"
1651 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1653 "OpDecorate %buf BufferBlock\n"
1654 "OpDecorate %indata1 DescriptorSet 0\n"
1655 "OpDecorate %indata1 Binding 0\n"
1656 "OpDecorate %indata2 DescriptorSet 0\n"
1657 "OpDecorate %indata2 Binding 1\n"
1658 "OpDecorate %outdata DescriptorSet 0\n"
1659 "OpDecorate %outdata Binding 2\n"
1660 "OpDecorate %i64arr ArrayStride 8\n"
1661 "OpMemberDecorate %buf 0 Offset 0\n"
1663 + string(getComputeAsmCommonTypes())
1664 + string(getComputeAsmCommonInt64Types()) +
1666 "%buf = OpTypeStruct %i64arr\n"
1667 "%bufptr = OpTypePointer Uniform %buf\n"
1668 "%indata1 = OpVariable %bufptr Uniform\n"
1669 "%indata2 = OpVariable %bufptr Uniform\n"
1670 "%outdata = OpVariable %bufptr Uniform\n"
1672 "%id = OpVariable %uvec3ptr Input\n"
1673 "%zero = OpConstant %i64 0\n"
1675 "%main = OpFunction %void None %voidf\n"
1676 "%label = OpLabel\n"
1677 "%idval = OpLoad %uvec3 %id\n"
1678 "%x = OpCompositeExtract %u32 %idval 0\n"
1679 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1680 "%inval1 = OpLoad %i64 %inloc1\n"
1681 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1682 "%inval2 = OpLoad %i64 %inloc2\n"
1683 "%rem = OpSMod %i64 %inval1 %inval2\n"
1684 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1685 " OpStore %outloc %rem\n"
1689 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1690 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1691 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1692 spec.numWorkGroups = IVec3(numElements, 1, 1);
1693 spec.failResult = params.failResult;
1694 spec.failMessage = params.failMessage;
1696 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1699 return group.release();
1702 // Copy contents in the input buffer to the output buffer.
1703 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
1705 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
1706 de::Random rnd (deStringHash(group->getName()));
1707 const int numElements = 100;
1709 // 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.
1710 ComputeShaderSpec spec1;
1711 vector<Vec4> inputFloats1 (numElements);
1712 vector<Vec4> outputFloats1 (numElements);
1714 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
1716 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1717 floorAll(inputFloats1);
1719 for (size_t ndx = 0; ndx < numElements; ++ndx)
1720 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
1723 string(getComputeAsmShaderPreamble()) +
1725 "OpName %main \"main\"\n"
1726 "OpName %id \"gl_GlobalInvocationID\"\n"
1728 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1729 "OpDecorate %vec4arr ArrayStride 16\n"
1731 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1733 "%vec4 = OpTypeVector %f32 4\n"
1734 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
1735 "%vec4ptr_f = OpTypePointer Function %vec4\n"
1736 "%vec4arr = OpTypeRuntimeArray %vec4\n"
1737 "%buf = OpTypeStruct %vec4arr\n"
1738 "%bufptr = OpTypePointer Uniform %buf\n"
1739 "%indata = OpVariable %bufptr Uniform\n"
1740 "%outdata = OpVariable %bufptr Uniform\n"
1742 "%id = OpVariable %uvec3ptr Input\n"
1743 "%zero = OpConstant %i32 0\n"
1744 "%c_f_0 = OpConstant %f32 0.\n"
1745 "%c_f_0_5 = OpConstant %f32 0.5\n"
1746 "%c_f_1_5 = OpConstant %f32 1.5\n"
1747 "%c_f_2_5 = OpConstant %f32 2.5\n"
1748 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
1750 "%main = OpFunction %void None %voidf\n"
1751 "%label = OpLabel\n"
1752 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
1753 "%idval = OpLoad %uvec3 %id\n"
1754 "%x = OpCompositeExtract %u32 %idval 0\n"
1755 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
1756 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
1757 " OpCopyMemory %v_vec4 %inloc\n"
1758 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
1759 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
1760 " OpStore %outloc %add\n"
1764 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
1765 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
1766 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1768 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
1770 // The following case copies a float[100] variable from the input buffer to the output buffer.
1771 ComputeShaderSpec spec2;
1772 vector<float> inputFloats2 (numElements);
1773 vector<float> outputFloats2 (numElements);
1775 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
1777 for (size_t ndx = 0; ndx < numElements; ++ndx)
1778 outputFloats2[ndx] = inputFloats2[ndx];
1781 string(getComputeAsmShaderPreamble()) +
1783 "OpName %main \"main\"\n"
1784 "OpName %id \"gl_GlobalInvocationID\"\n"
1786 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1787 "OpDecorate %f32arr100 ArrayStride 4\n"
1789 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1791 "%hundred = OpConstant %u32 100\n"
1792 "%f32arr100 = OpTypeArray %f32 %hundred\n"
1793 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
1794 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
1795 "%buf = OpTypeStruct %f32arr100\n"
1796 "%bufptr = OpTypePointer Uniform %buf\n"
1797 "%indata = OpVariable %bufptr Uniform\n"
1798 "%outdata = OpVariable %bufptr Uniform\n"
1800 "%id = OpVariable %uvec3ptr Input\n"
1801 "%zero = OpConstant %i32 0\n"
1803 "%main = OpFunction %void None %voidf\n"
1804 "%label = OpLabel\n"
1805 "%var = OpVariable %f32arr100ptr_f Function\n"
1806 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
1807 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
1808 " OpCopyMemory %var %inarr\n"
1809 " OpCopyMemory %outarr %var\n"
1813 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1814 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1815 spec2.numWorkGroups = IVec3(1, 1, 1);
1817 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
1819 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
1820 ComputeShaderSpec spec3;
1821 vector<float> inputFloats3 (16);
1822 vector<float> outputFloats3 (16);
1824 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
1826 for (size_t ndx = 0; ndx < 16; ++ndx)
1827 outputFloats3[ndx] = inputFloats3[ndx];
1830 string(getComputeAsmShaderPreamble()) +
1832 "OpName %main \"main\"\n"
1833 "OpName %id \"gl_GlobalInvocationID\"\n"
1835 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1836 "OpMemberDecorate %buf 0 Offset 0\n"
1837 "OpMemberDecorate %buf 1 Offset 16\n"
1838 "OpMemberDecorate %buf 2 Offset 32\n"
1839 "OpMemberDecorate %buf 3 Offset 48\n"
1841 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1843 "%vec4 = OpTypeVector %f32 4\n"
1844 "%buf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
1845 "%bufptr = OpTypePointer Uniform %buf\n"
1846 "%indata = OpVariable %bufptr Uniform\n"
1847 "%outdata = OpVariable %bufptr Uniform\n"
1848 "%vec4stptr = OpTypePointer Function %buf\n"
1850 "%id = OpVariable %uvec3ptr Input\n"
1851 "%zero = OpConstant %i32 0\n"
1853 "%main = OpFunction %void None %voidf\n"
1854 "%label = OpLabel\n"
1855 "%var = OpVariable %vec4stptr Function\n"
1856 " OpCopyMemory %var %indata\n"
1857 " OpCopyMemory %outdata %var\n"
1861 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1862 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
1863 spec3.numWorkGroups = IVec3(1, 1, 1);
1865 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
1867 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
1868 ComputeShaderSpec spec4;
1869 vector<float> inputFloats4 (numElements);
1870 vector<float> outputFloats4 (numElements);
1872 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
1874 for (size_t ndx = 0; ndx < numElements; ++ndx)
1875 outputFloats4[ndx] = -inputFloats4[ndx];
1878 string(getComputeAsmShaderPreamble()) +
1880 "OpName %main \"main\"\n"
1881 "OpName %id \"gl_GlobalInvocationID\"\n"
1883 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1885 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
1887 "%f32ptr_f = OpTypePointer Function %f32\n"
1888 "%id = OpVariable %uvec3ptr Input\n"
1889 "%zero = OpConstant %i32 0\n"
1891 "%main = OpFunction %void None %voidf\n"
1892 "%label = OpLabel\n"
1893 "%var = OpVariable %f32ptr_f Function\n"
1894 "%idval = OpLoad %uvec3 %id\n"
1895 "%x = OpCompositeExtract %u32 %idval 0\n"
1896 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1897 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1898 " OpCopyMemory %var %inloc\n"
1899 "%val = OpLoad %f32 %var\n"
1900 "%neg = OpFNegate %f32 %val\n"
1901 " OpStore %outloc %neg\n"
1905 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
1906 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
1907 spec4.numWorkGroups = IVec3(numElements, 1, 1);
1909 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
1911 return group.release();
1914 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
1916 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
1917 ComputeShaderSpec spec;
1918 de::Random rnd (deStringHash(group->getName()));
1919 const int numElements = 100;
1920 vector<float> inputFloats (numElements, 0);
1921 vector<float> outputFloats (numElements, 0);
1923 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
1925 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1926 floorAll(inputFloats);
1928 for (size_t ndx = 0; ndx < numElements; ++ndx)
1929 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
1932 string(getComputeAsmShaderPreamble()) +
1934 "OpName %main \"main\"\n"
1935 "OpName %id \"gl_GlobalInvocationID\"\n"
1937 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1939 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1941 "%fmat = OpTypeMatrix %fvec3 3\n"
1942 "%three = OpConstant %u32 3\n"
1943 "%farr = OpTypeArray %f32 %three\n"
1944 "%fst = OpTypeStruct %f32 %f32\n"
1946 + string(getComputeAsmInputOutputBuffer()) +
1948 "%id = OpVariable %uvec3ptr Input\n"
1949 "%zero = OpConstant %i32 0\n"
1950 "%c_f = OpConstant %f32 1.5\n"
1951 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
1952 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
1953 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
1954 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
1956 "%main = OpFunction %void None %voidf\n"
1957 "%label = OpLabel\n"
1958 "%c_f_copy = OpCopyObject %f32 %c_f\n"
1959 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
1960 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
1961 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
1962 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
1963 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
1964 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
1965 "%farr_elem = OpCompositeExtract %f32 %c_farr_copy 2\n"
1966 "%fst_elem = OpCompositeExtract %f32 %c_fst_copy 1\n"
1967 // Add up. 1.5 * 5 = 7.5.
1968 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
1969 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
1970 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
1971 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
1973 "%idval = OpLoad %uvec3 %id\n"
1974 "%x = OpCompositeExtract %u32 %idval 0\n"
1975 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1976 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1977 "%inval = OpLoad %f32 %inloc\n"
1978 "%add = OpFAdd %f32 %add4 %inval\n"
1979 " OpStore %outloc %add\n"
1982 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1983 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1984 spec.numWorkGroups = IVec3(numElements, 1, 1);
1986 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
1988 return group.release();
1990 // Assembly code used for testing OpUnreachable is based on GLSL source code:
1994 // layout(std140, set = 0, binding = 0) readonly buffer Input {
1995 // float elements[];
1997 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
1998 // float elements[];
2001 // void not_called_func() {
2002 // // place OpUnreachable here
2005 // uint modulo4(uint val) {
2006 // switch (val % uint(4)) {
2007 // case 0: return 3;
2008 // case 1: return 2;
2009 // case 2: return 1;
2010 // case 3: return 0;
2011 // default: return 100; // place OpUnreachable here
2017 // // place OpUnreachable here
2021 // uint x = gl_GlobalInvocationID.x;
2022 // if (const5() > modulo4(1000)) {
2023 // output_data.elements[x] = -input_data.elements[x];
2025 // // place OpUnreachable here
2026 // output_data.elements[x] = input_data.elements[x];
2030 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
2032 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
2033 ComputeShaderSpec spec;
2034 de::Random rnd (deStringHash(group->getName()));
2035 const int numElements = 100;
2036 vector<float> positiveFloats (numElements, 0);
2037 vector<float> negativeFloats (numElements, 0);
2039 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2041 for (size_t ndx = 0; ndx < numElements; ++ndx)
2042 negativeFloats[ndx] = -positiveFloats[ndx];
2045 string(getComputeAsmShaderPreamble()) +
2047 "OpSource GLSL 430\n"
2048 "OpName %main \"main\"\n"
2049 "OpName %func_not_called_func \"not_called_func(\"\n"
2050 "OpName %func_modulo4 \"modulo4(u1;\"\n"
2051 "OpName %func_const5 \"const5(\"\n"
2052 "OpName %id \"gl_GlobalInvocationID\"\n"
2054 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2056 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2058 "%u32ptr = OpTypePointer Function %u32\n"
2059 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
2060 "%unitf = OpTypeFunction %u32\n"
2062 "%id = OpVariable %uvec3ptr Input\n"
2063 "%zero = OpConstant %u32 0\n"
2064 "%one = OpConstant %u32 1\n"
2065 "%two = OpConstant %u32 2\n"
2066 "%three = OpConstant %u32 3\n"
2067 "%four = OpConstant %u32 4\n"
2068 "%five = OpConstant %u32 5\n"
2069 "%hundred = OpConstant %u32 100\n"
2070 "%thousand = OpConstant %u32 1000\n"
2072 + string(getComputeAsmInputOutputBuffer()) +
2075 "%main = OpFunction %void None %voidf\n"
2076 "%main_entry = OpLabel\n"
2077 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
2078 "%idval = OpLoad %uvec3 %id\n"
2079 "%x = OpCompositeExtract %u32 %idval 0\n"
2080 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2081 "%inval = OpLoad %f32 %inloc\n"
2082 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2083 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
2084 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
2085 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
2086 " OpSelectionMerge %if_end None\n"
2087 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2088 "%if_true = OpLabel\n"
2089 "%negate = OpFNegate %f32 %inval\n"
2090 " OpStore %outloc %negate\n"
2091 " OpBranch %if_end\n"
2092 "%if_false = OpLabel\n"
2093 " OpUnreachable\n" // Unreachable else branch for if statement
2094 "%if_end = OpLabel\n"
2098 // not_called_function()
2099 "%func_not_called_func = OpFunction %void None %voidf\n"
2100 "%not_called_func_entry = OpLabel\n"
2101 " OpUnreachable\n" // Unreachable entry block in not called static function
2105 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
2106 "%valptr = OpFunctionParameter %u32ptr\n"
2107 "%modulo4_entry = OpLabel\n"
2108 "%val = OpLoad %u32 %valptr\n"
2109 "%modulo = OpUMod %u32 %val %four\n"
2110 " OpSelectionMerge %switch_merge None\n"
2111 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
2112 "%case0 = OpLabel\n"
2113 " OpReturnValue %three\n"
2114 "%case1 = OpLabel\n"
2115 " OpReturnValue %two\n"
2116 "%case2 = OpLabel\n"
2117 " OpReturnValue %one\n"
2118 "%case3 = OpLabel\n"
2119 " OpReturnValue %zero\n"
2120 "%default = OpLabel\n"
2121 " OpUnreachable\n" // Unreachable default case for switch statement
2122 "%switch_merge = OpLabel\n"
2123 " OpUnreachable\n" // Unreachable merge block for switch statement
2127 "%func_const5 = OpFunction %u32 None %unitf\n"
2128 "%const5_entry = OpLabel\n"
2129 " OpReturnValue %five\n"
2130 "%unreachable = OpLabel\n"
2131 " OpUnreachable\n" // Unreachable block in function
2133 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2134 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2135 spec.numWorkGroups = IVec3(numElements, 1, 1);
2137 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
2139 return group.release();
2142 // Assembly code used for testing decoration group is based on GLSL source code:
2146 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
2147 // float elements[];
2149 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
2150 // float elements[];
2152 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
2153 // float elements[];
2155 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
2156 // float elements[];
2158 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
2159 // float elements[];
2161 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
2162 // float elements[];
2166 // uint x = gl_GlobalInvocationID.x;
2167 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
2169 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
2171 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
2172 ComputeShaderSpec spec;
2173 de::Random rnd (deStringHash(group->getName()));
2174 const int numElements = 100;
2175 vector<float> inputFloats0 (numElements, 0);
2176 vector<float> inputFloats1 (numElements, 0);
2177 vector<float> inputFloats2 (numElements, 0);
2178 vector<float> inputFloats3 (numElements, 0);
2179 vector<float> inputFloats4 (numElements, 0);
2180 vector<float> outputFloats (numElements, 0);
2182 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
2183 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
2184 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
2185 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
2186 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
2188 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2189 floorAll(inputFloats0);
2190 floorAll(inputFloats1);
2191 floorAll(inputFloats2);
2192 floorAll(inputFloats3);
2193 floorAll(inputFloats4);
2195 for (size_t ndx = 0; ndx < numElements; ++ndx)
2196 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
2199 string(getComputeAsmShaderPreamble()) +
2201 "OpSource GLSL 430\n"
2202 "OpName %main \"main\"\n"
2203 "OpName %id \"gl_GlobalInvocationID\"\n"
2205 // Not using group decoration on variable.
2206 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2207 // Not using group decoration on type.
2208 "OpDecorate %f32arr ArrayStride 4\n"
2210 "OpDecorate %groups BufferBlock\n"
2211 "OpDecorate %groupm Offset 0\n"
2212 "%groups = OpDecorationGroup\n"
2213 "%groupm = OpDecorationGroup\n"
2215 // Group decoration on multiple structs.
2216 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
2217 // Group decoration on multiple struct members.
2218 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
2220 "OpDecorate %group1 DescriptorSet 0\n"
2221 "OpDecorate %group3 DescriptorSet 0\n"
2222 "OpDecorate %group3 NonWritable\n"
2223 "OpDecorate %group3 Restrict\n"
2224 "%group0 = OpDecorationGroup\n"
2225 "%group1 = OpDecorationGroup\n"
2226 "%group3 = OpDecorationGroup\n"
2228 // Applying the same decoration group multiple times.
2229 "OpGroupDecorate %group1 %outdata\n"
2230 "OpGroupDecorate %group1 %outdata\n"
2231 "OpGroupDecorate %group1 %outdata\n"
2232 "OpDecorate %outdata DescriptorSet 0\n"
2233 "OpDecorate %outdata Binding 5\n"
2234 // Applying decoration group containing nothing.
2235 "OpGroupDecorate %group0 %indata0\n"
2236 "OpDecorate %indata0 DescriptorSet 0\n"
2237 "OpDecorate %indata0 Binding 0\n"
2238 // Applying decoration group containing one decoration.
2239 "OpGroupDecorate %group1 %indata1\n"
2240 "OpDecorate %indata1 Binding 1\n"
2241 // Applying decoration group containing multiple decorations.
2242 "OpGroupDecorate %group3 %indata2 %indata3\n"
2243 "OpDecorate %indata2 Binding 2\n"
2244 "OpDecorate %indata3 Binding 3\n"
2245 // Applying multiple decoration groups (with overlapping).
2246 "OpGroupDecorate %group0 %indata4\n"
2247 "OpGroupDecorate %group1 %indata4\n"
2248 "OpGroupDecorate %group3 %indata4\n"
2249 "OpDecorate %indata4 Binding 4\n"
2251 + string(getComputeAsmCommonTypes()) +
2253 "%id = OpVariable %uvec3ptr Input\n"
2254 "%zero = OpConstant %i32 0\n"
2256 "%outbuf = OpTypeStruct %f32arr\n"
2257 "%outbufptr = OpTypePointer Uniform %outbuf\n"
2258 "%outdata = OpVariable %outbufptr Uniform\n"
2259 "%inbuf0 = OpTypeStruct %f32arr\n"
2260 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
2261 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
2262 "%inbuf1 = OpTypeStruct %f32arr\n"
2263 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
2264 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
2265 "%inbuf2 = OpTypeStruct %f32arr\n"
2266 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
2267 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
2268 "%inbuf3 = OpTypeStruct %f32arr\n"
2269 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
2270 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
2271 "%inbuf4 = OpTypeStruct %f32arr\n"
2272 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
2273 "%indata4 = OpVariable %inbufptr Uniform\n"
2275 "%main = OpFunction %void None %voidf\n"
2276 "%label = OpLabel\n"
2277 "%idval = OpLoad %uvec3 %id\n"
2278 "%x = OpCompositeExtract %u32 %idval 0\n"
2279 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
2280 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
2281 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
2282 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
2283 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
2284 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2285 "%inval0 = OpLoad %f32 %inloc0\n"
2286 "%inval1 = OpLoad %f32 %inloc1\n"
2287 "%inval2 = OpLoad %f32 %inloc2\n"
2288 "%inval3 = OpLoad %f32 %inloc3\n"
2289 "%inval4 = OpLoad %f32 %inloc4\n"
2290 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
2291 "%add1 = OpFAdd %f32 %add0 %inval2\n"
2292 "%add2 = OpFAdd %f32 %add1 %inval3\n"
2293 "%add = OpFAdd %f32 %add2 %inval4\n"
2294 " OpStore %outloc %add\n"
2297 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
2298 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
2299 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2300 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2301 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2302 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2303 spec.numWorkGroups = IVec3(numElements, 1, 1);
2305 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
2307 return group.release();
2310 struct SpecConstantTwoIntCase
2312 const char* caseName;
2313 const char* scDefinition0;
2314 const char* scDefinition1;
2315 const char* scResultType;
2316 const char* scOperation;
2317 deInt32 scActualValue0;
2318 deInt32 scActualValue1;
2319 const char* resultOperation;
2320 vector<deInt32> expectedOutput;
2322 SpecConstantTwoIntCase (const char* name,
2323 const char* definition0,
2324 const char* definition1,
2325 const char* resultType,
2326 const char* operation,
2329 const char* resultOp,
2330 const vector<deInt32>& output)
2332 , scDefinition0 (definition0)
2333 , scDefinition1 (definition1)
2334 , scResultType (resultType)
2335 , scOperation (operation)
2336 , scActualValue0 (value0)
2337 , scActualValue1 (value1)
2338 , resultOperation (resultOp)
2339 , expectedOutput (output) {}
2342 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
2344 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
2345 vector<SpecConstantTwoIntCase> cases;
2346 de::Random rnd (deStringHash(group->getName()));
2347 const int numElements = 100;
2348 vector<deInt32> inputInts (numElements, 0);
2349 vector<deInt32> outputInts1 (numElements, 0);
2350 vector<deInt32> outputInts2 (numElements, 0);
2351 vector<deInt32> outputInts3 (numElements, 0);
2352 vector<deInt32> outputInts4 (numElements, 0);
2353 const StringTemplate shaderTemplate (
2354 string(getComputeAsmShaderPreamble()) +
2356 "OpName %main \"main\"\n"
2357 "OpName %id \"gl_GlobalInvocationID\"\n"
2359 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2360 "OpDecorate %sc_0 SpecId 0\n"
2361 "OpDecorate %sc_1 SpecId 1\n"
2362 "OpDecorate %i32arr ArrayStride 4\n"
2364 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2366 "%buf = OpTypeStruct %i32arr\n"
2367 "%bufptr = OpTypePointer Uniform %buf\n"
2368 "%indata = OpVariable %bufptr Uniform\n"
2369 "%outdata = OpVariable %bufptr Uniform\n"
2371 "%id = OpVariable %uvec3ptr Input\n"
2372 "%zero = OpConstant %i32 0\n"
2374 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
2375 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
2376 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
2378 "%main = OpFunction %void None %voidf\n"
2379 "%label = OpLabel\n"
2380 "%idval = OpLoad %uvec3 %id\n"
2381 "%x = OpCompositeExtract %u32 %idval 0\n"
2382 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2383 "%inval = OpLoad %i32 %inloc\n"
2384 "%final = ${GEN_RESULT}\n"
2385 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2386 " OpStore %outloc %final\n"
2388 " OpFunctionEnd\n");
2390 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
2392 for (size_t ndx = 0; ndx < numElements; ++ndx)
2394 outputInts1[ndx] = inputInts[ndx] + 42;
2395 outputInts2[ndx] = inputInts[ndx];
2396 outputInts3[ndx] = inputInts[ndx] - 11200;
2397 outputInts4[ndx] = inputInts[ndx] + 1;
2400 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
2401 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
2402 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
2404 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
2405 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
2406 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
2407 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
2408 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
2409 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2410 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2411 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
2412 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
2413 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
2414 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
2415 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2416 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2417 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
2418 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
2419 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
2420 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2421 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
2422 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
2423 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
2424 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2425 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
2426 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
2427 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2428 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2429 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2430 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2431 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
2432 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
2433 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
2434 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
2435 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
2437 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2439 map<string, string> specializations;
2440 ComputeShaderSpec spec;
2442 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
2443 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
2444 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
2445 specializations["SC_OP"] = cases[caseNdx].scOperation;
2446 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
2448 spec.assembly = shaderTemplate.specialize(specializations);
2449 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2450 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
2451 spec.numWorkGroups = IVec3(numElements, 1, 1);
2452 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
2453 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
2455 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec));
2458 ComputeShaderSpec spec;
2461 string(getComputeAsmShaderPreamble()) +
2463 "OpName %main \"main\"\n"
2464 "OpName %id \"gl_GlobalInvocationID\"\n"
2466 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2467 "OpDecorate %sc_0 SpecId 0\n"
2468 "OpDecorate %sc_1 SpecId 1\n"
2469 "OpDecorate %sc_2 SpecId 2\n"
2470 "OpDecorate %i32arr ArrayStride 4\n"
2472 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2474 "%ivec3 = OpTypeVector %i32 3\n"
2475 "%buf = OpTypeStruct %i32arr\n"
2476 "%bufptr = OpTypePointer Uniform %buf\n"
2477 "%indata = OpVariable %bufptr Uniform\n"
2478 "%outdata = OpVariable %bufptr Uniform\n"
2480 "%id = OpVariable %uvec3ptr Input\n"
2481 "%zero = OpConstant %i32 0\n"
2482 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
2483 "%vec3_undef = OpUndef %ivec3\n"
2485 "%sc_0 = OpSpecConstant %i32 0\n"
2486 "%sc_1 = OpSpecConstant %i32 0\n"
2487 "%sc_2 = OpSpecConstant %i32 0\n"
2488 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
2489 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
2490 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
2491 "%sc_vec3_0_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
2492 "%sc_vec3_1_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
2493 "%sc_vec3_2_s = OpSpecConstantOp %ivec3 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
2494 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
2495 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
2496 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
2497 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
2498 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
2499 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
2500 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
2502 "%main = OpFunction %void None %voidf\n"
2503 "%label = OpLabel\n"
2504 "%idval = OpLoad %uvec3 %id\n"
2505 "%x = OpCompositeExtract %u32 %idval 0\n"
2506 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2507 "%inval = OpLoad %i32 %inloc\n"
2508 "%final = OpIAdd %i32 %inval %sc_final\n"
2509 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2510 " OpStore %outloc %final\n"
2513 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2514 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
2515 spec.numWorkGroups = IVec3(numElements, 1, 1);
2516 spec.specConstants.push_back(123);
2517 spec.specConstants.push_back(56);
2518 spec.specConstants.push_back(-77);
2520 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
2522 return group.release();
2525 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
2527 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
2528 ComputeShaderSpec spec1;
2529 ComputeShaderSpec spec2;
2530 ComputeShaderSpec spec3;
2531 de::Random rnd (deStringHash(group->getName()));
2532 const int numElements = 100;
2533 vector<float> inputFloats (numElements, 0);
2534 vector<float> outputFloats1 (numElements, 0);
2535 vector<float> outputFloats2 (numElements, 0);
2536 vector<float> outputFloats3 (numElements, 0);
2538 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
2540 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2541 floorAll(inputFloats);
2543 for (size_t ndx = 0; ndx < numElements; ++ndx)
2547 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
2548 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
2549 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
2552 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
2553 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
2557 string(getComputeAsmShaderPreamble()) +
2559 "OpSource GLSL 430\n"
2560 "OpName %main \"main\"\n"
2561 "OpName %id \"gl_GlobalInvocationID\"\n"
2563 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2565 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2567 "%id = OpVariable %uvec3ptr Input\n"
2568 "%zero = OpConstant %i32 0\n"
2569 "%three = OpConstant %u32 3\n"
2570 "%constf5p5 = OpConstant %f32 5.5\n"
2571 "%constf20p5 = OpConstant %f32 20.5\n"
2572 "%constf1p75 = OpConstant %f32 1.75\n"
2573 "%constf8p5 = OpConstant %f32 8.5\n"
2574 "%constf6p5 = OpConstant %f32 6.5\n"
2576 "%main = OpFunction %void None %voidf\n"
2577 "%entry = OpLabel\n"
2578 "%idval = OpLoad %uvec3 %id\n"
2579 "%x = OpCompositeExtract %u32 %idval 0\n"
2580 "%selector = OpUMod %u32 %x %three\n"
2581 " OpSelectionMerge %phi None\n"
2582 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
2584 // Case 1 before OpPhi.
2585 "%case1 = OpLabel\n"
2588 "%default = OpLabel\n"
2592 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
2593 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2594 "%inval = OpLoad %f32 %inloc\n"
2595 "%add = OpFAdd %f32 %inval %operand\n"
2596 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2597 " OpStore %outloc %add\n"
2600 // Case 0 after OpPhi.
2601 "%case0 = OpLabel\n"
2605 // Case 2 after OpPhi.
2606 "%case2 = OpLabel\n"
2610 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2611 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
2612 spec1.numWorkGroups = IVec3(numElements, 1, 1);
2614 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
2617 string(getComputeAsmShaderPreamble()) +
2619 "OpName %main \"main\"\n"
2620 "OpName %id \"gl_GlobalInvocationID\"\n"
2622 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2624 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2626 "%id = OpVariable %uvec3ptr Input\n"
2627 "%zero = OpConstant %i32 0\n"
2628 "%one = OpConstant %i32 1\n"
2629 "%three = OpConstant %i32 3\n"
2630 "%constf6p5 = OpConstant %f32 6.5\n"
2632 "%main = OpFunction %void None %voidf\n"
2633 "%entry = OpLabel\n"
2634 "%idval = OpLoad %uvec3 %id\n"
2635 "%x = OpCompositeExtract %u32 %idval 0\n"
2636 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2637 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2638 "%inval = OpLoad %f32 %inloc\n"
2642 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
2643 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
2644 "%step_next = OpIAdd %i32 %step %one\n"
2645 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
2646 "%still_loop = OpSLessThan %bool %step %three\n"
2647 " OpLoopMerge %exit %phi None\n"
2648 " OpBranchConditional %still_loop %phi %exit\n"
2651 " OpStore %outloc %accum\n"
2654 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2655 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2656 spec2.numWorkGroups = IVec3(numElements, 1, 1);
2658 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
2661 string(getComputeAsmShaderPreamble()) +
2663 "OpName %main \"main\"\n"
2664 "OpName %id \"gl_GlobalInvocationID\"\n"
2666 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2668 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2670 "%f32ptr_f = OpTypePointer Function %f32\n"
2671 "%id = OpVariable %uvec3ptr Input\n"
2672 "%true = OpConstantTrue %bool\n"
2673 "%false = OpConstantFalse %bool\n"
2674 "%zero = OpConstant %i32 0\n"
2675 "%constf8p5 = OpConstant %f32 8.5\n"
2677 "%main = OpFunction %void None %voidf\n"
2678 "%entry = OpLabel\n"
2679 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
2680 "%idval = OpLoad %uvec3 %id\n"
2681 "%x = OpCompositeExtract %u32 %idval 0\n"
2682 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2683 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2684 "%a_init = OpLoad %f32 %inloc\n"
2685 "%b_init = OpLoad %f32 %b\n"
2689 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
2690 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
2691 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
2692 " OpLoopMerge %exit %phi None\n"
2693 " OpBranchConditional %still_loop %phi %exit\n"
2696 "%sub = OpFSub %f32 %a_next %b_next\n"
2697 " OpStore %outloc %sub\n"
2700 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2701 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
2702 spec3.numWorkGroups = IVec3(numElements, 1, 1);
2704 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
2706 return group.release();
2709 // Assembly code used for testing block order is based on GLSL source code:
2713 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2714 // float elements[];
2716 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2717 // float elements[];
2721 // uint x = gl_GlobalInvocationID.x;
2722 // output_data.elements[x] = input_data.elements[x];
2723 // if (x > uint(50)) {
2724 // switch (x % uint(3)) {
2725 // case 0: output_data.elements[x] += 1.5f; break;
2726 // case 1: output_data.elements[x] += 42.f; break;
2727 // case 2: output_data.elements[x] -= 27.f; break;
2731 // output_data.elements[x] = -input_data.elements[x];
2734 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
2736 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
2737 ComputeShaderSpec spec;
2738 de::Random rnd (deStringHash(group->getName()));
2739 const int numElements = 100;
2740 vector<float> inputFloats (numElements, 0);
2741 vector<float> outputFloats (numElements, 0);
2743 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2745 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2746 floorAll(inputFloats);
2748 for (size_t ndx = 0; ndx <= 50; ++ndx)
2749 outputFloats[ndx] = -inputFloats[ndx];
2751 for (size_t ndx = 51; ndx < numElements; ++ndx)
2755 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
2756 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
2757 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
2763 string(getComputeAsmShaderPreamble()) +
2765 "OpSource GLSL 430\n"
2766 "OpName %main \"main\"\n"
2767 "OpName %id \"gl_GlobalInvocationID\"\n"
2769 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2771 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2773 "%u32ptr = OpTypePointer Function %u32\n"
2774 "%u32ptr_input = OpTypePointer Input %u32\n"
2776 + string(getComputeAsmInputOutputBuffer()) +
2778 "%id = OpVariable %uvec3ptr Input\n"
2779 "%zero = OpConstant %i32 0\n"
2780 "%const3 = OpConstant %u32 3\n"
2781 "%const50 = OpConstant %u32 50\n"
2782 "%constf1p5 = OpConstant %f32 1.5\n"
2783 "%constf27 = OpConstant %f32 27.0\n"
2784 "%constf42 = OpConstant %f32 42.0\n"
2786 "%main = OpFunction %void None %voidf\n"
2789 "%entry = OpLabel\n"
2791 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
2792 "%xvar = OpVariable %u32ptr Function\n"
2793 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
2794 "%x = OpLoad %u32 %xptr\n"
2795 " OpStore %xvar %x\n"
2797 "%cmp = OpUGreaterThan %bool %x %const50\n"
2798 " OpSelectionMerge %if_merge None\n"
2799 " OpBranchConditional %cmp %if_true %if_false\n"
2801 // False branch for if-statement: placed in the middle of switch cases and before true branch.
2802 "%if_false = OpLabel\n"
2803 "%x_f = OpLoad %u32 %xvar\n"
2804 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
2805 "%inval_f = OpLoad %f32 %inloc_f\n"
2806 "%negate = OpFNegate %f32 %inval_f\n"
2807 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
2808 " OpStore %outloc_f %negate\n"
2809 " OpBranch %if_merge\n"
2811 // Merge block for if-statement: placed in the middle of true and false branch.
2812 "%if_merge = OpLabel\n"
2815 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
2816 "%if_true = OpLabel\n"
2817 "%xval_t = OpLoad %u32 %xvar\n"
2818 "%mod = OpUMod %u32 %xval_t %const3\n"
2819 " OpSelectionMerge %switch_merge None\n"
2820 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
2822 // Merge block for switch-statement: placed before the case
2823 // bodies. But it must follow OpSwitch which dominates it.
2824 "%switch_merge = OpLabel\n"
2825 " OpBranch %if_merge\n"
2827 // Case 1 for switch-statement: placed before case 0.
2828 // It must follow the OpSwitch that dominates it.
2829 "%case1 = OpLabel\n"
2830 "%x_1 = OpLoad %u32 %xvar\n"
2831 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
2832 "%inval_1 = OpLoad %f32 %inloc_1\n"
2833 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
2834 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
2835 " OpStore %outloc_1 %addf42\n"
2836 " OpBranch %switch_merge\n"
2838 // Case 2 for switch-statement.
2839 "%case2 = OpLabel\n"
2840 "%x_2 = OpLoad %u32 %xvar\n"
2841 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
2842 "%inval_2 = OpLoad %f32 %inloc_2\n"
2843 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
2844 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
2845 " OpStore %outloc_2 %subf27\n"
2846 " OpBranch %switch_merge\n"
2848 // Default case for switch-statement: placed in the middle of normal cases.
2849 "%default = OpLabel\n"
2850 " OpBranch %switch_merge\n"
2852 // Case 0 for switch-statement: out of order.
2853 "%case0 = OpLabel\n"
2854 "%x_0 = OpLoad %u32 %xvar\n"
2855 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
2856 "%inval_0 = OpLoad %f32 %inloc_0\n"
2857 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
2858 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
2859 " OpStore %outloc_0 %addf1p5\n"
2860 " OpBranch %switch_merge\n"
2863 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2864 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2865 spec.numWorkGroups = IVec3(numElements, 1, 1);
2867 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
2869 return group.release();
2872 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
2874 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
2875 ComputeShaderSpec spec1;
2876 ComputeShaderSpec spec2;
2877 de::Random rnd (deStringHash(group->getName()));
2878 const int numElements = 100;
2879 vector<float> inputFloats (numElements, 0);
2880 vector<float> outputFloats1 (numElements, 0);
2881 vector<float> outputFloats2 (numElements, 0);
2882 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
2884 for (size_t ndx = 0; ndx < numElements; ++ndx)
2886 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
2887 outputFloats2[ndx] = -inputFloats[ndx];
2890 const string assembly(
2891 "OpCapability Shader\n"
2892 "OpCapability ClipDistance\n"
2893 "OpMemoryModel Logical GLSL450\n"
2894 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
2895 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
2896 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
2897 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
2898 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
2899 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
2901 "OpName %comp_main1 \"entrypoint1\"\n"
2902 "OpName %comp_main2 \"entrypoint2\"\n"
2903 "OpName %vert_main \"entrypoint2\"\n"
2904 "OpName %id \"gl_GlobalInvocationID\"\n"
2905 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
2906 "OpName %vertexIndex \"gl_VertexIndex\"\n"
2907 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
2908 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
2909 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
2910 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
2912 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2913 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
2914 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
2915 "OpDecorate %vert_builtin_st Block\n"
2916 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
2917 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
2918 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
2920 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2922 "%zero = OpConstant %i32 0\n"
2923 "%one = OpConstant %u32 1\n"
2924 "%c_f32_1 = OpConstant %f32 1\n"
2926 "%i32inputptr = OpTypePointer Input %i32\n"
2927 "%vec4 = OpTypeVector %f32 4\n"
2928 "%vec4ptr = OpTypePointer Output %vec4\n"
2929 "%f32arr1 = OpTypeArray %f32 %one\n"
2930 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
2931 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
2932 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
2934 "%id = OpVariable %uvec3ptr Input\n"
2935 "%vertexIndex = OpVariable %i32inputptr Input\n"
2936 "%instanceIndex = OpVariable %i32inputptr Input\n"
2937 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
2939 // gl_Position = vec4(1.);
2940 "%vert_main = OpFunction %void None %voidf\n"
2941 "%vert_entry = OpLabel\n"
2942 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
2943 " OpStore %position %c_vec4_1\n"
2948 "%comp_main1 = OpFunction %void None %voidf\n"
2949 "%comp1_entry = OpLabel\n"
2950 "%idval1 = OpLoad %uvec3 %id\n"
2951 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
2952 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
2953 "%inval1 = OpLoad %f32 %inloc1\n"
2954 "%add = OpFAdd %f32 %inval1 %inval1\n"
2955 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
2956 " OpStore %outloc1 %add\n"
2961 "%comp_main2 = OpFunction %void None %voidf\n"
2962 "%comp2_entry = OpLabel\n"
2963 "%idval2 = OpLoad %uvec3 %id\n"
2964 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
2965 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
2966 "%inval2 = OpLoad %f32 %inloc2\n"
2967 "%neg = OpFNegate %f32 %inval2\n"
2968 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
2969 " OpStore %outloc2 %neg\n"
2971 " OpFunctionEnd\n");
2973 spec1.assembly = assembly;
2974 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2975 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
2976 spec1.numWorkGroups = IVec3(numElements, 1, 1);
2977 spec1.entryPoint = "entrypoint1";
2979 spec2.assembly = assembly;
2980 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2981 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2982 spec2.numWorkGroups = IVec3(numElements, 1, 1);
2983 spec2.entryPoint = "entrypoint2";
2985 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
2986 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
2988 return group.release();
2991 inline std::string makeLongUTF8String (size_t num4ByteChars)
2993 // An example of a longest valid UTF-8 character. Be explicit about the
2994 // character type because Microsoft compilers can otherwise interpret the
2995 // character string as being over wide (16-bit) characters. Ideally, we
2996 // would just use a C++11 UTF-8 string literal, but we want to support older
2997 // Microsoft compilers.
2998 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
2999 std::string longString;
3000 longString.reserve(num4ByteChars * 4);
3001 for (size_t count = 0; count < num4ByteChars; count++)
3003 longString += earthAfrica;
3008 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
3010 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
3011 vector<CaseParameter> cases;
3012 de::Random rnd (deStringHash(group->getName()));
3013 const int numElements = 100;
3014 vector<float> positiveFloats (numElements, 0);
3015 vector<float> negativeFloats (numElements, 0);
3016 const StringTemplate shaderTemplate (
3017 "OpCapability Shader\n"
3018 "OpMemoryModel Logical GLSL450\n"
3020 "OpEntryPoint GLCompute %main \"main\" %id\n"
3021 "OpExecutionMode %main LocalSize 1 1 1\n"
3025 "OpName %main \"main\"\n"
3026 "OpName %id \"gl_GlobalInvocationID\"\n"
3028 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3030 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3032 "%id = OpVariable %uvec3ptr Input\n"
3033 "%zero = OpConstant %i32 0\n"
3035 "%main = OpFunction %void None %voidf\n"
3036 "%label = OpLabel\n"
3037 "%idval = OpLoad %uvec3 %id\n"
3038 "%x = OpCompositeExtract %u32 %idval 0\n"
3039 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3040 "%inval = OpLoad %f32 %inloc\n"
3041 "%neg = OpFNegate %f32 %inval\n"
3042 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3043 " OpStore %outloc %neg\n"
3045 " OpFunctionEnd\n");
3047 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
3048 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
3049 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
3050 "OpSource GLSL 430 %fname"));
3051 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
3052 "OpSource GLSL 430 %fname"));
3053 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
3054 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
3055 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
3056 "OpSource GLSL 430 %fname \"\""));
3057 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
3058 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
3059 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
3060 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
3061 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
3062 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
3063 "OpSourceContinued \"id main() {}\""));
3064 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
3065 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3066 "OpSourceContinued \"\""));
3067 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
3068 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3069 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
3070 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
3071 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3072 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
3073 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
3074 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
3075 "OpSourceContinued \"void\"\n"
3076 "OpSourceContinued \"main()\"\n"
3077 "OpSourceContinued \"{}\""));
3078 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
3079 "OpSource GLSL 430 %fname \"\"\n"
3080 "OpSourceContinued \"#version 430\nvoid main() {}\""));
3082 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3084 for (size_t ndx = 0; ndx < numElements; ++ndx)
3085 negativeFloats[ndx] = -positiveFloats[ndx];
3087 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3089 map<string, string> specializations;
3090 ComputeShaderSpec spec;
3092 specializations["SOURCE"] = cases[caseNdx].param;
3093 spec.assembly = shaderTemplate.specialize(specializations);
3094 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3095 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3096 spec.numWorkGroups = IVec3(numElements, 1, 1);
3098 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3101 return group.release();
3104 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
3106 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
3107 vector<CaseParameter> cases;
3108 de::Random rnd (deStringHash(group->getName()));
3109 const int numElements = 100;
3110 vector<float> inputFloats (numElements, 0);
3111 vector<float> outputFloats (numElements, 0);
3112 const StringTemplate shaderTemplate (
3113 string(getComputeAsmShaderPreamble()) +
3115 "OpSourceExtension \"${EXTENSION}\"\n"
3117 "OpName %main \"main\"\n"
3118 "OpName %id \"gl_GlobalInvocationID\"\n"
3120 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3122 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3124 "%id = OpVariable %uvec3ptr Input\n"
3125 "%zero = OpConstant %i32 0\n"
3127 "%main = OpFunction %void None %voidf\n"
3128 "%label = OpLabel\n"
3129 "%idval = OpLoad %uvec3 %id\n"
3130 "%x = OpCompositeExtract %u32 %idval 0\n"
3131 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3132 "%inval = OpLoad %f32 %inloc\n"
3133 "%neg = OpFNegate %f32 %inval\n"
3134 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3135 " OpStore %outloc %neg\n"
3137 " OpFunctionEnd\n");
3139 cases.push_back(CaseParameter("empty_extension", ""));
3140 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
3141 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
3142 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
3143 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
3145 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
3147 for (size_t ndx = 0; ndx < numElements; ++ndx)
3148 outputFloats[ndx] = -inputFloats[ndx];
3150 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3152 map<string, string> specializations;
3153 ComputeShaderSpec spec;
3155 specializations["EXTENSION"] = cases[caseNdx].param;
3156 spec.assembly = shaderTemplate.specialize(specializations);
3157 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3158 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3159 spec.numWorkGroups = IVec3(numElements, 1, 1);
3161 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3164 return group.release();
3167 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
3168 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
3170 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
3171 vector<CaseParameter> cases;
3172 de::Random rnd (deStringHash(group->getName()));
3173 const int numElements = 100;
3174 vector<float> positiveFloats (numElements, 0);
3175 vector<float> negativeFloats (numElements, 0);
3176 const StringTemplate shaderTemplate (
3177 string(getComputeAsmShaderPreamble()) +
3179 "OpSource GLSL 430\n"
3180 "OpName %main \"main\"\n"
3181 "OpName %id \"gl_GlobalInvocationID\"\n"
3183 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3185 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3186 "%uvec2 = OpTypeVector %u32 2\n"
3187 "%bvec3 = OpTypeVector %bool 3\n"
3188 "%fvec4 = OpTypeVector %f32 4\n"
3189 "%fmat33 = OpTypeMatrix %fvec3 3\n"
3190 "%const100 = OpConstant %u32 100\n"
3191 "%uarr100 = OpTypeArray %i32 %const100\n"
3192 "%struct = OpTypeStruct %f32 %i32 %u32\n"
3193 "%pointer = OpTypePointer Function %i32\n"
3194 + string(getComputeAsmInputOutputBuffer()) +
3196 "%null = OpConstantNull ${TYPE}\n"
3198 "%id = OpVariable %uvec3ptr Input\n"
3199 "%zero = OpConstant %i32 0\n"
3201 "%main = OpFunction %void None %voidf\n"
3202 "%label = OpLabel\n"
3203 "%idval = OpLoad %uvec3 %id\n"
3204 "%x = OpCompositeExtract %u32 %idval 0\n"
3205 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3206 "%inval = OpLoad %f32 %inloc\n"
3207 "%neg = OpFNegate %f32 %inval\n"
3208 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3209 " OpStore %outloc %neg\n"
3211 " OpFunctionEnd\n");
3213 cases.push_back(CaseParameter("bool", "%bool"));
3214 cases.push_back(CaseParameter("sint32", "%i32"));
3215 cases.push_back(CaseParameter("uint32", "%u32"));
3216 cases.push_back(CaseParameter("float32", "%f32"));
3217 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
3218 cases.push_back(CaseParameter("vec3bool", "%bvec3"));
3219 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
3220 cases.push_back(CaseParameter("matrix", "%fmat33"));
3221 cases.push_back(CaseParameter("array", "%uarr100"));
3222 cases.push_back(CaseParameter("struct", "%struct"));
3223 cases.push_back(CaseParameter("pointer", "%pointer"));
3225 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3227 for (size_t ndx = 0; ndx < numElements; ++ndx)
3228 negativeFloats[ndx] = -positiveFloats[ndx];
3230 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3232 map<string, string> specializations;
3233 ComputeShaderSpec spec;
3235 specializations["TYPE"] = cases[caseNdx].param;
3236 spec.assembly = shaderTemplate.specialize(specializations);
3237 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3238 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3239 spec.numWorkGroups = IVec3(numElements, 1, 1);
3241 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3244 return group.release();
3247 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
3248 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
3250 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
3251 vector<CaseParameter> cases;
3252 de::Random rnd (deStringHash(group->getName()));
3253 const int numElements = 100;
3254 vector<float> positiveFloats (numElements, 0);
3255 vector<float> negativeFloats (numElements, 0);
3256 const StringTemplate shaderTemplate (
3257 string(getComputeAsmShaderPreamble()) +
3259 "OpSource GLSL 430\n"
3260 "OpName %main \"main\"\n"
3261 "OpName %id \"gl_GlobalInvocationID\"\n"
3263 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3265 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3267 "%id = OpVariable %uvec3ptr Input\n"
3268 "%zero = OpConstant %i32 0\n"
3272 "%main = OpFunction %void None %voidf\n"
3273 "%label = OpLabel\n"
3274 "%idval = OpLoad %uvec3 %id\n"
3275 "%x = OpCompositeExtract %u32 %idval 0\n"
3276 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3277 "%inval = OpLoad %f32 %inloc\n"
3278 "%neg = OpFNegate %f32 %inval\n"
3279 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3280 " OpStore %outloc %neg\n"
3282 " OpFunctionEnd\n");
3284 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
3285 "%const = OpConstantComposite %uvec3 %five %zero %five"));
3286 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
3287 "%ten = OpConstant %f32 10.\n"
3288 "%fzero = OpConstant %f32 0.\n"
3289 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
3290 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
3291 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
3292 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
3293 "%fzero = OpConstant %f32 0.\n"
3294 "%one = OpConstant %f32 1.\n"
3295 "%point5 = OpConstant %f32 0.5\n"
3296 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
3297 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
3298 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
3299 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
3300 "%st2 = OpTypeStruct %i32 %i32\n"
3301 "%struct = OpTypeStruct %st1 %st2\n"
3302 "%point5 = OpConstant %f32 0.5\n"
3303 "%one = OpConstant %u32 1\n"
3304 "%ten = OpConstant %i32 10\n"
3305 "%st1val = OpConstantComposite %st1 %one %point5\n"
3306 "%st2val = OpConstantComposite %st2 %ten %ten\n"
3307 "%const = OpConstantComposite %struct %st1val %st2val"));
3309 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3311 for (size_t ndx = 0; ndx < numElements; ++ndx)
3312 negativeFloats[ndx] = -positiveFloats[ndx];
3314 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3316 map<string, string> specializations;
3317 ComputeShaderSpec spec;
3319 specializations["CONSTANT"] = cases[caseNdx].param;
3320 spec.assembly = shaderTemplate.specialize(specializations);
3321 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3322 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3323 spec.numWorkGroups = IVec3(numElements, 1, 1);
3325 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3328 return group.release();
3331 // Creates a floating point number with the given exponent, and significand
3332 // bits set. It can only create normalized numbers. Only the least significant
3333 // 24 bits of the significand will be examined. The final bit of the
3334 // significand will also be ignored. This allows alignment to be written
3335 // similarly to C99 hex-floats.
3336 // For example if you wanted to write 0x1.7f34p-12 you would call
3337 // constructNormalizedFloat(-12, 0x7f3400)
3338 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
3342 for (deInt32 idx = 0; idx < 23; ++idx)
3344 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
3348 return std::ldexp(f, exponent);
3351 // Compare instruction for the OpQuantizeF16 compute exact case.
3352 // Returns true if the output is what is expected from the test case.
3353 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
3355 if (outputAllocs.size() != 1)
3358 // We really just need this for size because we cannot compare Nans.
3359 const BufferSp& expectedOutput = expectedOutputs[0];
3360 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
3362 if (expectedOutput->getNumBytes() != 4*sizeof(float)) {
3366 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
3367 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
3372 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
3373 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
3378 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
3379 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
3384 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
3385 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
3392 // Checks that every output from a test-case is a float NaN.
3393 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
3395 if (outputAllocs.size() != 1)
3398 // We really just need this for size because we cannot compare Nans.
3399 const BufferSp& expectedOutput = expectedOutputs[0];
3400 const float* output_as_float = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
3402 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
3404 if (!deFloatIsNaN(output_as_float[idx]))
3413 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
3414 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
3416 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
3418 const std::string shader (
3419 string(getComputeAsmShaderPreamble()) +
3421 "OpSource GLSL 430\n"
3422 "OpName %main \"main\"\n"
3423 "OpName %id \"gl_GlobalInvocationID\"\n"
3425 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3427 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3429 "%id = OpVariable %uvec3ptr Input\n"
3430 "%zero = OpConstant %i32 0\n"
3432 "%main = OpFunction %void None %voidf\n"
3433 "%label = OpLabel\n"
3434 "%idval = OpLoad %uvec3 %id\n"
3435 "%x = OpCompositeExtract %u32 %idval 0\n"
3436 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3437 "%inval = OpLoad %f32 %inloc\n"
3438 "%quant = OpQuantizeToF16 %f32 %inval\n"
3439 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3440 " OpStore %outloc %quant\n"
3442 " OpFunctionEnd\n");
3445 ComputeShaderSpec spec;
3446 const deUint32 numElements = 100;
3447 vector<float> infinities;
3448 vector<float> results;
3450 infinities.reserve(numElements);
3451 results.reserve(numElements);
3453 for (size_t idx = 0; idx < numElements; ++idx)
3458 infinities.push_back(std::numeric_limits<float>::infinity());
3459 results.push_back(std::numeric_limits<float>::infinity());
3462 infinities.push_back(-std::numeric_limits<float>::infinity());
3463 results.push_back(-std::numeric_limits<float>::infinity());
3466 infinities.push_back(std::ldexp(1.0f, 16));
3467 results.push_back(std::numeric_limits<float>::infinity());
3470 infinities.push_back(std::ldexp(-1.0f, 32));
3471 results.push_back(-std::numeric_limits<float>::infinity());
3476 spec.assembly = shader;
3477 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
3478 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
3479 spec.numWorkGroups = IVec3(numElements, 1, 1);
3481 group->addChild(new SpvAsmComputeShaderCase(
3482 testCtx, "infinities", "Check that infinities propagated and created", spec));
3486 ComputeShaderSpec spec;
3488 const deUint32 numElements = 100;
3490 nans.reserve(numElements);
3492 for (size_t idx = 0; idx < numElements; ++idx)
3496 nans.push_back(std::numeric_limits<float>::quiet_NaN());
3500 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
3504 spec.assembly = shader;
3505 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
3506 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
3507 spec.numWorkGroups = IVec3(numElements, 1, 1);
3508 spec.verifyIO = &compareNan;
3510 group->addChild(new SpvAsmComputeShaderCase(
3511 testCtx, "propagated_nans", "Check that nans are propagated", spec));
3515 ComputeShaderSpec spec;
3516 vector<float> small;
3517 vector<float> zeros;
3518 const deUint32 numElements = 100;
3520 small.reserve(numElements);
3521 zeros.reserve(numElements);
3523 for (size_t idx = 0; idx < numElements; ++idx)
3528 small.push_back(0.f);
3529 zeros.push_back(0.f);
3532 small.push_back(-0.f);
3533 zeros.push_back(-0.f);
3536 small.push_back(std::ldexp(1.0f, -16));
3537 zeros.push_back(0.f);
3540 small.push_back(std::ldexp(-1.0f, -32));
3541 zeros.push_back(-0.f);
3544 small.push_back(std::ldexp(1.0f, -127));
3545 zeros.push_back(0.f);
3548 small.push_back(-std::ldexp(1.0f, -128));
3549 zeros.push_back(-0.f);
3554 spec.assembly = shader;
3555 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
3556 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
3557 spec.numWorkGroups = IVec3(numElements, 1, 1);
3559 group->addChild(new SpvAsmComputeShaderCase(
3560 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
3564 ComputeShaderSpec spec;
3565 vector<float> exact;
3566 const deUint32 numElements = 200;
3568 exact.reserve(numElements);
3570 for (size_t idx = 0; idx < numElements; ++idx)
3571 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
3573 spec.assembly = shader;
3574 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
3575 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
3576 spec.numWorkGroups = IVec3(numElements, 1, 1);
3578 group->addChild(new SpvAsmComputeShaderCase(
3579 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
3583 ComputeShaderSpec spec;
3584 vector<float> inputs;
3585 const deUint32 numElements = 4;
3587 inputs.push_back(constructNormalizedFloat(8, 0x300300));
3588 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
3589 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
3590 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
3592 spec.assembly = shader;
3593 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
3594 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3595 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
3596 spec.numWorkGroups = IVec3(numElements, 1, 1);
3598 group->addChild(new SpvAsmComputeShaderCase(
3599 testCtx, "rounded", "Check that are rounded when needed", spec));
3602 return group.release();
3605 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
3607 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
3609 const std::string shader (
3610 string(getComputeAsmShaderPreamble()) +
3612 "OpName %main \"main\"\n"
3613 "OpName %id \"gl_GlobalInvocationID\"\n"
3615 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3617 "OpDecorate %sc_0 SpecId 0\n"
3618 "OpDecorate %sc_1 SpecId 1\n"
3619 "OpDecorate %sc_2 SpecId 2\n"
3620 "OpDecorate %sc_3 SpecId 3\n"
3621 "OpDecorate %sc_4 SpecId 4\n"
3622 "OpDecorate %sc_5 SpecId 5\n"
3624 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3626 "%id = OpVariable %uvec3ptr Input\n"
3627 "%zero = OpConstant %i32 0\n"
3628 "%c_u32_6 = OpConstant %u32 6\n"
3630 "%sc_0 = OpSpecConstant %f32 0.\n"
3631 "%sc_1 = OpSpecConstant %f32 0.\n"
3632 "%sc_2 = OpSpecConstant %f32 0.\n"
3633 "%sc_3 = OpSpecConstant %f32 0.\n"
3634 "%sc_4 = OpSpecConstant %f32 0.\n"
3635 "%sc_5 = OpSpecConstant %f32 0.\n"
3637 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
3638 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
3639 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
3640 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
3641 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
3642 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
3644 "%main = OpFunction %void None %voidf\n"
3645 "%label = OpLabel\n"
3646 "%idval = OpLoad %uvec3 %id\n"
3647 "%x = OpCompositeExtract %u32 %idval 0\n"
3648 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3649 "%selector = OpUMod %u32 %x %c_u32_6\n"
3650 " OpSelectionMerge %exit None\n"
3651 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
3653 "%case0 = OpLabel\n"
3654 " OpStore %outloc %sc_0_quant\n"
3657 "%case1 = OpLabel\n"
3658 " OpStore %outloc %sc_1_quant\n"
3661 "%case2 = OpLabel\n"
3662 " OpStore %outloc %sc_2_quant\n"
3665 "%case3 = OpLabel\n"
3666 " OpStore %outloc %sc_3_quant\n"
3669 "%case4 = OpLabel\n"
3670 " OpStore %outloc %sc_4_quant\n"
3673 "%case5 = OpLabel\n"
3674 " OpStore %outloc %sc_5_quant\n"
3680 " OpFunctionEnd\n");
3683 ComputeShaderSpec spec;
3684 const deUint8 numCases = 4;
3685 vector<float> inputs (numCases, 0.f);
3686 vector<float> outputs;
3688 spec.assembly = shader;
3689 spec.numWorkGroups = IVec3(numCases, 1, 1);
3691 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
3692 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
3693 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
3694 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
3696 outputs.push_back(std::numeric_limits<float>::infinity());
3697 outputs.push_back(-std::numeric_limits<float>::infinity());
3698 outputs.push_back(std::numeric_limits<float>::infinity());
3699 outputs.push_back(-std::numeric_limits<float>::infinity());
3701 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3702 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3704 group->addChild(new SpvAsmComputeShaderCase(
3705 testCtx, "infinities", "Check that infinities propagated and created", spec));
3709 ComputeShaderSpec spec;
3710 const deUint8 numCases = 2;
3711 vector<float> inputs (numCases, 0.f);
3712 vector<float> outputs;
3714 spec.assembly = shader;
3715 spec.numWorkGroups = IVec3(numCases, 1, 1);
3716 spec.verifyIO = &compareNan;
3718 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
3719 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
3721 for (deUint8 idx = 0; idx < numCases; ++idx)
3722 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
3724 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3725 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3727 group->addChild(new SpvAsmComputeShaderCase(
3728 testCtx, "propagated_nans", "Check that nans are propagated", spec));
3732 ComputeShaderSpec spec;
3733 const deUint8 numCases = 6;
3734 vector<float> inputs (numCases, 0.f);
3735 vector<float> outputs;
3737 spec.assembly = shader;
3738 spec.numWorkGroups = IVec3(numCases, 1, 1);
3740 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
3741 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
3742 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
3743 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
3744 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
3745 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
3747 outputs.push_back(0.f);
3748 outputs.push_back(-0.f);
3749 outputs.push_back(0.f);
3750 outputs.push_back(-0.f);
3751 outputs.push_back(0.f);
3752 outputs.push_back(-0.f);
3754 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3755 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3757 group->addChild(new SpvAsmComputeShaderCase(
3758 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
3762 ComputeShaderSpec spec;
3763 const deUint8 numCases = 6;
3764 vector<float> inputs (numCases, 0.f);
3765 vector<float> outputs;
3767 spec.assembly = shader;
3768 spec.numWorkGroups = IVec3(numCases, 1, 1);
3770 for (deUint8 idx = 0; idx < 6; ++idx)
3772 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
3773 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
3774 outputs.push_back(f);
3777 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3778 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3780 group->addChild(new SpvAsmComputeShaderCase(
3781 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
3785 ComputeShaderSpec spec;
3786 const deUint8 numCases = 4;
3787 vector<float> inputs (numCases, 0.f);
3788 vector<float> outputs;
3790 spec.assembly = shader;
3791 spec.numWorkGroups = IVec3(numCases, 1, 1);
3792 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
3794 outputs.push_back(constructNormalizedFloat(8, 0x300300));
3795 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
3796 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
3797 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
3799 for (deUint8 idx = 0; idx < numCases; ++idx)
3800 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
3802 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3803 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3805 group->addChild(new SpvAsmComputeShaderCase(
3806 testCtx, "rounded", "Check that are rounded when needed", spec));
3809 return group.release();
3812 // Checks that constant null/composite values can be used in computation.
3813 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
3815 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
3816 ComputeShaderSpec spec;
3817 de::Random rnd (deStringHash(group->getName()));
3818 const int numElements = 100;
3819 vector<float> positiveFloats (numElements, 0);
3820 vector<float> negativeFloats (numElements, 0);
3822 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3824 for (size_t ndx = 0; ndx < numElements; ++ndx)
3825 negativeFloats[ndx] = -positiveFloats[ndx];
3828 "OpCapability Shader\n"
3829 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
3830 "OpMemoryModel Logical GLSL450\n"
3831 "OpEntryPoint GLCompute %main \"main\" %id\n"
3832 "OpExecutionMode %main LocalSize 1 1 1\n"
3834 "OpSource GLSL 430\n"
3835 "OpName %main \"main\"\n"
3836 "OpName %id \"gl_GlobalInvocationID\"\n"
3838 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3840 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3842 "%fmat = OpTypeMatrix %fvec3 3\n"
3843 "%ten = OpConstant %u32 10\n"
3844 "%f32arr10 = OpTypeArray %f32 %ten\n"
3845 "%fst = OpTypeStruct %f32 %f32\n"
3847 + string(getComputeAsmInputOutputBuffer()) +
3849 "%id = OpVariable %uvec3ptr Input\n"
3850 "%zero = OpConstant %i32 0\n"
3852 // Create a bunch of null values
3853 "%unull = OpConstantNull %u32\n"
3854 "%fnull = OpConstantNull %f32\n"
3855 "%vnull = OpConstantNull %fvec3\n"
3856 "%mnull = OpConstantNull %fmat\n"
3857 "%anull = OpConstantNull %f32arr10\n"
3858 "%snull = OpConstantComposite %fst %fnull %fnull\n"
3860 "%main = OpFunction %void None %voidf\n"
3861 "%label = OpLabel\n"
3862 "%idval = OpLoad %uvec3 %id\n"
3863 "%x = OpCompositeExtract %u32 %idval 0\n"
3864 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3865 "%inval = OpLoad %f32 %inloc\n"
3866 "%neg = OpFNegate %f32 %inval\n"
3868 // Get the abs() of (a certain element of) those null values
3869 "%unull_cov = OpConvertUToF %f32 %unull\n"
3870 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
3871 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
3872 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
3873 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
3874 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
3875 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
3876 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
3877 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
3878 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
3879 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
3882 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
3883 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
3884 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
3885 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
3886 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
3887 "%final = OpFAdd %f32 %add5 %snull_abs\n"
3889 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3890 " OpStore %outloc %final\n" // write to output
3893 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3894 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3895 spec.numWorkGroups = IVec3(numElements, 1, 1);
3897 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
3899 return group.release();
3902 // Assembly code used for testing loop control is based on GLSL source code:
3905 // layout(std140, set = 0, binding = 0) readonly buffer Input {
3906 // float elements[];
3908 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
3909 // float elements[];
3913 // uint x = gl_GlobalInvocationID.x;
3914 // output_data.elements[x] = input_data.elements[x];
3915 // for (uint i = 0; i < 4; ++i)
3916 // output_data.elements[x] += 1.f;
3918 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
3920 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
3921 vector<CaseParameter> cases;
3922 de::Random rnd (deStringHash(group->getName()));
3923 const int numElements = 100;
3924 vector<float> inputFloats (numElements, 0);
3925 vector<float> outputFloats (numElements, 0);
3926 const StringTemplate shaderTemplate (
3927 string(getComputeAsmShaderPreamble()) +
3929 "OpSource GLSL 430\n"
3930 "OpName %main \"main\"\n"
3931 "OpName %id \"gl_GlobalInvocationID\"\n"
3933 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3935 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3937 "%u32ptr = OpTypePointer Function %u32\n"
3939 "%id = OpVariable %uvec3ptr Input\n"
3940 "%zero = OpConstant %i32 0\n"
3941 "%uzero = OpConstant %u32 0\n"
3942 "%one = OpConstant %i32 1\n"
3943 "%constf1 = OpConstant %f32 1.0\n"
3944 "%four = OpConstant %u32 4\n"
3946 "%main = OpFunction %void None %voidf\n"
3947 "%entry = OpLabel\n"
3948 "%i = OpVariable %u32ptr Function\n"
3949 " OpStore %i %uzero\n"
3951 "%idval = OpLoad %uvec3 %id\n"
3952 "%x = OpCompositeExtract %u32 %idval 0\n"
3953 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3954 "%inval = OpLoad %f32 %inloc\n"
3955 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3956 " OpStore %outloc %inval\n"
3957 " OpBranch %loop_entry\n"
3959 "%loop_entry = OpLabel\n"
3960 "%i_val = OpLoad %u32 %i\n"
3961 "%cmp_lt = OpULessThan %bool %i_val %four\n"
3962 " OpLoopMerge %loop_merge %loop_body ${CONTROL}\n"
3963 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
3964 "%loop_body = OpLabel\n"
3965 "%outval = OpLoad %f32 %outloc\n"
3966 "%addf1 = OpFAdd %f32 %outval %constf1\n"
3967 " OpStore %outloc %addf1\n"
3968 "%new_i = OpIAdd %u32 %i_val %one\n"
3969 " OpStore %i %new_i\n"
3970 " OpBranch %loop_entry\n"
3971 "%loop_merge = OpLabel\n"
3973 " OpFunctionEnd\n");
3975 cases.push_back(CaseParameter("none", "None"));
3976 cases.push_back(CaseParameter("unroll", "Unroll"));
3977 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
3978 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
3980 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3982 for (size_t ndx = 0; ndx < numElements; ++ndx)
3983 outputFloats[ndx] = inputFloats[ndx] + 4.f;
3985 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3987 map<string, string> specializations;
3988 ComputeShaderSpec spec;
3990 specializations["CONTROL"] = cases[caseNdx].param;
3991 spec.assembly = shaderTemplate.specialize(specializations);
3992 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3993 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3994 spec.numWorkGroups = IVec3(numElements, 1, 1);
3996 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3999 return group.release();
4002 // Assembly code used for testing selection control is based on GLSL source code:
4005 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4006 // float elements[];
4008 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4009 // float elements[];
4013 // uint x = gl_GlobalInvocationID.x;
4014 // float val = input_data.elements[x];
4016 // output_data.elements[x] = val + 1.f;
4018 // output_data.elements[x] = val - 1.f;
4020 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
4022 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
4023 vector<CaseParameter> cases;
4024 de::Random rnd (deStringHash(group->getName()));
4025 const int numElements = 100;
4026 vector<float> inputFloats (numElements, 0);
4027 vector<float> outputFloats (numElements, 0);
4028 const StringTemplate shaderTemplate (
4029 string(getComputeAsmShaderPreamble()) +
4031 "OpSource GLSL 430\n"
4032 "OpName %main \"main\"\n"
4033 "OpName %id \"gl_GlobalInvocationID\"\n"
4035 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4037 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4039 "%id = OpVariable %uvec3ptr Input\n"
4040 "%zero = OpConstant %i32 0\n"
4041 "%constf1 = OpConstant %f32 1.0\n"
4042 "%constf10 = OpConstant %f32 10.0\n"
4044 "%main = OpFunction %void None %voidf\n"
4045 "%entry = OpLabel\n"
4046 "%idval = OpLoad %uvec3 %id\n"
4047 "%x = OpCompositeExtract %u32 %idval 0\n"
4048 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4049 "%inval = OpLoad %f32 %inloc\n"
4050 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4051 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
4053 " OpSelectionMerge %if_end ${CONTROL}\n"
4054 " OpBranchConditional %cmp_gt %if_true %if_false\n"
4055 "%if_true = OpLabel\n"
4056 "%addf1 = OpFAdd %f32 %inval %constf1\n"
4057 " OpStore %outloc %addf1\n"
4058 " OpBranch %if_end\n"
4059 "%if_false = OpLabel\n"
4060 "%subf1 = OpFSub %f32 %inval %constf1\n"
4061 " OpStore %outloc %subf1\n"
4062 " OpBranch %if_end\n"
4063 "%if_end = OpLabel\n"
4065 " OpFunctionEnd\n");
4067 cases.push_back(CaseParameter("none", "None"));
4068 cases.push_back(CaseParameter("flatten", "Flatten"));
4069 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
4070 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
4072 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4074 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4075 floorAll(inputFloats);
4077 for (size_t ndx = 0; ndx < numElements; ++ndx)
4078 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
4080 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4082 map<string, string> specializations;
4083 ComputeShaderSpec spec;
4085 specializations["CONTROL"] = cases[caseNdx].param;
4086 spec.assembly = shaderTemplate.specialize(specializations);
4087 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4088 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4089 spec.numWorkGroups = IVec3(numElements, 1, 1);
4091 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4094 return group.release();
4097 // Assembly code used for testing function control is based on GLSL source code:
4101 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4102 // float elements[];
4104 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4105 // float elements[];
4108 // float const10() { return 10.f; }
4111 // uint x = gl_GlobalInvocationID.x;
4112 // output_data.elements[x] = input_data.elements[x] + const10();
4114 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
4116 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
4117 vector<CaseParameter> cases;
4118 de::Random rnd (deStringHash(group->getName()));
4119 const int numElements = 100;
4120 vector<float> inputFloats (numElements, 0);
4121 vector<float> outputFloats (numElements, 0);
4122 const StringTemplate shaderTemplate (
4123 string(getComputeAsmShaderPreamble()) +
4125 "OpSource GLSL 430\n"
4126 "OpName %main \"main\"\n"
4127 "OpName %func_const10 \"const10(\"\n"
4128 "OpName %id \"gl_GlobalInvocationID\"\n"
4130 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4132 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4134 "%f32f = OpTypeFunction %f32\n"
4135 "%id = OpVariable %uvec3ptr Input\n"
4136 "%zero = OpConstant %i32 0\n"
4137 "%constf10 = OpConstant %f32 10.0\n"
4139 "%main = OpFunction %void None %voidf\n"
4140 "%entry = OpLabel\n"
4141 "%idval = OpLoad %uvec3 %id\n"
4142 "%x = OpCompositeExtract %u32 %idval 0\n"
4143 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4144 "%inval = OpLoad %f32 %inloc\n"
4145 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
4146 "%fadd = OpFAdd %f32 %inval %ret_10\n"
4147 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4148 " OpStore %outloc %fadd\n"
4152 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
4153 "%label = OpLabel\n"
4154 " OpReturnValue %constf10\n"
4155 " OpFunctionEnd\n");
4157 cases.push_back(CaseParameter("none", "None"));
4158 cases.push_back(CaseParameter("inline", "Inline"));
4159 cases.push_back(CaseParameter("dont_inline", "DontInline"));
4160 cases.push_back(CaseParameter("pure", "Pure"));
4161 cases.push_back(CaseParameter("const", "Const"));
4162 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
4163 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
4164 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
4165 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
4167 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4169 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4170 floorAll(inputFloats);
4172 for (size_t ndx = 0; ndx < numElements; ++ndx)
4173 outputFloats[ndx] = inputFloats[ndx] + 10.f;
4175 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4177 map<string, string> specializations;
4178 ComputeShaderSpec spec;
4180 specializations["CONTROL"] = cases[caseNdx].param;
4181 spec.assembly = shaderTemplate.specialize(specializations);
4182 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4183 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4184 spec.numWorkGroups = IVec3(numElements, 1, 1);
4186 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4189 return group.release();
4192 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
4194 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
4195 vector<CaseParameter> cases;
4196 de::Random rnd (deStringHash(group->getName()));
4197 const int numElements = 100;
4198 vector<float> inputFloats (numElements, 0);
4199 vector<float> outputFloats (numElements, 0);
4200 const StringTemplate shaderTemplate (
4201 string(getComputeAsmShaderPreamble()) +
4203 "OpSource GLSL 430\n"
4204 "OpName %main \"main\"\n"
4205 "OpName %id \"gl_GlobalInvocationID\"\n"
4207 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4209 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4211 "%f32ptr_f = OpTypePointer Function %f32\n"
4213 "%id = OpVariable %uvec3ptr Input\n"
4214 "%zero = OpConstant %i32 0\n"
4215 "%four = OpConstant %i32 4\n"
4217 "%main = OpFunction %void None %voidf\n"
4218 "%label = OpLabel\n"
4219 "%copy = OpVariable %f32ptr_f Function\n"
4220 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
4221 "%x = OpCompositeExtract %u32 %idval 0\n"
4222 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4223 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4224 " OpCopyMemory %copy %inloc ${ACCESS}\n"
4225 "%val1 = OpLoad %f32 %copy\n"
4226 "%val2 = OpLoad %f32 %inloc\n"
4227 "%add = OpFAdd %f32 %val1 %val2\n"
4228 " OpStore %outloc %add ${ACCESS}\n"
4230 " OpFunctionEnd\n");
4232 cases.push_back(CaseParameter("null", ""));
4233 cases.push_back(CaseParameter("none", "None"));
4234 cases.push_back(CaseParameter("volatile", "Volatile"));
4235 cases.push_back(CaseParameter("aligned", "Aligned 4"));
4236 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
4237 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
4238 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
4240 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4242 for (size_t ndx = 0; ndx < numElements; ++ndx)
4243 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
4245 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4247 map<string, string> specializations;
4248 ComputeShaderSpec spec;
4250 specializations["ACCESS"] = cases[caseNdx].param;
4251 spec.assembly = shaderTemplate.specialize(specializations);
4252 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4253 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4254 spec.numWorkGroups = IVec3(numElements, 1, 1);
4256 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4259 return group.release();
4262 // Checks that we can get undefined values for various types, without exercising a computation with it.
4263 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
4265 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
4266 vector<CaseParameter> cases;
4267 de::Random rnd (deStringHash(group->getName()));
4268 const int numElements = 100;
4269 vector<float> positiveFloats (numElements, 0);
4270 vector<float> negativeFloats (numElements, 0);
4271 const StringTemplate shaderTemplate (
4272 string(getComputeAsmShaderPreamble()) +
4274 "OpSource GLSL 430\n"
4275 "OpName %main \"main\"\n"
4276 "OpName %id \"gl_GlobalInvocationID\"\n"
4278 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4280 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4281 "%uvec2 = OpTypeVector %u32 2\n"
4282 "%fvec4 = OpTypeVector %f32 4\n"
4283 "%fmat33 = OpTypeMatrix %fvec3 3\n"
4284 "%image = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
4285 "%sampler = OpTypeSampler\n"
4286 "%simage = OpTypeSampledImage %image\n"
4287 "%const100 = OpConstant %u32 100\n"
4288 "%uarr100 = OpTypeArray %i32 %const100\n"
4289 "%struct = OpTypeStruct %f32 %i32 %u32\n"
4290 "%pointer = OpTypePointer Function %i32\n"
4291 + string(getComputeAsmInputOutputBuffer()) +
4293 "%id = OpVariable %uvec3ptr Input\n"
4294 "%zero = OpConstant %i32 0\n"
4296 "%main = OpFunction %void None %voidf\n"
4297 "%label = OpLabel\n"
4299 "%undef = OpUndef ${TYPE}\n"
4301 "%idval = OpLoad %uvec3 %id\n"
4302 "%x = OpCompositeExtract %u32 %idval 0\n"
4304 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4305 "%inval = OpLoad %f32 %inloc\n"
4306 "%neg = OpFNegate %f32 %inval\n"
4307 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4308 " OpStore %outloc %neg\n"
4310 " OpFunctionEnd\n");
4312 cases.push_back(CaseParameter("bool", "%bool"));
4313 cases.push_back(CaseParameter("sint32", "%i32"));
4314 cases.push_back(CaseParameter("uint32", "%u32"));
4315 cases.push_back(CaseParameter("float32", "%f32"));
4316 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
4317 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
4318 cases.push_back(CaseParameter("matrix", "%fmat33"));
4319 cases.push_back(CaseParameter("image", "%image"));
4320 cases.push_back(CaseParameter("sampler", "%sampler"));
4321 cases.push_back(CaseParameter("sampledimage", "%simage"));
4322 cases.push_back(CaseParameter("array", "%uarr100"));
4323 cases.push_back(CaseParameter("runtimearray", "%f32arr"));
4324 cases.push_back(CaseParameter("struct", "%struct"));
4325 cases.push_back(CaseParameter("pointer", "%pointer"));
4327 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4329 for (size_t ndx = 0; ndx < numElements; ++ndx)
4330 negativeFloats[ndx] = -positiveFloats[ndx];
4332 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4334 map<string, string> specializations;
4335 ComputeShaderSpec spec;
4337 specializations["TYPE"] = cases[caseNdx].param;
4338 spec.assembly = shaderTemplate.specialize(specializations);
4339 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4340 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4341 spec.numWorkGroups = IVec3(numElements, 1, 1);
4343 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4346 return group.release();
4351 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
4353 struct NameCodePair { string name, code; };
4354 RGBA defaultColors[4];
4355 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
4356 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
4357 map<string, string> fragments = passthruFragments();
4358 const NameCodePair tests[] =
4360 {"unknown", "OpSource Unknown 321"},
4361 {"essl", "OpSource ESSL 310"},
4362 {"glsl", "OpSource GLSL 450"},
4363 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
4364 {"opencl_c", "OpSource OpenCL_C 120"},
4365 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
4366 {"file", opsourceGLSLWithFile},
4367 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
4368 // Longest possible source string: SPIR-V limits instructions to 65535
4369 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
4370 // contain 65530 UTF8 characters (one word each) plus one last word
4371 // containing 3 ASCII characters and \0.
4372 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
4375 getDefaultColors(defaultColors);
4376 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
4378 fragments["debug"] = tests[testNdx].code;
4379 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
4382 return opSourceTests.release();
4385 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
4387 struct NameCodePair { string name, code; };
4388 RGBA defaultColors[4];
4389 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
4390 map<string, string> fragments = passthruFragments();
4391 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
4392 const NameCodePair tests[] =
4394 {"empty", opsource + "OpSourceContinued \"\""},
4395 {"short", opsource + "OpSourceContinued \"abcde\""},
4396 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
4397 // Longest possible source string: SPIR-V limits instructions to 65535
4398 // words, of which the first one is OpSourceContinued/length; the rest
4399 // will contain 65533 UTF8 characters (one word each) plus one last word
4400 // containing 3 ASCII characters and \0.
4401 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
4404 getDefaultColors(defaultColors);
4405 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
4407 fragments["debug"] = tests[testNdx].code;
4408 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
4411 return opSourceTests.release();
4414 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
4416 RGBA defaultColors[4];
4417 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
4418 map<string, string> fragments;
4419 getDefaultColors(defaultColors);
4420 fragments["debug"] =
4421 "%name = OpString \"name\"\n";
4423 fragments["pre_main"] =
4426 "OpLine %name 1 1\n"
4428 "OpLine %name 1 1\n"
4429 "OpLine %name 1 1\n"
4430 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
4432 "OpLine %name 1 1\n"
4434 "OpLine %name 1 1\n"
4435 "OpLine %name 1 1\n"
4436 "%second_param1 = OpFunctionParameter %v4f32\n"
4439 "%label_secondfunction = OpLabel\n"
4441 "OpReturnValue %second_param1\n"
4446 fragments["testfun"] =
4447 // A %test_code function that returns its argument unchanged.
4450 "OpLine %name 1 1\n"
4451 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4453 "%param1 = OpFunctionParameter %v4f32\n"
4456 "%label_testfun = OpLabel\n"
4458 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
4459 "OpReturnValue %val1\n"
4461 "OpLine %name 1 1\n"
4464 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
4466 return opLineTests.release();
4470 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
4472 RGBA defaultColors[4];
4473 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
4474 map<string, string> fragments;
4475 std::vector<std::pair<std::string, std::string> > problemStrings;
4477 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
4478 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
4479 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
4480 getDefaultColors(defaultColors);
4482 fragments["debug"] =
4483 "%other_name = OpString \"other_name\"\n";
4485 fragments["pre_main"] =
4486 "OpLine %file_name 32 0\n"
4487 "OpLine %file_name 32 32\n"
4488 "OpLine %file_name 32 40\n"
4489 "OpLine %other_name 32 40\n"
4490 "OpLine %other_name 0 100\n"
4491 "OpLine %other_name 0 4294967295\n"
4492 "OpLine %other_name 4294967295 0\n"
4493 "OpLine %other_name 32 40\n"
4494 "OpLine %file_name 0 0\n"
4495 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
4496 "OpLine %file_name 1 0\n"
4497 "%second_param1 = OpFunctionParameter %v4f32\n"
4498 "OpLine %file_name 1 3\n"
4499 "OpLine %file_name 1 2\n"
4500 "%label_secondfunction = OpLabel\n"
4501 "OpLine %file_name 0 2\n"
4502 "OpReturnValue %second_param1\n"
4504 "OpLine %file_name 0 2\n"
4505 "OpLine %file_name 0 2\n";
4507 fragments["testfun"] =
4508 // A %test_code function that returns its argument unchanged.
4509 "OpLine %file_name 1 0\n"
4510 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4511 "OpLine %file_name 16 330\n"
4512 "%param1 = OpFunctionParameter %v4f32\n"
4513 "OpLine %file_name 14 442\n"
4514 "%label_testfun = OpLabel\n"
4515 "OpLine %file_name 11 1024\n"
4516 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
4517 "OpLine %file_name 2 97\n"
4518 "OpReturnValue %val1\n"
4520 "OpLine %file_name 5 32\n";
4522 for (size_t i = 0; i < problemStrings.size(); ++i)
4524 map<string, string> testFragments = fragments;
4525 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
4526 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
4529 return opLineTests.release();
4532 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
4534 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
4538 const char functionStart[] =
4539 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4540 "%param1 = OpFunctionParameter %v4f32\n"
4543 const char functionEnd[] =
4544 "OpReturnValue %transformed_param\n"
4547 struct NameConstantsCode
4554 NameConstantsCode tests[] =
4558 "%cnull = OpConstantNull %v4f32\n",
4559 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
4563 "%cnull = OpConstantNull %f32\n",
4564 "%vp = OpVariable %fp_v4f32 Function\n"
4565 "%v = OpLoad %v4f32 %vp\n"
4566 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
4567 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
4568 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
4569 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
4570 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
4574 "%cnull = OpConstantNull %bool\n",
4575 "%v = OpVariable %fp_v4f32 Function\n"
4576 " OpStore %v %param1\n"
4577 " OpSelectionMerge %false_label None\n"
4578 " OpBranchConditional %cnull %true_label %false_label\n"
4579 "%true_label = OpLabel\n"
4580 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
4581 " OpBranch %false_label\n"
4582 "%false_label = OpLabel\n"
4583 "%transformed_param = OpLoad %v4f32 %v\n"
4587 "%cnull = OpConstantNull %i32\n",
4588 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
4589 "%b = OpIEqual %bool %cnull %c_i32_0\n"
4590 " OpSelectionMerge %false_label None\n"
4591 " OpBranchConditional %b %true_label %false_label\n"
4592 "%true_label = OpLabel\n"
4593 " OpStore %v %param1\n"
4594 " OpBranch %false_label\n"
4595 "%false_label = OpLabel\n"
4596 "%transformed_param = OpLoad %v4f32 %v\n"
4600 "%stype = OpTypeStruct %f32 %v4f32\n"
4601 "%fp_stype = OpTypePointer Function %stype\n"
4602 "%cnull = OpConstantNull %stype\n",
4603 "%v = OpVariable %fp_stype Function %cnull\n"
4604 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
4605 "%f_val = OpLoad %v4f32 %f\n"
4606 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
4610 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
4611 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
4612 "%cnull = OpConstantNull %a4_v4f32\n",
4613 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
4614 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
4615 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
4616 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
4617 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
4618 "%f_val = OpLoad %v4f32 %f\n"
4619 "%f1_val = OpLoad %v4f32 %f1\n"
4620 "%f2_val = OpLoad %v4f32 %f2\n"
4621 "%f3_val = OpLoad %v4f32 %f3\n"
4622 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
4623 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
4624 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
4625 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
4629 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
4630 "%cnull = OpConstantNull %mat4x4_f32\n",
4631 // Our null matrix * any vector should result in a zero vector.
4632 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
4633 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
4637 getHalfColorsFullAlpha(colors);
4639 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
4641 map<string, string> fragments;
4642 fragments["pre_main"] = tests[testNdx].constants;
4643 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
4644 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
4646 return opConstantNullTests.release();
4648 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
4650 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
4651 RGBA inputColors[4];
4652 RGBA outputColors[4];
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[] =
4676 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
4677 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
4682 "%stype = OpTypeStruct %v4f32 %f32\n"
4683 "%fp_stype = OpTypePointer Function %stype\n"
4684 "%f32_n_1 = OpConstant %f32 -1.0\n"
4685 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
4686 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
4687 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
4689 "%v = OpVariable %fp_stype Function %cval\n"
4690 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
4691 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
4692 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
4693 "%f32_val = OpLoad %f32 %f32_ptr\n"
4694 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
4695 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
4696 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
4699 // [1|0|0|0.5] [x] = x + 0.5
4700 // [0|1|0|0.5] [y] = y + 0.5
4701 // [0|0|1|0.5] [z] = z + 0.5
4702 // [0|0|0|1 ] [1] = 1
4705 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
4706 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
4707 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
4708 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
4709 "%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"
4710 "%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",
4712 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
4717 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4718 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
4719 "%f32_n_1 = OpConstant %f32 -1.0\n"
4720 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
4721 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
4723 "%v = OpVariable %fp_a4f32 Function %carr\n"
4724 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
4725 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
4726 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
4727 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
4728 "%f_val = OpLoad %f32 %f\n"
4729 "%f1_val = OpLoad %f32 %f1\n"
4730 "%f2_val = OpLoad %f32 %f2\n"
4731 "%f3_val = OpLoad %f32 %f3\n"
4732 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
4733 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
4734 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
4735 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
4736 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
4743 // [ 1.0, 1.0, 1.0, 1.0]
4747 // [ 0.0, 0.5, 0.0, 0.0]
4751 // [ 1.0, 1.0, 1.0, 1.0]
4754 "array_of_struct_of_array",
4756 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4757 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
4758 "%stype = OpTypeStruct %f32 %a4f32\n"
4759 "%a3stype = OpTypeArray %stype %c_u32_3\n"
4760 "%fp_a3stype = OpTypePointer Function %a3stype\n"
4761 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
4762 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
4763 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
4764 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
4765 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
4767 "%v = OpVariable %fp_a3stype Function %carr\n"
4768 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
4769 "%f_l = OpLoad %f32 %f\n"
4770 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
4771 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
4775 getHalfColorsFullAlpha(inputColors);
4776 outputColors[0] = RGBA(255, 255, 255, 255);
4777 outputColors[1] = RGBA(255, 127, 127, 255);
4778 outputColors[2] = RGBA(127, 255, 127, 255);
4779 outputColors[3] = RGBA(127, 127, 255, 255);
4781 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
4783 map<string, string> fragments;
4784 fragments["pre_main"] = tests[testNdx].constants;
4785 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
4786 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
4788 return opConstantCompositeTests.release();
4791 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
4793 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
4794 RGBA inputColors[4];
4795 RGBA outputColors[4];
4796 map<string, string> fragments;
4798 // vec4 test_code(vec4 param) {
4799 // vec4 result = param;
4800 // for (int i = 0; i < 4; ++i) {
4801 // if (i == 0) result[i] = 0.;
4802 // else result[i] = 1. - result[i];
4806 const char function[] =
4807 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4808 "%param1 = OpFunctionParameter %v4f32\n"
4810 "%iptr = OpVariable %fp_i32 Function\n"
4811 "%result = OpVariable %fp_v4f32 Function\n"
4812 " OpStore %iptr %c_i32_0\n"
4813 " OpStore %result %param1\n"
4816 // Loop entry block.
4818 "%ival = OpLoad %i32 %iptr\n"
4819 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
4820 " OpLoopMerge %exit %if_entry None\n"
4821 " OpBranchConditional %lt_4 %if_entry %exit\n"
4823 // Merge block for loop.
4825 "%ret = OpLoad %v4f32 %result\n"
4826 " OpReturnValue %ret\n"
4828 // If-statement entry block.
4829 "%if_entry = OpLabel\n"
4830 "%loc = OpAccessChain %fp_f32 %result %ival\n"
4831 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
4832 " OpSelectionMerge %if_exit None\n"
4833 " OpBranchConditional %eq_0 %if_true %if_false\n"
4835 // False branch for if-statement.
4836 "%if_false = OpLabel\n"
4837 "%val = OpLoad %f32 %loc\n"
4838 "%sub = OpFSub %f32 %c_f32_1 %val\n"
4839 " OpStore %loc %sub\n"
4840 " OpBranch %if_exit\n"
4842 // Merge block for if-statement.
4843 "%if_exit = OpLabel\n"
4844 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
4845 " OpStore %iptr %ival_next\n"
4848 // True branch for if-statement.
4849 "%if_true = OpLabel\n"
4850 " OpStore %loc %c_f32_0\n"
4851 " OpBranch %if_exit\n"
4855 fragments["testfun"] = function;
4857 inputColors[0] = RGBA(127, 127, 127, 0);
4858 inputColors[1] = RGBA(127, 0, 0, 0);
4859 inputColors[2] = RGBA(0, 127, 0, 0);
4860 inputColors[3] = RGBA(0, 0, 127, 0);
4862 outputColors[0] = RGBA(0, 128, 128, 255);
4863 outputColors[1] = RGBA(0, 255, 255, 255);
4864 outputColors[2] = RGBA(0, 128, 255, 255);
4865 outputColors[3] = RGBA(0, 255, 128, 255);
4867 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
4869 return group.release();
4872 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
4874 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
4875 RGBA inputColors[4];
4876 RGBA outputColors[4];
4877 map<string, string> fragments;
4879 const char typesAndConstants[] =
4880 "%c_f32_p2 = OpConstant %f32 0.2\n"
4881 "%c_f32_p4 = OpConstant %f32 0.4\n"
4882 "%c_f32_p6 = OpConstant %f32 0.6\n"
4883 "%c_f32_p8 = OpConstant %f32 0.8\n";
4885 // vec4 test_code(vec4 param) {
4886 // vec4 result = param;
4887 // for (int i = 0; i < 4; ++i) {
4889 // case 0: result[i] += .2; break;
4890 // case 1: result[i] += .6; break;
4891 // case 2: result[i] += .4; break;
4892 // case 3: result[i] += .8; break;
4893 // default: break; // unreachable
4898 const char function[] =
4899 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4900 "%param1 = OpFunctionParameter %v4f32\n"
4902 "%iptr = OpVariable %fp_i32 Function\n"
4903 "%result = OpVariable %fp_v4f32 Function\n"
4904 " OpStore %iptr %c_i32_0\n"
4905 " OpStore %result %param1\n"
4908 // Loop entry block.
4910 "%ival = OpLoad %i32 %iptr\n"
4911 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
4912 " OpLoopMerge %exit %switch_exit None\n"
4913 " OpBranchConditional %lt_4 %switch_entry %exit\n"
4915 // Merge block for loop.
4917 "%ret = OpLoad %v4f32 %result\n"
4918 " OpReturnValue %ret\n"
4920 // Switch-statement entry block.
4921 "%switch_entry = OpLabel\n"
4922 "%loc = OpAccessChain %fp_f32 %result %ival\n"
4923 "%val = OpLoad %f32 %loc\n"
4924 " OpSelectionMerge %switch_exit None\n"
4925 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
4927 "%case2 = OpLabel\n"
4928 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
4929 " OpStore %loc %addp4\n"
4930 " OpBranch %switch_exit\n"
4932 "%switch_default = OpLabel\n"
4935 "%case3 = OpLabel\n"
4936 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
4937 " OpStore %loc %addp8\n"
4938 " OpBranch %switch_exit\n"
4940 "%case0 = OpLabel\n"
4941 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
4942 " OpStore %loc %addp2\n"
4943 " OpBranch %switch_exit\n"
4945 // Merge block for switch-statement.
4946 "%switch_exit = OpLabel\n"
4947 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
4948 " OpStore %iptr %ival_next\n"
4951 "%case1 = OpLabel\n"
4952 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
4953 " OpStore %loc %addp6\n"
4954 " OpBranch %switch_exit\n"
4958 fragments["pre_main"] = typesAndConstants;
4959 fragments["testfun"] = function;
4961 inputColors[0] = RGBA(127, 27, 127, 51);
4962 inputColors[1] = RGBA(127, 0, 0, 51);
4963 inputColors[2] = RGBA(0, 27, 0, 51);
4964 inputColors[3] = RGBA(0, 0, 127, 51);
4966 outputColors[0] = RGBA(178, 180, 229, 255);
4967 outputColors[1] = RGBA(178, 153, 102, 255);
4968 outputColors[2] = RGBA(51, 180, 102, 255);
4969 outputColors[3] = RGBA(51, 153, 229, 255);
4971 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
4973 return group.release();
4976 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
4978 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
4979 RGBA inputColors[4];
4980 RGBA outputColors[4];
4981 map<string, string> fragments;
4983 const char decorations[] =
4984 "OpDecorate %array_group ArrayStride 4\n"
4985 "OpDecorate %struct_member_group Offset 0\n"
4986 "%array_group = OpDecorationGroup\n"
4987 "%struct_member_group = OpDecorationGroup\n"
4989 "OpDecorate %group1 RelaxedPrecision\n"
4990 "OpDecorate %group3 RelaxedPrecision\n"
4991 "OpDecorate %group3 Invariant\n"
4992 "OpDecorate %group3 Restrict\n"
4993 "%group0 = OpDecorationGroup\n"
4994 "%group1 = OpDecorationGroup\n"
4995 "%group3 = OpDecorationGroup\n";
4997 const char typesAndConstants[] =
4998 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
4999 "%struct1 = OpTypeStruct %a3f32\n"
5000 "%struct2 = OpTypeStruct %a3f32\n"
5001 "%fp_struct1 = OpTypePointer Function %struct1\n"
5002 "%fp_struct2 = OpTypePointer Function %struct2\n"
5003 "%c_f32_2 = OpConstant %f32 2.\n"
5004 "%c_f32_n2 = OpConstant %f32 -2.\n"
5006 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
5007 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
5008 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
5009 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
5011 const char function[] =
5012 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5013 "%param = OpFunctionParameter %v4f32\n"
5014 "%entry = OpLabel\n"
5015 "%result = OpVariable %fp_v4f32 Function\n"
5016 "%v_struct1 = OpVariable %fp_struct1 Function\n"
5017 "%v_struct2 = OpVariable %fp_struct2 Function\n"
5018 " OpStore %result %param\n"
5019 " OpStore %v_struct1 %c_struct1\n"
5020 " OpStore %v_struct2 %c_struct2\n"
5021 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
5022 "%val1 = OpLoad %f32 %ptr1\n"
5023 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
5024 "%val2 = OpLoad %f32 %ptr2\n"
5025 "%addvalues = OpFAdd %f32 %val1 %val2\n"
5026 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
5027 "%val = OpLoad %f32 %ptr\n"
5028 "%addresult = OpFAdd %f32 %addvalues %val\n"
5029 " OpStore %ptr %addresult\n"
5030 "%ret = OpLoad %v4f32 %result\n"
5031 " OpReturnValue %ret\n"
5034 struct CaseNameDecoration
5040 CaseNameDecoration tests[] =
5043 "same_decoration_group_on_multiple_types",
5044 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
5047 "empty_decoration_group",
5048 "OpGroupDecorate %group0 %a3f32\n"
5049 "OpGroupDecorate %group0 %result\n"
5052 "one_element_decoration_group",
5053 "OpGroupDecorate %array_group %a3f32\n"
5056 "multiple_elements_decoration_group",
5057 "OpGroupDecorate %group3 %v_struct1\n"
5060 "multiple_decoration_groups_on_same_variable",
5061 "OpGroupDecorate %group0 %v_struct2\n"
5062 "OpGroupDecorate %group1 %v_struct2\n"
5063 "OpGroupDecorate %group3 %v_struct2\n"
5066 "same_decoration_group_multiple_times",
5067 "OpGroupDecorate %group1 %addvalues\n"
5068 "OpGroupDecorate %group1 %addvalues\n"
5069 "OpGroupDecorate %group1 %addvalues\n"
5074 getHalfColorsFullAlpha(inputColors);
5075 getHalfColorsFullAlpha(outputColors);
5077 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
5079 fragments["decoration"] = decorations + tests[idx].decoration;
5080 fragments["pre_main"] = typesAndConstants;
5081 fragments["testfun"] = function;
5083 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
5086 return group.release();
5089 struct SpecConstantTwoIntGraphicsCase
5091 const char* caseName;
5092 const char* scDefinition0;
5093 const char* scDefinition1;
5094 const char* scResultType;
5095 const char* scOperation;
5096 deInt32 scActualValue0;
5097 deInt32 scActualValue1;
5098 const char* resultOperation;
5099 RGBA expectedColors[4];
5101 SpecConstantTwoIntGraphicsCase (const char* name,
5102 const char* definition0,
5103 const char* definition1,
5104 const char* resultType,
5105 const char* operation,
5108 const char* resultOp,
5109 const RGBA (&output)[4])
5111 , scDefinition0 (definition0)
5112 , scDefinition1 (definition1)
5113 , scResultType (resultType)
5114 , scOperation (operation)
5115 , scActualValue0 (value0)
5116 , scActualValue1 (value1)
5117 , resultOperation (resultOp)
5119 expectedColors[0] = output[0];
5120 expectedColors[1] = output[1];
5121 expectedColors[2] = output[2];
5122 expectedColors[3] = output[3];
5126 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
5128 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
5129 vector<SpecConstantTwoIntGraphicsCase> cases;
5130 RGBA inputColors[4];
5131 RGBA outputColors0[4];
5132 RGBA outputColors1[4];
5133 RGBA outputColors2[4];
5135 const char decorations1[] =
5136 "OpDecorate %sc_0 SpecId 0\n"
5137 "OpDecorate %sc_1 SpecId 1\n";
5139 const char typesAndConstants1[] =
5140 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
5141 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
5142 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
5144 const char function1[] =
5145 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5146 "%param = OpFunctionParameter %v4f32\n"
5147 "%label = OpLabel\n"
5148 "%result = OpVariable %fp_v4f32 Function\n"
5149 " OpStore %result %param\n"
5150 "%gen = ${GEN_RESULT}\n"
5151 "%index = OpIAdd %i32 %gen %c_i32_1\n"
5152 "%loc = OpAccessChain %fp_f32 %result %index\n"
5153 "%val = OpLoad %f32 %loc\n"
5154 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
5155 " OpStore %loc %add\n"
5156 "%ret = OpLoad %v4f32 %result\n"
5157 " OpReturnValue %ret\n"
5160 inputColors[0] = RGBA(127, 127, 127, 255);
5161 inputColors[1] = RGBA(127, 0, 0, 255);
5162 inputColors[2] = RGBA(0, 127, 0, 255);
5163 inputColors[3] = RGBA(0, 0, 127, 255);
5165 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
5166 outputColors0[0] = RGBA(255, 127, 127, 255);
5167 outputColors0[1] = RGBA(255, 0, 0, 255);
5168 outputColors0[2] = RGBA(128, 127, 0, 255);
5169 outputColors0[3] = RGBA(128, 0, 127, 255);
5171 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
5172 outputColors1[0] = RGBA(127, 255, 127, 255);
5173 outputColors1[1] = RGBA(127, 128, 0, 255);
5174 outputColors1[2] = RGBA(0, 255, 0, 255);
5175 outputColors1[3] = RGBA(0, 128, 127, 255);
5177 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
5178 outputColors2[0] = RGBA(127, 127, 255, 255);
5179 outputColors2[1] = RGBA(127, 0, 128, 255);
5180 outputColors2[2] = RGBA(0, 127, 128, 255);
5181 outputColors2[3] = RGBA(0, 0, 255, 255);
5183 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
5184 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
5185 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
5187 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
5188 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
5189 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
5190 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
5191 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
5192 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
5193 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
5194 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
5195 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
5196 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
5197 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
5198 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
5199 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
5200 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
5201 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
5202 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
5203 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
5204 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
5205 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
5206 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
5207 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
5208 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
5209 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
5210 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
5211 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
5212 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
5213 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
5214 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
5215 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
5216 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
5217 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
5218 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
5219 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
5221 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5223 map<string, string> specializations;
5224 map<string, string> fragments;
5225 vector<deInt32> specConstants;
5227 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
5228 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
5229 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
5230 specializations["SC_OP"] = cases[caseNdx].scOperation;
5231 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
5233 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
5234 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
5235 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
5237 specConstants.push_back(cases[caseNdx].scActualValue0);
5238 specConstants.push_back(cases[caseNdx].scActualValue1);
5240 createTestsForAllStages(cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants, group.get());
5243 const char decorations2[] =
5244 "OpDecorate %sc_0 SpecId 0\n"
5245 "OpDecorate %sc_1 SpecId 1\n"
5246 "OpDecorate %sc_2 SpecId 2\n";
5248 const char typesAndConstants2[] =
5249 "%v3i32 = OpTypeVector %i32 3\n"
5250 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
5251 "%vec3_undef = OpUndef %v3i32\n"
5253 "%sc_0 = OpSpecConstant %i32 0\n"
5254 "%sc_1 = OpSpecConstant %i32 0\n"
5255 "%sc_2 = OpSpecConstant %i32 0\n"
5256 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
5257 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
5258 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
5259 "%sc_vec3_0_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
5260 "%sc_vec3_1_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
5261 "%sc_vec3_2_s = OpSpecConstantOp %v3i32 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
5262 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
5263 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
5264 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
5265 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
5266 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
5267 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
5268 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
5270 const char function2[] =
5271 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5272 "%param = OpFunctionParameter %v4f32\n"
5273 "%label = OpLabel\n"
5274 "%result = OpVariable %fp_v4f32 Function\n"
5275 " OpStore %result %param\n"
5276 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
5277 "%val = OpLoad %f32 %loc\n"
5278 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
5279 " OpStore %loc %add\n"
5280 "%ret = OpLoad %v4f32 %result\n"
5281 " OpReturnValue %ret\n"
5284 map<string, string> fragments;
5285 vector<deInt32> specConstants;
5287 fragments["decoration"] = decorations2;
5288 fragments["pre_main"] = typesAndConstants2;
5289 fragments["testfun"] = function2;
5291 specConstants.push_back(56789);
5292 specConstants.push_back(-2);
5293 specConstants.push_back(56788);
5295 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
5297 return group.release();
5300 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
5302 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
5303 RGBA inputColors[4];
5304 RGBA outputColors1[4];
5305 RGBA outputColors2[4];
5306 RGBA outputColors3[4];
5307 map<string, string> fragments1;
5308 map<string, string> fragments2;
5309 map<string, string> fragments3;
5311 const char typesAndConstants1[] =
5312 "%c_f32_p2 = OpConstant %f32 0.2\n"
5313 "%c_f32_p4 = OpConstant %f32 0.4\n"
5314 "%c_f32_p5 = OpConstant %f32 0.5\n"
5315 "%c_f32_p8 = OpConstant %f32 0.8\n";
5317 // vec4 test_code(vec4 param) {
5318 // vec4 result = param;
5319 // for (int i = 0; i < 4; ++i) {
5322 // case 0: operand = .2; break;
5323 // case 1: operand = .5; break;
5324 // case 2: operand = .4; break;
5325 // case 3: operand = .0; break;
5326 // default: break; // unreachable
5328 // result[i] += operand;
5332 const char function1[] =
5333 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5334 "%param1 = OpFunctionParameter %v4f32\n"
5336 "%iptr = OpVariable %fp_i32 Function\n"
5337 "%result = OpVariable %fp_v4f32 Function\n"
5338 " OpStore %iptr %c_i32_0\n"
5339 " OpStore %result %param1\n"
5343 "%ival = OpLoad %i32 %iptr\n"
5344 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5345 " OpLoopMerge %exit %phi None\n"
5346 " OpBranchConditional %lt_4 %entry %exit\n"
5348 "%entry = OpLabel\n"
5349 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5350 "%val = OpLoad %f32 %loc\n"
5351 " OpSelectionMerge %phi None\n"
5352 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5354 "%case0 = OpLabel\n"
5356 "%case1 = OpLabel\n"
5358 "%case2 = OpLabel\n"
5360 "%case3 = OpLabel\n"
5363 "%default = OpLabel\n"
5367 "%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
5368 "%add = OpFAdd %f32 %val %operand\n"
5369 " OpStore %loc %add\n"
5370 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5371 " OpStore %iptr %ival_next\n"
5375 "%ret = OpLoad %v4f32 %result\n"
5376 " OpReturnValue %ret\n"
5380 fragments1["pre_main"] = typesAndConstants1;
5381 fragments1["testfun"] = function1;
5383 getHalfColorsFullAlpha(inputColors);
5385 outputColors1[0] = RGBA(178, 255, 229, 255);
5386 outputColors1[1] = RGBA(178, 127, 102, 255);
5387 outputColors1[2] = RGBA(51, 255, 102, 255);
5388 outputColors1[3] = RGBA(51, 127, 229, 255);
5390 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
5392 const char typesAndConstants2[] =
5393 "%c_f32_p2 = OpConstant %f32 0.2\n";
5395 // Add .4 to the second element of the given parameter.
5396 const char function2[] =
5397 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5398 "%param = OpFunctionParameter %v4f32\n"
5399 "%entry = OpLabel\n"
5400 "%result = OpVariable %fp_v4f32 Function\n"
5401 " OpStore %result %param\n"
5402 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
5403 "%val = OpLoad %f32 %loc\n"
5407 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
5408 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
5409 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
5410 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
5411 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
5412 " OpLoopMerge %exit %phi None\n"
5413 " OpBranchConditional %still_loop %phi %exit\n"
5416 " OpStore %loc %accum\n"
5417 "%ret = OpLoad %v4f32 %result\n"
5418 " OpReturnValue %ret\n"
5422 fragments2["pre_main"] = typesAndConstants2;
5423 fragments2["testfun"] = function2;
5425 outputColors2[0] = RGBA(127, 229, 127, 255);
5426 outputColors2[1] = RGBA(127, 102, 0, 255);
5427 outputColors2[2] = RGBA(0, 229, 0, 255);
5428 outputColors2[3] = RGBA(0, 102, 127, 255);
5430 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
5432 const char typesAndConstants3[] =
5433 "%true = OpConstantTrue %bool\n"
5434 "%false = OpConstantFalse %bool\n"
5435 "%c_f32_p2 = OpConstant %f32 0.2\n";
5437 // Swap the second and the third element of the given parameter.
5438 const char function3[] =
5439 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5440 "%param = OpFunctionParameter %v4f32\n"
5441 "%entry = OpLabel\n"
5442 "%result = OpVariable %fp_v4f32 Function\n"
5443 " OpStore %result %param\n"
5444 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
5445 "%a_init = OpLoad %f32 %a_loc\n"
5446 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
5447 "%b_init = OpLoad %f32 %b_loc\n"
5451 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
5452 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
5453 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
5454 " OpLoopMerge %exit %phi None\n"
5455 " OpBranchConditional %still_loop %phi %exit\n"
5458 " OpStore %a_loc %a_next\n"
5459 " OpStore %b_loc %b_next\n"
5460 "%ret = OpLoad %v4f32 %result\n"
5461 " OpReturnValue %ret\n"
5465 fragments3["pre_main"] = typesAndConstants3;
5466 fragments3["testfun"] = function3;
5468 outputColors3[0] = RGBA(127, 127, 127, 255);
5469 outputColors3[1] = RGBA(127, 0, 0, 255);
5470 outputColors3[2] = RGBA(0, 0, 127, 255);
5471 outputColors3[3] = RGBA(0, 127, 0, 255);
5473 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
5475 return group.release();
5478 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
5480 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
5481 RGBA inputColors[4];
5482 RGBA outputColors[4];
5484 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
5485 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
5486 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
5487 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
5488 const char constantsAndTypes[] =
5489 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
5490 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5491 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
5492 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
5493 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n"
5496 const char function[] =
5497 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5498 "%param = OpFunctionParameter %v4f32\n"
5499 "%label = OpLabel\n"
5500 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
5501 "%var2 = OpVariable %fp_f32 Function\n"
5502 "%red = OpCompositeExtract %f32 %param 0\n"
5503 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
5504 " OpStore %var2 %plus_red\n"
5505 "%val1 = OpLoad %f32 %var1\n"
5506 "%val2 = OpLoad %f32 %var2\n"
5507 "%mul = OpFMul %f32 %val1 %val2\n"
5508 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
5509 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
5510 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
5511 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
5512 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
5513 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
5514 " OpReturnValue %ret\n"
5517 struct CaseNameDecoration
5524 CaseNameDecoration tests[] = {
5525 {"multiplication", "OpDecorate %mul NoContraction"},
5526 {"addition", "OpDecorate %add NoContraction"},
5527 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
5530 getHalfColorsFullAlpha(inputColors);
5532 for (deUint8 idx = 0; idx < 4; ++idx)
5534 inputColors[idx].setRed(0);
5535 outputColors[idx] = RGBA(0, 0, 0, 255);
5538 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
5540 map<string, string> fragments;
5542 fragments["decoration"] = tests[testNdx].decoration;
5543 fragments["pre_main"] = constantsAndTypes;
5544 fragments["testfun"] = function;
5546 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
5549 return group.release();
5552 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
5554 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
5557 const char constantsAndTypes[] =
5558 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
5559 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
5560 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
5561 "%fp_stype = OpTypePointer Function %stype\n";
5563 const char function[] =
5564 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5565 "%param1 = OpFunctionParameter %v4f32\n"
5567 "%v1 = OpVariable %fp_v4f32 Function\n"
5568 "%v2 = OpVariable %fp_a2f32 Function\n"
5569 "%v3 = OpVariable %fp_f32 Function\n"
5570 "%v = OpVariable %fp_stype Function\n"
5571 "%vv = OpVariable %fp_stype Function\n"
5572 "%vvv = OpVariable %fp_f32 Function\n"
5574 " OpStore %v1 %c_v4f32_1_1_1_1\n"
5575 " OpStore %v2 %c_a2f32_1\n"
5576 " OpStore %v3 %c_f32_1\n"
5578 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5579 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
5580 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5581 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
5582 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
5583 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
5585 " OpStore %p_v4f32 %v1_v ${access_type}\n"
5586 " OpStore %p_a2f32 %v2_v ${access_type}\n"
5587 " OpStore %p_f32 %v3_v ${access_type}\n"
5589 " OpCopyMemory %vv %v ${access_type}\n"
5590 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
5592 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
5593 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
5594 "%v_f32_3 = OpLoad %f32 %vvv\n"
5596 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
5597 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
5598 " OpReturnValue %ret2\n"
5601 struct NameMemoryAccess
5608 NameMemoryAccess tests[] =
5611 { "volatile", "Volatile" },
5612 { "aligned", "Aligned 1" },
5613 { "volatile_aligned", "Volatile|Aligned 1" },
5614 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
5615 { "volatile_nontemporal", "Volatile|Nontemporal" },
5616 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
5619 getHalfColorsFullAlpha(colors);
5621 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
5623 map<string, string> fragments;
5624 map<string, string> memoryAccess;
5625 memoryAccess["access_type"] = tests[testNdx].accessType;
5627 fragments["pre_main"] = constantsAndTypes;
5628 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
5629 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
5631 return memoryAccessTests.release();
5633 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
5635 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
5636 RGBA defaultColors[4];
5637 map<string, string> fragments;
5638 getDefaultColors(defaultColors);
5640 // First, simple cases that don't do anything with the OpUndef result.
5641 struct NameCodePair { string name, decl, type; };
5642 const NameCodePair tests[] =
5644 {"bool", "", "%bool"},
5645 {"vec2uint32", "%type = OpTypeVector %u32 2", "%type"},
5646 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown", "%type"},
5647 {"sampler", "%type = OpTypeSampler", "%type"},
5648 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img", "%type"},
5649 {"pointer", "", "%fp_i32"},
5650 {"runtimearray", "%type = OpTypeRuntimeArray %f32", "%type"},
5651 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100", "%type"},
5652 {"struct", "%type = OpTypeStruct %f32 %i32 %u32", "%type"}};
5653 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5655 fragments["undef_type"] = tests[testNdx].type;
5656 fragments["testfun"] = StringTemplate(
5657 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5658 "%param1 = OpFunctionParameter %v4f32\n"
5659 "%label_testfun = OpLabel\n"
5660 "%undef = OpUndef ${undef_type}\n"
5661 "OpReturnValue %param1\n"
5662 "OpFunctionEnd\n").specialize(fragments);
5663 fragments["pre_main"] = tests[testNdx].decl;
5664 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
5668 fragments["testfun"] =
5669 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5670 "%param1 = OpFunctionParameter %v4f32\n"
5671 "%label_testfun = OpLabel\n"
5672 "%undef = OpUndef %f32\n"
5673 "%zero = OpFMul %f32 %undef %c_f32_0\n"
5674 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
5675 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
5676 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5677 "%b = OpFAdd %f32 %a %actually_zero\n"
5678 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
5679 "OpReturnValue %ret\n"
5682 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
5684 fragments["testfun"] =
5685 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5686 "%param1 = OpFunctionParameter %v4f32\n"
5687 "%label_testfun = OpLabel\n"
5688 "%undef = OpUndef %i32\n"
5689 "%zero = OpIMul %i32 %undef %c_i32_0\n"
5690 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
5691 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
5692 "OpReturnValue %ret\n"
5695 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
5697 fragments["testfun"] =
5698 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5699 "%param1 = OpFunctionParameter %v4f32\n"
5700 "%label_testfun = OpLabel\n"
5701 "%undef = OpUndef %u32\n"
5702 "%zero = OpIMul %u32 %undef %c_i32_0\n"
5703 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
5704 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
5705 "OpReturnValue %ret\n"
5708 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
5710 fragments["testfun"] =
5711 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5712 "%param1 = OpFunctionParameter %v4f32\n"
5713 "%label_testfun = OpLabel\n"
5714 "%undef = OpUndef %v4f32\n"
5715 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
5716 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
5717 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
5718 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
5719 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
5720 "%is_nan_0 = OpIsNan %bool %zero_0\n"
5721 "%is_nan_1 = OpIsNan %bool %zero_1\n"
5722 "%is_nan_2 = OpIsNan %bool %zero_2\n"
5723 "%is_nan_3 = OpIsNan %bool %zero_3\n"
5724 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
5725 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
5726 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
5727 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
5728 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5729 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
5730 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
5731 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
5732 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
5733 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
5734 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
5735 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
5736 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
5737 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
5738 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
5739 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
5740 "OpReturnValue %ret\n"
5743 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
5745 fragments["pre_main"] =
5746 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
5747 fragments["testfun"] =
5748 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5749 "%param1 = OpFunctionParameter %v4f32\n"
5750 "%label_testfun = OpLabel\n"
5751 "%undef = OpUndef %m2x2f32\n"
5752 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
5753 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
5754 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
5755 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
5756 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
5757 "%is_nan_0 = OpIsNan %bool %zero_0\n"
5758 "%is_nan_1 = OpIsNan %bool %zero_1\n"
5759 "%is_nan_2 = OpIsNan %bool %zero_2\n"
5760 "%is_nan_3 = OpIsNan %bool %zero_3\n"
5761 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
5762 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
5763 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
5764 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
5765 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5766 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
5767 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
5768 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
5769 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
5770 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
5771 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
5772 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
5773 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
5774 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
5775 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
5776 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
5777 "OpReturnValue %ret\n"
5780 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
5782 return opUndefTests.release();
5785 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
5787 const RGBA inputColors[4] =
5790 RGBA(0, 0, 255, 255),
5791 RGBA(0, 255, 0, 255),
5792 RGBA(0, 255, 255, 255)
5795 const RGBA expectedColors[4] =
5797 RGBA(255, 0, 0, 255),
5798 RGBA(255, 0, 0, 255),
5799 RGBA(255, 0, 0, 255),
5800 RGBA(255, 0, 0, 255)
5803 const struct SingleFP16Possibility
5806 const char* constant; // Value to assign to %test_constant.
5808 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
5814 -constructNormalizedFloat(1, 0x300000),
5815 "%cond = OpFOrdEqual %bool %c %test_constant\n"
5820 constructNormalizedFloat(7, 0x000000),
5821 "%cond = OpFOrdEqual %bool %c %test_constant\n"
5823 // SPIR-V requires that OpQuantizeToF16 flushes
5824 // any numbers that would end up denormalized in F16 to zero.
5828 std::ldexp(1.5f, -140),
5829 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5834 -std::ldexp(1.5f, -140),
5835 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5840 std::ldexp(1.0f, -16),
5841 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5842 }, // too small positive
5844 "negative_too_small",
5846 -std::ldexp(1.0f, -32),
5847 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5848 }, // too small negative
5852 -std::ldexp(1.0f, 128),
5854 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
5855 "%inf = OpIsInf %bool %c\n"
5856 "%cond = OpLogicalAnd %bool %gz %inf\n"
5861 std::ldexp(1.0f, 128),
5863 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
5864 "%inf = OpIsInf %bool %c\n"
5865 "%cond = OpLogicalAnd %bool %gz %inf\n"
5868 "round_to_negative_inf",
5870 -std::ldexp(1.0f, 32),
5872 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
5873 "%inf = OpIsInf %bool %c\n"
5874 "%cond = OpLogicalAnd %bool %gz %inf\n"
5879 std::ldexp(1.0f, 16),
5881 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
5882 "%inf = OpIsInf %bool %c\n"
5883 "%cond = OpLogicalAnd %bool %gz %inf\n"
5888 std::numeric_limits<float>::quiet_NaN(),
5890 // Test for any NaN value, as NaNs are not preserved
5891 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
5892 "%cond = OpIsNan %bool %direct_quant\n"
5897 std::numeric_limits<float>::quiet_NaN(),
5899 // Test for any NaN value, as NaNs are not preserved
5900 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
5901 "%cond = OpIsNan %bool %direct_quant\n"
5904 const char* constants =
5905 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
5907 StringTemplate function (
5908 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5909 "%param1 = OpFunctionParameter %v4f32\n"
5910 "%label_testfun = OpLabel\n"
5911 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5912 "%b = OpFAdd %f32 %test_constant %a\n"
5913 "%c = OpQuantizeToF16 %f32 %b\n"
5915 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
5916 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
5917 " OpReturnValue %retval\n"
5921 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
5922 const char* specConstants =
5923 "%test_constant = OpSpecConstant %f32 0.\n"
5924 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
5926 StringTemplate specConstantFunction(
5927 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5928 "%param1 = OpFunctionParameter %v4f32\n"
5929 "%label_testfun = OpLabel\n"
5931 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
5932 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
5933 " OpReturnValue %retval\n"
5937 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
5939 map<string, string> codeSpecialization;
5940 map<string, string> fragments;
5941 codeSpecialization["condition"] = tests[idx].condition;
5942 fragments["testfun"] = function.specialize(codeSpecialization);
5943 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
5944 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
5947 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
5949 map<string, string> codeSpecialization;
5950 map<string, string> fragments;
5951 vector<deInt32> passConstants;
5952 deInt32 specConstant;
5954 codeSpecialization["condition"] = tests[idx].condition;
5955 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
5956 fragments["decoration"] = specDecorations;
5957 fragments["pre_main"] = specConstants;
5959 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
5960 passConstants.push_back(specConstant);
5962 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
5966 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
5968 RGBA inputColors[4] = {
5970 RGBA(0, 0, 255, 255),
5971 RGBA(0, 255, 0, 255),
5972 RGBA(0, 255, 255, 255)
5975 RGBA expectedColors[4] =
5977 RGBA(255, 0, 0, 255),
5978 RGBA(255, 0, 0, 255),
5979 RGBA(255, 0, 0, 255),
5980 RGBA(255, 0, 0, 255)
5983 struct DualFP16Possibility
5988 const char* possibleOutput1;
5989 const char* possibleOutput2;
5992 "positive_round_up_or_round_down",
5994 constructNormalizedFloat(8, 0x300300),
5999 "negative_round_up_or_round_down",
6001 -constructNormalizedFloat(-7, 0x600800),
6008 constructNormalizedFloat(2, 0x01e000),
6013 "carry_to_exponent",
6015 constructNormalizedFloat(1, 0xffe000),
6020 StringTemplate constants (
6021 "%input_const = OpConstant %f32 ${input}\n"
6022 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6023 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6026 StringTemplate specConstants (
6027 "%input_const = OpSpecConstant %f32 0.\n"
6028 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6029 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6032 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
6034 const char* function =
6035 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6036 "%param1 = OpFunctionParameter %v4f32\n"
6037 "%label_testfun = OpLabel\n"
6038 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6039 // For the purposes of this test we assume that 0.f will always get
6040 // faithfully passed through the pipeline stages.
6041 "%b = OpFAdd %f32 %input_const %a\n"
6042 "%c = OpQuantizeToF16 %f32 %b\n"
6043 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
6044 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
6045 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
6046 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6047 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
6048 " OpReturnValue %retval\n"
6051 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6052 map<string, string> fragments;
6053 map<string, string> constantSpecialization;
6055 constantSpecialization["input"] = tests[idx].input;
6056 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6057 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6058 fragments["testfun"] = function;
6059 fragments["pre_main"] = constants.specialize(constantSpecialization);
6060 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6063 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6064 map<string, string> fragments;
6065 map<string, string> constantSpecialization;
6066 vector<deInt32> passConstants;
6067 deInt32 specConstant;
6069 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6070 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6071 fragments["testfun"] = function;
6072 fragments["decoration"] = specDecorations;
6073 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
6075 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
6076 passConstants.push_back(specConstant);
6078 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6082 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
6084 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
6085 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
6086 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
6087 return opQuantizeTests.release();
6090 struct ShaderPermutation
6092 deUint8 vertexPermutation;
6093 deUint8 geometryPermutation;
6094 deUint8 tesscPermutation;
6095 deUint8 tessePermutation;
6096 deUint8 fragmentPermutation;
6099 ShaderPermutation getShaderPermutation(deUint8 inputValue)
6101 ShaderPermutation permutation =
6103 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
6104 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
6105 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
6106 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
6107 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
6112 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
6114 RGBA defaultColors[4];
6115 RGBA invertedColors[4];
6116 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
6118 const ShaderElement combinedPipeline[] =
6120 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
6121 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
6122 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
6123 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
6124 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
6127 getDefaultColors(defaultColors);
6128 getInvertedDefaultColors(invertedColors);
6129 addFunctionCaseWithPrograms<InstanceContext>(
6130 moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline,
6131 createInstanceContext(combinedPipeline, map<string, string>()));
6133 const char* numbers[] =
6138 for (deInt8 idx = 0; idx < 32; ++idx)
6140 ShaderPermutation permutation = getShaderPermutation(idx);
6141 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
6142 const ShaderElement pipeline[] =
6144 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
6145 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
6146 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
6147 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
6148 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
6151 // If there are an even number of swaps, then it should be no-op.
6152 // If there are an odd number, the color should be flipped.
6153 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
6155 addFunctionCaseWithPrograms<InstanceContext>(
6156 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
6157 createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
6161 addFunctionCaseWithPrograms<InstanceContext>(
6162 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
6163 createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
6166 return moduleTests.release();
6169 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
6171 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
6172 RGBA defaultColors[4];
6173 getDefaultColors(defaultColors);
6174 map<string, string> fragments;
6175 fragments["pre_main"] =
6176 "%c_f32_5 = OpConstant %f32 5.\n";
6178 // A loop with a single block. The Continue Target is the loop block
6179 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
6180 // -- the "continue construct" forms the entire loop.
6181 fragments["testfun"] =
6182 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6183 "%param1 = OpFunctionParameter %v4f32\n"
6185 "%entry = OpLabel\n"
6186 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6189 ";adds and subtracts 1.0 to %val in alternate iterations\n"
6191 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
6192 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
6193 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
6194 "%val = OpFAdd %f32 %val1 %delta\n"
6195 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
6196 "%count__ = OpISub %i32 %count %c_i32_1\n"
6197 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6198 "OpLoopMerge %exit %loop None\n"
6199 "OpBranchConditional %again %loop %exit\n"
6202 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
6203 "OpReturnValue %result\n"
6207 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
6209 // Body comprised of multiple basic blocks.
6210 const StringTemplate multiBlock(
6211 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6212 "%param1 = OpFunctionParameter %v4f32\n"
6214 "%entry = OpLabel\n"
6215 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6218 ";adds and subtracts 1.0 to %val in alternate iterations\n"
6220 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
6221 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
6222 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
6223 // There are several possibilities for the Continue Target below. Each
6224 // will be specialized into a separate test case.
6225 "OpLoopMerge %exit ${continue_target} None\n"
6229 ";delta_next = (delta > 0) ? -1 : 1;\n"
6230 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
6231 "OpSelectionMerge %gather DontFlatten\n"
6232 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
6235 "OpBranch %gather\n"
6238 "OpBranch %gather\n"
6240 "%gather = OpLabel\n"
6241 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
6242 "%val = OpFAdd %f32 %val1 %delta\n"
6243 "%count__ = OpISub %i32 %count %c_i32_1\n"
6244 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6245 "OpBranchConditional %again %loop %exit\n"
6248 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
6249 "OpReturnValue %result\n"
6253 map<string, string> continue_target;
6255 // The Continue Target is the loop block itself.
6256 continue_target["continue_target"] = "%loop";
6257 fragments["testfun"] = multiBlock.specialize(continue_target);
6258 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
6260 // The Continue Target is at the end of the loop.
6261 continue_target["continue_target"] = "%gather";
6262 fragments["testfun"] = multiBlock.specialize(continue_target);
6263 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
6265 // A loop with continue statement.
6266 fragments["testfun"] =
6267 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6268 "%param1 = OpFunctionParameter %v4f32\n"
6270 "%entry = OpLabel\n"
6271 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6274 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
6276 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6277 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
6278 "OpLoopMerge %exit %continue None\n"
6282 ";skip if %count==2\n"
6283 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
6284 "OpSelectionMerge %continue DontFlatten\n"
6285 "OpBranchConditional %eq2 %continue %body\n"
6288 "%fcount = OpConvertSToF %f32 %count\n"
6289 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6290 "OpBranch %continue\n"
6292 "%continue = OpLabel\n"
6293 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
6294 "%count__ = OpISub %i32 %count %c_i32_1\n"
6295 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6296 "OpBranchConditional %again %loop %exit\n"
6299 "%same = OpFSub %f32 %val %c_f32_8\n"
6300 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6301 "OpReturnValue %result\n"
6303 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
6305 // A loop with break.
6306 fragments["testfun"] =
6307 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6308 "%param1 = OpFunctionParameter %v4f32\n"
6310 "%entry = OpLabel\n"
6311 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6312 "%dot = OpDot %f32 %param1 %param1\n"
6313 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6314 "%zero = OpConvertFToU %u32 %div\n"
6315 "%two = OpIAdd %i32 %zero %c_i32_2\n"
6316 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6319 ";adds 4 and 3 to %val0 (exits early)\n"
6321 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6322 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
6323 "OpLoopMerge %exit %continue None\n"
6327 ";end loop if %count==%two\n"
6328 "%above2 = OpSGreaterThan %bool %count %two\n"
6329 "OpSelectionMerge %continue DontFlatten\n"
6330 "OpBranchConditional %above2 %body %exit\n"
6333 "%fcount = OpConvertSToF %f32 %count\n"
6334 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6335 "OpBranch %continue\n"
6337 "%continue = OpLabel\n"
6338 "%count__ = OpISub %i32 %count %c_i32_1\n"
6339 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6340 "OpBranchConditional %again %loop %exit\n"
6343 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
6344 "%same = OpFSub %f32 %val_post %c_f32_7\n"
6345 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6346 "OpReturnValue %result\n"
6348 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
6350 // A loop with return.
6351 fragments["testfun"] =
6352 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6353 "%param1 = OpFunctionParameter %v4f32\n"
6355 "%entry = OpLabel\n"
6356 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6357 "%dot = OpDot %f32 %param1 %param1\n"
6358 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6359 "%zero = OpConvertFToU %u32 %div\n"
6360 "%two = OpIAdd %i32 %zero %c_i32_2\n"
6361 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6364 ";returns early without modifying %param1\n"
6366 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6367 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
6368 "OpLoopMerge %exit %continue None\n"
6372 ";return if %count==%two\n"
6373 "%above2 = OpSGreaterThan %bool %count %two\n"
6374 "OpSelectionMerge %continue DontFlatten\n"
6375 "OpBranchConditional %above2 %body %early_exit\n"
6377 "%early_exit = OpLabel\n"
6378 "OpReturnValue %param1\n"
6381 "%fcount = OpConvertSToF %f32 %count\n"
6382 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6383 "OpBranch %continue\n"
6385 "%continue = OpLabel\n"
6386 "%count__ = OpISub %i32 %count %c_i32_1\n"
6387 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6388 "OpBranchConditional %again %loop %exit\n"
6391 ";should never get here, so return an incorrect result\n"
6392 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
6393 "OpReturnValue %result\n"
6395 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
6397 return testGroup.release();
6400 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
6401 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
6403 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
6404 map<string, string> fragments;
6406 // A barrier inside a function body.
6407 fragments["pre_main"] =
6408 "%Workgroup = OpConstant %i32 2\n"
6409 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
6410 fragments["testfun"] =
6411 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6412 "%param1 = OpFunctionParameter %v4f32\n"
6413 "%label_testfun = OpLabel\n"
6414 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6415 "OpReturnValue %param1\n"
6417 addTessCtrlTest(testGroup.get(), "in_function", fragments);
6419 // Common setup code for the following tests.
6420 fragments["pre_main"] =
6421 "%Workgroup = OpConstant %i32 2\n"
6422 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
6423 "%c_f32_5 = OpConstant %f32 5.\n";
6424 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
6425 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6426 "%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";
6433 // Barriers inside OpSwitch branches.
6434 fragments["testfun"] =
6436 "OpSelectionMerge %switch_exit None\n"
6437 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
6439 "%case1 = OpLabel\n"
6440 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6441 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6442 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6443 "OpBranch %switch_exit\n"
6445 "%switch_default = OpLabel\n"
6446 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6447 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6448 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6449 "OpBranch %switch_exit\n"
6451 "%case0 = OpLabel\n"
6452 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6453 "OpBranch %switch_exit\n"
6455 "%switch_exit = OpLabel\n"
6456 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
6457 "OpReturnValue %ret\n"
6459 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
6461 // Barriers inside if-then-else.
6462 fragments["testfun"] =
6464 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
6465 "OpSelectionMerge %exit DontFlatten\n"
6466 "OpBranchConditional %eq0 %then %else\n"
6469 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6470 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6471 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6475 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6479 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
6480 "OpReturnValue %ret\n"
6482 addTessCtrlTest(testGroup.get(), "in_if", fragments);
6484 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
6485 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
6486 fragments["testfun"] =
6488 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
6489 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
6490 "OpSelectionMerge %exit DontFlatten\n"
6491 "OpBranchConditional %thread0 %then %else\n"
6494 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6498 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
6502 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
6503 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6504 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
6505 "OpReturnValue %ret\n"
6507 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
6509 // A barrier inside a loop.
6510 fragments["pre_main"] =
6511 "%Workgroup = OpConstant %i32 2\n"
6512 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
6513 "%c_f32_10 = OpConstant %f32 10.\n";
6514 fragments["testfun"] =
6515 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6516 "%param1 = OpFunctionParameter %v4f32\n"
6517 "%entry = OpLabel\n"
6518 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6521 ";adds 4, 3, 2, and 1 to %val0\n"
6523 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
6524 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
6525 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6526 "%fcount = OpConvertSToF %f32 %count\n"
6527 "%val = OpFAdd %f32 %val1 %fcount\n"
6528 "%count__ = OpISub %i32 %count %c_i32_1\n"
6529 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6530 "OpLoopMerge %exit %loop None\n"
6531 "OpBranchConditional %again %loop %exit\n"
6534 "%same = OpFSub %f32 %val %c_f32_10\n"
6535 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6536 "OpReturnValue %ret\n"
6538 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
6540 return testGroup.release();
6543 // Test for the OpFRem instruction.
6544 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
6546 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
6547 map<string, string> fragments;
6548 RGBA inputColors[4];
6549 RGBA outputColors[4];
6551 fragments["pre_main"] =
6552 "%c_f32_3 = OpConstant %f32 3.0\n"
6553 "%c_f32_n3 = OpConstant %f32 -3.0\n"
6554 "%c_f32_4 = OpConstant %f32 4.0\n"
6555 "%c_f32_p75 = OpConstant %f32 0.75\n"
6556 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
6557 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
6558 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
6560 // The test does the following.
6561 // vec4 result = (param1 * 8.0) - 4.0;
6562 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
6563 fragments["testfun"] =
6564 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6565 "%param1 = OpFunctionParameter %v4f32\n"
6566 "%label_testfun = OpLabel\n"
6567 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
6568 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
6569 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
6570 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
6571 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
6572 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
6573 "OpReturnValue %xy_0_1\n"
6577 inputColors[0] = RGBA(16, 16, 0, 255);
6578 inputColors[1] = RGBA(232, 232, 0, 255);
6579 inputColors[2] = RGBA(232, 16, 0, 255);
6580 inputColors[3] = RGBA(16, 232, 0, 255);
6582 outputColors[0] = RGBA(64, 64, 0, 255);
6583 outputColors[1] = RGBA(255, 255, 0, 255);
6584 outputColors[2] = RGBA(255, 64, 0, 255);
6585 outputColors[3] = RGBA(64, 255, 0, 255);
6587 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
6588 return testGroup.release();
6591 // Test for the OpSRem instruction.
6592 tcu::TestCaseGroup* createOpSRemGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
6594 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "srem", "OpSRem"));
6595 map<string, string> fragments;
6597 fragments["pre_main"] =
6598 "%c_f32_255 = OpConstant %f32 255.0\n"
6599 "%c_i32_128 = OpConstant %i32 128\n"
6600 "%c_i32_255 = OpConstant %i32 255\n"
6601 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
6602 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
6603 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
6605 // The test does the following.
6606 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
6607 // ivec4 result = ivec4(srem(ints.x, ints.y), srem(ints.y, ints.z), srem(ints.z, ints.x), 255);
6608 // return float(result + 128) / 255.0;
6609 fragments["testfun"] =
6610 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6611 "%param1 = OpFunctionParameter %v4f32\n"
6612 "%label_testfun = OpLabel\n"
6613 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
6614 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
6615 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
6616 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
6617 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
6618 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
6619 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
6620 "%x_out = OpSRem %i32 %x_in %y_in\n"
6621 "%y_out = OpSRem %i32 %y_in %z_in\n"
6622 "%z_out = OpSRem %i32 %z_in %x_in\n"
6623 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
6624 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
6625 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
6626 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
6627 "OpReturnValue %float_out\n"
6630 const struct CaseParams
6633 const char* failMessageTemplate; // customized status message
6634 qpTestResult failResult; // override status on failure
6635 int operands[4][3]; // four (x, y, z) vectors of operands
6636 int results[4][3]; // four (x, y, z) vectors of results
6642 QP_TEST_RESULT_FAIL,
6643 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
6644 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
6648 "Inconsistent results, but within specification: ${reason}",
6649 negFailResult, // negative operands, not required by the spec
6650 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
6651 { { 5, 12, -2 }, { 0, -5, 2 }, { 3, 8, -6 }, { 25, -60, 0 } }, // results
6654 // If either operand is negative the result is undefined. Some implementations may still return correct values.
6656 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
6658 const CaseParams& params = cases[caseNdx];
6659 RGBA inputColors[4];
6660 RGBA outputColors[4];
6662 for (int i = 0; i < 4; ++i)
6664 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
6665 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
6668 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
6671 return testGroup.release();
6674 // Test for the OpSMod instruction.
6675 tcu::TestCaseGroup* createOpSModGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
6677 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "smod", "OpSMod"));
6678 map<string, string> fragments;
6680 fragments["pre_main"] =
6681 "%c_f32_255 = OpConstant %f32 255.0\n"
6682 "%c_i32_128 = OpConstant %i32 128\n"
6683 "%c_i32_255 = OpConstant %i32 255\n"
6684 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
6685 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
6686 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
6688 // The test does the following.
6689 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
6690 // ivec4 result = ivec4(smod(ints.x, ints.y), smod(ints.y, ints.z), smod(ints.z, ints.x), 255);
6691 // return float(result + 128) / 255.0;
6692 fragments["testfun"] =
6693 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6694 "%param1 = OpFunctionParameter %v4f32\n"
6695 "%label_testfun = OpLabel\n"
6696 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
6697 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
6698 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
6699 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
6700 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
6701 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
6702 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
6703 "%x_out = OpSMod %i32 %x_in %y_in\n"
6704 "%y_out = OpSMod %i32 %y_in %z_in\n"
6705 "%z_out = OpSMod %i32 %z_in %x_in\n"
6706 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
6707 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
6708 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
6709 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
6710 "OpReturnValue %float_out\n"
6713 const struct CaseParams
6716 const char* failMessageTemplate; // customized status message
6717 qpTestResult failResult; // override status on failure
6718 int operands[4][3]; // four (x, y, z) vectors of operands
6719 int results[4][3]; // four (x, y, z) vectors of results
6725 QP_TEST_RESULT_FAIL,
6726 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
6727 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
6731 "Inconsistent results, but within specification: ${reason}",
6732 negFailResult, // negative operands, not required by the spec
6733 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
6734 { { 5, -5, 3 }, { 0, 2, -3 }, { 3, -73, 69 }, { -35, 40, 0 } }, // results
6737 // If either operand is negative the result is undefined. Some implementations may still return correct values.
6739 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
6741 const CaseParams& params = cases[caseNdx];
6742 RGBA inputColors[4];
6743 RGBA outputColors[4];
6745 for (int i = 0; i < 4; ++i)
6747 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
6748 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
6751 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
6754 return testGroup.release();
6759 INTEGER_TYPE_SIGNED_16,
6760 INTEGER_TYPE_SIGNED_32,
6761 INTEGER_TYPE_SIGNED_64,
6763 INTEGER_TYPE_UNSIGNED_16,
6764 INTEGER_TYPE_UNSIGNED_32,
6765 INTEGER_TYPE_UNSIGNED_64,
6768 const string getBitWidthStr (IntegerType type)
6772 case INTEGER_TYPE_SIGNED_16:
6773 case INTEGER_TYPE_UNSIGNED_16: return "16";
6775 case INTEGER_TYPE_SIGNED_32:
6776 case INTEGER_TYPE_UNSIGNED_32: return "32";
6778 case INTEGER_TYPE_SIGNED_64:
6779 case INTEGER_TYPE_UNSIGNED_64: return "64";
6781 default: DE_ASSERT(false);
6786 const string getByteWidthStr (IntegerType type)
6790 case INTEGER_TYPE_SIGNED_16:
6791 case INTEGER_TYPE_UNSIGNED_16: return "2";
6793 case INTEGER_TYPE_SIGNED_32:
6794 case INTEGER_TYPE_UNSIGNED_32: return "4";
6796 case INTEGER_TYPE_SIGNED_64:
6797 case INTEGER_TYPE_UNSIGNED_64: return "8";
6799 default: DE_ASSERT(false);
6804 bool isSigned (IntegerType type)
6806 return (type <= INTEGER_TYPE_SIGNED_64);
6809 const string getTypeName (IntegerType type)
6811 string prefix = isSigned(type) ? "" : "u";
6812 return prefix + "int" + getBitWidthStr(type);
6815 const string getTestName (IntegerType from, IntegerType to)
6817 return getTypeName(from) + "_to_" + getTypeName(to);
6820 const string getAsmTypeDeclaration (IntegerType type)
6822 string sign = isSigned(type) ? " 1" : " 0";
6823 return "OpTypeInt " + getBitWidthStr(type) + sign;
6826 template<typename T>
6827 BufferSp getSpecializedBuffer (deInt64 number)
6829 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
6832 BufferSp getBuffer (IntegerType type, deInt64 number)
6836 case INTEGER_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
6837 case INTEGER_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
6838 case INTEGER_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
6840 case INTEGER_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
6841 case INTEGER_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
6842 case INTEGER_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
6844 default: DE_ASSERT(false);
6845 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
6849 bool usesInt16 (IntegerType from, IntegerType to)
6851 return (from == INTEGER_TYPE_SIGNED_16 || from == INTEGER_TYPE_UNSIGNED_16
6852 || to == INTEGER_TYPE_SIGNED_16 || to == INTEGER_TYPE_UNSIGNED_16);
6855 bool usesInt64 (IntegerType from, IntegerType to)
6857 return (from == INTEGER_TYPE_SIGNED_64 || from == INTEGER_TYPE_UNSIGNED_64
6858 || to == INTEGER_TYPE_SIGNED_64 || to == INTEGER_TYPE_UNSIGNED_64);
6861 ComputeTestFeatures getConversionUsedFeatures (IntegerType from, IntegerType to)
6863 if (usesInt16(from, to))
6865 if (usesInt64(from, to))
6867 return COMPUTE_TEST_USES_INT16_INT64;
6871 return COMPUTE_TEST_USES_INT16;
6876 return COMPUTE_TEST_USES_INT64;
6882 ConvertCase (IntegerType from, IntegerType to, deInt64 number)
6885 , m_features (getConversionUsedFeatures(from, to))
6886 , m_name (getTestName(from, to))
6887 , m_inputBuffer (getBuffer(from, number))
6888 , m_outputBuffer (getBuffer(to, number))
6890 m_asmTypes["inputType"] = getAsmTypeDeclaration(from);
6891 m_asmTypes["outputType"] = getAsmTypeDeclaration(to);
6893 if (m_features == COMPUTE_TEST_USES_INT16)
6895 m_asmTypes["int_capabilities"] = "OpCapability Int16\n";
6897 else if (m_features == COMPUTE_TEST_USES_INT64)
6899 m_asmTypes["int_capabilities"] = "OpCapability Int64\n";
6901 else if (m_features == COMPUTE_TEST_USES_INT16_INT64)
6903 m_asmTypes["int_capabilities"] = string("OpCapability Int16\n") +
6904 "OpCapability Int64\n";
6912 IntegerType m_fromType;
6913 IntegerType m_toType;
6914 ComputeTestFeatures m_features;
6916 map<string, string> m_asmTypes;
6917 BufferSp m_inputBuffer;
6918 BufferSp m_outputBuffer;
6921 const string getConvertCaseShaderStr (const string& instruction, const ConvertCase& convertCase)
6923 map<string, string> params = convertCase.m_asmTypes;
6925 params["instruction"] = instruction;
6927 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
6928 params["outDecorator"] = getByteWidthStr(convertCase.m_toType);
6930 const StringTemplate shader (
6931 "OpCapability Shader\n"
6932 "${int_capabilities}"
6933 "OpMemoryModel Logical GLSL450\n"
6934 "OpEntryPoint GLCompute %main \"main\" %id\n"
6935 "OpExecutionMode %main LocalSize 1 1 1\n"
6936 "OpSource GLSL 430\n"
6937 "OpName %main \"main\"\n"
6938 "OpName %id \"gl_GlobalInvocationID\"\n"
6940 "OpDecorate %id BuiltIn GlobalInvocationId\n"
6941 "OpDecorate %indata DescriptorSet 0\n"
6942 "OpDecorate %indata Binding 0\n"
6943 "OpDecorate %outdata DescriptorSet 0\n"
6944 "OpDecorate %outdata Binding 1\n"
6945 "OpDecorate %in_arr ArrayStride ${inDecorator}\n"
6946 "OpDecorate %out_arr ArrayStride ${outDecorator}\n"
6947 "OpDecorate %in_buf BufferBlock\n"
6948 "OpDecorate %out_buf BufferBlock\n"
6949 "OpMemberDecorate %in_buf 0 Offset 0\n"
6950 "OpMemberDecorate %out_buf 0 Offset 0\n"
6952 "%void = OpTypeVoid\n"
6953 "%voidf = OpTypeFunction %void\n"
6954 "%u32 = OpTypeInt 32 0\n"
6955 "%i32 = OpTypeInt 32 1\n"
6956 "%uvec3 = OpTypeVector %u32 3\n"
6957 "%uvec3ptr = OpTypePointer Input %uvec3\n"
6959 "%in_type = ${inputType}\n"
6960 "%out_type = ${outputType}\n"
6962 "%in_ptr = OpTypePointer Uniform %in_type\n"
6963 "%out_ptr = OpTypePointer Uniform %out_type\n"
6964 "%in_arr = OpTypeRuntimeArray %in_type\n"
6965 "%out_arr = OpTypeRuntimeArray %out_type\n"
6966 "%in_buf = OpTypeStruct %in_arr\n"
6967 "%out_buf = OpTypeStruct %out_arr\n"
6968 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
6969 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
6970 "%indata = OpVariable %in_bufptr Uniform\n"
6971 "%outdata = OpVariable %out_bufptr Uniform\n"
6972 "%inputptr = OpTypePointer Input %in_type\n"
6973 "%id = OpVariable %uvec3ptr Input\n"
6975 "%zero = OpConstant %i32 0\n"
6977 "%main = OpFunction %void None %voidf\n"
6978 "%label = OpLabel\n"
6979 "%idval = OpLoad %uvec3 %id\n"
6980 "%x = OpCompositeExtract %u32 %idval 0\n"
6981 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
6982 "%outloc = OpAccessChain %out_ptr %outdata %zero %x\n"
6983 "%inval = OpLoad %in_type %inloc\n"
6984 "%conv = ${instruction} %out_type %inval\n"
6985 " OpStore %outloc %conv\n"
6990 return shader.specialize(params);
6993 void createSConvertCases (vector<ConvertCase>& testCases)
6995 // Convert int to int
6996 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_32, 14669));
6997 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_64, 3341));
6999 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_SIGNED_64, 973610259));
7001 // Convert int to unsigned int
7002 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_32, 9288));
7003 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_64, 15460));
7005 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_UNSIGNED_64, 346213461));
7008 // Test for the OpSConvert instruction.
7009 tcu::TestCaseGroup* createSConvertTests (tcu::TestContext& testCtx)
7011 const string instruction ("OpSConvert");
7012 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "sconvert", "OpSConvert"));
7013 vector<ConvertCase> testCases;
7014 createSConvertCases(testCases);
7016 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7018 ComputeShaderSpec spec;
7020 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7021 spec.inputs.push_back(test->m_inputBuffer);
7022 spec.outputs.push_back(test->m_outputBuffer);
7023 spec.numWorkGroups = IVec3(1, 1, 1);
7025 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpSConvert.", spec, test->m_features));
7028 return group.release();
7031 void createUConvertCases (vector<ConvertCase>& testCases)
7033 // Convert unsigned int to unsigned int
7034 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_32, 60653));
7035 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_64, 17991));
7037 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_UNSIGNED_64, 904256275));
7039 // Convert unsigned int to int
7040 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_32, 38002));
7041 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_64, 64921));
7043 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_SIGNED_64, 4294956295ll));
7046 // Test for the OpUConvert instruction.
7047 tcu::TestCaseGroup* createUConvertTests (tcu::TestContext& testCtx)
7049 const string instruction ("OpUConvert");
7050 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "uconvert", "OpUConvert"));
7051 vector<ConvertCase> testCases;
7052 createUConvertCases(testCases);
7054 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7056 ComputeShaderSpec spec;
7058 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7059 spec.inputs.push_back(test->m_inputBuffer);
7060 spec.outputs.push_back(test->m_outputBuffer);
7061 spec.numWorkGroups = IVec3(1, 1, 1);
7063 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpUConvert.", spec, test->m_features));
7065 return group.release();
7068 const string getNumberTypeName (const NumberType type)
7070 if (type == NUMBERTYPE_INT32)
7074 else if (type == NUMBERTYPE_UINT32)
7078 else if (type == NUMBERTYPE_FLOAT32)
7089 deInt32 getInt(de::Random& rnd)
7091 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
7094 const string repeatString (const string& str, int times)
7097 for (int i = 0; i < times; ++i)
7104 const string getRandomConstantString (const NumberType type, de::Random& rnd)
7106 if (type == NUMBERTYPE_INT32)
7108 return numberToString<deInt32>(getInt(rnd));
7110 else if (type == NUMBERTYPE_UINT32)
7112 return numberToString<deUint32>(rnd.getUint32());
7114 else if (type == NUMBERTYPE_FLOAT32)
7116 return numberToString<float>(rnd.getFloat());
7125 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7127 map<string, string> params;
7130 for (int width = 2; width <= 4; ++width)
7132 string randomConst = numberToString(getInt(rnd));
7133 string widthStr = numberToString(width);
7134 int index = rnd.getInt(0, width-1);
7136 params["type"] = "vec";
7137 params["name"] = params["type"] + "_" + widthStr;
7138 params["compositeType"] = "%composite = OpTypeVector %custom " + widthStr +"\n";
7139 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7140 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7141 params["indexes"] = numberToString(index);
7142 testCases.push_back(params);
7146 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7148 const int limit = 10;
7149 map<string, string> params;
7151 for (int width = 2; width <= limit; ++width)
7153 string randomConst = numberToString(getInt(rnd));
7154 string widthStr = numberToString(width);
7155 int index = rnd.getInt(0, width-1);
7157 params["type"] = "array";
7158 params["name"] = params["type"] + "_" + widthStr;
7159 params["compositeType"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
7160 + "%composite = OpTypeArray %custom %arraywidth\n";
7162 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7163 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7164 params["indexes"] = numberToString(index);
7165 testCases.push_back(params);
7169 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7171 const int limit = 10;
7172 map<string, string> params;
7174 for (int width = 2; width <= limit; ++width)
7176 string randomConst = numberToString(getInt(rnd));
7177 int index = rnd.getInt(0, width-1);
7179 params["type"] = "struct";
7180 params["name"] = params["type"] + "_" + numberToString(width);
7181 params["compositeType"] = "%composite = OpTypeStruct" + repeatString(" %custom", width) + "\n";
7182 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7183 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7184 params["indexes"] = numberToString(index);
7185 testCases.push_back(params);
7189 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7191 map<string, string> params;
7194 for (int width = 2; width <= 4; ++width)
7196 string widthStr = numberToString(width);
7198 for (int column = 2 ; column <= 4; ++column)
7200 int index_0 = rnd.getInt(0, column-1);
7201 int index_1 = rnd.getInt(0, width-1);
7202 string columnStr = numberToString(column);
7204 params["type"] = "matrix";
7205 params["name"] = params["type"] + "_" + widthStr + "x" + columnStr;
7206 params["compositeType"] = string("%vectype = OpTypeVector %custom " + widthStr + "\n")
7207 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
7209 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n"
7210 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
7212 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
7213 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
7214 testCases.push_back(params);
7219 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7221 createVectorCompositeCases(testCases, rnd, type);
7222 createArrayCompositeCases(testCases, rnd, type);
7223 createStructCompositeCases(testCases, rnd, type);
7224 // Matrix only supports float types
7225 if (type == NUMBERTYPE_FLOAT32)
7227 createMatrixCompositeCases(testCases, rnd, type);
7231 const string getAssemblyTypeDeclaration (const NumberType type)
7235 case NUMBERTYPE_INT32: return "OpTypeInt 32 1";
7236 case NUMBERTYPE_UINT32: return "OpTypeInt 32 0";
7237 case NUMBERTYPE_FLOAT32: return "OpTypeFloat 32";
7238 default: DE_ASSERT(false); return "";
7242 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
7244 map<string, string> parameters(params);
7246 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7248 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
7250 return StringTemplate (
7251 "OpCapability Shader\n"
7252 "OpCapability Matrix\n"
7253 "OpMemoryModel Logical GLSL450\n"
7254 "OpEntryPoint GLCompute %main \"main\" %id\n"
7255 "OpExecutionMode %main LocalSize 1 1 1\n"
7257 "OpSource GLSL 430\n"
7258 "OpName %main \"main\"\n"
7259 "OpName %id \"gl_GlobalInvocationID\"\n"
7262 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7263 "OpDecorate %buf BufferBlock\n"
7264 "OpDecorate %indata DescriptorSet 0\n"
7265 "OpDecorate %indata Binding 0\n"
7266 "OpDecorate %outdata DescriptorSet 0\n"
7267 "OpDecorate %outdata Binding 1\n"
7268 "OpDecorate %customarr ArrayStride 4\n"
7269 "${compositeDecorator}"
7270 "OpMemberDecorate %buf 0 Offset 0\n"
7273 "%void = OpTypeVoid\n"
7274 "%voidf = OpTypeFunction %void\n"
7275 "%u32 = OpTypeInt 32 0\n"
7276 "%i32 = OpTypeInt 32 1\n"
7277 "%uvec3 = OpTypeVector %u32 3\n"
7278 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7281 "%custom = ${typeDeclaration}\n"
7287 // Inherited from custom
7288 "%customptr = OpTypePointer Uniform %custom\n"
7289 "%customarr = OpTypeRuntimeArray %custom\n"
7290 "%buf = OpTypeStruct %customarr\n"
7291 "%bufptr = OpTypePointer Uniform %buf\n"
7293 "%indata = OpVariable %bufptr Uniform\n"
7294 "%outdata = OpVariable %bufptr Uniform\n"
7296 "%id = OpVariable %uvec3ptr Input\n"
7297 "%zero = OpConstant %i32 0\n"
7299 "%main = OpFunction %void None %voidf\n"
7300 "%label = OpLabel\n"
7301 "%idval = OpLoad %uvec3 %id\n"
7302 "%x = OpCompositeExtract %u32 %idval 0\n"
7304 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
7305 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
7306 // Read the input value
7307 "%inval = OpLoad %custom %inloc\n"
7308 // Create the composite and fill it
7309 "${compositeConstruct}"
7310 // Insert the input value to a place
7311 "%instance2 = OpCompositeInsert %composite %inval %instance ${indexes}\n"
7312 // Read back the value from the position
7313 "%out_val = OpCompositeExtract %custom %instance2 ${indexes}\n"
7314 // Store it in the output position
7315 " OpStore %outloc %out_val\n"
7318 ).specialize(parameters);
7321 template<typename T>
7322 BufferSp createCompositeBuffer(T number)
7324 return BufferSp(new Buffer<T>(vector<T>(1, number)));
7327 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
7329 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
7330 de::Random rnd (deStringHash(group->getName()));
7332 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7334 NumberType numberType = NumberType(type);
7335 const string typeName = getNumberTypeName(numberType);
7336 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
7337 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7338 vector<map<string, string> > testCases;
7340 createCompositeCases(testCases, rnd, numberType);
7342 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7344 ComputeShaderSpec spec;
7346 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
7350 case NUMBERTYPE_INT32:
7352 deInt32 number = getInt(rnd);
7353 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7354 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7357 case NUMBERTYPE_UINT32:
7359 deUint32 number = rnd.getUint32();
7360 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7361 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7364 case NUMBERTYPE_FLOAT32:
7366 float number = rnd.getFloat();
7367 spec.inputs.push_back(createCompositeBuffer<float>(number));
7368 spec.outputs.push_back(createCompositeBuffer<float>(number));
7375 spec.numWorkGroups = IVec3(1, 1, 1);
7376 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
7378 group->addChild(subGroup.release());
7380 return group.release();
7383 struct AssemblyStructInfo
7385 AssemblyStructInfo (const deUint32 comp, const deUint32 idx)
7390 deUint32 components;
7394 const string specializeInBoundsShaderTemplate (const NumberType type, const AssemblyStructInfo& structInfo, const map<string, string>& params)
7396 // Create the full index string
7397 string fullIndex = numberToString(structInfo.index) + " " + params.at("indexes");
7398 // Convert it to list of indexes
7399 vector<string> indexes = de::splitString(fullIndex, ' ');
7401 map<string, string> parameters (params);
7402 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7403 parameters["structType"] = repeatString(" %composite", structInfo.components);
7404 parameters["structConstruct"] = repeatString(" %instance", structInfo.components);
7405 parameters["insertIndexes"] = fullIndex;
7407 // In matrix cases the last two index is the CompositeExtract indexes
7408 const deUint32 extractIndexes = (parameters["type"] == "matrix") ? 2 : 1;
7410 // Construct the extractIndex
7411 for (vector<string>::const_iterator index = indexes.end() - extractIndexes; index != indexes.end(); ++index)
7413 parameters["extractIndexes"] += " " + *index;
7416 // Remove the last 1 or 2 element depends on matrix case or not
7417 indexes.erase(indexes.end() - extractIndexes, indexes.end());
7420 // Generate AccessChain index expressions (except for the last one, because we use ptr to the composite)
7421 for (vector<string>::const_iterator index = indexes.begin(); index != indexes.end(); ++index)
7423 string indexId = "%index_" + numberToString(id++);
7424 parameters["accessChainConstDeclaration"] += indexId + " = OpConstant %u32 " + *index + "\n";
7425 parameters["accessChainIndexes"] += " " + indexId;
7428 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
7430 return StringTemplate (
7431 "OpCapability Shader\n"
7432 "OpCapability Matrix\n"
7433 "OpMemoryModel Logical GLSL450\n"
7434 "OpEntryPoint GLCompute %main \"main\" %id\n"
7435 "OpExecutionMode %main LocalSize 1 1 1\n"
7437 "OpSource GLSL 430\n"
7438 "OpName %main \"main\"\n"
7439 "OpName %id \"gl_GlobalInvocationID\"\n"
7441 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7442 "OpDecorate %buf BufferBlock\n"
7443 "OpDecorate %indata DescriptorSet 0\n"
7444 "OpDecorate %indata Binding 0\n"
7445 "OpDecorate %outdata DescriptorSet 0\n"
7446 "OpDecorate %outdata Binding 1\n"
7447 "OpDecorate %customarr ArrayStride 4\n"
7448 "${compositeDecorator}"
7449 "OpMemberDecorate %buf 0 Offset 0\n"
7451 "%void = OpTypeVoid\n"
7452 "%voidf = OpTypeFunction %void\n"
7453 "%u32 = OpTypeInt 32 0\n"
7454 "%uvec3 = OpTypeVector %u32 3\n"
7455 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7457 "%custom = ${typeDeclaration}\n"
7460 // Inherited from composite
7461 "%composite_p = OpTypePointer Function %composite\n"
7462 "%struct_t = OpTypeStruct${structType}\n"
7463 "%struct_p = OpTypePointer Function %struct_t\n"
7466 "${accessChainConstDeclaration}"
7467 // Inherited from custom
7468 "%customptr = OpTypePointer Uniform %custom\n"
7469 "%customarr = OpTypeRuntimeArray %custom\n"
7470 "%buf = OpTypeStruct %customarr\n"
7471 "%bufptr = OpTypePointer Uniform %buf\n"
7472 "%indata = OpVariable %bufptr Uniform\n"
7473 "%outdata = OpVariable %bufptr Uniform\n"
7475 "%id = OpVariable %uvec3ptr Input\n"
7476 "%zero = OpConstant %u32 0\n"
7477 "%main = OpFunction %void None %voidf\n"
7478 "%label = OpLabel\n"
7479 "%struct_v = OpVariable %struct_p Function\n"
7480 "%idval = OpLoad %uvec3 %id\n"
7481 "%x = OpCompositeExtract %u32 %idval 0\n"
7482 // Create the input/output type
7483 "%inloc = OpInBoundsAccessChain %customptr %indata %zero %x\n"
7484 "%outloc = OpInBoundsAccessChain %customptr %outdata %zero %x\n"
7485 // Read the input value
7486 "%inval = OpLoad %custom %inloc\n"
7487 // Create the composite and fill it
7488 "${compositeConstruct}"
7489 // Create the struct and fill it with the composite
7490 "%struct = OpCompositeConstruct %struct_t${structConstruct}\n"
7492 "%comp_obj = OpCompositeInsert %struct_t %inval %struct ${insertIndexes}\n"
7494 " OpStore %struct_v %comp_obj\n"
7495 // Get deepest possible composite pointer
7496 "%inner_ptr = OpInBoundsAccessChain %composite_p %struct_v${accessChainIndexes}\n"
7497 "%read_obj = OpLoad %composite %inner_ptr\n"
7498 // Read back the stored value
7499 "%read_val = OpCompositeExtract %custom %read_obj${extractIndexes}\n"
7500 " OpStore %outloc %read_val\n"
7502 " OpFunctionEnd\n").specialize(parameters);
7505 tcu::TestCaseGroup* createOpInBoundsAccessChainGroup (tcu::TestContext& testCtx)
7507 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opinboundsaccesschain", "Test the OpInBoundsAccessChain instruction"));
7508 de::Random rnd (deStringHash(group->getName()));
7510 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7512 NumberType numberType = NumberType(type);
7513 const string typeName = getNumberTypeName(numberType);
7514 const string description = "Test the OpInBoundsAccessChain instruction with " + typeName + "s";
7515 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7517 vector<map<string, string> > testCases;
7518 createCompositeCases(testCases, rnd, numberType);
7520 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7522 ComputeShaderSpec spec;
7524 // Number of components inside of a struct
7525 deUint32 structComponents = rnd.getInt(2, 8);
7526 // Component index value
7527 deUint32 structIndex = rnd.getInt(0, structComponents - 1);
7528 AssemblyStructInfo structInfo(structComponents, structIndex);
7530 spec.assembly = specializeInBoundsShaderTemplate(numberType, structInfo, *test);
7534 case NUMBERTYPE_INT32:
7536 deInt32 number = getInt(rnd);
7537 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7538 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7541 case NUMBERTYPE_UINT32:
7543 deUint32 number = rnd.getUint32();
7544 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7545 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7548 case NUMBERTYPE_FLOAT32:
7550 float number = rnd.getFloat();
7551 spec.inputs.push_back(createCompositeBuffer<float>(number));
7552 spec.outputs.push_back(createCompositeBuffer<float>(number));
7558 spec.numWorkGroups = IVec3(1, 1, 1);
7559 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpInBoundsAccessChain test", spec));
7561 group->addChild(subGroup.release());
7563 return group.release();
7566 // If the params missing, uninitialized case
7567 const string specializeDefaultOutputShaderTemplate (const NumberType type, const map<string, string>& params = map<string, string>())
7569 map<string, string> parameters(params);
7571 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7573 // Declare the const value, and use it in the initializer
7574 if (params.find("constValue") != params.end())
7576 parameters["constDeclaration"] = "%const = OpConstant %in_type " + params.at("constValue") + "\n";
7577 parameters["variableInitializer"] = "%const";
7579 // Uninitialized case
7582 parameters["constDeclaration"] = "";
7583 parameters["variableInitializer"] = "";
7586 return StringTemplate(
7587 "OpCapability Shader\n"
7588 "OpMemoryModel Logical GLSL450\n"
7589 "OpEntryPoint GLCompute %main \"main\" %id\n"
7590 "OpExecutionMode %main LocalSize 1 1 1\n"
7591 "OpSource GLSL 430\n"
7592 "OpName %main \"main\"\n"
7593 "OpName %id \"gl_GlobalInvocationID\"\n"
7595 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7596 "OpDecorate %indata DescriptorSet 0\n"
7597 "OpDecorate %indata Binding 0\n"
7598 "OpDecorate %outdata DescriptorSet 0\n"
7599 "OpDecorate %outdata Binding 1\n"
7600 "OpDecorate %in_arr ArrayStride 4\n"
7601 "OpDecorate %in_buf BufferBlock\n"
7602 "OpMemberDecorate %in_buf 0 Offset 0\n"
7604 "%void = OpTypeVoid\n"
7605 "%voidf = OpTypeFunction %void\n"
7606 "%u32 = OpTypeInt 32 0\n"
7607 "%i32 = OpTypeInt 32 1\n"
7608 "%uvec3 = OpTypeVector %u32 3\n"
7609 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7611 "%in_type = ${typeDeclaration}\n"
7612 // "%const = OpConstant %in_type ${constValue}\n"
7613 "${constDeclaration}\n"
7615 "%in_ptr = OpTypePointer Uniform %in_type\n"
7616 "%in_arr = OpTypeRuntimeArray %in_type\n"
7617 "%in_buf = OpTypeStruct %in_arr\n"
7618 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
7619 "%indata = OpVariable %in_bufptr Uniform\n"
7620 "%outdata = OpVariable %in_bufptr Uniform\n"
7621 "%id = OpVariable %uvec3ptr Input\n"
7622 "%var_ptr = OpTypePointer Function %in_type\n"
7624 "%zero = OpConstant %i32 0\n"
7626 "%main = OpFunction %void None %voidf\n"
7627 "%label = OpLabel\n"
7628 "%out_var = OpVariable %var_ptr Function ${variableInitializer}\n"
7629 "%idval = OpLoad %uvec3 %id\n"
7630 "%x = OpCompositeExtract %u32 %idval 0\n"
7631 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
7632 "%outloc = OpAccessChain %in_ptr %outdata %zero %x\n"
7634 "%outval = OpLoad %in_type %out_var\n"
7635 " OpStore %outloc %outval\n"
7638 ).specialize(parameters);
7641 bool compareFloats (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
7643 DE_ASSERT(outputAllocs.size() != 0);
7644 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
7646 // Use custom epsilon because of the float->string conversion
7647 const float epsilon = 0.00001f;
7649 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
7652 memcpy(&expected, expectedOutputs[outputNdx]->data(), expectedOutputs[outputNdx]->getNumBytes());
7655 memcpy(&actual, outputAllocs[outputNdx]->getHostPtr(), expectedOutputs[outputNdx]->getNumBytes());
7657 // Test with epsilon
7658 if (fabs(expected - actual) > epsilon)
7660 log << TestLog::Message << "Error: The actual and expected values not matching."
7661 << " Expected: " << expected << " Actual: " << actual << " Epsilon: " << epsilon << TestLog::EndMessage;
7668 // Checks if the driver crash with uninitialized cases
7669 bool passthruVerify (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
7671 DE_ASSERT(outputAllocs.size() != 0);
7672 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
7674 // Copy and discard the result.
7675 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
7677 size_t width = expectedOutputs[outputNdx]->getNumBytes();
7679 vector<char> data(width);
7680 memcpy(&data[0], outputAllocs[outputNdx]->getHostPtr(), width);
7685 tcu::TestCaseGroup* createShaderDefaultOutputGroup (tcu::TestContext& testCtx)
7687 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "shader_default_output", "Test shader default output."));
7688 de::Random rnd (deStringHash(group->getName()));
7690 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7692 NumberType numberType = NumberType(type);
7693 const string typeName = getNumberTypeName(numberType);
7694 const string description = "Test the OpVariable initializer with " + typeName + ".";
7695 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7697 // 2 similar subcases (initialized and uninitialized)
7698 for (int subCase = 0; subCase < 2; ++subCase)
7700 ComputeShaderSpec spec;
7701 spec.numWorkGroups = IVec3(1, 1, 1);
7703 map<string, string> params;
7707 case NUMBERTYPE_INT32:
7709 deInt32 number = getInt(rnd);
7710 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7711 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7712 params["constValue"] = numberToString(number);
7715 case NUMBERTYPE_UINT32:
7717 deUint32 number = rnd.getUint32();
7718 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7719 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7720 params["constValue"] = numberToString(number);
7723 case NUMBERTYPE_FLOAT32:
7725 float number = rnd.getFloat();
7726 spec.inputs.push_back(createCompositeBuffer<float>(number));
7727 spec.outputs.push_back(createCompositeBuffer<float>(number));
7728 spec.verifyIO = &compareFloats;
7729 params["constValue"] = numberToString(number);
7736 // Initialized subcase
7739 spec.assembly = specializeDefaultOutputShaderTemplate(numberType, params);
7740 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "initialized", "OpVariable initializer tests.", spec));
7742 // Uninitialized subcase
7745 spec.assembly = specializeDefaultOutputShaderTemplate(numberType);
7746 spec.verifyIO = &passthruVerify;
7747 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "uninitialized", "OpVariable initializer tests.", spec));
7750 group->addChild(subGroup.release());
7752 return group.release();
7755 tcu::TestCaseGroup* createOpNopTests (tcu::TestContext& testCtx)
7757 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
7758 RGBA defaultColors[4];
7759 map<string, string> opNopFragments;
7761 getDefaultColors(defaultColors);
7763 opNopFragments["testfun"] =
7764 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7765 "%param1 = OpFunctionParameter %v4f32\n"
7766 "%label_testfun = OpLabel\n"
7775 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7776 "%b = OpFAdd %f32 %a %a\n"
7778 "%c = OpFSub %f32 %b %a\n"
7779 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
7782 "OpReturnValue %ret\n"
7785 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, testGroup.get());
7787 return testGroup.release();
7790 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
7792 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
7793 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
7794 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
7796 computeTests->addChild(createOpNopGroup(testCtx));
7797 computeTests->addChild(createOpFUnordGroup(testCtx));
7798 computeTests->addChild(createOpAtomicGroup(testCtx, false));
7799 computeTests->addChild(createOpAtomicGroup(testCtx, true)); // Using new StorageBuffer decoration
7800 computeTests->addChild(createOpLineGroup(testCtx));
7801 computeTests->addChild(createOpNoLineGroup(testCtx));
7802 computeTests->addChild(createOpConstantNullGroup(testCtx));
7803 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
7804 computeTests->addChild(createOpConstantUsageGroup(testCtx));
7805 computeTests->addChild(createSpecConstantGroup(testCtx));
7806 computeTests->addChild(createOpSourceGroup(testCtx));
7807 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
7808 computeTests->addChild(createDecorationGroupGroup(testCtx));
7809 computeTests->addChild(createOpPhiGroup(testCtx));
7810 computeTests->addChild(createLoopControlGroup(testCtx));
7811 computeTests->addChild(createFunctionControlGroup(testCtx));
7812 computeTests->addChild(createSelectionControlGroup(testCtx));
7813 computeTests->addChild(createBlockOrderGroup(testCtx));
7814 computeTests->addChild(createMultipleShaderGroup(testCtx));
7815 computeTests->addChild(createMemoryAccessGroup(testCtx));
7816 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
7817 computeTests->addChild(createOpCopyObjectGroup(testCtx));
7818 computeTests->addChild(createNoContractionGroup(testCtx));
7819 computeTests->addChild(createOpUndefGroup(testCtx));
7820 computeTests->addChild(createOpUnreachableGroup(testCtx));
7821 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
7822 computeTests ->addChild(createOpFRemGroup(testCtx));
7823 computeTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_PASS));
7824 computeTests->addChild(createOpSRemComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
7825 computeTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_PASS));
7826 computeTests->addChild(createOpSModComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
7827 computeTests->addChild(createSConvertTests(testCtx));
7828 computeTests->addChild(createUConvertTests(testCtx));
7829 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
7830 computeTests->addChild(createOpInBoundsAccessChainGroup(testCtx));
7831 computeTests->addChild(createShaderDefaultOutputGroup(testCtx));
7832 computeTests->addChild(createOpNMinGroup(testCtx));
7833 computeTests->addChild(createOpNMaxGroup(testCtx));
7834 computeTests->addChild(createOpNClampGroup(testCtx));
7836 de::MovePtr<tcu::TestCaseGroup> computeAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
7838 computeAndroidTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7839 computeAndroidTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7841 computeTests->addChild(computeAndroidTests.release());
7844 computeTests->addChild(create16BitStorageComputeGroup(testCtx));
7845 computeTests->addChild(createUboMatrixPaddingComputeGroup(testCtx));
7846 computeTests->addChild(createVariablePointersComputeGroup(testCtx));
7847 graphicsTests->addChild(createOpNopTests(testCtx));
7848 graphicsTests->addChild(createOpSourceTests(testCtx));
7849 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
7850 graphicsTests->addChild(createOpLineTests(testCtx));
7851 graphicsTests->addChild(createOpNoLineTests(testCtx));
7852 graphicsTests->addChild(createOpConstantNullTests(testCtx));
7853 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
7854 graphicsTests->addChild(createMemoryAccessTests(testCtx));
7855 graphicsTests->addChild(createOpUndefTests(testCtx));
7856 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
7857 graphicsTests->addChild(createModuleTests(testCtx));
7858 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
7859 graphicsTests->addChild(createOpPhiTests(testCtx));
7860 graphicsTests->addChild(createNoContractionTests(testCtx));
7861 graphicsTests->addChild(createOpQuantizeTests(testCtx));
7862 graphicsTests->addChild(createLoopTests(testCtx));
7863 graphicsTests->addChild(createSpecConstantTests(testCtx));
7864 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
7865 graphicsTests->addChild(createBarrierTests(testCtx));
7866 graphicsTests->addChild(createDecorationGroupTests(testCtx));
7867 graphicsTests->addChild(createFRemTests(testCtx));
7868 graphicsTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
7869 graphicsTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
7872 de::MovePtr<tcu::TestCaseGroup> graphicsAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
7874 graphicsAndroidTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7875 graphicsAndroidTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7877 graphicsTests->addChild(graphicsAndroidTests.release());
7880 graphicsTests->addChild(create16BitStorageGraphicsGroup(testCtx));
7881 graphicsTests->addChild(createUboMatrixPaddingGraphicsGroup(testCtx));
7882 graphicsTests->addChild(createVariablePointersGraphicsGroup(testCtx));
7884 instructionTests->addChild(computeTests.release());
7885 instructionTests->addChild(graphicsTests.release());
7887 return instructionTests.release();