1 /*-------------------------------------------------------------------------
2 * Vulkan Conformance Tests
3 * ------------------------
5 * Copyright (c) 2015 Google Inc.
6 * Copyright (c) 2016 The Khronos Group Inc.
8 * Licensed under the Apache License, Version 2.0 (the "License");
9 * you may not use this file except in compliance with the License.
10 * You may obtain a copy of the License at
12 * http://www.apache.org/licenses/LICENSE-2.0
14 * Unless required by applicable law or agreed to in writing, software
15 * distributed under the License is distributed on an "AS IS" BASIS,
16 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17 * See the License for the specific language governing permissions and
18 * limitations under the License.
22 * \brief SPIR-V Assembly Tests for Instructions (special opcode/operand)
23 *//*--------------------------------------------------------------------*/
25 #include "vktSpvAsmInstructionTests.hpp"
27 #include "tcuCommandLine.hpp"
28 #include "tcuFormatUtil.hpp"
29 #include "tcuFloat.hpp"
30 #include "tcuRGBA.hpp"
31 #include "tcuStringTemplate.hpp"
32 #include "tcuTestLog.hpp"
33 #include "tcuVectorUtil.hpp"
34 #include "tcuInterval.hpp"
37 #include "vkDeviceUtil.hpp"
38 #include "vkMemUtil.hpp"
39 #include "vkPlatform.hpp"
40 #include "vkPrograms.hpp"
41 #include "vkQueryUtil.hpp"
43 #include "vkRefUtil.hpp"
44 #include "vkStrUtil.hpp"
45 #include "vkTypeUtil.hpp"
47 #include "deRandom.hpp"
48 #include "deStringUtil.hpp"
49 #include "deUniquePtr.hpp"
51 #include "tcuStringTemplate.hpp"
53 #include "vktSpvAsm16bitStorageTests.hpp"
54 #include "vktSpvAsmUboMatrixPaddingTests.hpp"
55 #include "vktSpvAsmConditionalBranchTests.hpp"
56 #include "vktSpvAsmIndexingTests.hpp"
57 #include "vktSpvAsmComputeShaderCase.hpp"
58 #include "vktSpvAsmComputeShaderTestUtil.hpp"
59 #include "vktSpvAsmGraphicsShaderTestUtil.hpp"
60 #include "vktSpvAsmVariablePointersTests.hpp"
61 #include "vktTestCaseUtil.hpp"
72 namespace SpirVAssembly
86 using tcu::TestStatus;
89 using tcu::StringTemplate;
93 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0)
95 T* const typedPtr = (T*)dst;
96 for (int ndx = 0; ndx < numValues; ndx++)
97 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue);
100 // Filter is a function that returns true if a value should pass, false otherwise.
101 template<typename T, typename FilterT>
102 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, FilterT filter, int offset = 0)
104 T* const typedPtr = (T*)dst;
106 for (int ndx = 0; ndx < numValues; ndx++)
109 value = randomScalar<T>(rnd, minValue, maxValue);
110 while (!filter(value));
112 typedPtr[offset + ndx] = value;
116 // Gets a 64-bit integer with a more logarithmic distribution
117 deInt64 randomInt64LogDistributed (de::Random& rnd)
119 deInt64 val = rnd.getUint64();
120 val &= (1ull << rnd.getInt(1, 63)) - 1;
126 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues)
128 for (int ndx = 0; ndx < numValues; ndx++)
129 dst[ndx] = randomInt64LogDistributed(rnd);
132 template<typename FilterT>
133 static void fillRandomInt64sLogDistributed (de::Random& rnd, vector<deInt64>& dst, int numValues, FilterT filter)
135 for (int ndx = 0; ndx < numValues; ndx++)
139 value = randomInt64LogDistributed(rnd);
140 } while (!filter(value));
145 inline bool filterNonNegative (const deInt64 value)
150 inline bool filterPositive (const deInt64 value)
155 inline bool filterNotZero (const deInt64 value)
160 static void floorAll (vector<float>& values)
162 for (size_t i = 0; i < values.size(); i++)
163 values[i] = deFloatFloor(values[i]);
166 static void floorAll (vector<Vec4>& values)
168 for (size_t i = 0; i < values.size(); i++)
169 values[i] = floor(values[i]);
177 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {}
180 // Assembly code used for testing OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code:
184 // layout(std140, set = 0, binding = 0) readonly buffer Input {
187 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
191 // layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
194 // uint x = gl_GlobalInvocationID.x;
195 // output_data.elements[x] = -input_data.elements[x];
198 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
200 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
201 ComputeShaderSpec spec;
202 de::Random rnd (deStringHash(group->getName()));
203 const int numElements = 100;
204 vector<float> positiveFloats (numElements, 0);
205 vector<float> negativeFloats (numElements, 0);
207 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
209 for (size_t ndx = 0; ndx < numElements; ++ndx)
210 negativeFloats[ndx] = -positiveFloats[ndx];
213 string(getComputeAsmShaderPreamble()) +
215 "OpSource GLSL 430\n"
216 "OpName %main \"main\"\n"
217 "OpName %id \"gl_GlobalInvocationID\"\n"
219 "OpDecorate %id BuiltIn GlobalInvocationId\n"
221 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes())
223 + string(getComputeAsmInputOutputBuffer()) +
225 "%id = OpVariable %uvec3ptr Input\n"
226 "%zero = OpConstant %i32 0\n"
228 "%main = OpFunction %void None %voidf\n"
230 "%idval = OpLoad %uvec3 %id\n"
231 "%x = OpCompositeExtract %u32 %idval 0\n"
233 " OpNop\n" // Inside a function body
235 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
236 "%inval = OpLoad %f32 %inloc\n"
237 "%neg = OpFNegate %f32 %inval\n"
238 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
239 " OpStore %outloc %neg\n"
242 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
243 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
244 spec.numWorkGroups = IVec3(numElements, 1, 1);
246 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
248 return group.release();
251 bool compareFUnord (const std::vector<BufferSp>& inputs, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
253 if (outputAllocs.size() != 1)
256 const BufferSp& expectedOutput = expectedOutputs[0];
257 const deInt32* expectedOutputAsInt = static_cast<const deInt32*>(expectedOutputs[0]->data());
258 const deInt32* outputAsInt = static_cast<const deInt32*>(outputAllocs[0]->getHostPtr());
259 const float* input1AsFloat = static_cast<const float*>(inputs[0]->data());
260 const float* input2AsFloat = static_cast<const float*>(inputs[1]->data());
261 bool returnValue = true;
263 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(deInt32); ++idx)
265 if (outputAsInt[idx] != expectedOutputAsInt[idx])
267 log << TestLog::Message << "ERROR: Sub-case failed. inputs: " << input1AsFloat[idx] << "," << input2AsFloat[idx] << " output: " << outputAsInt[idx]<< " expected output: " << expectedOutputAsInt[idx] << TestLog::EndMessage;
274 typedef VkBool32 (*compareFuncType) (float, float);
280 compareFuncType compareFunc;
282 OpFUnordCase (const char* _name, const char* _opCode, compareFuncType _compareFunc)
285 , compareFunc (_compareFunc) {}
288 #define ADD_OPFUNORD_CASE(NAME, OPCODE, OPERATOR) \
290 struct compare_##NAME { static VkBool32 compare(float x, float y) { return (x OPERATOR y) ? VK_TRUE : VK_FALSE; } }; \
291 cases.push_back(OpFUnordCase(#NAME, OPCODE, compare_##NAME::compare)); \
292 } while (deGetFalse())
294 tcu::TestCaseGroup* createOpFUnordGroup (tcu::TestContext& testCtx)
296 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfunord", "Test the OpFUnord* opcodes"));
297 de::Random rnd (deStringHash(group->getName()));
298 const int numElements = 100;
299 vector<OpFUnordCase> cases;
301 const StringTemplate shaderTemplate (
303 string(getComputeAsmShaderPreamble()) +
305 "OpSource GLSL 430\n"
306 "OpName %main \"main\"\n"
307 "OpName %id \"gl_GlobalInvocationID\"\n"
309 "OpDecorate %id BuiltIn GlobalInvocationId\n"
311 "OpDecorate %buf BufferBlock\n"
312 "OpDecorate %buf2 BufferBlock\n"
313 "OpDecorate %indata1 DescriptorSet 0\n"
314 "OpDecorate %indata1 Binding 0\n"
315 "OpDecorate %indata2 DescriptorSet 0\n"
316 "OpDecorate %indata2 Binding 1\n"
317 "OpDecorate %outdata DescriptorSet 0\n"
318 "OpDecorate %outdata Binding 2\n"
319 "OpDecorate %f32arr ArrayStride 4\n"
320 "OpDecorate %i32arr ArrayStride 4\n"
321 "OpMemberDecorate %buf 0 Offset 0\n"
322 "OpMemberDecorate %buf2 0 Offset 0\n"
324 + string(getComputeAsmCommonTypes()) +
326 "%buf = OpTypeStruct %f32arr\n"
327 "%bufptr = OpTypePointer Uniform %buf\n"
328 "%indata1 = OpVariable %bufptr Uniform\n"
329 "%indata2 = OpVariable %bufptr Uniform\n"
331 "%buf2 = OpTypeStruct %i32arr\n"
332 "%buf2ptr = OpTypePointer Uniform %buf2\n"
333 "%outdata = OpVariable %buf2ptr Uniform\n"
335 "%id = OpVariable %uvec3ptr Input\n"
336 "%zero = OpConstant %i32 0\n"
337 "%consti1 = OpConstant %i32 1\n"
338 "%constf1 = OpConstant %f32 1.0\n"
340 "%main = OpFunction %void None %voidf\n"
342 "%idval = OpLoad %uvec3 %id\n"
343 "%x = OpCompositeExtract %u32 %idval 0\n"
345 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
346 "%inval1 = OpLoad %f32 %inloc1\n"
347 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
348 "%inval2 = OpLoad %f32 %inloc2\n"
349 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
351 "%result = ${OPCODE} %bool %inval1 %inval2\n"
352 "%int_res = OpSelect %i32 %result %consti1 %zero\n"
353 " OpStore %outloc %int_res\n"
358 ADD_OPFUNORD_CASE(equal, "OpFUnordEqual", ==);
359 ADD_OPFUNORD_CASE(less, "OpFUnordLessThan", <);
360 ADD_OPFUNORD_CASE(lessequal, "OpFUnordLessThanEqual", <=);
361 ADD_OPFUNORD_CASE(greater, "OpFUnordGreaterThan", >);
362 ADD_OPFUNORD_CASE(greaterequal, "OpFUnordGreaterThanEqual", >=);
363 ADD_OPFUNORD_CASE(notequal, "OpFUnordNotEqual", !=);
365 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
367 map<string, string> specializations;
368 ComputeShaderSpec spec;
369 const float NaN = std::numeric_limits<float>::quiet_NaN();
370 vector<float> inputFloats1 (numElements, 0);
371 vector<float> inputFloats2 (numElements, 0);
372 vector<deInt32> expectedInts (numElements, 0);
374 specializations["OPCODE"] = cases[caseNdx].opCode;
375 spec.assembly = shaderTemplate.specialize(specializations);
377 fillRandomScalars(rnd, 1.f, 100.f, &inputFloats1[0], numElements);
378 for (size_t ndx = 0; ndx < numElements; ++ndx)
382 case 0: inputFloats2[ndx] = inputFloats1[ndx] + 1.0f; break;
383 case 1: inputFloats2[ndx] = inputFloats1[ndx] - 1.0f; break;
384 case 2: inputFloats2[ndx] = inputFloats1[ndx]; break;
385 case 3: inputFloats2[ndx] = NaN; break;
386 case 4: inputFloats2[ndx] = inputFloats1[ndx]; inputFloats1[ndx] = NaN; break;
387 case 5: inputFloats2[ndx] = NaN; inputFloats1[ndx] = NaN; break;
389 expectedInts[ndx] = tcu::Float32(inputFloats1[ndx]).isNaN() || tcu::Float32(inputFloats2[ndx]).isNaN() || cases[caseNdx].compareFunc(inputFloats1[ndx], inputFloats2[ndx]);
392 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
393 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
394 spec.outputs.push_back(BufferSp(new Int32Buffer(expectedInts)));
395 spec.numWorkGroups = IVec3(numElements, 1, 1);
396 spec.verifyIO = &compareFUnord;
397 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
400 return group.release();
406 const char* assembly;
407 void (*calculateExpected)(deInt32&, deInt32);
408 deInt32 numOutputElements;
410 OpAtomicCase (const char* _name, const char* _assembly, void (*_calculateExpected)(deInt32&, deInt32), deInt32 _numOutputElements)
412 , assembly (_assembly)
413 , calculateExpected (_calculateExpected)
414 , numOutputElements (_numOutputElements) {}
417 tcu::TestCaseGroup* createOpAtomicGroup (tcu::TestContext& testCtx, bool useStorageBuffer)
419 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx,
420 useStorageBuffer ? "opatomic_storage_buffer" : "opatomic",
421 "Test the OpAtomic* opcodes"));
422 de::Random rnd (deStringHash(group->getName()));
423 const int numElements = 65535;
424 vector<OpAtomicCase> cases;
426 const StringTemplate shaderTemplate (
428 string("OpCapability Shader\n") +
429 (useStorageBuffer ? "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" : "") +
430 "OpMemoryModel Logical GLSL450\n"
431 "OpEntryPoint GLCompute %main \"main\" %id\n"
432 "OpExecutionMode %main LocalSize 1 1 1\n" +
434 "OpSource GLSL 430\n"
435 "OpName %main \"main\"\n"
436 "OpName %id \"gl_GlobalInvocationID\"\n"
438 "OpDecorate %id BuiltIn GlobalInvocationId\n"
440 "OpDecorate %buf ${BLOCK_DECORATION}\n"
441 "OpDecorate %indata DescriptorSet 0\n"
442 "OpDecorate %indata Binding 0\n"
443 "OpDecorate %i32arr ArrayStride 4\n"
444 "OpMemberDecorate %buf 0 Offset 0\n"
446 "OpDecorate %sumbuf ${BLOCK_DECORATION}\n"
447 "OpDecorate %sum DescriptorSet 0\n"
448 "OpDecorate %sum Binding 1\n"
449 "OpMemberDecorate %sumbuf 0 Coherent\n"
450 "OpMemberDecorate %sumbuf 0 Offset 0\n"
452 + getComputeAsmCommonTypes("${BLOCK_POINTER_TYPE}") +
454 "%buf = OpTypeStruct %i32arr\n"
455 "%bufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %buf\n"
456 "%indata = OpVariable %bufptr ${BLOCK_POINTER_TYPE}\n"
458 "%sumbuf = OpTypeStruct %i32arr\n"
459 "%sumbufptr = OpTypePointer ${BLOCK_POINTER_TYPE} %sumbuf\n"
460 "%sum = OpVariable %sumbufptr ${BLOCK_POINTER_TYPE}\n"
462 "%id = OpVariable %uvec3ptr Input\n"
463 "%minusone = OpConstant %i32 -1\n"
464 "%zero = OpConstant %i32 0\n"
465 "%one = OpConstant %u32 1\n"
466 "%two = OpConstant %i32 2\n"
468 "%main = OpFunction %void None %voidf\n"
470 "%idval = OpLoad %uvec3 %id\n"
471 "%x = OpCompositeExtract %u32 %idval 0\n"
473 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
474 "%inval = OpLoad %i32 %inloc\n"
476 "%outloc = OpAccessChain %i32ptr %sum %zero ${INDEX}\n"
482 #define ADD_OPATOMIC_CASE(NAME, ASSEMBLY, CALCULATE_EXPECTED, NUM_OUTPUT_ELEMENTS) \
484 DE_STATIC_ASSERT((NUM_OUTPUT_ELEMENTS) == 1 || (NUM_OUTPUT_ELEMENTS) == numElements); \
485 struct calculateExpected_##NAME { static void calculateExpected(deInt32& expected, deInt32 input) CALCULATE_EXPECTED }; /* NOLINT(CALCULATE_EXPECTED) */ \
486 cases.push_back(OpAtomicCase(#NAME, ASSEMBLY, calculateExpected_##NAME::calculateExpected, NUM_OUTPUT_ELEMENTS)); \
487 } while (deGetFalse())
488 #define ADD_OPATOMIC_CASE_1(NAME, ASSEMBLY, CALCULATE_EXPECTED) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, CALCULATE_EXPECTED, 1)
489 #define ADD_OPATOMIC_CASE_N(NAME, ASSEMBLY, CALCULATE_EXPECTED) ADD_OPATOMIC_CASE(NAME, ASSEMBLY, CALCULATE_EXPECTED, numElements)
491 ADD_OPATOMIC_CASE_1(iadd, "%unused = OpAtomicIAdd %i32 %outloc %one %zero %inval\n", { expected += input; } );
492 ADD_OPATOMIC_CASE_1(isub, "%unused = OpAtomicISub %i32 %outloc %one %zero %inval\n", { expected -= input; } );
493 ADD_OPATOMIC_CASE_1(iinc, "%unused = OpAtomicIIncrement %i32 %outloc %one %zero\n", { ++expected; (void)input;} );
494 ADD_OPATOMIC_CASE_1(idec, "%unused = OpAtomicIDecrement %i32 %outloc %one %zero\n", { --expected; (void)input;} );
495 ADD_OPATOMIC_CASE_N(load, "%inval2 = OpAtomicLoad %i32 %inloc %zero %zero\n"
496 " OpStore %outloc %inval2\n", { expected = input;} );
497 ADD_OPATOMIC_CASE_N(store, " OpAtomicStore %outloc %zero %zero %inval\n", { expected = input;} );
498 ADD_OPATOMIC_CASE_N(compex, "%even = OpSMod %i32 %inval %two\n"
499 " OpStore %outloc %even\n"
500 "%unused = OpAtomicCompareExchange %i32 %outloc %one %zero %zero %minusone %zero\n", { expected = (input % 2) == 0 ? -1 : 1;} );
502 #undef ADD_OPATOMIC_CASE
503 #undef ADD_OPATOMIC_CASE_1
504 #undef ADD_OPATOMIC_CASE_N
506 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
508 map<string, string> specializations;
509 ComputeShaderSpec spec;
510 vector<deInt32> inputInts (numElements, 0);
511 vector<deInt32> expected (cases[caseNdx].numOutputElements, -1);
513 specializations["INDEX"] = (cases[caseNdx].numOutputElements == 1) ? "%zero" : "%x";
514 specializations["INSTRUCTION"] = cases[caseNdx].assembly;
515 specializations["BLOCK_DECORATION"] = useStorageBuffer ? "Block" : "BufferBlock";
516 specializations["BLOCK_POINTER_TYPE"] = useStorageBuffer ? "StorageBuffer" : "Uniform";
517 spec.assembly = shaderTemplate.specialize(specializations);
519 if (useStorageBuffer)
520 spec.extensions.push_back("VK_KHR_storage_buffer_storage_class");
522 fillRandomScalars(rnd, 1, 100, &inputInts[0], numElements);
523 for (size_t ndx = 0; ndx < numElements; ++ndx)
525 cases[caseNdx].calculateExpected((cases[caseNdx].numOutputElements == 1) ? expected[0] : expected[ndx], inputInts[ndx]);
528 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
529 spec.outputs.push_back(BufferSp(new Int32Buffer(expected)));
530 spec.numWorkGroups = IVec3(numElements, 1, 1);
531 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
534 return group.release();
537 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
539 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
540 ComputeShaderSpec spec;
541 de::Random rnd (deStringHash(group->getName()));
542 const int numElements = 100;
543 vector<float> positiveFloats (numElements, 0);
544 vector<float> negativeFloats (numElements, 0);
546 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
548 for (size_t ndx = 0; ndx < numElements; ++ndx)
549 negativeFloats[ndx] = -positiveFloats[ndx];
552 string(getComputeAsmShaderPreamble()) +
554 "%fname1 = OpString \"negateInputs.comp\"\n"
555 "%fname2 = OpString \"negateInputs\"\n"
557 "OpSource GLSL 430\n"
558 "OpName %main \"main\"\n"
559 "OpName %id \"gl_GlobalInvocationID\"\n"
561 "OpDecorate %id BuiltIn GlobalInvocationId\n"
563 + string(getComputeAsmInputOutputBufferTraits()) +
565 "OpLine %fname1 0 0\n" // At the earliest possible position
567 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
569 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
570 "OpLine %fname2 1 0\n" // Different filenames
571 "OpLine %fname1 1000 100000\n"
573 "%id = OpVariable %uvec3ptr Input\n"
574 "%zero = OpConstant %i32 0\n"
576 "OpLine %fname1 1 1\n" // Before a function
578 "%main = OpFunction %void None %voidf\n"
581 "OpLine %fname1 1 1\n" // In a function
583 "%idval = OpLoad %uvec3 %id\n"
584 "%x = OpCompositeExtract %u32 %idval 0\n"
585 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
586 "%inval = OpLoad %f32 %inloc\n"
587 "%neg = OpFNegate %f32 %inval\n"
588 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
589 " OpStore %outloc %neg\n"
592 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
593 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
594 spec.numWorkGroups = IVec3(numElements, 1, 1);
596 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
598 return group.release();
601 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
603 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
604 ComputeShaderSpec spec;
605 de::Random rnd (deStringHash(group->getName()));
606 const int numElements = 100;
607 vector<float> positiveFloats (numElements, 0);
608 vector<float> negativeFloats (numElements, 0);
610 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
612 for (size_t ndx = 0; ndx < numElements; ++ndx)
613 negativeFloats[ndx] = -positiveFloats[ndx];
616 string(getComputeAsmShaderPreamble()) +
618 "%fname = OpString \"negateInputs.comp\"\n"
620 "OpSource GLSL 430\n"
621 "OpName %main \"main\"\n"
622 "OpName %id \"gl_GlobalInvocationID\"\n"
624 "OpDecorate %id BuiltIn GlobalInvocationId\n"
626 + string(getComputeAsmInputOutputBufferTraits()) +
628 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
630 + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
632 "OpLine %fname 0 1\n"
633 "OpNoLine\n" // Immediately following a preceding OpLine
635 "OpLine %fname 1000 1\n"
637 "%id = OpVariable %uvec3ptr Input\n"
638 "%zero = OpConstant %i32 0\n"
640 "OpNoLine\n" // Contents after the previous OpLine
642 "%main = OpFunction %void None %voidf\n"
644 "%idval = OpLoad %uvec3 %id\n"
645 "%x = OpCompositeExtract %u32 %idval 0\n"
647 "OpNoLine\n" // Multiple OpNoLine
651 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
652 "%inval = OpLoad %f32 %inloc\n"
653 "%neg = OpFNegate %f32 %inval\n"
654 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
655 " OpStore %outloc %neg\n"
658 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
659 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
660 spec.numWorkGroups = IVec3(numElements, 1, 1);
662 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
664 return group.release();
667 // Compare instruction for the contraction compute case.
668 // Returns true if the output is what is expected from the test case.
669 bool compareNoContractCase(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
671 if (outputAllocs.size() != 1)
674 // We really just need this for size because we are not comparing the exact values.
675 const BufferSp& expectedOutput = expectedOutputs[0];
676 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
678 for(size_t i = 0; i < expectedOutput->getNumBytes() / sizeof(float); ++i) {
679 if (outputAsFloat[i] != 0.f &&
680 outputAsFloat[i] != -ldexp(1, -24)) {
688 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
690 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
691 vector<CaseParameter> cases;
692 const int numElements = 100;
693 vector<float> inputFloats1 (numElements, 0);
694 vector<float> inputFloats2 (numElements, 0);
695 vector<float> outputFloats (numElements, 0);
696 const StringTemplate shaderTemplate (
697 string(getComputeAsmShaderPreamble()) +
699 "OpName %main \"main\"\n"
700 "OpName %id \"gl_GlobalInvocationID\"\n"
702 "OpDecorate %id BuiltIn GlobalInvocationId\n"
706 "OpDecorate %buf BufferBlock\n"
707 "OpDecorate %indata1 DescriptorSet 0\n"
708 "OpDecorate %indata1 Binding 0\n"
709 "OpDecorate %indata2 DescriptorSet 0\n"
710 "OpDecorate %indata2 Binding 1\n"
711 "OpDecorate %outdata DescriptorSet 0\n"
712 "OpDecorate %outdata Binding 2\n"
713 "OpDecorate %f32arr ArrayStride 4\n"
714 "OpMemberDecorate %buf 0 Offset 0\n"
716 + string(getComputeAsmCommonTypes()) +
718 "%buf = OpTypeStruct %f32arr\n"
719 "%bufptr = OpTypePointer Uniform %buf\n"
720 "%indata1 = OpVariable %bufptr Uniform\n"
721 "%indata2 = OpVariable %bufptr Uniform\n"
722 "%outdata = OpVariable %bufptr Uniform\n"
724 "%id = OpVariable %uvec3ptr Input\n"
725 "%zero = OpConstant %i32 0\n"
726 "%c_f_m1 = OpConstant %f32 -1.\n"
728 "%main = OpFunction %void None %voidf\n"
730 "%idval = OpLoad %uvec3 %id\n"
731 "%x = OpCompositeExtract %u32 %idval 0\n"
732 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
733 "%inval1 = OpLoad %f32 %inloc1\n"
734 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
735 "%inval2 = OpLoad %f32 %inloc2\n"
736 "%mul = OpFMul %f32 %inval1 %inval2\n"
737 "%add = OpFAdd %f32 %mul %c_f_m1\n"
738 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
739 " OpStore %outloc %add\n"
743 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
744 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
745 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
747 for (size_t ndx = 0; ndx < numElements; ++ndx)
749 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
750 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
751 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
752 // conducted separately and the result is rounded to 1, or 0x1.fffffcp-1
753 // So the final result will be 0.f or 0x1p-24.
754 // If the operation is combined into a precise fused multiply-add, then the result would be
755 // 2^-46 (0xa8800000).
756 outputFloats[ndx] = 0.f;
759 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
761 map<string, string> specializations;
762 ComputeShaderSpec spec;
764 specializations["DECORATION"] = cases[caseNdx].param;
765 spec.assembly = shaderTemplate.specialize(specializations);
766 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
767 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
768 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
769 spec.numWorkGroups = IVec3(numElements, 1, 1);
770 // Check against the two possible answers based on rounding mode.
771 spec.verifyIO = &compareNoContractCase;
773 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
775 return group.release();
778 bool compareFRem(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
780 if (outputAllocs.size() != 1)
783 const BufferSp& expectedOutput = expectedOutputs[0];
784 const float *expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
785 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
787 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
789 const float f0 = expectedOutputAsFloat[idx];
790 const float f1 = outputAsFloat[idx];
791 // \todo relative error needs to be fairly high because FRem may be implemented as
792 // (roughly) frac(a/b)*b, so LSB errors can be magnified. But this should be fine for now.
793 if (deFloatAbs((f1 - f0) / f0) > 0.02)
800 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
802 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
803 ComputeShaderSpec spec;
804 de::Random rnd (deStringHash(group->getName()));
805 const int numElements = 200;
806 vector<float> inputFloats1 (numElements, 0);
807 vector<float> inputFloats2 (numElements, 0);
808 vector<float> outputFloats (numElements, 0);
810 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
811 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
813 for (size_t ndx = 0; ndx < numElements; ++ndx)
815 // Guard against divisors near zero.
816 if (std::fabs(inputFloats2[ndx]) < 1e-3)
817 inputFloats2[ndx] = 8.f;
819 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
820 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
824 string(getComputeAsmShaderPreamble()) +
826 "OpName %main \"main\"\n"
827 "OpName %id \"gl_GlobalInvocationID\"\n"
829 "OpDecorate %id BuiltIn GlobalInvocationId\n"
831 "OpDecorate %buf BufferBlock\n"
832 "OpDecorate %indata1 DescriptorSet 0\n"
833 "OpDecorate %indata1 Binding 0\n"
834 "OpDecorate %indata2 DescriptorSet 0\n"
835 "OpDecorate %indata2 Binding 1\n"
836 "OpDecorate %outdata DescriptorSet 0\n"
837 "OpDecorate %outdata Binding 2\n"
838 "OpDecorate %f32arr ArrayStride 4\n"
839 "OpMemberDecorate %buf 0 Offset 0\n"
841 + string(getComputeAsmCommonTypes()) +
843 "%buf = OpTypeStruct %f32arr\n"
844 "%bufptr = OpTypePointer Uniform %buf\n"
845 "%indata1 = OpVariable %bufptr Uniform\n"
846 "%indata2 = OpVariable %bufptr Uniform\n"
847 "%outdata = OpVariable %bufptr Uniform\n"
849 "%id = OpVariable %uvec3ptr Input\n"
850 "%zero = OpConstant %i32 0\n"
852 "%main = OpFunction %void None %voidf\n"
854 "%idval = OpLoad %uvec3 %id\n"
855 "%x = OpCompositeExtract %u32 %idval 0\n"
856 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
857 "%inval1 = OpLoad %f32 %inloc1\n"
858 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
859 "%inval2 = OpLoad %f32 %inloc2\n"
860 "%rem = OpFRem %f32 %inval1 %inval2\n"
861 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
862 " OpStore %outloc %rem\n"
866 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
867 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
868 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
869 spec.numWorkGroups = IVec3(numElements, 1, 1);
870 spec.verifyIO = &compareFRem;
872 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
874 return group.release();
877 bool compareNMin (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
879 if (outputAllocs.size() != 1)
882 const BufferSp& expectedOutput = expectedOutputs[0];
883 const float* const expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
884 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
886 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
888 const float f0 = expectedOutputAsFloat[idx];
889 const float f1 = outputAsFloat[idx];
891 // For NMin, we accept NaN as output if both inputs were NaN.
892 // Otherwise the NaN is the wrong choise, as on architectures that
893 // do not handle NaN, those are huge values.
894 if (!(tcu::Float32(f1).isNaN() && tcu::Float32(f0).isNaN()) && deFloatAbs(f1 - f0) > 0.00001f)
901 tcu::TestCaseGroup* createOpNMinGroup (tcu::TestContext& testCtx)
903 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnmin", "Test the OpNMin instruction"));
904 ComputeShaderSpec spec;
905 de::Random rnd (deStringHash(group->getName()));
906 const int numElements = 200;
907 vector<float> inputFloats1 (numElements, 0);
908 vector<float> inputFloats2 (numElements, 0);
909 vector<float> outputFloats (numElements, 0);
911 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
912 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
914 // Make the first case a full-NAN case.
915 inputFloats1[0] = TCU_NAN;
916 inputFloats2[0] = TCU_NAN;
918 for (size_t ndx = 0; ndx < numElements; ++ndx)
920 // By default, pick the smallest
921 outputFloats[ndx] = std::min(inputFloats1[ndx], inputFloats2[ndx]);
923 // Make half of the cases NaN cases
926 // Alternate between the NaN operand
929 outputFloats[ndx] = inputFloats2[ndx];
930 inputFloats1[ndx] = TCU_NAN;
934 outputFloats[ndx] = inputFloats1[ndx];
935 inputFloats2[ndx] = TCU_NAN;
941 "OpCapability Shader\n"
942 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
943 "OpMemoryModel Logical GLSL450\n"
944 "OpEntryPoint GLCompute %main \"main\" %id\n"
945 "OpExecutionMode %main LocalSize 1 1 1\n"
947 "OpName %main \"main\"\n"
948 "OpName %id \"gl_GlobalInvocationID\"\n"
950 "OpDecorate %id BuiltIn GlobalInvocationId\n"
952 "OpDecorate %buf BufferBlock\n"
953 "OpDecorate %indata1 DescriptorSet 0\n"
954 "OpDecorate %indata1 Binding 0\n"
955 "OpDecorate %indata2 DescriptorSet 0\n"
956 "OpDecorate %indata2 Binding 1\n"
957 "OpDecorate %outdata DescriptorSet 0\n"
958 "OpDecorate %outdata Binding 2\n"
959 "OpDecorate %f32arr ArrayStride 4\n"
960 "OpMemberDecorate %buf 0 Offset 0\n"
962 + string(getComputeAsmCommonTypes()) +
964 "%buf = OpTypeStruct %f32arr\n"
965 "%bufptr = OpTypePointer Uniform %buf\n"
966 "%indata1 = OpVariable %bufptr Uniform\n"
967 "%indata2 = OpVariable %bufptr Uniform\n"
968 "%outdata = OpVariable %bufptr Uniform\n"
970 "%id = OpVariable %uvec3ptr Input\n"
971 "%zero = OpConstant %i32 0\n"
973 "%main = OpFunction %void None %voidf\n"
975 "%idval = OpLoad %uvec3 %id\n"
976 "%x = OpCompositeExtract %u32 %idval 0\n"
977 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
978 "%inval1 = OpLoad %f32 %inloc1\n"
979 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
980 "%inval2 = OpLoad %f32 %inloc2\n"
981 "%rem = OpExtInst %f32 %std450 NMin %inval1 %inval2\n"
982 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
983 " OpStore %outloc %rem\n"
987 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
988 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
989 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
990 spec.numWorkGroups = IVec3(numElements, 1, 1);
991 spec.verifyIO = &compareNMin;
993 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
995 return group.release();
998 bool compareNMax (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1000 if (outputAllocs.size() != 1)
1003 const BufferSp& expectedOutput = expectedOutputs[0];
1004 const float* const expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
1005 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
1007 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
1009 const float f0 = expectedOutputAsFloat[idx];
1010 const float f1 = outputAsFloat[idx];
1012 // For NMax, NaN is considered acceptable result, since in
1013 // architectures that do not handle NaNs, those are huge values.
1014 if (!tcu::Float32(f1).isNaN() && deFloatAbs(f1 - f0) > 0.00001f)
1021 tcu::TestCaseGroup* createOpNMaxGroup (tcu::TestContext& testCtx)
1023 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, "opnmax", "Test the OpNMax instruction"));
1024 ComputeShaderSpec spec;
1025 de::Random rnd (deStringHash(group->getName()));
1026 const int numElements = 200;
1027 vector<float> inputFloats1 (numElements, 0);
1028 vector<float> inputFloats2 (numElements, 0);
1029 vector<float> outputFloats (numElements, 0);
1031 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1032 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1034 // Make the first case a full-NAN case.
1035 inputFloats1[0] = TCU_NAN;
1036 inputFloats2[0] = TCU_NAN;
1038 for (size_t ndx = 0; ndx < numElements; ++ndx)
1040 // By default, pick the biggest
1041 outputFloats[ndx] = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1043 // Make half of the cases NaN cases
1046 // Alternate between the NaN operand
1049 outputFloats[ndx] = inputFloats2[ndx];
1050 inputFloats1[ndx] = TCU_NAN;
1054 outputFloats[ndx] = inputFloats1[ndx];
1055 inputFloats2[ndx] = TCU_NAN;
1061 "OpCapability Shader\n"
1062 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1063 "OpMemoryModel Logical GLSL450\n"
1064 "OpEntryPoint GLCompute %main \"main\" %id\n"
1065 "OpExecutionMode %main LocalSize 1 1 1\n"
1067 "OpName %main \"main\"\n"
1068 "OpName %id \"gl_GlobalInvocationID\"\n"
1070 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1072 "OpDecorate %buf BufferBlock\n"
1073 "OpDecorate %indata1 DescriptorSet 0\n"
1074 "OpDecorate %indata1 Binding 0\n"
1075 "OpDecorate %indata2 DescriptorSet 0\n"
1076 "OpDecorate %indata2 Binding 1\n"
1077 "OpDecorate %outdata DescriptorSet 0\n"
1078 "OpDecorate %outdata Binding 2\n"
1079 "OpDecorate %f32arr ArrayStride 4\n"
1080 "OpMemberDecorate %buf 0 Offset 0\n"
1082 + string(getComputeAsmCommonTypes()) +
1084 "%buf = OpTypeStruct %f32arr\n"
1085 "%bufptr = OpTypePointer Uniform %buf\n"
1086 "%indata1 = OpVariable %bufptr Uniform\n"
1087 "%indata2 = OpVariable %bufptr Uniform\n"
1088 "%outdata = OpVariable %bufptr Uniform\n"
1090 "%id = OpVariable %uvec3ptr Input\n"
1091 "%zero = OpConstant %i32 0\n"
1093 "%main = OpFunction %void None %voidf\n"
1094 "%label = OpLabel\n"
1095 "%idval = OpLoad %uvec3 %id\n"
1096 "%x = OpCompositeExtract %u32 %idval 0\n"
1097 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1098 "%inval1 = OpLoad %f32 %inloc1\n"
1099 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1100 "%inval2 = OpLoad %f32 %inloc2\n"
1101 "%rem = OpExtInst %f32 %std450 NMax %inval1 %inval2\n"
1102 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1103 " OpStore %outloc %rem\n"
1107 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1108 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1109 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1110 spec.numWorkGroups = IVec3(numElements, 1, 1);
1111 spec.verifyIO = &compareNMax;
1113 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1115 return group.release();
1118 bool compareNClamp (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
1120 if (outputAllocs.size() != 1)
1123 const BufferSp& expectedOutput = expectedOutputs[0];
1124 const float* const expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
1125 const float* const outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
1127 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float) / 2; ++idx)
1129 const float e0 = expectedOutputAsFloat[idx * 2];
1130 const float e1 = expectedOutputAsFloat[idx * 2 + 1];
1131 const float res = outputAsFloat[idx];
1133 // For NClamp, we have two possible outcomes based on
1134 // whether NaNs are handled or not.
1135 // If either min or max value is NaN, the result is undefined,
1136 // so this test doesn't stress those. If the clamped value is
1137 // NaN, and NaNs are handled, the result is min; if NaNs are not
1138 // handled, they are big values that result in max.
1139 // If all three parameters are NaN, the result should be NaN.
1140 if (!((tcu::Float32(e0).isNaN() && tcu::Float32(res).isNaN()) ||
1141 (deFloatAbs(e0 - res) < 0.00001f) ||
1142 (deFloatAbs(e1 - res) < 0.00001f)))
1149 tcu::TestCaseGroup* createOpNClampGroup (tcu::TestContext& testCtx)
1151 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnclamp", "Test the OpNClamp instruction"));
1152 ComputeShaderSpec spec;
1153 de::Random rnd (deStringHash(group->getName()));
1154 const int numElements = 200;
1155 vector<float> inputFloats1 (numElements, 0);
1156 vector<float> inputFloats2 (numElements, 0);
1157 vector<float> inputFloats3 (numElements, 0);
1158 vector<float> outputFloats (numElements * 2, 0);
1160 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
1161 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats2[0], numElements);
1162 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats3[0], numElements);
1164 for (size_t ndx = 0; ndx < numElements; ++ndx)
1166 // Results are only defined if max value is bigger than min value.
1167 if (inputFloats2[ndx] > inputFloats3[ndx])
1169 float t = inputFloats2[ndx];
1170 inputFloats2[ndx] = inputFloats3[ndx];
1171 inputFloats3[ndx] = t;
1174 // By default, do the clamp, setting both possible answers
1175 float defaultRes = std::min(std::max(inputFloats1[ndx], inputFloats2[ndx]), inputFloats3[ndx]);
1177 float maxResA = std::max(inputFloats1[ndx], inputFloats2[ndx]);
1178 float maxResB = maxResA;
1180 // Alternate between the NaN cases
1183 inputFloats1[ndx] = TCU_NAN;
1184 // If NaN is handled, the result should be same as the clamp minimum.
1185 // If NaN is not handled, the result should clamp to the clamp maximum.
1186 maxResA = inputFloats2[ndx];
1187 maxResB = inputFloats3[ndx];
1191 // Not a NaN case - only one legal result.
1192 maxResA = defaultRes;
1193 maxResB = defaultRes;
1196 outputFloats[ndx * 2] = maxResA;
1197 outputFloats[ndx * 2 + 1] = maxResB;
1200 // Make the first case a full-NAN case.
1201 inputFloats1[0] = TCU_NAN;
1202 inputFloats2[0] = TCU_NAN;
1203 inputFloats3[0] = TCU_NAN;
1204 outputFloats[0] = TCU_NAN;
1205 outputFloats[1] = TCU_NAN;
1208 "OpCapability Shader\n"
1209 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
1210 "OpMemoryModel Logical GLSL450\n"
1211 "OpEntryPoint GLCompute %main \"main\" %id\n"
1212 "OpExecutionMode %main LocalSize 1 1 1\n"
1214 "OpName %main \"main\"\n"
1215 "OpName %id \"gl_GlobalInvocationID\"\n"
1217 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1219 "OpDecorate %buf BufferBlock\n"
1220 "OpDecorate %indata1 DescriptorSet 0\n"
1221 "OpDecorate %indata1 Binding 0\n"
1222 "OpDecorate %indata2 DescriptorSet 0\n"
1223 "OpDecorate %indata2 Binding 1\n"
1224 "OpDecorate %indata3 DescriptorSet 0\n"
1225 "OpDecorate %indata3 Binding 2\n"
1226 "OpDecorate %outdata DescriptorSet 0\n"
1227 "OpDecorate %outdata Binding 3\n"
1228 "OpDecorate %f32arr ArrayStride 4\n"
1229 "OpMemberDecorate %buf 0 Offset 0\n"
1231 + string(getComputeAsmCommonTypes()) +
1233 "%buf = OpTypeStruct %f32arr\n"
1234 "%bufptr = OpTypePointer Uniform %buf\n"
1235 "%indata1 = OpVariable %bufptr Uniform\n"
1236 "%indata2 = OpVariable %bufptr Uniform\n"
1237 "%indata3 = OpVariable %bufptr Uniform\n"
1238 "%outdata = OpVariable %bufptr Uniform\n"
1240 "%id = OpVariable %uvec3ptr Input\n"
1241 "%zero = OpConstant %i32 0\n"
1243 "%main = OpFunction %void None %voidf\n"
1244 "%label = OpLabel\n"
1245 "%idval = OpLoad %uvec3 %id\n"
1246 "%x = OpCompositeExtract %u32 %idval 0\n"
1247 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1248 "%inval1 = OpLoad %f32 %inloc1\n"
1249 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1250 "%inval2 = OpLoad %f32 %inloc2\n"
1251 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1252 "%inval3 = OpLoad %f32 %inloc3\n"
1253 "%rem = OpExtInst %f32 %std450 NClamp %inval1 %inval2 %inval3\n"
1254 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1255 " OpStore %outloc %rem\n"
1259 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1260 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1261 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1262 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1263 spec.numWorkGroups = IVec3(numElements, 1, 1);
1264 spec.verifyIO = &compareNClamp;
1266 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
1268 return group.release();
1271 tcu::TestCaseGroup* createOpSRemComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1273 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem", "Test the OpSRem instruction"));
1274 de::Random rnd (deStringHash(group->getName()));
1275 const int numElements = 200;
1277 const struct CaseParams
1280 const char* failMessage; // customized status message
1281 qpTestResult failResult; // override status on failure
1282 int op1Min, op1Max; // operand ranges
1286 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1287 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1289 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1291 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1293 const CaseParams& params = cases[caseNdx];
1294 ComputeShaderSpec spec;
1295 vector<deInt32> inputInts1 (numElements, 0);
1296 vector<deInt32> inputInts2 (numElements, 0);
1297 vector<deInt32> outputInts (numElements, 0);
1299 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1300 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1302 for (int ndx = 0; ndx < numElements; ++ndx)
1304 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1305 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1309 string(getComputeAsmShaderPreamble()) +
1311 "OpName %main \"main\"\n"
1312 "OpName %id \"gl_GlobalInvocationID\"\n"
1314 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1316 "OpDecorate %buf BufferBlock\n"
1317 "OpDecorate %indata1 DescriptorSet 0\n"
1318 "OpDecorate %indata1 Binding 0\n"
1319 "OpDecorate %indata2 DescriptorSet 0\n"
1320 "OpDecorate %indata2 Binding 1\n"
1321 "OpDecorate %outdata DescriptorSet 0\n"
1322 "OpDecorate %outdata Binding 2\n"
1323 "OpDecorate %i32arr ArrayStride 4\n"
1324 "OpMemberDecorate %buf 0 Offset 0\n"
1326 + string(getComputeAsmCommonTypes()) +
1328 "%buf = OpTypeStruct %i32arr\n"
1329 "%bufptr = OpTypePointer Uniform %buf\n"
1330 "%indata1 = OpVariable %bufptr Uniform\n"
1331 "%indata2 = OpVariable %bufptr Uniform\n"
1332 "%outdata = OpVariable %bufptr Uniform\n"
1334 "%id = OpVariable %uvec3ptr Input\n"
1335 "%zero = OpConstant %i32 0\n"
1337 "%main = OpFunction %void None %voidf\n"
1338 "%label = OpLabel\n"
1339 "%idval = OpLoad %uvec3 %id\n"
1340 "%x = OpCompositeExtract %u32 %idval 0\n"
1341 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1342 "%inval1 = OpLoad %i32 %inloc1\n"
1343 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1344 "%inval2 = OpLoad %i32 %inloc2\n"
1345 "%rem = OpSRem %i32 %inval1 %inval2\n"
1346 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1347 " OpStore %outloc %rem\n"
1351 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1352 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1353 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1354 spec.numWorkGroups = IVec3(numElements, 1, 1);
1355 spec.failResult = params.failResult;
1356 spec.failMessage = params.failMessage;
1358 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1361 return group.release();
1364 tcu::TestCaseGroup* createOpSRemComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1366 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsrem64", "Test the 64-bit OpSRem instruction"));
1367 de::Random rnd (deStringHash(group->getName()));
1368 const int numElements = 200;
1370 const struct CaseParams
1373 const char* failMessage; // customized status message
1374 qpTestResult failResult; // override status on failure
1378 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1379 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1381 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1383 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1385 const CaseParams& params = cases[caseNdx];
1386 ComputeShaderSpec spec;
1387 vector<deInt64> inputInts1 (numElements, 0);
1388 vector<deInt64> inputInts2 (numElements, 0);
1389 vector<deInt64> outputInts (numElements, 0);
1391 if (params.positive)
1393 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1394 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1398 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1399 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1402 for (int ndx = 0; ndx < numElements; ++ndx)
1404 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
1405 outputInts[ndx] = inputInts1[ndx] % inputInts2[ndx];
1409 "OpCapability Int64\n"
1411 + string(getComputeAsmShaderPreamble()) +
1413 "OpName %main \"main\"\n"
1414 "OpName %id \"gl_GlobalInvocationID\"\n"
1416 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1418 "OpDecorate %buf BufferBlock\n"
1419 "OpDecorate %indata1 DescriptorSet 0\n"
1420 "OpDecorate %indata1 Binding 0\n"
1421 "OpDecorate %indata2 DescriptorSet 0\n"
1422 "OpDecorate %indata2 Binding 1\n"
1423 "OpDecorate %outdata DescriptorSet 0\n"
1424 "OpDecorate %outdata Binding 2\n"
1425 "OpDecorate %i64arr ArrayStride 8\n"
1426 "OpMemberDecorate %buf 0 Offset 0\n"
1428 + string(getComputeAsmCommonTypes())
1429 + string(getComputeAsmCommonInt64Types()) +
1431 "%buf = OpTypeStruct %i64arr\n"
1432 "%bufptr = OpTypePointer Uniform %buf\n"
1433 "%indata1 = OpVariable %bufptr Uniform\n"
1434 "%indata2 = OpVariable %bufptr Uniform\n"
1435 "%outdata = OpVariable %bufptr Uniform\n"
1437 "%id = OpVariable %uvec3ptr Input\n"
1438 "%zero = OpConstant %i64 0\n"
1440 "%main = OpFunction %void None %voidf\n"
1441 "%label = OpLabel\n"
1442 "%idval = OpLoad %uvec3 %id\n"
1443 "%x = OpCompositeExtract %u32 %idval 0\n"
1444 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1445 "%inval1 = OpLoad %i64 %inloc1\n"
1446 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1447 "%inval2 = OpLoad %i64 %inloc2\n"
1448 "%rem = OpSRem %i64 %inval1 %inval2\n"
1449 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1450 " OpStore %outloc %rem\n"
1454 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1455 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1456 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1457 spec.numWorkGroups = IVec3(numElements, 1, 1);
1458 spec.failResult = params.failResult;
1459 spec.failMessage = params.failMessage;
1461 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1464 return group.release();
1467 tcu::TestCaseGroup* createOpSModComputeGroup (tcu::TestContext& testCtx, qpTestResult negFailResult)
1469 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod", "Test the OpSMod instruction"));
1470 de::Random rnd (deStringHash(group->getName()));
1471 const int numElements = 200;
1473 const struct CaseParams
1476 const char* failMessage; // customized status message
1477 qpTestResult failResult; // override status on failure
1478 int op1Min, op1Max; // operand ranges
1482 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, 0, 65536, 0, 100 },
1483 { "all", "Inconsistent results, but within specification", negFailResult, -65536, 65536, -100, 100 }, // see below
1485 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1487 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1489 const CaseParams& params = cases[caseNdx];
1491 ComputeShaderSpec spec;
1492 vector<deInt32> inputInts1 (numElements, 0);
1493 vector<deInt32> inputInts2 (numElements, 0);
1494 vector<deInt32> outputInts (numElements, 0);
1496 fillRandomScalars(rnd, params.op1Min, params.op1Max, &inputInts1[0], numElements);
1497 fillRandomScalars(rnd, params.op2Min, params.op2Max, &inputInts2[0], numElements, filterNotZero);
1499 for (int ndx = 0; ndx < numElements; ++ndx)
1501 deInt32 rem = inputInts1[ndx] % inputInts2[ndx];
1504 outputInts[ndx] = 0;
1506 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1508 // They have the same sign
1509 outputInts[ndx] = rem;
1513 // They have opposite sign. The remainder operation takes the
1514 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1515 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1516 // the result has the correct sign and that it is still
1517 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1519 // See also http://mathforum.org/library/drmath/view/52343.html
1520 outputInts[ndx] = rem + inputInts2[ndx];
1525 string(getComputeAsmShaderPreamble()) +
1527 "OpName %main \"main\"\n"
1528 "OpName %id \"gl_GlobalInvocationID\"\n"
1530 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1532 "OpDecorate %buf BufferBlock\n"
1533 "OpDecorate %indata1 DescriptorSet 0\n"
1534 "OpDecorate %indata1 Binding 0\n"
1535 "OpDecorate %indata2 DescriptorSet 0\n"
1536 "OpDecorate %indata2 Binding 1\n"
1537 "OpDecorate %outdata DescriptorSet 0\n"
1538 "OpDecorate %outdata Binding 2\n"
1539 "OpDecorate %i32arr ArrayStride 4\n"
1540 "OpMemberDecorate %buf 0 Offset 0\n"
1542 + string(getComputeAsmCommonTypes()) +
1544 "%buf = OpTypeStruct %i32arr\n"
1545 "%bufptr = OpTypePointer Uniform %buf\n"
1546 "%indata1 = OpVariable %bufptr Uniform\n"
1547 "%indata2 = OpVariable %bufptr Uniform\n"
1548 "%outdata = OpVariable %bufptr Uniform\n"
1550 "%id = OpVariable %uvec3ptr Input\n"
1551 "%zero = OpConstant %i32 0\n"
1553 "%main = OpFunction %void None %voidf\n"
1554 "%label = OpLabel\n"
1555 "%idval = OpLoad %uvec3 %id\n"
1556 "%x = OpCompositeExtract %u32 %idval 0\n"
1557 "%inloc1 = OpAccessChain %i32ptr %indata1 %zero %x\n"
1558 "%inval1 = OpLoad %i32 %inloc1\n"
1559 "%inloc2 = OpAccessChain %i32ptr %indata2 %zero %x\n"
1560 "%inval2 = OpLoad %i32 %inloc2\n"
1561 "%rem = OpSMod %i32 %inval1 %inval2\n"
1562 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1563 " OpStore %outloc %rem\n"
1567 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts1)));
1568 spec.inputs.push_back (BufferSp(new Int32Buffer(inputInts2)));
1569 spec.outputs.push_back (BufferSp(new Int32Buffer(outputInts)));
1570 spec.numWorkGroups = IVec3(numElements, 1, 1);
1571 spec.failResult = params.failResult;
1572 spec.failMessage = params.failMessage;
1574 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec));
1577 return group.release();
1580 tcu::TestCaseGroup* createOpSModComputeGroup64 (tcu::TestContext& testCtx, qpTestResult negFailResult)
1582 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsmod64", "Test the OpSMod instruction"));
1583 de::Random rnd (deStringHash(group->getName()));
1584 const int numElements = 200;
1586 const struct CaseParams
1589 const char* failMessage; // customized status message
1590 qpTestResult failResult; // override status on failure
1594 { "positive", "Output doesn't match with expected", QP_TEST_RESULT_FAIL, true },
1595 { "all", "Inconsistent results, but within specification", negFailResult, false }, // see below
1597 // If either operand is negative the result is undefined. Some implementations may still return correct values.
1599 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
1601 const CaseParams& params = cases[caseNdx];
1603 ComputeShaderSpec spec;
1604 vector<deInt64> inputInts1 (numElements, 0);
1605 vector<deInt64> inputInts2 (numElements, 0);
1606 vector<deInt64> outputInts (numElements, 0);
1609 if (params.positive)
1611 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements, filterNonNegative);
1612 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterPositive);
1616 fillRandomInt64sLogDistributed(rnd, inputInts1, numElements);
1617 fillRandomInt64sLogDistributed(rnd, inputInts2, numElements, filterNotZero);
1620 for (int ndx = 0; ndx < numElements; ++ndx)
1622 deInt64 rem = inputInts1[ndx] % inputInts2[ndx];
1625 outputInts[ndx] = 0;
1627 else if ((inputInts1[ndx] >= 0) == (inputInts2[ndx] >= 0))
1629 // They have the same sign
1630 outputInts[ndx] = rem;
1634 // They have opposite sign. The remainder operation takes the
1635 // sign inputInts1[ndx] but OpSMod is supposed to take ths sign
1636 // of inputInts2[ndx]. Adding inputInts2[ndx] will ensure that
1637 // the result has the correct sign and that it is still
1638 // congruent to inputInts1[ndx] modulo inputInts2[ndx]
1640 // See also http://mathforum.org/library/drmath/view/52343.html
1641 outputInts[ndx] = rem + inputInts2[ndx];
1646 "OpCapability Int64\n"
1648 + string(getComputeAsmShaderPreamble()) +
1650 "OpName %main \"main\"\n"
1651 "OpName %id \"gl_GlobalInvocationID\"\n"
1653 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1655 "OpDecorate %buf BufferBlock\n"
1656 "OpDecorate %indata1 DescriptorSet 0\n"
1657 "OpDecorate %indata1 Binding 0\n"
1658 "OpDecorate %indata2 DescriptorSet 0\n"
1659 "OpDecorate %indata2 Binding 1\n"
1660 "OpDecorate %outdata DescriptorSet 0\n"
1661 "OpDecorate %outdata Binding 2\n"
1662 "OpDecorate %i64arr ArrayStride 8\n"
1663 "OpMemberDecorate %buf 0 Offset 0\n"
1665 + string(getComputeAsmCommonTypes())
1666 + string(getComputeAsmCommonInt64Types()) +
1668 "%buf = OpTypeStruct %i64arr\n"
1669 "%bufptr = OpTypePointer Uniform %buf\n"
1670 "%indata1 = OpVariable %bufptr Uniform\n"
1671 "%indata2 = OpVariable %bufptr Uniform\n"
1672 "%outdata = OpVariable %bufptr Uniform\n"
1674 "%id = OpVariable %uvec3ptr Input\n"
1675 "%zero = OpConstant %i64 0\n"
1677 "%main = OpFunction %void None %voidf\n"
1678 "%label = OpLabel\n"
1679 "%idval = OpLoad %uvec3 %id\n"
1680 "%x = OpCompositeExtract %u32 %idval 0\n"
1681 "%inloc1 = OpAccessChain %i64ptr %indata1 %zero %x\n"
1682 "%inval1 = OpLoad %i64 %inloc1\n"
1683 "%inloc2 = OpAccessChain %i64ptr %indata2 %zero %x\n"
1684 "%inval2 = OpLoad %i64 %inloc2\n"
1685 "%rem = OpSMod %i64 %inval1 %inval2\n"
1686 "%outloc = OpAccessChain %i64ptr %outdata %zero %x\n"
1687 " OpStore %outloc %rem\n"
1691 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts1)));
1692 spec.inputs.push_back (BufferSp(new Int64Buffer(inputInts2)));
1693 spec.outputs.push_back (BufferSp(new Int64Buffer(outputInts)));
1694 spec.numWorkGroups = IVec3(numElements, 1, 1);
1695 spec.failResult = params.failResult;
1696 spec.failMessage = params.failMessage;
1698 group->addChild(new SpvAsmComputeShaderCase(testCtx, params.name, "", spec, COMPUTE_TEST_USES_INT64));
1701 return group.release();
1704 // Copy contents in the input buffer to the output buffer.
1705 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
1707 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
1708 de::Random rnd (deStringHash(group->getName()));
1709 const int numElements = 100;
1711 // 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.
1712 ComputeShaderSpec spec1;
1713 vector<Vec4> inputFloats1 (numElements);
1714 vector<Vec4> outputFloats1 (numElements);
1716 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
1718 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1719 floorAll(inputFloats1);
1721 for (size_t ndx = 0; ndx < numElements; ++ndx)
1722 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
1725 string(getComputeAsmShaderPreamble()) +
1727 "OpName %main \"main\"\n"
1728 "OpName %id \"gl_GlobalInvocationID\"\n"
1730 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1731 "OpDecorate %vec4arr ArrayStride 16\n"
1733 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1735 "%vec4 = OpTypeVector %f32 4\n"
1736 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
1737 "%vec4ptr_f = OpTypePointer Function %vec4\n"
1738 "%vec4arr = OpTypeRuntimeArray %vec4\n"
1739 "%buf = OpTypeStruct %vec4arr\n"
1740 "%bufptr = OpTypePointer Uniform %buf\n"
1741 "%indata = OpVariable %bufptr Uniform\n"
1742 "%outdata = OpVariable %bufptr Uniform\n"
1744 "%id = OpVariable %uvec3ptr Input\n"
1745 "%zero = OpConstant %i32 0\n"
1746 "%c_f_0 = OpConstant %f32 0.\n"
1747 "%c_f_0_5 = OpConstant %f32 0.5\n"
1748 "%c_f_1_5 = OpConstant %f32 1.5\n"
1749 "%c_f_2_5 = OpConstant %f32 2.5\n"
1750 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
1752 "%main = OpFunction %void None %voidf\n"
1753 "%label = OpLabel\n"
1754 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
1755 "%idval = OpLoad %uvec3 %id\n"
1756 "%x = OpCompositeExtract %u32 %idval 0\n"
1757 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
1758 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
1759 " OpCopyMemory %v_vec4 %inloc\n"
1760 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
1761 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
1762 " OpStore %outloc %add\n"
1766 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
1767 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
1768 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1770 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
1772 // The following case copies a float[100] variable from the input buffer to the output buffer.
1773 ComputeShaderSpec spec2;
1774 vector<float> inputFloats2 (numElements);
1775 vector<float> outputFloats2 (numElements);
1777 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
1779 for (size_t ndx = 0; ndx < numElements; ++ndx)
1780 outputFloats2[ndx] = inputFloats2[ndx];
1783 string(getComputeAsmShaderPreamble()) +
1785 "OpName %main \"main\"\n"
1786 "OpName %id \"gl_GlobalInvocationID\"\n"
1788 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1789 "OpDecorate %f32arr100 ArrayStride 4\n"
1791 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1793 "%hundred = OpConstant %u32 100\n"
1794 "%f32arr100 = OpTypeArray %f32 %hundred\n"
1795 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
1796 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
1797 "%buf = OpTypeStruct %f32arr100\n"
1798 "%bufptr = OpTypePointer Uniform %buf\n"
1799 "%indata = OpVariable %bufptr Uniform\n"
1800 "%outdata = OpVariable %bufptr Uniform\n"
1802 "%id = OpVariable %uvec3ptr Input\n"
1803 "%zero = OpConstant %i32 0\n"
1805 "%main = OpFunction %void None %voidf\n"
1806 "%label = OpLabel\n"
1807 "%var = OpVariable %f32arr100ptr_f Function\n"
1808 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
1809 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
1810 " OpCopyMemory %var %inarr\n"
1811 " OpCopyMemory %outarr %var\n"
1815 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1816 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1817 spec2.numWorkGroups = IVec3(1, 1, 1);
1819 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
1821 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
1822 ComputeShaderSpec spec3;
1823 vector<float> inputFloats3 (16);
1824 vector<float> outputFloats3 (16);
1826 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
1828 for (size_t ndx = 0; ndx < 16; ++ndx)
1829 outputFloats3[ndx] = inputFloats3[ndx];
1832 string(getComputeAsmShaderPreamble()) +
1834 "OpName %main \"main\"\n"
1835 "OpName %id \"gl_GlobalInvocationID\"\n"
1837 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1838 "OpMemberDecorate %buf 0 Offset 0\n"
1839 "OpMemberDecorate %buf 1 Offset 16\n"
1840 "OpMemberDecorate %buf 2 Offset 32\n"
1841 "OpMemberDecorate %buf 3 Offset 48\n"
1843 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1845 "%vec4 = OpTypeVector %f32 4\n"
1846 "%buf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
1847 "%bufptr = OpTypePointer Uniform %buf\n"
1848 "%indata = OpVariable %bufptr Uniform\n"
1849 "%outdata = OpVariable %bufptr Uniform\n"
1850 "%vec4stptr = OpTypePointer Function %buf\n"
1852 "%id = OpVariable %uvec3ptr Input\n"
1853 "%zero = OpConstant %i32 0\n"
1855 "%main = OpFunction %void None %voidf\n"
1856 "%label = OpLabel\n"
1857 "%var = OpVariable %vec4stptr Function\n"
1858 " OpCopyMemory %var %indata\n"
1859 " OpCopyMemory %outdata %var\n"
1863 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1864 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
1865 spec3.numWorkGroups = IVec3(1, 1, 1);
1867 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
1869 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
1870 ComputeShaderSpec spec4;
1871 vector<float> inputFloats4 (numElements);
1872 vector<float> outputFloats4 (numElements);
1874 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
1876 for (size_t ndx = 0; ndx < numElements; ++ndx)
1877 outputFloats4[ndx] = -inputFloats4[ndx];
1880 string(getComputeAsmShaderPreamble()) +
1882 "OpName %main \"main\"\n"
1883 "OpName %id \"gl_GlobalInvocationID\"\n"
1885 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1887 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
1889 "%f32ptr_f = OpTypePointer Function %f32\n"
1890 "%id = OpVariable %uvec3ptr Input\n"
1891 "%zero = OpConstant %i32 0\n"
1893 "%main = OpFunction %void None %voidf\n"
1894 "%label = OpLabel\n"
1895 "%var = OpVariable %f32ptr_f Function\n"
1896 "%idval = OpLoad %uvec3 %id\n"
1897 "%x = OpCompositeExtract %u32 %idval 0\n"
1898 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1899 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1900 " OpCopyMemory %var %inloc\n"
1901 "%val = OpLoad %f32 %var\n"
1902 "%neg = OpFNegate %f32 %val\n"
1903 " OpStore %outloc %neg\n"
1907 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
1908 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
1909 spec4.numWorkGroups = IVec3(numElements, 1, 1);
1911 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
1913 return group.release();
1916 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
1918 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
1919 ComputeShaderSpec spec;
1920 de::Random rnd (deStringHash(group->getName()));
1921 const int numElements = 100;
1922 vector<float> inputFloats (numElements, 0);
1923 vector<float> outputFloats (numElements, 0);
1925 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
1927 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1928 floorAll(inputFloats);
1930 for (size_t ndx = 0; ndx < numElements; ++ndx)
1931 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
1934 string(getComputeAsmShaderPreamble()) +
1936 "OpName %main \"main\"\n"
1937 "OpName %id \"gl_GlobalInvocationID\"\n"
1939 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1941 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
1943 "%fmat = OpTypeMatrix %fvec3 3\n"
1944 "%three = OpConstant %u32 3\n"
1945 "%farr = OpTypeArray %f32 %three\n"
1946 "%fst = OpTypeStruct %f32 %f32\n"
1948 + string(getComputeAsmInputOutputBuffer()) +
1950 "%id = OpVariable %uvec3ptr Input\n"
1951 "%zero = OpConstant %i32 0\n"
1952 "%c_f = OpConstant %f32 1.5\n"
1953 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
1954 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
1955 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
1956 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
1958 "%main = OpFunction %void None %voidf\n"
1959 "%label = OpLabel\n"
1960 "%c_f_copy = OpCopyObject %f32 %c_f\n"
1961 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
1962 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
1963 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
1964 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
1965 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
1966 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
1967 "%farr_elem = OpCompositeExtract %f32 %c_farr_copy 2\n"
1968 "%fst_elem = OpCompositeExtract %f32 %c_fst_copy 1\n"
1969 // Add up. 1.5 * 5 = 7.5.
1970 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
1971 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
1972 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
1973 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
1975 "%idval = OpLoad %uvec3 %id\n"
1976 "%x = OpCompositeExtract %u32 %idval 0\n"
1977 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1978 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1979 "%inval = OpLoad %f32 %inloc\n"
1980 "%add = OpFAdd %f32 %add4 %inval\n"
1981 " OpStore %outloc %add\n"
1984 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1985 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1986 spec.numWorkGroups = IVec3(numElements, 1, 1);
1988 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
1990 return group.release();
1992 // Assembly code used for testing OpUnreachable is based on GLSL source code:
1996 // layout(std140, set = 0, binding = 0) readonly buffer Input {
1997 // float elements[];
1999 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2000 // float elements[];
2003 // void not_called_func() {
2004 // // place OpUnreachable here
2007 // uint modulo4(uint val) {
2008 // switch (val % uint(4)) {
2009 // case 0: return 3;
2010 // case 1: return 2;
2011 // case 2: return 1;
2012 // case 3: return 0;
2013 // default: return 100; // place OpUnreachable here
2019 // // place OpUnreachable here
2023 // uint x = gl_GlobalInvocationID.x;
2024 // if (const5() > modulo4(1000)) {
2025 // output_data.elements[x] = -input_data.elements[x];
2027 // // place OpUnreachable here
2028 // output_data.elements[x] = input_data.elements[x];
2032 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
2034 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
2035 ComputeShaderSpec spec;
2036 de::Random rnd (deStringHash(group->getName()));
2037 const int numElements = 100;
2038 vector<float> positiveFloats (numElements, 0);
2039 vector<float> negativeFloats (numElements, 0);
2041 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2043 for (size_t ndx = 0; ndx < numElements; ++ndx)
2044 negativeFloats[ndx] = -positiveFloats[ndx];
2047 string(getComputeAsmShaderPreamble()) +
2049 "OpSource GLSL 430\n"
2050 "OpName %main \"main\"\n"
2051 "OpName %func_not_called_func \"not_called_func(\"\n"
2052 "OpName %func_modulo4 \"modulo4(u1;\"\n"
2053 "OpName %func_const5 \"const5(\"\n"
2054 "OpName %id \"gl_GlobalInvocationID\"\n"
2056 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2058 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2060 "%u32ptr = OpTypePointer Function %u32\n"
2061 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
2062 "%unitf = OpTypeFunction %u32\n"
2064 "%id = OpVariable %uvec3ptr Input\n"
2065 "%zero = OpConstant %u32 0\n"
2066 "%one = OpConstant %u32 1\n"
2067 "%two = OpConstant %u32 2\n"
2068 "%three = OpConstant %u32 3\n"
2069 "%four = OpConstant %u32 4\n"
2070 "%five = OpConstant %u32 5\n"
2071 "%hundred = OpConstant %u32 100\n"
2072 "%thousand = OpConstant %u32 1000\n"
2074 + string(getComputeAsmInputOutputBuffer()) +
2077 "%main = OpFunction %void None %voidf\n"
2078 "%main_entry = OpLabel\n"
2079 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
2080 "%idval = OpLoad %uvec3 %id\n"
2081 "%x = OpCompositeExtract %u32 %idval 0\n"
2082 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2083 "%inval = OpLoad %f32 %inloc\n"
2084 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2085 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
2086 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
2087 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
2088 " OpSelectionMerge %if_end None\n"
2089 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2090 "%if_true = OpLabel\n"
2091 "%negate = OpFNegate %f32 %inval\n"
2092 " OpStore %outloc %negate\n"
2093 " OpBranch %if_end\n"
2094 "%if_false = OpLabel\n"
2095 " OpUnreachable\n" // Unreachable else branch for if statement
2096 "%if_end = OpLabel\n"
2100 // not_called_function()
2101 "%func_not_called_func = OpFunction %void None %voidf\n"
2102 "%not_called_func_entry = OpLabel\n"
2103 " OpUnreachable\n" // Unreachable entry block in not called static function
2107 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
2108 "%valptr = OpFunctionParameter %u32ptr\n"
2109 "%modulo4_entry = OpLabel\n"
2110 "%val = OpLoad %u32 %valptr\n"
2111 "%modulo = OpUMod %u32 %val %four\n"
2112 " OpSelectionMerge %switch_merge None\n"
2113 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
2114 "%case0 = OpLabel\n"
2115 " OpReturnValue %three\n"
2116 "%case1 = OpLabel\n"
2117 " OpReturnValue %two\n"
2118 "%case2 = OpLabel\n"
2119 " OpReturnValue %one\n"
2120 "%case3 = OpLabel\n"
2121 " OpReturnValue %zero\n"
2122 "%default = OpLabel\n"
2123 " OpUnreachable\n" // Unreachable default case for switch statement
2124 "%switch_merge = OpLabel\n"
2125 " OpUnreachable\n" // Unreachable merge block for switch statement
2129 "%func_const5 = OpFunction %u32 None %unitf\n"
2130 "%const5_entry = OpLabel\n"
2131 " OpReturnValue %five\n"
2132 "%unreachable = OpLabel\n"
2133 " OpUnreachable\n" // Unreachable block in function
2135 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2136 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2137 spec.numWorkGroups = IVec3(numElements, 1, 1);
2139 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
2141 return group.release();
2144 // Assembly code used for testing decoration group is based on GLSL source code:
2148 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
2149 // float elements[];
2151 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
2152 // float elements[];
2154 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
2155 // float elements[];
2157 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
2158 // float elements[];
2160 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
2161 // float elements[];
2163 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
2164 // float elements[];
2168 // uint x = gl_GlobalInvocationID.x;
2169 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
2171 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
2173 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
2174 ComputeShaderSpec spec;
2175 de::Random rnd (deStringHash(group->getName()));
2176 const int numElements = 100;
2177 vector<float> inputFloats0 (numElements, 0);
2178 vector<float> inputFloats1 (numElements, 0);
2179 vector<float> inputFloats2 (numElements, 0);
2180 vector<float> inputFloats3 (numElements, 0);
2181 vector<float> inputFloats4 (numElements, 0);
2182 vector<float> outputFloats (numElements, 0);
2184 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
2185 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
2186 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
2187 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
2188 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
2190 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2191 floorAll(inputFloats0);
2192 floorAll(inputFloats1);
2193 floorAll(inputFloats2);
2194 floorAll(inputFloats3);
2195 floorAll(inputFloats4);
2197 for (size_t ndx = 0; ndx < numElements; ++ndx)
2198 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
2201 string(getComputeAsmShaderPreamble()) +
2203 "OpSource GLSL 430\n"
2204 "OpName %main \"main\"\n"
2205 "OpName %id \"gl_GlobalInvocationID\"\n"
2207 // Not using group decoration on variable.
2208 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2209 // Not using group decoration on type.
2210 "OpDecorate %f32arr ArrayStride 4\n"
2212 "OpDecorate %groups BufferBlock\n"
2213 "OpDecorate %groupm Offset 0\n"
2214 "%groups = OpDecorationGroup\n"
2215 "%groupm = OpDecorationGroup\n"
2217 // Group decoration on multiple structs.
2218 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
2219 // Group decoration on multiple struct members.
2220 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
2222 "OpDecorate %group1 DescriptorSet 0\n"
2223 "OpDecorate %group3 DescriptorSet 0\n"
2224 "OpDecorate %group3 NonWritable\n"
2225 "OpDecorate %group3 Restrict\n"
2226 "%group0 = OpDecorationGroup\n"
2227 "%group1 = OpDecorationGroup\n"
2228 "%group3 = OpDecorationGroup\n"
2230 // Applying the same decoration group multiple times.
2231 "OpGroupDecorate %group1 %outdata\n"
2232 "OpGroupDecorate %group1 %outdata\n"
2233 "OpGroupDecorate %group1 %outdata\n"
2234 "OpDecorate %outdata DescriptorSet 0\n"
2235 "OpDecorate %outdata Binding 5\n"
2236 // Applying decoration group containing nothing.
2237 "OpGroupDecorate %group0 %indata0\n"
2238 "OpDecorate %indata0 DescriptorSet 0\n"
2239 "OpDecorate %indata0 Binding 0\n"
2240 // Applying decoration group containing one decoration.
2241 "OpGroupDecorate %group1 %indata1\n"
2242 "OpDecorate %indata1 Binding 1\n"
2243 // Applying decoration group containing multiple decorations.
2244 "OpGroupDecorate %group3 %indata2 %indata3\n"
2245 "OpDecorate %indata2 Binding 2\n"
2246 "OpDecorate %indata3 Binding 3\n"
2247 // Applying multiple decoration groups (with overlapping).
2248 "OpGroupDecorate %group0 %indata4\n"
2249 "OpGroupDecorate %group1 %indata4\n"
2250 "OpGroupDecorate %group3 %indata4\n"
2251 "OpDecorate %indata4 Binding 4\n"
2253 + string(getComputeAsmCommonTypes()) +
2255 "%id = OpVariable %uvec3ptr Input\n"
2256 "%zero = OpConstant %i32 0\n"
2258 "%outbuf = OpTypeStruct %f32arr\n"
2259 "%outbufptr = OpTypePointer Uniform %outbuf\n"
2260 "%outdata = OpVariable %outbufptr Uniform\n"
2261 "%inbuf0 = OpTypeStruct %f32arr\n"
2262 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
2263 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
2264 "%inbuf1 = OpTypeStruct %f32arr\n"
2265 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
2266 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
2267 "%inbuf2 = OpTypeStruct %f32arr\n"
2268 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
2269 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
2270 "%inbuf3 = OpTypeStruct %f32arr\n"
2271 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
2272 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
2273 "%inbuf4 = OpTypeStruct %f32arr\n"
2274 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
2275 "%indata4 = OpVariable %inbufptr Uniform\n"
2277 "%main = OpFunction %void None %voidf\n"
2278 "%label = OpLabel\n"
2279 "%idval = OpLoad %uvec3 %id\n"
2280 "%x = OpCompositeExtract %u32 %idval 0\n"
2281 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
2282 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
2283 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
2284 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
2285 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
2286 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2287 "%inval0 = OpLoad %f32 %inloc0\n"
2288 "%inval1 = OpLoad %f32 %inloc1\n"
2289 "%inval2 = OpLoad %f32 %inloc2\n"
2290 "%inval3 = OpLoad %f32 %inloc3\n"
2291 "%inval4 = OpLoad %f32 %inloc4\n"
2292 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
2293 "%add1 = OpFAdd %f32 %add0 %inval2\n"
2294 "%add2 = OpFAdd %f32 %add1 %inval3\n"
2295 "%add = OpFAdd %f32 %add2 %inval4\n"
2296 " OpStore %outloc %add\n"
2299 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
2300 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
2301 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
2302 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
2303 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
2304 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2305 spec.numWorkGroups = IVec3(numElements, 1, 1);
2307 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
2309 return group.release();
2312 struct SpecConstantTwoIntCase
2314 const char* caseName;
2315 const char* scDefinition0;
2316 const char* scDefinition1;
2317 const char* scResultType;
2318 const char* scOperation;
2319 deInt32 scActualValue0;
2320 deInt32 scActualValue1;
2321 const char* resultOperation;
2322 vector<deInt32> expectedOutput;
2324 SpecConstantTwoIntCase (const char* name,
2325 const char* definition0,
2326 const char* definition1,
2327 const char* resultType,
2328 const char* operation,
2331 const char* resultOp,
2332 const vector<deInt32>& output)
2334 , scDefinition0 (definition0)
2335 , scDefinition1 (definition1)
2336 , scResultType (resultType)
2337 , scOperation (operation)
2338 , scActualValue0 (value0)
2339 , scActualValue1 (value1)
2340 , resultOperation (resultOp)
2341 , expectedOutput (output) {}
2344 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
2346 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
2347 vector<SpecConstantTwoIntCase> cases;
2348 de::Random rnd (deStringHash(group->getName()));
2349 const int numElements = 100;
2350 vector<deInt32> inputInts (numElements, 0);
2351 vector<deInt32> outputInts1 (numElements, 0);
2352 vector<deInt32> outputInts2 (numElements, 0);
2353 vector<deInt32> outputInts3 (numElements, 0);
2354 vector<deInt32> outputInts4 (numElements, 0);
2355 const StringTemplate shaderTemplate (
2356 string(getComputeAsmShaderPreamble()) +
2358 "OpName %main \"main\"\n"
2359 "OpName %id \"gl_GlobalInvocationID\"\n"
2361 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2362 "OpDecorate %sc_0 SpecId 0\n"
2363 "OpDecorate %sc_1 SpecId 1\n"
2364 "OpDecorate %i32arr ArrayStride 4\n"
2366 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2368 "%buf = OpTypeStruct %i32arr\n"
2369 "%bufptr = OpTypePointer Uniform %buf\n"
2370 "%indata = OpVariable %bufptr Uniform\n"
2371 "%outdata = OpVariable %bufptr Uniform\n"
2373 "%id = OpVariable %uvec3ptr Input\n"
2374 "%zero = OpConstant %i32 0\n"
2376 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
2377 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
2378 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
2380 "%main = OpFunction %void None %voidf\n"
2381 "%label = OpLabel\n"
2382 "%idval = OpLoad %uvec3 %id\n"
2383 "%x = OpCompositeExtract %u32 %idval 0\n"
2384 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2385 "%inval = OpLoad %i32 %inloc\n"
2386 "%final = ${GEN_RESULT}\n"
2387 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2388 " OpStore %outloc %final\n"
2390 " OpFunctionEnd\n");
2392 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
2394 for (size_t ndx = 0; ndx < numElements; ++ndx)
2396 outputInts1[ndx] = inputInts[ndx] + 42;
2397 outputInts2[ndx] = inputInts[ndx];
2398 outputInts3[ndx] = inputInts[ndx] - 11200;
2399 outputInts4[ndx] = inputInts[ndx] + 1;
2402 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
2403 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
2404 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
2406 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
2407 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
2408 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
2409 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
2410 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
2411 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2412 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
2413 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
2414 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
2415 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
2416 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
2417 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2418 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
2419 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
2420 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
2421 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
2422 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2423 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
2424 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
2425 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
2426 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
2427 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
2428 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
2429 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2430 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2431 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
2432 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
2433 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
2434 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
2435 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
2436 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
2437 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
2439 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2441 map<string, string> specializations;
2442 ComputeShaderSpec spec;
2444 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
2445 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
2446 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
2447 specializations["SC_OP"] = cases[caseNdx].scOperation;
2448 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
2450 spec.assembly = shaderTemplate.specialize(specializations);
2451 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2452 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
2453 spec.numWorkGroups = IVec3(numElements, 1, 1);
2454 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
2455 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
2457 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec));
2460 ComputeShaderSpec spec;
2463 string(getComputeAsmShaderPreamble()) +
2465 "OpName %main \"main\"\n"
2466 "OpName %id \"gl_GlobalInvocationID\"\n"
2468 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2469 "OpDecorate %sc_0 SpecId 0\n"
2470 "OpDecorate %sc_1 SpecId 1\n"
2471 "OpDecorate %sc_2 SpecId 2\n"
2472 "OpDecorate %i32arr ArrayStride 4\n"
2474 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2476 "%ivec3 = OpTypeVector %i32 3\n"
2477 "%buf = OpTypeStruct %i32arr\n"
2478 "%bufptr = OpTypePointer Uniform %buf\n"
2479 "%indata = OpVariable %bufptr Uniform\n"
2480 "%outdata = OpVariable %bufptr Uniform\n"
2482 "%id = OpVariable %uvec3ptr Input\n"
2483 "%zero = OpConstant %i32 0\n"
2484 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
2485 "%vec3_undef = OpUndef %ivec3\n"
2487 "%sc_0 = OpSpecConstant %i32 0\n"
2488 "%sc_1 = OpSpecConstant %i32 0\n"
2489 "%sc_2 = OpSpecConstant %i32 0\n"
2490 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
2491 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
2492 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
2493 "%sc_vec3_0_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
2494 "%sc_vec3_1_s = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
2495 "%sc_vec3_2_s = OpSpecConstantOp %ivec3 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
2496 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
2497 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
2498 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
2499 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
2500 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
2501 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
2502 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
2504 "%main = OpFunction %void None %voidf\n"
2505 "%label = OpLabel\n"
2506 "%idval = OpLoad %uvec3 %id\n"
2507 "%x = OpCompositeExtract %u32 %idval 0\n"
2508 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
2509 "%inval = OpLoad %i32 %inloc\n"
2510 "%final = OpIAdd %i32 %inval %sc_final\n"
2511 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
2512 " OpStore %outloc %final\n"
2515 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
2516 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
2517 spec.numWorkGroups = IVec3(numElements, 1, 1);
2518 spec.specConstants.push_back(123);
2519 spec.specConstants.push_back(56);
2520 spec.specConstants.push_back(-77);
2522 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
2524 return group.release();
2527 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
2529 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
2530 ComputeShaderSpec spec1;
2531 ComputeShaderSpec spec2;
2532 ComputeShaderSpec spec3;
2533 de::Random rnd (deStringHash(group->getName()));
2534 const int numElements = 100;
2535 vector<float> inputFloats (numElements, 0);
2536 vector<float> outputFloats1 (numElements, 0);
2537 vector<float> outputFloats2 (numElements, 0);
2538 vector<float> outputFloats3 (numElements, 0);
2540 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
2542 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2543 floorAll(inputFloats);
2545 for (size_t ndx = 0; ndx < numElements; ++ndx)
2549 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
2550 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
2551 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
2554 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
2555 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
2559 string(getComputeAsmShaderPreamble()) +
2561 "OpSource GLSL 430\n"
2562 "OpName %main \"main\"\n"
2563 "OpName %id \"gl_GlobalInvocationID\"\n"
2565 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2567 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2569 "%id = OpVariable %uvec3ptr Input\n"
2570 "%zero = OpConstant %i32 0\n"
2571 "%three = OpConstant %u32 3\n"
2572 "%constf5p5 = OpConstant %f32 5.5\n"
2573 "%constf20p5 = OpConstant %f32 20.5\n"
2574 "%constf1p75 = OpConstant %f32 1.75\n"
2575 "%constf8p5 = OpConstant %f32 8.5\n"
2576 "%constf6p5 = OpConstant %f32 6.5\n"
2578 "%main = OpFunction %void None %voidf\n"
2579 "%entry = OpLabel\n"
2580 "%idval = OpLoad %uvec3 %id\n"
2581 "%x = OpCompositeExtract %u32 %idval 0\n"
2582 "%selector = OpUMod %u32 %x %three\n"
2583 " OpSelectionMerge %phi None\n"
2584 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
2586 // Case 1 before OpPhi.
2587 "%case1 = OpLabel\n"
2590 "%default = OpLabel\n"
2594 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
2595 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2596 "%inval = OpLoad %f32 %inloc\n"
2597 "%add = OpFAdd %f32 %inval %operand\n"
2598 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2599 " OpStore %outloc %add\n"
2602 // Case 0 after OpPhi.
2603 "%case0 = OpLabel\n"
2607 // Case 2 after OpPhi.
2608 "%case2 = OpLabel\n"
2612 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2613 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
2614 spec1.numWorkGroups = IVec3(numElements, 1, 1);
2616 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
2619 string(getComputeAsmShaderPreamble()) +
2621 "OpName %main \"main\"\n"
2622 "OpName %id \"gl_GlobalInvocationID\"\n"
2624 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2626 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2628 "%id = OpVariable %uvec3ptr Input\n"
2629 "%zero = OpConstant %i32 0\n"
2630 "%one = OpConstant %i32 1\n"
2631 "%three = OpConstant %i32 3\n"
2632 "%constf6p5 = OpConstant %f32 6.5\n"
2634 "%main = OpFunction %void None %voidf\n"
2635 "%entry = OpLabel\n"
2636 "%idval = OpLoad %uvec3 %id\n"
2637 "%x = OpCompositeExtract %u32 %idval 0\n"
2638 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2639 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2640 "%inval = OpLoad %f32 %inloc\n"
2644 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
2645 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
2646 "%step_next = OpIAdd %i32 %step %one\n"
2647 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
2648 "%still_loop = OpSLessThan %bool %step %three\n"
2649 " OpLoopMerge %exit %phi None\n"
2650 " OpBranchConditional %still_loop %phi %exit\n"
2653 " OpStore %outloc %accum\n"
2656 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2657 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2658 spec2.numWorkGroups = IVec3(numElements, 1, 1);
2660 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
2663 string(getComputeAsmShaderPreamble()) +
2665 "OpName %main \"main\"\n"
2666 "OpName %id \"gl_GlobalInvocationID\"\n"
2668 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2670 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2672 "%f32ptr_f = OpTypePointer Function %f32\n"
2673 "%id = OpVariable %uvec3ptr Input\n"
2674 "%true = OpConstantTrue %bool\n"
2675 "%false = OpConstantFalse %bool\n"
2676 "%zero = OpConstant %i32 0\n"
2677 "%constf8p5 = OpConstant %f32 8.5\n"
2679 "%main = OpFunction %void None %voidf\n"
2680 "%entry = OpLabel\n"
2681 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
2682 "%idval = OpLoad %uvec3 %id\n"
2683 "%x = OpCompositeExtract %u32 %idval 0\n"
2684 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2685 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2686 "%a_init = OpLoad %f32 %inloc\n"
2687 "%b_init = OpLoad %f32 %b\n"
2691 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
2692 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
2693 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
2694 " OpLoopMerge %exit %phi None\n"
2695 " OpBranchConditional %still_loop %phi %exit\n"
2698 "%sub = OpFSub %f32 %a_next %b_next\n"
2699 " OpStore %outloc %sub\n"
2702 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2703 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
2704 spec3.numWorkGroups = IVec3(numElements, 1, 1);
2706 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
2708 return group.release();
2711 // Assembly code used for testing block order is based on GLSL source code:
2715 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2716 // float elements[];
2718 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2719 // float elements[];
2723 // uint x = gl_GlobalInvocationID.x;
2724 // output_data.elements[x] = input_data.elements[x];
2725 // if (x > uint(50)) {
2726 // switch (x % uint(3)) {
2727 // case 0: output_data.elements[x] += 1.5f; break;
2728 // case 1: output_data.elements[x] += 42.f; break;
2729 // case 2: output_data.elements[x] -= 27.f; break;
2733 // output_data.elements[x] = -input_data.elements[x];
2736 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
2738 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
2739 ComputeShaderSpec spec;
2740 de::Random rnd (deStringHash(group->getName()));
2741 const int numElements = 100;
2742 vector<float> inputFloats (numElements, 0);
2743 vector<float> outputFloats (numElements, 0);
2745 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2747 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2748 floorAll(inputFloats);
2750 for (size_t ndx = 0; ndx <= 50; ++ndx)
2751 outputFloats[ndx] = -inputFloats[ndx];
2753 for (size_t ndx = 51; ndx < numElements; ++ndx)
2757 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
2758 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
2759 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
2765 string(getComputeAsmShaderPreamble()) +
2767 "OpSource GLSL 430\n"
2768 "OpName %main \"main\"\n"
2769 "OpName %id \"gl_GlobalInvocationID\"\n"
2771 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2773 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2775 "%u32ptr = OpTypePointer Function %u32\n"
2776 "%u32ptr_input = OpTypePointer Input %u32\n"
2778 + string(getComputeAsmInputOutputBuffer()) +
2780 "%id = OpVariable %uvec3ptr Input\n"
2781 "%zero = OpConstant %i32 0\n"
2782 "%const3 = OpConstant %u32 3\n"
2783 "%const50 = OpConstant %u32 50\n"
2784 "%constf1p5 = OpConstant %f32 1.5\n"
2785 "%constf27 = OpConstant %f32 27.0\n"
2786 "%constf42 = OpConstant %f32 42.0\n"
2788 "%main = OpFunction %void None %voidf\n"
2791 "%entry = OpLabel\n"
2793 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
2794 "%xvar = OpVariable %u32ptr Function\n"
2795 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
2796 "%x = OpLoad %u32 %xptr\n"
2797 " OpStore %xvar %x\n"
2799 "%cmp = OpUGreaterThan %bool %x %const50\n"
2800 " OpSelectionMerge %if_merge None\n"
2801 " OpBranchConditional %cmp %if_true %if_false\n"
2803 // False branch for if-statement: placed in the middle of switch cases and before true branch.
2804 "%if_false = OpLabel\n"
2805 "%x_f = OpLoad %u32 %xvar\n"
2806 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
2807 "%inval_f = OpLoad %f32 %inloc_f\n"
2808 "%negate = OpFNegate %f32 %inval_f\n"
2809 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
2810 " OpStore %outloc_f %negate\n"
2811 " OpBranch %if_merge\n"
2813 // Merge block for if-statement: placed in the middle of true and false branch.
2814 "%if_merge = OpLabel\n"
2817 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
2818 "%if_true = OpLabel\n"
2819 "%xval_t = OpLoad %u32 %xvar\n"
2820 "%mod = OpUMod %u32 %xval_t %const3\n"
2821 " OpSelectionMerge %switch_merge None\n"
2822 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
2824 // Merge block for switch-statement: placed before the case
2825 // bodies. But it must follow OpSwitch which dominates it.
2826 "%switch_merge = OpLabel\n"
2827 " OpBranch %if_merge\n"
2829 // Case 1 for switch-statement: placed before case 0.
2830 // It must follow the OpSwitch that dominates it.
2831 "%case1 = OpLabel\n"
2832 "%x_1 = OpLoad %u32 %xvar\n"
2833 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
2834 "%inval_1 = OpLoad %f32 %inloc_1\n"
2835 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
2836 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
2837 " OpStore %outloc_1 %addf42\n"
2838 " OpBranch %switch_merge\n"
2840 // Case 2 for switch-statement.
2841 "%case2 = OpLabel\n"
2842 "%x_2 = OpLoad %u32 %xvar\n"
2843 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
2844 "%inval_2 = OpLoad %f32 %inloc_2\n"
2845 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
2846 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
2847 " OpStore %outloc_2 %subf27\n"
2848 " OpBranch %switch_merge\n"
2850 // Default case for switch-statement: placed in the middle of normal cases.
2851 "%default = OpLabel\n"
2852 " OpBranch %switch_merge\n"
2854 // Case 0 for switch-statement: out of order.
2855 "%case0 = OpLabel\n"
2856 "%x_0 = OpLoad %u32 %xvar\n"
2857 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
2858 "%inval_0 = OpLoad %f32 %inloc_0\n"
2859 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
2860 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
2861 " OpStore %outloc_0 %addf1p5\n"
2862 " OpBranch %switch_merge\n"
2865 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2866 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2867 spec.numWorkGroups = IVec3(numElements, 1, 1);
2869 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
2871 return group.release();
2874 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
2876 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
2877 ComputeShaderSpec spec1;
2878 ComputeShaderSpec spec2;
2879 de::Random rnd (deStringHash(group->getName()));
2880 const int numElements = 100;
2881 vector<float> inputFloats (numElements, 0);
2882 vector<float> outputFloats1 (numElements, 0);
2883 vector<float> outputFloats2 (numElements, 0);
2884 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
2886 for (size_t ndx = 0; ndx < numElements; ++ndx)
2888 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
2889 outputFloats2[ndx] = -inputFloats[ndx];
2892 const string assembly(
2893 "OpCapability Shader\n"
2894 "OpCapability ClipDistance\n"
2895 "OpMemoryModel Logical GLSL450\n"
2896 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
2897 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
2898 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
2899 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
2900 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
2901 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
2903 "OpName %comp_main1 \"entrypoint1\"\n"
2904 "OpName %comp_main2 \"entrypoint2\"\n"
2905 "OpName %vert_main \"entrypoint2\"\n"
2906 "OpName %id \"gl_GlobalInvocationID\"\n"
2907 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
2908 "OpName %vertexIndex \"gl_VertexIndex\"\n"
2909 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
2910 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
2911 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
2912 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
2914 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2915 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
2916 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
2917 "OpDecorate %vert_builtin_st Block\n"
2918 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
2919 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
2920 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
2922 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2924 "%zero = OpConstant %i32 0\n"
2925 "%one = OpConstant %u32 1\n"
2926 "%c_f32_1 = OpConstant %f32 1\n"
2928 "%i32inputptr = OpTypePointer Input %i32\n"
2929 "%vec4 = OpTypeVector %f32 4\n"
2930 "%vec4ptr = OpTypePointer Output %vec4\n"
2931 "%f32arr1 = OpTypeArray %f32 %one\n"
2932 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
2933 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
2934 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
2936 "%id = OpVariable %uvec3ptr Input\n"
2937 "%vertexIndex = OpVariable %i32inputptr Input\n"
2938 "%instanceIndex = OpVariable %i32inputptr Input\n"
2939 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
2941 // gl_Position = vec4(1.);
2942 "%vert_main = OpFunction %void None %voidf\n"
2943 "%vert_entry = OpLabel\n"
2944 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
2945 " OpStore %position %c_vec4_1\n"
2950 "%comp_main1 = OpFunction %void None %voidf\n"
2951 "%comp1_entry = OpLabel\n"
2952 "%idval1 = OpLoad %uvec3 %id\n"
2953 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
2954 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
2955 "%inval1 = OpLoad %f32 %inloc1\n"
2956 "%add = OpFAdd %f32 %inval1 %inval1\n"
2957 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
2958 " OpStore %outloc1 %add\n"
2963 "%comp_main2 = OpFunction %void None %voidf\n"
2964 "%comp2_entry = OpLabel\n"
2965 "%idval2 = OpLoad %uvec3 %id\n"
2966 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
2967 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
2968 "%inval2 = OpLoad %f32 %inloc2\n"
2969 "%neg = OpFNegate %f32 %inval2\n"
2970 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
2971 " OpStore %outloc2 %neg\n"
2973 " OpFunctionEnd\n");
2975 spec1.assembly = assembly;
2976 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2977 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
2978 spec1.numWorkGroups = IVec3(numElements, 1, 1);
2979 spec1.entryPoint = "entrypoint1";
2981 spec2.assembly = assembly;
2982 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2983 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2984 spec2.numWorkGroups = IVec3(numElements, 1, 1);
2985 spec2.entryPoint = "entrypoint2";
2987 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
2988 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
2990 return group.release();
2993 inline std::string makeLongUTF8String (size_t num4ByteChars)
2995 // An example of a longest valid UTF-8 character. Be explicit about the
2996 // character type because Microsoft compilers can otherwise interpret the
2997 // character string as being over wide (16-bit) characters. Ideally, we
2998 // would just use a C++11 UTF-8 string literal, but we want to support older
2999 // Microsoft compilers.
3000 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
3001 std::string longString;
3002 longString.reserve(num4ByteChars * 4);
3003 for (size_t count = 0; count < num4ByteChars; count++)
3005 longString += earthAfrica;
3010 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
3012 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
3013 vector<CaseParameter> cases;
3014 de::Random rnd (deStringHash(group->getName()));
3015 const int numElements = 100;
3016 vector<float> positiveFloats (numElements, 0);
3017 vector<float> negativeFloats (numElements, 0);
3018 const StringTemplate shaderTemplate (
3019 "OpCapability Shader\n"
3020 "OpMemoryModel Logical GLSL450\n"
3022 "OpEntryPoint GLCompute %main \"main\" %id\n"
3023 "OpExecutionMode %main LocalSize 1 1 1\n"
3027 "OpName %main \"main\"\n"
3028 "OpName %id \"gl_GlobalInvocationID\"\n"
3030 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3032 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3034 "%id = OpVariable %uvec3ptr Input\n"
3035 "%zero = OpConstant %i32 0\n"
3037 "%main = OpFunction %void None %voidf\n"
3038 "%label = OpLabel\n"
3039 "%idval = OpLoad %uvec3 %id\n"
3040 "%x = OpCompositeExtract %u32 %idval 0\n"
3041 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3042 "%inval = OpLoad %f32 %inloc\n"
3043 "%neg = OpFNegate %f32 %inval\n"
3044 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3045 " OpStore %outloc %neg\n"
3047 " OpFunctionEnd\n");
3049 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
3050 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
3051 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
3052 "OpSource GLSL 430 %fname"));
3053 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
3054 "OpSource GLSL 430 %fname"));
3055 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
3056 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
3057 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
3058 "OpSource GLSL 430 %fname \"\""));
3059 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
3060 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
3061 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
3062 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
3063 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
3064 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
3065 "OpSourceContinued \"id main() {}\""));
3066 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
3067 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3068 "OpSourceContinued \"\""));
3069 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
3070 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3071 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
3072 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
3073 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3074 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
3075 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
3076 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
3077 "OpSourceContinued \"void\"\n"
3078 "OpSourceContinued \"main()\"\n"
3079 "OpSourceContinued \"{}\""));
3080 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
3081 "OpSource GLSL 430 %fname \"\"\n"
3082 "OpSourceContinued \"#version 430\nvoid main() {}\""));
3084 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3086 for (size_t ndx = 0; ndx < numElements; ++ndx)
3087 negativeFloats[ndx] = -positiveFloats[ndx];
3089 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3091 map<string, string> specializations;
3092 ComputeShaderSpec spec;
3094 specializations["SOURCE"] = cases[caseNdx].param;
3095 spec.assembly = shaderTemplate.specialize(specializations);
3096 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3097 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3098 spec.numWorkGroups = IVec3(numElements, 1, 1);
3100 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3103 return group.release();
3106 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
3108 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
3109 vector<CaseParameter> cases;
3110 de::Random rnd (deStringHash(group->getName()));
3111 const int numElements = 100;
3112 vector<float> inputFloats (numElements, 0);
3113 vector<float> outputFloats (numElements, 0);
3114 const StringTemplate shaderTemplate (
3115 string(getComputeAsmShaderPreamble()) +
3117 "OpSourceExtension \"${EXTENSION}\"\n"
3119 "OpName %main \"main\"\n"
3120 "OpName %id \"gl_GlobalInvocationID\"\n"
3122 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3124 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3126 "%id = OpVariable %uvec3ptr Input\n"
3127 "%zero = OpConstant %i32 0\n"
3129 "%main = OpFunction %void None %voidf\n"
3130 "%label = OpLabel\n"
3131 "%idval = OpLoad %uvec3 %id\n"
3132 "%x = OpCompositeExtract %u32 %idval 0\n"
3133 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3134 "%inval = OpLoad %f32 %inloc\n"
3135 "%neg = OpFNegate %f32 %inval\n"
3136 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3137 " OpStore %outloc %neg\n"
3139 " OpFunctionEnd\n");
3141 cases.push_back(CaseParameter("empty_extension", ""));
3142 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
3143 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
3144 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
3145 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
3147 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
3149 for (size_t ndx = 0; ndx < numElements; ++ndx)
3150 outputFloats[ndx] = -inputFloats[ndx];
3152 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3154 map<string, string> specializations;
3155 ComputeShaderSpec spec;
3157 specializations["EXTENSION"] = cases[caseNdx].param;
3158 spec.assembly = shaderTemplate.specialize(specializations);
3159 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3160 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3161 spec.numWorkGroups = IVec3(numElements, 1, 1);
3163 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3166 return group.release();
3169 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
3170 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
3172 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
3173 vector<CaseParameter> cases;
3174 de::Random rnd (deStringHash(group->getName()));
3175 const int numElements = 100;
3176 vector<float> positiveFloats (numElements, 0);
3177 vector<float> negativeFloats (numElements, 0);
3178 const StringTemplate shaderTemplate (
3179 string(getComputeAsmShaderPreamble()) +
3181 "OpSource GLSL 430\n"
3182 "OpName %main \"main\"\n"
3183 "OpName %id \"gl_GlobalInvocationID\"\n"
3185 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3187 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3188 "%uvec2 = OpTypeVector %u32 2\n"
3189 "%bvec3 = OpTypeVector %bool 3\n"
3190 "%fvec4 = OpTypeVector %f32 4\n"
3191 "%fmat33 = OpTypeMatrix %fvec3 3\n"
3192 "%const100 = OpConstant %u32 100\n"
3193 "%uarr100 = OpTypeArray %i32 %const100\n"
3194 "%struct = OpTypeStruct %f32 %i32 %u32\n"
3195 "%pointer = OpTypePointer Function %i32\n"
3196 + string(getComputeAsmInputOutputBuffer()) +
3198 "%null = OpConstantNull ${TYPE}\n"
3200 "%id = OpVariable %uvec3ptr Input\n"
3201 "%zero = OpConstant %i32 0\n"
3203 "%main = OpFunction %void None %voidf\n"
3204 "%label = OpLabel\n"
3205 "%idval = OpLoad %uvec3 %id\n"
3206 "%x = OpCompositeExtract %u32 %idval 0\n"
3207 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3208 "%inval = OpLoad %f32 %inloc\n"
3209 "%neg = OpFNegate %f32 %inval\n"
3210 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3211 " OpStore %outloc %neg\n"
3213 " OpFunctionEnd\n");
3215 cases.push_back(CaseParameter("bool", "%bool"));
3216 cases.push_back(CaseParameter("sint32", "%i32"));
3217 cases.push_back(CaseParameter("uint32", "%u32"));
3218 cases.push_back(CaseParameter("float32", "%f32"));
3219 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
3220 cases.push_back(CaseParameter("vec3bool", "%bvec3"));
3221 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
3222 cases.push_back(CaseParameter("matrix", "%fmat33"));
3223 cases.push_back(CaseParameter("array", "%uarr100"));
3224 cases.push_back(CaseParameter("struct", "%struct"));
3225 cases.push_back(CaseParameter("pointer", "%pointer"));
3227 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3229 for (size_t ndx = 0; ndx < numElements; ++ndx)
3230 negativeFloats[ndx] = -positiveFloats[ndx];
3232 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3234 map<string, string> specializations;
3235 ComputeShaderSpec spec;
3237 specializations["TYPE"] = cases[caseNdx].param;
3238 spec.assembly = shaderTemplate.specialize(specializations);
3239 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3240 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3241 spec.numWorkGroups = IVec3(numElements, 1, 1);
3243 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3246 return group.release();
3249 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
3250 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
3252 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
3253 vector<CaseParameter> cases;
3254 de::Random rnd (deStringHash(group->getName()));
3255 const int numElements = 100;
3256 vector<float> positiveFloats (numElements, 0);
3257 vector<float> negativeFloats (numElements, 0);
3258 const StringTemplate shaderTemplate (
3259 string(getComputeAsmShaderPreamble()) +
3261 "OpSource GLSL 430\n"
3262 "OpName %main \"main\"\n"
3263 "OpName %id \"gl_GlobalInvocationID\"\n"
3265 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3267 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3269 "%id = OpVariable %uvec3ptr Input\n"
3270 "%zero = OpConstant %i32 0\n"
3274 "%main = OpFunction %void None %voidf\n"
3275 "%label = OpLabel\n"
3276 "%idval = OpLoad %uvec3 %id\n"
3277 "%x = OpCompositeExtract %u32 %idval 0\n"
3278 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3279 "%inval = OpLoad %f32 %inloc\n"
3280 "%neg = OpFNegate %f32 %inval\n"
3281 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3282 " OpStore %outloc %neg\n"
3284 " OpFunctionEnd\n");
3286 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
3287 "%const = OpConstantComposite %uvec3 %five %zero %five"));
3288 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
3289 "%ten = OpConstant %f32 10.\n"
3290 "%fzero = OpConstant %f32 0.\n"
3291 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
3292 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
3293 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
3294 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
3295 "%fzero = OpConstant %f32 0.\n"
3296 "%one = OpConstant %f32 1.\n"
3297 "%point5 = OpConstant %f32 0.5\n"
3298 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
3299 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
3300 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
3301 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
3302 "%st2 = OpTypeStruct %i32 %i32\n"
3303 "%struct = OpTypeStruct %st1 %st2\n"
3304 "%point5 = OpConstant %f32 0.5\n"
3305 "%one = OpConstant %u32 1\n"
3306 "%ten = OpConstant %i32 10\n"
3307 "%st1val = OpConstantComposite %st1 %one %point5\n"
3308 "%st2val = OpConstantComposite %st2 %ten %ten\n"
3309 "%const = OpConstantComposite %struct %st1val %st2val"));
3311 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3313 for (size_t ndx = 0; ndx < numElements; ++ndx)
3314 negativeFloats[ndx] = -positiveFloats[ndx];
3316 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3318 map<string, string> specializations;
3319 ComputeShaderSpec spec;
3321 specializations["CONSTANT"] = cases[caseNdx].param;
3322 spec.assembly = shaderTemplate.specialize(specializations);
3323 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3324 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3325 spec.numWorkGroups = IVec3(numElements, 1, 1);
3327 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3330 return group.release();
3333 // Creates a floating point number with the given exponent, and significand
3334 // bits set. It can only create normalized numbers. Only the least significant
3335 // 24 bits of the significand will be examined. The final bit of the
3336 // significand will also be ignored. This allows alignment to be written
3337 // similarly to C99 hex-floats.
3338 // For example if you wanted to write 0x1.7f34p-12 you would call
3339 // constructNormalizedFloat(-12, 0x7f3400)
3340 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
3344 for (deInt32 idx = 0; idx < 23; ++idx)
3346 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
3350 return std::ldexp(f, exponent);
3353 // Compare instruction for the OpQuantizeF16 compute exact case.
3354 // Returns true if the output is what is expected from the test case.
3355 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
3357 if (outputAllocs.size() != 1)
3360 // We really just need this for size because we cannot compare Nans.
3361 const BufferSp& expectedOutput = expectedOutputs[0];
3362 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
3364 if (expectedOutput->getNumBytes() != 4*sizeof(float)) {
3368 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
3369 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
3374 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
3375 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
3380 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
3381 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
3386 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
3387 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
3394 // Checks that every output from a test-case is a float NaN.
3395 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
3397 if (outputAllocs.size() != 1)
3400 // We really just need this for size because we cannot compare Nans.
3401 const BufferSp& expectedOutput = expectedOutputs[0];
3402 const float* output_as_float = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
3404 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
3406 if (!deFloatIsNaN(output_as_float[idx]))
3415 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
3416 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
3418 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
3420 const std::string shader (
3421 string(getComputeAsmShaderPreamble()) +
3423 "OpSource GLSL 430\n"
3424 "OpName %main \"main\"\n"
3425 "OpName %id \"gl_GlobalInvocationID\"\n"
3427 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3429 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3431 "%id = OpVariable %uvec3ptr Input\n"
3432 "%zero = OpConstant %i32 0\n"
3434 "%main = OpFunction %void None %voidf\n"
3435 "%label = OpLabel\n"
3436 "%idval = OpLoad %uvec3 %id\n"
3437 "%x = OpCompositeExtract %u32 %idval 0\n"
3438 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3439 "%inval = OpLoad %f32 %inloc\n"
3440 "%quant = OpQuantizeToF16 %f32 %inval\n"
3441 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3442 " OpStore %outloc %quant\n"
3444 " OpFunctionEnd\n");
3447 ComputeShaderSpec spec;
3448 const deUint32 numElements = 100;
3449 vector<float> infinities;
3450 vector<float> results;
3452 infinities.reserve(numElements);
3453 results.reserve(numElements);
3455 for (size_t idx = 0; idx < numElements; ++idx)
3460 infinities.push_back(std::numeric_limits<float>::infinity());
3461 results.push_back(std::numeric_limits<float>::infinity());
3464 infinities.push_back(-std::numeric_limits<float>::infinity());
3465 results.push_back(-std::numeric_limits<float>::infinity());
3468 infinities.push_back(std::ldexp(1.0f, 16));
3469 results.push_back(std::numeric_limits<float>::infinity());
3472 infinities.push_back(std::ldexp(-1.0f, 32));
3473 results.push_back(-std::numeric_limits<float>::infinity());
3478 spec.assembly = shader;
3479 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
3480 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
3481 spec.numWorkGroups = IVec3(numElements, 1, 1);
3483 group->addChild(new SpvAsmComputeShaderCase(
3484 testCtx, "infinities", "Check that infinities propagated and created", spec));
3488 ComputeShaderSpec spec;
3490 const deUint32 numElements = 100;
3492 nans.reserve(numElements);
3494 for (size_t idx = 0; idx < numElements; ++idx)
3498 nans.push_back(std::numeric_limits<float>::quiet_NaN());
3502 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
3506 spec.assembly = shader;
3507 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
3508 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
3509 spec.numWorkGroups = IVec3(numElements, 1, 1);
3510 spec.verifyIO = &compareNan;
3512 group->addChild(new SpvAsmComputeShaderCase(
3513 testCtx, "propagated_nans", "Check that nans are propagated", spec));
3517 ComputeShaderSpec spec;
3518 vector<float> small;
3519 vector<float> zeros;
3520 const deUint32 numElements = 100;
3522 small.reserve(numElements);
3523 zeros.reserve(numElements);
3525 for (size_t idx = 0; idx < numElements; ++idx)
3530 small.push_back(0.f);
3531 zeros.push_back(0.f);
3534 small.push_back(-0.f);
3535 zeros.push_back(-0.f);
3538 small.push_back(std::ldexp(1.0f, -16));
3539 zeros.push_back(0.f);
3542 small.push_back(std::ldexp(-1.0f, -32));
3543 zeros.push_back(-0.f);
3546 small.push_back(std::ldexp(1.0f, -127));
3547 zeros.push_back(0.f);
3550 small.push_back(-std::ldexp(1.0f, -128));
3551 zeros.push_back(-0.f);
3556 spec.assembly = shader;
3557 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
3558 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
3559 spec.numWorkGroups = IVec3(numElements, 1, 1);
3561 group->addChild(new SpvAsmComputeShaderCase(
3562 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
3566 ComputeShaderSpec spec;
3567 vector<float> exact;
3568 const deUint32 numElements = 200;
3570 exact.reserve(numElements);
3572 for (size_t idx = 0; idx < numElements; ++idx)
3573 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
3575 spec.assembly = shader;
3576 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
3577 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
3578 spec.numWorkGroups = IVec3(numElements, 1, 1);
3580 group->addChild(new SpvAsmComputeShaderCase(
3581 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
3585 ComputeShaderSpec spec;
3586 vector<float> inputs;
3587 const deUint32 numElements = 4;
3589 inputs.push_back(constructNormalizedFloat(8, 0x300300));
3590 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
3591 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
3592 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
3594 spec.assembly = shader;
3595 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
3596 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3597 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
3598 spec.numWorkGroups = IVec3(numElements, 1, 1);
3600 group->addChild(new SpvAsmComputeShaderCase(
3601 testCtx, "rounded", "Check that are rounded when needed", spec));
3604 return group.release();
3607 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
3609 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
3611 const std::string shader (
3612 string(getComputeAsmShaderPreamble()) +
3614 "OpName %main \"main\"\n"
3615 "OpName %id \"gl_GlobalInvocationID\"\n"
3617 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3619 "OpDecorate %sc_0 SpecId 0\n"
3620 "OpDecorate %sc_1 SpecId 1\n"
3621 "OpDecorate %sc_2 SpecId 2\n"
3622 "OpDecorate %sc_3 SpecId 3\n"
3623 "OpDecorate %sc_4 SpecId 4\n"
3624 "OpDecorate %sc_5 SpecId 5\n"
3626 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3628 "%id = OpVariable %uvec3ptr Input\n"
3629 "%zero = OpConstant %i32 0\n"
3630 "%c_u32_6 = OpConstant %u32 6\n"
3632 "%sc_0 = OpSpecConstant %f32 0.\n"
3633 "%sc_1 = OpSpecConstant %f32 0.\n"
3634 "%sc_2 = OpSpecConstant %f32 0.\n"
3635 "%sc_3 = OpSpecConstant %f32 0.\n"
3636 "%sc_4 = OpSpecConstant %f32 0.\n"
3637 "%sc_5 = OpSpecConstant %f32 0.\n"
3639 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
3640 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
3641 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
3642 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
3643 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
3644 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
3646 "%main = OpFunction %void None %voidf\n"
3647 "%label = OpLabel\n"
3648 "%idval = OpLoad %uvec3 %id\n"
3649 "%x = OpCompositeExtract %u32 %idval 0\n"
3650 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3651 "%selector = OpUMod %u32 %x %c_u32_6\n"
3652 " OpSelectionMerge %exit None\n"
3653 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
3655 "%case0 = OpLabel\n"
3656 " OpStore %outloc %sc_0_quant\n"
3659 "%case1 = OpLabel\n"
3660 " OpStore %outloc %sc_1_quant\n"
3663 "%case2 = OpLabel\n"
3664 " OpStore %outloc %sc_2_quant\n"
3667 "%case3 = OpLabel\n"
3668 " OpStore %outloc %sc_3_quant\n"
3671 "%case4 = OpLabel\n"
3672 " OpStore %outloc %sc_4_quant\n"
3675 "%case5 = OpLabel\n"
3676 " OpStore %outloc %sc_5_quant\n"
3682 " OpFunctionEnd\n");
3685 ComputeShaderSpec spec;
3686 const deUint8 numCases = 4;
3687 vector<float> inputs (numCases, 0.f);
3688 vector<float> outputs;
3690 spec.assembly = shader;
3691 spec.numWorkGroups = IVec3(numCases, 1, 1);
3693 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
3694 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
3695 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
3696 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
3698 outputs.push_back(std::numeric_limits<float>::infinity());
3699 outputs.push_back(-std::numeric_limits<float>::infinity());
3700 outputs.push_back(std::numeric_limits<float>::infinity());
3701 outputs.push_back(-std::numeric_limits<float>::infinity());
3703 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3704 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3706 group->addChild(new SpvAsmComputeShaderCase(
3707 testCtx, "infinities", "Check that infinities propagated and created", spec));
3711 ComputeShaderSpec spec;
3712 const deUint8 numCases = 2;
3713 vector<float> inputs (numCases, 0.f);
3714 vector<float> outputs;
3716 spec.assembly = shader;
3717 spec.numWorkGroups = IVec3(numCases, 1, 1);
3718 spec.verifyIO = &compareNan;
3720 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
3721 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
3723 for (deUint8 idx = 0; idx < numCases; ++idx)
3724 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
3726 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3727 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3729 group->addChild(new SpvAsmComputeShaderCase(
3730 testCtx, "propagated_nans", "Check that nans are propagated", spec));
3734 ComputeShaderSpec spec;
3735 const deUint8 numCases = 6;
3736 vector<float> inputs (numCases, 0.f);
3737 vector<float> outputs;
3739 spec.assembly = shader;
3740 spec.numWorkGroups = IVec3(numCases, 1, 1);
3742 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
3743 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
3744 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
3745 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
3746 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
3747 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
3749 outputs.push_back(0.f);
3750 outputs.push_back(-0.f);
3751 outputs.push_back(0.f);
3752 outputs.push_back(-0.f);
3753 outputs.push_back(0.f);
3754 outputs.push_back(-0.f);
3756 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3757 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3759 group->addChild(new SpvAsmComputeShaderCase(
3760 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
3764 ComputeShaderSpec spec;
3765 const deUint8 numCases = 6;
3766 vector<float> inputs (numCases, 0.f);
3767 vector<float> outputs;
3769 spec.assembly = shader;
3770 spec.numWorkGroups = IVec3(numCases, 1, 1);
3772 for (deUint8 idx = 0; idx < 6; ++idx)
3774 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
3775 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
3776 outputs.push_back(f);
3779 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3780 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3782 group->addChild(new SpvAsmComputeShaderCase(
3783 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
3787 ComputeShaderSpec spec;
3788 const deUint8 numCases = 4;
3789 vector<float> inputs (numCases, 0.f);
3790 vector<float> outputs;
3792 spec.assembly = shader;
3793 spec.numWorkGroups = IVec3(numCases, 1, 1);
3794 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
3796 outputs.push_back(constructNormalizedFloat(8, 0x300300));
3797 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
3798 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
3799 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
3801 for (deUint8 idx = 0; idx < numCases; ++idx)
3802 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
3804 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3805 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3807 group->addChild(new SpvAsmComputeShaderCase(
3808 testCtx, "rounded", "Check that are rounded when needed", spec));
3811 return group.release();
3814 // Checks that constant null/composite values can be used in computation.
3815 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
3817 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
3818 ComputeShaderSpec spec;
3819 de::Random rnd (deStringHash(group->getName()));
3820 const int numElements = 100;
3821 vector<float> positiveFloats (numElements, 0);
3822 vector<float> negativeFloats (numElements, 0);
3824 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3826 for (size_t ndx = 0; ndx < numElements; ++ndx)
3827 negativeFloats[ndx] = -positiveFloats[ndx];
3830 "OpCapability Shader\n"
3831 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
3832 "OpMemoryModel Logical GLSL450\n"
3833 "OpEntryPoint GLCompute %main \"main\" %id\n"
3834 "OpExecutionMode %main LocalSize 1 1 1\n"
3836 "OpSource GLSL 430\n"
3837 "OpName %main \"main\"\n"
3838 "OpName %id \"gl_GlobalInvocationID\"\n"
3840 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3842 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3844 "%fmat = OpTypeMatrix %fvec3 3\n"
3845 "%ten = OpConstant %u32 10\n"
3846 "%f32arr10 = OpTypeArray %f32 %ten\n"
3847 "%fst = OpTypeStruct %f32 %f32\n"
3849 + string(getComputeAsmInputOutputBuffer()) +
3851 "%id = OpVariable %uvec3ptr Input\n"
3852 "%zero = OpConstant %i32 0\n"
3854 // Create a bunch of null values
3855 "%unull = OpConstantNull %u32\n"
3856 "%fnull = OpConstantNull %f32\n"
3857 "%vnull = OpConstantNull %fvec3\n"
3858 "%mnull = OpConstantNull %fmat\n"
3859 "%anull = OpConstantNull %f32arr10\n"
3860 "%snull = OpConstantComposite %fst %fnull %fnull\n"
3862 "%main = OpFunction %void None %voidf\n"
3863 "%label = OpLabel\n"
3864 "%idval = OpLoad %uvec3 %id\n"
3865 "%x = OpCompositeExtract %u32 %idval 0\n"
3866 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3867 "%inval = OpLoad %f32 %inloc\n"
3868 "%neg = OpFNegate %f32 %inval\n"
3870 // Get the abs() of (a certain element of) those null values
3871 "%unull_cov = OpConvertUToF %f32 %unull\n"
3872 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
3873 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
3874 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
3875 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
3876 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
3877 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
3878 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
3879 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
3880 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
3881 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
3884 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
3885 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
3886 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
3887 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
3888 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
3889 "%final = OpFAdd %f32 %add5 %snull_abs\n"
3891 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3892 " OpStore %outloc %final\n" // write to output
3895 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3896 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3897 spec.numWorkGroups = IVec3(numElements, 1, 1);
3899 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
3901 return group.release();
3904 // Assembly code used for testing loop control is based on GLSL source code:
3907 // layout(std140, set = 0, binding = 0) readonly buffer Input {
3908 // float elements[];
3910 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
3911 // float elements[];
3915 // uint x = gl_GlobalInvocationID.x;
3916 // output_data.elements[x] = input_data.elements[x];
3917 // for (uint i = 0; i < 4; ++i)
3918 // output_data.elements[x] += 1.f;
3920 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
3922 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
3923 vector<CaseParameter> cases;
3924 de::Random rnd (deStringHash(group->getName()));
3925 const int numElements = 100;
3926 vector<float> inputFloats (numElements, 0);
3927 vector<float> outputFloats (numElements, 0);
3928 const StringTemplate shaderTemplate (
3929 string(getComputeAsmShaderPreamble()) +
3931 "OpSource GLSL 430\n"
3932 "OpName %main \"main\"\n"
3933 "OpName %id \"gl_GlobalInvocationID\"\n"
3935 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3937 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3939 "%u32ptr = OpTypePointer Function %u32\n"
3941 "%id = OpVariable %uvec3ptr Input\n"
3942 "%zero = OpConstant %i32 0\n"
3943 "%uzero = OpConstant %u32 0\n"
3944 "%one = OpConstant %i32 1\n"
3945 "%constf1 = OpConstant %f32 1.0\n"
3946 "%four = OpConstant %u32 4\n"
3948 "%main = OpFunction %void None %voidf\n"
3949 "%entry = OpLabel\n"
3950 "%i = OpVariable %u32ptr Function\n"
3951 " OpStore %i %uzero\n"
3953 "%idval = OpLoad %uvec3 %id\n"
3954 "%x = OpCompositeExtract %u32 %idval 0\n"
3955 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3956 "%inval = OpLoad %f32 %inloc\n"
3957 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3958 " OpStore %outloc %inval\n"
3959 " OpBranch %loop_entry\n"
3961 "%loop_entry = OpLabel\n"
3962 "%i_val = OpLoad %u32 %i\n"
3963 "%cmp_lt = OpULessThan %bool %i_val %four\n"
3964 " OpLoopMerge %loop_merge %loop_body ${CONTROL}\n"
3965 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
3966 "%loop_body = OpLabel\n"
3967 "%outval = OpLoad %f32 %outloc\n"
3968 "%addf1 = OpFAdd %f32 %outval %constf1\n"
3969 " OpStore %outloc %addf1\n"
3970 "%new_i = OpIAdd %u32 %i_val %one\n"
3971 " OpStore %i %new_i\n"
3972 " OpBranch %loop_entry\n"
3973 "%loop_merge = OpLabel\n"
3975 " OpFunctionEnd\n");
3977 cases.push_back(CaseParameter("none", "None"));
3978 cases.push_back(CaseParameter("unroll", "Unroll"));
3979 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
3980 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
3982 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3984 for (size_t ndx = 0; ndx < numElements; ++ndx)
3985 outputFloats[ndx] = inputFloats[ndx] + 4.f;
3987 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3989 map<string, string> specializations;
3990 ComputeShaderSpec spec;
3992 specializations["CONTROL"] = cases[caseNdx].param;
3993 spec.assembly = shaderTemplate.specialize(specializations);
3994 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3995 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3996 spec.numWorkGroups = IVec3(numElements, 1, 1);
3998 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4001 return group.release();
4004 // Assembly code used for testing selection control is based on GLSL source code:
4007 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4008 // float elements[];
4010 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4011 // float elements[];
4015 // uint x = gl_GlobalInvocationID.x;
4016 // float val = input_data.elements[x];
4018 // output_data.elements[x] = val + 1.f;
4020 // output_data.elements[x] = val - 1.f;
4022 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
4024 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
4025 vector<CaseParameter> cases;
4026 de::Random rnd (deStringHash(group->getName()));
4027 const int numElements = 100;
4028 vector<float> inputFloats (numElements, 0);
4029 vector<float> outputFloats (numElements, 0);
4030 const StringTemplate shaderTemplate (
4031 string(getComputeAsmShaderPreamble()) +
4033 "OpSource GLSL 430\n"
4034 "OpName %main \"main\"\n"
4035 "OpName %id \"gl_GlobalInvocationID\"\n"
4037 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4039 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4041 "%id = OpVariable %uvec3ptr Input\n"
4042 "%zero = OpConstant %i32 0\n"
4043 "%constf1 = OpConstant %f32 1.0\n"
4044 "%constf10 = OpConstant %f32 10.0\n"
4046 "%main = OpFunction %void None %voidf\n"
4047 "%entry = OpLabel\n"
4048 "%idval = OpLoad %uvec3 %id\n"
4049 "%x = OpCompositeExtract %u32 %idval 0\n"
4050 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4051 "%inval = OpLoad %f32 %inloc\n"
4052 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4053 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
4055 " OpSelectionMerge %if_end ${CONTROL}\n"
4056 " OpBranchConditional %cmp_gt %if_true %if_false\n"
4057 "%if_true = OpLabel\n"
4058 "%addf1 = OpFAdd %f32 %inval %constf1\n"
4059 " OpStore %outloc %addf1\n"
4060 " OpBranch %if_end\n"
4061 "%if_false = OpLabel\n"
4062 "%subf1 = OpFSub %f32 %inval %constf1\n"
4063 " OpStore %outloc %subf1\n"
4064 " OpBranch %if_end\n"
4065 "%if_end = OpLabel\n"
4067 " OpFunctionEnd\n");
4069 cases.push_back(CaseParameter("none", "None"));
4070 cases.push_back(CaseParameter("flatten", "Flatten"));
4071 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
4072 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
4074 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4076 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4077 floorAll(inputFloats);
4079 for (size_t ndx = 0; ndx < numElements; ++ndx)
4080 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
4082 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4084 map<string, string> specializations;
4085 ComputeShaderSpec spec;
4087 specializations["CONTROL"] = cases[caseNdx].param;
4088 spec.assembly = shaderTemplate.specialize(specializations);
4089 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4090 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4091 spec.numWorkGroups = IVec3(numElements, 1, 1);
4093 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4096 return group.release();
4099 // Assembly code used for testing function control is based on GLSL source code:
4103 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4104 // float elements[];
4106 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4107 // float elements[];
4110 // float const10() { return 10.f; }
4113 // uint x = gl_GlobalInvocationID.x;
4114 // output_data.elements[x] = input_data.elements[x] + const10();
4116 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
4118 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
4119 vector<CaseParameter> cases;
4120 de::Random rnd (deStringHash(group->getName()));
4121 const int numElements = 100;
4122 vector<float> inputFloats (numElements, 0);
4123 vector<float> outputFloats (numElements, 0);
4124 const StringTemplate shaderTemplate (
4125 string(getComputeAsmShaderPreamble()) +
4127 "OpSource GLSL 430\n"
4128 "OpName %main \"main\"\n"
4129 "OpName %func_const10 \"const10(\"\n"
4130 "OpName %id \"gl_GlobalInvocationID\"\n"
4132 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4134 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4136 "%f32f = OpTypeFunction %f32\n"
4137 "%id = OpVariable %uvec3ptr Input\n"
4138 "%zero = OpConstant %i32 0\n"
4139 "%constf10 = OpConstant %f32 10.0\n"
4141 "%main = OpFunction %void None %voidf\n"
4142 "%entry = OpLabel\n"
4143 "%idval = OpLoad %uvec3 %id\n"
4144 "%x = OpCompositeExtract %u32 %idval 0\n"
4145 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4146 "%inval = OpLoad %f32 %inloc\n"
4147 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
4148 "%fadd = OpFAdd %f32 %inval %ret_10\n"
4149 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4150 " OpStore %outloc %fadd\n"
4154 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
4155 "%label = OpLabel\n"
4156 " OpReturnValue %constf10\n"
4157 " OpFunctionEnd\n");
4159 cases.push_back(CaseParameter("none", "None"));
4160 cases.push_back(CaseParameter("inline", "Inline"));
4161 cases.push_back(CaseParameter("dont_inline", "DontInline"));
4162 cases.push_back(CaseParameter("pure", "Pure"));
4163 cases.push_back(CaseParameter("const", "Const"));
4164 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
4165 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
4166 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
4167 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
4169 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4171 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4172 floorAll(inputFloats);
4174 for (size_t ndx = 0; ndx < numElements; ++ndx)
4175 outputFloats[ndx] = inputFloats[ndx] + 10.f;
4177 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4179 map<string, string> specializations;
4180 ComputeShaderSpec spec;
4182 specializations["CONTROL"] = cases[caseNdx].param;
4183 spec.assembly = shaderTemplate.specialize(specializations);
4184 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4185 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4186 spec.numWorkGroups = IVec3(numElements, 1, 1);
4188 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4191 return group.release();
4194 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
4196 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
4197 vector<CaseParameter> cases;
4198 de::Random rnd (deStringHash(group->getName()));
4199 const int numElements = 100;
4200 vector<float> inputFloats (numElements, 0);
4201 vector<float> outputFloats (numElements, 0);
4202 const StringTemplate shaderTemplate (
4203 string(getComputeAsmShaderPreamble()) +
4205 "OpSource GLSL 430\n"
4206 "OpName %main \"main\"\n"
4207 "OpName %id \"gl_GlobalInvocationID\"\n"
4209 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4211 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4213 "%f32ptr_f = OpTypePointer Function %f32\n"
4215 "%id = OpVariable %uvec3ptr Input\n"
4216 "%zero = OpConstant %i32 0\n"
4217 "%four = OpConstant %i32 4\n"
4219 "%main = OpFunction %void None %voidf\n"
4220 "%label = OpLabel\n"
4221 "%copy = OpVariable %f32ptr_f Function\n"
4222 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
4223 "%x = OpCompositeExtract %u32 %idval 0\n"
4224 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4225 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4226 " OpCopyMemory %copy %inloc ${ACCESS}\n"
4227 "%val1 = OpLoad %f32 %copy\n"
4228 "%val2 = OpLoad %f32 %inloc\n"
4229 "%add = OpFAdd %f32 %val1 %val2\n"
4230 " OpStore %outloc %add ${ACCESS}\n"
4232 " OpFunctionEnd\n");
4234 cases.push_back(CaseParameter("null", ""));
4235 cases.push_back(CaseParameter("none", "None"));
4236 cases.push_back(CaseParameter("volatile", "Volatile"));
4237 cases.push_back(CaseParameter("aligned", "Aligned 4"));
4238 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
4239 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
4240 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
4242 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4244 for (size_t ndx = 0; ndx < numElements; ++ndx)
4245 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
4247 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4249 map<string, string> specializations;
4250 ComputeShaderSpec spec;
4252 specializations["ACCESS"] = cases[caseNdx].param;
4253 spec.assembly = shaderTemplate.specialize(specializations);
4254 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4255 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4256 spec.numWorkGroups = IVec3(numElements, 1, 1);
4258 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4261 return group.release();
4264 // Checks that we can get undefined values for various types, without exercising a computation with it.
4265 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
4267 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
4268 vector<CaseParameter> cases;
4269 de::Random rnd (deStringHash(group->getName()));
4270 const int numElements = 100;
4271 vector<float> positiveFloats (numElements, 0);
4272 vector<float> negativeFloats (numElements, 0);
4273 const StringTemplate shaderTemplate (
4274 string(getComputeAsmShaderPreamble()) +
4276 "OpSource GLSL 430\n"
4277 "OpName %main \"main\"\n"
4278 "OpName %id \"gl_GlobalInvocationID\"\n"
4280 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4282 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4283 "%uvec2 = OpTypeVector %u32 2\n"
4284 "%fvec4 = OpTypeVector %f32 4\n"
4285 "%fmat33 = OpTypeMatrix %fvec3 3\n"
4286 "%image = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
4287 "%sampler = OpTypeSampler\n"
4288 "%simage = OpTypeSampledImage %image\n"
4289 "%const100 = OpConstant %u32 100\n"
4290 "%uarr100 = OpTypeArray %i32 %const100\n"
4291 "%struct = OpTypeStruct %f32 %i32 %u32\n"
4292 "%pointer = OpTypePointer Function %i32\n"
4293 + string(getComputeAsmInputOutputBuffer()) +
4295 "%id = OpVariable %uvec3ptr Input\n"
4296 "%zero = OpConstant %i32 0\n"
4298 "%main = OpFunction %void None %voidf\n"
4299 "%label = OpLabel\n"
4301 "%undef = OpUndef ${TYPE}\n"
4303 "%idval = OpLoad %uvec3 %id\n"
4304 "%x = OpCompositeExtract %u32 %idval 0\n"
4306 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4307 "%inval = OpLoad %f32 %inloc\n"
4308 "%neg = OpFNegate %f32 %inval\n"
4309 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4310 " OpStore %outloc %neg\n"
4312 " OpFunctionEnd\n");
4314 cases.push_back(CaseParameter("bool", "%bool"));
4315 cases.push_back(CaseParameter("sint32", "%i32"));
4316 cases.push_back(CaseParameter("uint32", "%u32"));
4317 cases.push_back(CaseParameter("float32", "%f32"));
4318 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
4319 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
4320 cases.push_back(CaseParameter("matrix", "%fmat33"));
4321 cases.push_back(CaseParameter("image", "%image"));
4322 cases.push_back(CaseParameter("sampler", "%sampler"));
4323 cases.push_back(CaseParameter("sampledimage", "%simage"));
4324 cases.push_back(CaseParameter("array", "%uarr100"));
4325 cases.push_back(CaseParameter("runtimearray", "%f32arr"));
4326 cases.push_back(CaseParameter("struct", "%struct"));
4327 cases.push_back(CaseParameter("pointer", "%pointer"));
4329 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4331 for (size_t ndx = 0; ndx < numElements; ++ndx)
4332 negativeFloats[ndx] = -positiveFloats[ndx];
4334 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4336 map<string, string> specializations;
4337 ComputeShaderSpec spec;
4339 specializations["TYPE"] = cases[caseNdx].param;
4340 spec.assembly = shaderTemplate.specialize(specializations);
4341 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4342 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4343 spec.numWorkGroups = IVec3(numElements, 1, 1);
4345 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4348 return group.release();
4353 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
4355 struct NameCodePair { string name, code; };
4356 RGBA defaultColors[4];
4357 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
4358 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
4359 map<string, string> fragments = passthruFragments();
4360 const NameCodePair tests[] =
4362 {"unknown", "OpSource Unknown 321"},
4363 {"essl", "OpSource ESSL 310"},
4364 {"glsl", "OpSource GLSL 450"},
4365 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
4366 {"opencl_c", "OpSource OpenCL_C 120"},
4367 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
4368 {"file", opsourceGLSLWithFile},
4369 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
4370 // Longest possible source string: SPIR-V limits instructions to 65535
4371 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
4372 // contain 65530 UTF8 characters (one word each) plus one last word
4373 // containing 3 ASCII characters and \0.
4374 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
4377 getDefaultColors(defaultColors);
4378 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
4380 fragments["debug"] = tests[testNdx].code;
4381 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
4384 return opSourceTests.release();
4387 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
4389 struct NameCodePair { string name, code; };
4390 RGBA defaultColors[4];
4391 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
4392 map<string, string> fragments = passthruFragments();
4393 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
4394 const NameCodePair tests[] =
4396 {"empty", opsource + "OpSourceContinued \"\""},
4397 {"short", opsource + "OpSourceContinued \"abcde\""},
4398 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
4399 // Longest possible source string: SPIR-V limits instructions to 65535
4400 // words, of which the first one is OpSourceContinued/length; the rest
4401 // will contain 65533 UTF8 characters (one word each) plus one last word
4402 // containing 3 ASCII characters and \0.
4403 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
4406 getDefaultColors(defaultColors);
4407 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
4409 fragments["debug"] = tests[testNdx].code;
4410 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
4413 return opSourceTests.release();
4416 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
4418 RGBA defaultColors[4];
4419 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
4420 map<string, string> fragments;
4421 getDefaultColors(defaultColors);
4422 fragments["debug"] =
4423 "%name = OpString \"name\"\n";
4425 fragments["pre_main"] =
4428 "OpLine %name 1 1\n"
4430 "OpLine %name 1 1\n"
4431 "OpLine %name 1 1\n"
4432 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
4434 "OpLine %name 1 1\n"
4436 "OpLine %name 1 1\n"
4437 "OpLine %name 1 1\n"
4438 "%second_param1 = OpFunctionParameter %v4f32\n"
4441 "%label_secondfunction = OpLabel\n"
4443 "OpReturnValue %second_param1\n"
4448 fragments["testfun"] =
4449 // A %test_code function that returns its argument unchanged.
4452 "OpLine %name 1 1\n"
4453 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4455 "%param1 = OpFunctionParameter %v4f32\n"
4458 "%label_testfun = OpLabel\n"
4460 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
4461 "OpReturnValue %val1\n"
4463 "OpLine %name 1 1\n"
4466 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
4468 return opLineTests.release();
4472 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
4474 RGBA defaultColors[4];
4475 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
4476 map<string, string> fragments;
4477 std::vector<std::pair<std::string, std::string> > problemStrings;
4479 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
4480 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
4481 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
4482 getDefaultColors(defaultColors);
4484 fragments["debug"] =
4485 "%other_name = OpString \"other_name\"\n";
4487 fragments["pre_main"] =
4488 "OpLine %file_name 32 0\n"
4489 "OpLine %file_name 32 32\n"
4490 "OpLine %file_name 32 40\n"
4491 "OpLine %other_name 32 40\n"
4492 "OpLine %other_name 0 100\n"
4493 "OpLine %other_name 0 4294967295\n"
4494 "OpLine %other_name 4294967295 0\n"
4495 "OpLine %other_name 32 40\n"
4496 "OpLine %file_name 0 0\n"
4497 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
4498 "OpLine %file_name 1 0\n"
4499 "%second_param1 = OpFunctionParameter %v4f32\n"
4500 "OpLine %file_name 1 3\n"
4501 "OpLine %file_name 1 2\n"
4502 "%label_secondfunction = OpLabel\n"
4503 "OpLine %file_name 0 2\n"
4504 "OpReturnValue %second_param1\n"
4506 "OpLine %file_name 0 2\n"
4507 "OpLine %file_name 0 2\n";
4509 fragments["testfun"] =
4510 // A %test_code function that returns its argument unchanged.
4511 "OpLine %file_name 1 0\n"
4512 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4513 "OpLine %file_name 16 330\n"
4514 "%param1 = OpFunctionParameter %v4f32\n"
4515 "OpLine %file_name 14 442\n"
4516 "%label_testfun = OpLabel\n"
4517 "OpLine %file_name 11 1024\n"
4518 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
4519 "OpLine %file_name 2 97\n"
4520 "OpReturnValue %val1\n"
4522 "OpLine %file_name 5 32\n";
4524 for (size_t i = 0; i < problemStrings.size(); ++i)
4526 map<string, string> testFragments = fragments;
4527 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
4528 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
4531 return opLineTests.release();
4534 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
4536 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
4540 const char functionStart[] =
4541 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4542 "%param1 = OpFunctionParameter %v4f32\n"
4545 const char functionEnd[] =
4546 "OpReturnValue %transformed_param\n"
4549 struct NameConstantsCode
4556 NameConstantsCode tests[] =
4560 "%cnull = OpConstantNull %v4f32\n",
4561 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
4565 "%cnull = OpConstantNull %f32\n",
4566 "%vp = OpVariable %fp_v4f32 Function\n"
4567 "%v = OpLoad %v4f32 %vp\n"
4568 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
4569 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
4570 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
4571 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
4572 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
4576 "%cnull = OpConstantNull %bool\n",
4577 "%v = OpVariable %fp_v4f32 Function\n"
4578 " OpStore %v %param1\n"
4579 " OpSelectionMerge %false_label None\n"
4580 " OpBranchConditional %cnull %true_label %false_label\n"
4581 "%true_label = OpLabel\n"
4582 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
4583 " OpBranch %false_label\n"
4584 "%false_label = OpLabel\n"
4585 "%transformed_param = OpLoad %v4f32 %v\n"
4589 "%cnull = OpConstantNull %i32\n",
4590 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
4591 "%b = OpIEqual %bool %cnull %c_i32_0\n"
4592 " OpSelectionMerge %false_label None\n"
4593 " OpBranchConditional %b %true_label %false_label\n"
4594 "%true_label = OpLabel\n"
4595 " OpStore %v %param1\n"
4596 " OpBranch %false_label\n"
4597 "%false_label = OpLabel\n"
4598 "%transformed_param = OpLoad %v4f32 %v\n"
4602 "%stype = OpTypeStruct %f32 %v4f32\n"
4603 "%fp_stype = OpTypePointer Function %stype\n"
4604 "%cnull = OpConstantNull %stype\n",
4605 "%v = OpVariable %fp_stype Function %cnull\n"
4606 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
4607 "%f_val = OpLoad %v4f32 %f\n"
4608 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
4612 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
4613 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
4614 "%cnull = OpConstantNull %a4_v4f32\n",
4615 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
4616 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
4617 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
4618 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
4619 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
4620 "%f_val = OpLoad %v4f32 %f\n"
4621 "%f1_val = OpLoad %v4f32 %f1\n"
4622 "%f2_val = OpLoad %v4f32 %f2\n"
4623 "%f3_val = OpLoad %v4f32 %f3\n"
4624 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
4625 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
4626 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
4627 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
4631 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
4632 "%cnull = OpConstantNull %mat4x4_f32\n",
4633 // Our null matrix * any vector should result in a zero vector.
4634 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
4635 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
4639 getHalfColorsFullAlpha(colors);
4641 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
4643 map<string, string> fragments;
4644 fragments["pre_main"] = tests[testNdx].constants;
4645 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
4646 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
4648 return opConstantNullTests.release();
4650 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
4652 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
4653 RGBA inputColors[4];
4654 RGBA outputColors[4];
4657 const char functionStart[] =
4658 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4659 "%param1 = OpFunctionParameter %v4f32\n"
4662 const char functionEnd[] =
4663 "OpReturnValue %transformed_param\n"
4666 struct NameConstantsCode
4673 NameConstantsCode tests[] =
4678 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
4679 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
4684 "%stype = OpTypeStruct %v4f32 %f32\n"
4685 "%fp_stype = OpTypePointer Function %stype\n"
4686 "%f32_n_1 = OpConstant %f32 -1.0\n"
4687 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
4688 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
4689 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
4691 "%v = OpVariable %fp_stype Function %cval\n"
4692 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
4693 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
4694 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
4695 "%f32_val = OpLoad %f32 %f32_ptr\n"
4696 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
4697 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
4698 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
4701 // [1|0|0|0.5] [x] = x + 0.5
4702 // [0|1|0|0.5] [y] = y + 0.5
4703 // [0|0|1|0.5] [z] = z + 0.5
4704 // [0|0|0|1 ] [1] = 1
4707 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
4708 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
4709 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
4710 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
4711 "%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"
4712 "%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",
4714 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
4719 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4720 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
4721 "%f32_n_1 = OpConstant %f32 -1.0\n"
4722 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
4723 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
4725 "%v = OpVariable %fp_a4f32 Function %carr\n"
4726 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
4727 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
4728 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
4729 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
4730 "%f_val = OpLoad %f32 %f\n"
4731 "%f1_val = OpLoad %f32 %f1\n"
4732 "%f2_val = OpLoad %f32 %f2\n"
4733 "%f3_val = OpLoad %f32 %f3\n"
4734 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
4735 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
4736 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
4737 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
4738 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
4745 // [ 1.0, 1.0, 1.0, 1.0]
4749 // [ 0.0, 0.5, 0.0, 0.0]
4753 // [ 1.0, 1.0, 1.0, 1.0]
4756 "array_of_struct_of_array",
4758 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4759 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
4760 "%stype = OpTypeStruct %f32 %a4f32\n"
4761 "%a3stype = OpTypeArray %stype %c_u32_3\n"
4762 "%fp_a3stype = OpTypePointer Function %a3stype\n"
4763 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
4764 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
4765 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
4766 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
4767 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
4769 "%v = OpVariable %fp_a3stype Function %carr\n"
4770 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
4771 "%f_l = OpLoad %f32 %f\n"
4772 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
4773 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
4777 getHalfColorsFullAlpha(inputColors);
4778 outputColors[0] = RGBA(255, 255, 255, 255);
4779 outputColors[1] = RGBA(255, 127, 127, 255);
4780 outputColors[2] = RGBA(127, 255, 127, 255);
4781 outputColors[3] = RGBA(127, 127, 255, 255);
4783 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
4785 map<string, string> fragments;
4786 fragments["pre_main"] = tests[testNdx].constants;
4787 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
4788 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
4790 return opConstantCompositeTests.release();
4793 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
4795 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
4796 RGBA inputColors[4];
4797 RGBA outputColors[4];
4798 map<string, string> fragments;
4800 // vec4 test_code(vec4 param) {
4801 // vec4 result = param;
4802 // for (int i = 0; i < 4; ++i) {
4803 // if (i == 0) result[i] = 0.;
4804 // else result[i] = 1. - result[i];
4808 const char function[] =
4809 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4810 "%param1 = OpFunctionParameter %v4f32\n"
4812 "%iptr = OpVariable %fp_i32 Function\n"
4813 "%result = OpVariable %fp_v4f32 Function\n"
4814 " OpStore %iptr %c_i32_0\n"
4815 " OpStore %result %param1\n"
4818 // Loop entry block.
4820 "%ival = OpLoad %i32 %iptr\n"
4821 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
4822 " OpLoopMerge %exit %if_entry None\n"
4823 " OpBranchConditional %lt_4 %if_entry %exit\n"
4825 // Merge block for loop.
4827 "%ret = OpLoad %v4f32 %result\n"
4828 " OpReturnValue %ret\n"
4830 // If-statement entry block.
4831 "%if_entry = OpLabel\n"
4832 "%loc = OpAccessChain %fp_f32 %result %ival\n"
4833 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
4834 " OpSelectionMerge %if_exit None\n"
4835 " OpBranchConditional %eq_0 %if_true %if_false\n"
4837 // False branch for if-statement.
4838 "%if_false = OpLabel\n"
4839 "%val = OpLoad %f32 %loc\n"
4840 "%sub = OpFSub %f32 %c_f32_1 %val\n"
4841 " OpStore %loc %sub\n"
4842 " OpBranch %if_exit\n"
4844 // Merge block for if-statement.
4845 "%if_exit = OpLabel\n"
4846 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
4847 " OpStore %iptr %ival_next\n"
4850 // True branch for if-statement.
4851 "%if_true = OpLabel\n"
4852 " OpStore %loc %c_f32_0\n"
4853 " OpBranch %if_exit\n"
4857 fragments["testfun"] = function;
4859 inputColors[0] = RGBA(127, 127, 127, 0);
4860 inputColors[1] = RGBA(127, 0, 0, 0);
4861 inputColors[2] = RGBA(0, 127, 0, 0);
4862 inputColors[3] = RGBA(0, 0, 127, 0);
4864 outputColors[0] = RGBA(0, 128, 128, 255);
4865 outputColors[1] = RGBA(0, 255, 255, 255);
4866 outputColors[2] = RGBA(0, 128, 255, 255);
4867 outputColors[3] = RGBA(0, 255, 128, 255);
4869 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
4871 return group.release();
4874 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
4876 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
4877 RGBA inputColors[4];
4878 RGBA outputColors[4];
4879 map<string, string> fragments;
4881 const char typesAndConstants[] =
4882 "%c_f32_p2 = OpConstant %f32 0.2\n"
4883 "%c_f32_p4 = OpConstant %f32 0.4\n"
4884 "%c_f32_p6 = OpConstant %f32 0.6\n"
4885 "%c_f32_p8 = OpConstant %f32 0.8\n";
4887 // vec4 test_code(vec4 param) {
4888 // vec4 result = param;
4889 // for (int i = 0; i < 4; ++i) {
4891 // case 0: result[i] += .2; break;
4892 // case 1: result[i] += .6; break;
4893 // case 2: result[i] += .4; break;
4894 // case 3: result[i] += .8; break;
4895 // default: break; // unreachable
4900 const char function[] =
4901 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4902 "%param1 = OpFunctionParameter %v4f32\n"
4904 "%iptr = OpVariable %fp_i32 Function\n"
4905 "%result = OpVariable %fp_v4f32 Function\n"
4906 " OpStore %iptr %c_i32_0\n"
4907 " OpStore %result %param1\n"
4910 // Loop entry block.
4912 "%ival = OpLoad %i32 %iptr\n"
4913 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
4914 " OpLoopMerge %exit %switch_exit None\n"
4915 " OpBranchConditional %lt_4 %switch_entry %exit\n"
4917 // Merge block for loop.
4919 "%ret = OpLoad %v4f32 %result\n"
4920 " OpReturnValue %ret\n"
4922 // Switch-statement entry block.
4923 "%switch_entry = OpLabel\n"
4924 "%loc = OpAccessChain %fp_f32 %result %ival\n"
4925 "%val = OpLoad %f32 %loc\n"
4926 " OpSelectionMerge %switch_exit None\n"
4927 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
4929 "%case2 = OpLabel\n"
4930 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
4931 " OpStore %loc %addp4\n"
4932 " OpBranch %switch_exit\n"
4934 "%switch_default = OpLabel\n"
4937 "%case3 = OpLabel\n"
4938 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
4939 " OpStore %loc %addp8\n"
4940 " OpBranch %switch_exit\n"
4942 "%case0 = OpLabel\n"
4943 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
4944 " OpStore %loc %addp2\n"
4945 " OpBranch %switch_exit\n"
4947 // Merge block for switch-statement.
4948 "%switch_exit = OpLabel\n"
4949 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
4950 " OpStore %iptr %ival_next\n"
4953 "%case1 = OpLabel\n"
4954 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
4955 " OpStore %loc %addp6\n"
4956 " OpBranch %switch_exit\n"
4960 fragments["pre_main"] = typesAndConstants;
4961 fragments["testfun"] = function;
4963 inputColors[0] = RGBA(127, 27, 127, 51);
4964 inputColors[1] = RGBA(127, 0, 0, 51);
4965 inputColors[2] = RGBA(0, 27, 0, 51);
4966 inputColors[3] = RGBA(0, 0, 127, 51);
4968 outputColors[0] = RGBA(178, 180, 229, 255);
4969 outputColors[1] = RGBA(178, 153, 102, 255);
4970 outputColors[2] = RGBA(51, 180, 102, 255);
4971 outputColors[3] = RGBA(51, 153, 229, 255);
4973 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
4975 return group.release();
4978 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
4980 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
4981 RGBA inputColors[4];
4982 RGBA outputColors[4];
4983 map<string, string> fragments;
4985 const char decorations[] =
4986 "OpDecorate %array_group ArrayStride 4\n"
4987 "OpDecorate %struct_member_group Offset 0\n"
4988 "%array_group = OpDecorationGroup\n"
4989 "%struct_member_group = OpDecorationGroup\n"
4991 "OpDecorate %group1 RelaxedPrecision\n"
4992 "OpDecorate %group3 RelaxedPrecision\n"
4993 "OpDecorate %group3 Invariant\n"
4994 "OpDecorate %group3 Restrict\n"
4995 "%group0 = OpDecorationGroup\n"
4996 "%group1 = OpDecorationGroup\n"
4997 "%group3 = OpDecorationGroup\n";
4999 const char typesAndConstants[] =
5000 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
5001 "%struct1 = OpTypeStruct %a3f32\n"
5002 "%struct2 = OpTypeStruct %a3f32\n"
5003 "%fp_struct1 = OpTypePointer Function %struct1\n"
5004 "%fp_struct2 = OpTypePointer Function %struct2\n"
5005 "%c_f32_2 = OpConstant %f32 2.\n"
5006 "%c_f32_n2 = OpConstant %f32 -2.\n"
5008 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
5009 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
5010 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
5011 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
5013 const char function[] =
5014 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5015 "%param = OpFunctionParameter %v4f32\n"
5016 "%entry = OpLabel\n"
5017 "%result = OpVariable %fp_v4f32 Function\n"
5018 "%v_struct1 = OpVariable %fp_struct1 Function\n"
5019 "%v_struct2 = OpVariable %fp_struct2 Function\n"
5020 " OpStore %result %param\n"
5021 " OpStore %v_struct1 %c_struct1\n"
5022 " OpStore %v_struct2 %c_struct2\n"
5023 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
5024 "%val1 = OpLoad %f32 %ptr1\n"
5025 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
5026 "%val2 = OpLoad %f32 %ptr2\n"
5027 "%addvalues = OpFAdd %f32 %val1 %val2\n"
5028 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
5029 "%val = OpLoad %f32 %ptr\n"
5030 "%addresult = OpFAdd %f32 %addvalues %val\n"
5031 " OpStore %ptr %addresult\n"
5032 "%ret = OpLoad %v4f32 %result\n"
5033 " OpReturnValue %ret\n"
5036 struct CaseNameDecoration
5042 CaseNameDecoration tests[] =
5045 "same_decoration_group_on_multiple_types",
5046 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
5049 "empty_decoration_group",
5050 "OpGroupDecorate %group0 %a3f32\n"
5051 "OpGroupDecorate %group0 %result\n"
5054 "one_element_decoration_group",
5055 "OpGroupDecorate %array_group %a3f32\n"
5058 "multiple_elements_decoration_group",
5059 "OpGroupDecorate %group3 %v_struct1\n"
5062 "multiple_decoration_groups_on_same_variable",
5063 "OpGroupDecorate %group0 %v_struct2\n"
5064 "OpGroupDecorate %group1 %v_struct2\n"
5065 "OpGroupDecorate %group3 %v_struct2\n"
5068 "same_decoration_group_multiple_times",
5069 "OpGroupDecorate %group1 %addvalues\n"
5070 "OpGroupDecorate %group1 %addvalues\n"
5071 "OpGroupDecorate %group1 %addvalues\n"
5076 getHalfColorsFullAlpha(inputColors);
5077 getHalfColorsFullAlpha(outputColors);
5079 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
5081 fragments["decoration"] = decorations + tests[idx].decoration;
5082 fragments["pre_main"] = typesAndConstants;
5083 fragments["testfun"] = function;
5085 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
5088 return group.release();
5091 struct SpecConstantTwoIntGraphicsCase
5093 const char* caseName;
5094 const char* scDefinition0;
5095 const char* scDefinition1;
5096 const char* scResultType;
5097 const char* scOperation;
5098 deInt32 scActualValue0;
5099 deInt32 scActualValue1;
5100 const char* resultOperation;
5101 RGBA expectedColors[4];
5103 SpecConstantTwoIntGraphicsCase (const char* name,
5104 const char* definition0,
5105 const char* definition1,
5106 const char* resultType,
5107 const char* operation,
5110 const char* resultOp,
5111 const RGBA (&output)[4])
5113 , scDefinition0 (definition0)
5114 , scDefinition1 (definition1)
5115 , scResultType (resultType)
5116 , scOperation (operation)
5117 , scActualValue0 (value0)
5118 , scActualValue1 (value1)
5119 , resultOperation (resultOp)
5121 expectedColors[0] = output[0];
5122 expectedColors[1] = output[1];
5123 expectedColors[2] = output[2];
5124 expectedColors[3] = output[3];
5128 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
5130 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
5131 vector<SpecConstantTwoIntGraphicsCase> cases;
5132 RGBA inputColors[4];
5133 RGBA outputColors0[4];
5134 RGBA outputColors1[4];
5135 RGBA outputColors2[4];
5137 const char decorations1[] =
5138 "OpDecorate %sc_0 SpecId 0\n"
5139 "OpDecorate %sc_1 SpecId 1\n";
5141 const char typesAndConstants1[] =
5142 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
5143 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
5144 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
5146 const char function1[] =
5147 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5148 "%param = OpFunctionParameter %v4f32\n"
5149 "%label = OpLabel\n"
5150 "%result = OpVariable %fp_v4f32 Function\n"
5151 " OpStore %result %param\n"
5152 "%gen = ${GEN_RESULT}\n"
5153 "%index = OpIAdd %i32 %gen %c_i32_1\n"
5154 "%loc = OpAccessChain %fp_f32 %result %index\n"
5155 "%val = OpLoad %f32 %loc\n"
5156 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
5157 " OpStore %loc %add\n"
5158 "%ret = OpLoad %v4f32 %result\n"
5159 " OpReturnValue %ret\n"
5162 inputColors[0] = RGBA(127, 127, 127, 255);
5163 inputColors[1] = RGBA(127, 0, 0, 255);
5164 inputColors[2] = RGBA(0, 127, 0, 255);
5165 inputColors[3] = RGBA(0, 0, 127, 255);
5167 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
5168 outputColors0[0] = RGBA(255, 127, 127, 255);
5169 outputColors0[1] = RGBA(255, 0, 0, 255);
5170 outputColors0[2] = RGBA(128, 127, 0, 255);
5171 outputColors0[3] = RGBA(128, 0, 127, 255);
5173 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
5174 outputColors1[0] = RGBA(127, 255, 127, 255);
5175 outputColors1[1] = RGBA(127, 128, 0, 255);
5176 outputColors1[2] = RGBA(0, 255, 0, 255);
5177 outputColors1[3] = RGBA(0, 128, 127, 255);
5179 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
5180 outputColors2[0] = RGBA(127, 127, 255, 255);
5181 outputColors2[1] = RGBA(127, 0, 128, 255);
5182 outputColors2[2] = RGBA(0, 127, 128, 255);
5183 outputColors2[3] = RGBA(0, 0, 255, 255);
5185 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
5186 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
5187 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
5189 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
5190 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
5191 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
5192 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
5193 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
5194 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
5195 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
5196 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
5197 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
5198 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
5199 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
5200 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
5201 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
5202 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
5203 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
5204 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
5205 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
5206 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
5207 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
5208 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
5209 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
5210 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
5211 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
5212 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
5213 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
5214 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
5215 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
5216 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
5217 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
5218 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
5219 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
5220 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
5221 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
5223 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5225 map<string, string> specializations;
5226 map<string, string> fragments;
5227 vector<deInt32> specConstants;
5229 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
5230 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
5231 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
5232 specializations["SC_OP"] = cases[caseNdx].scOperation;
5233 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
5235 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
5236 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
5237 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
5239 specConstants.push_back(cases[caseNdx].scActualValue0);
5240 specConstants.push_back(cases[caseNdx].scActualValue1);
5242 createTestsForAllStages(cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants, group.get());
5245 const char decorations2[] =
5246 "OpDecorate %sc_0 SpecId 0\n"
5247 "OpDecorate %sc_1 SpecId 1\n"
5248 "OpDecorate %sc_2 SpecId 2\n";
5250 const char typesAndConstants2[] =
5251 "%v3i32 = OpTypeVector %i32 3\n"
5252 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
5253 "%vec3_undef = OpUndef %v3i32\n"
5255 "%sc_0 = OpSpecConstant %i32 0\n"
5256 "%sc_1 = OpSpecConstant %i32 0\n"
5257 "%sc_2 = OpSpecConstant %i32 0\n"
5258 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
5259 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
5260 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
5261 "%sc_vec3_0_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
5262 "%sc_vec3_1_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
5263 "%sc_vec3_2_s = OpSpecConstantOp %v3i32 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
5264 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
5265 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
5266 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
5267 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
5268 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
5269 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
5270 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
5272 const char function2[] =
5273 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5274 "%param = OpFunctionParameter %v4f32\n"
5275 "%label = OpLabel\n"
5276 "%result = OpVariable %fp_v4f32 Function\n"
5277 " OpStore %result %param\n"
5278 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
5279 "%val = OpLoad %f32 %loc\n"
5280 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
5281 " OpStore %loc %add\n"
5282 "%ret = OpLoad %v4f32 %result\n"
5283 " OpReturnValue %ret\n"
5286 map<string, string> fragments;
5287 vector<deInt32> specConstants;
5289 fragments["decoration"] = decorations2;
5290 fragments["pre_main"] = typesAndConstants2;
5291 fragments["testfun"] = function2;
5293 specConstants.push_back(56789);
5294 specConstants.push_back(-2);
5295 specConstants.push_back(56788);
5297 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
5299 return group.release();
5302 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
5304 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
5305 RGBA inputColors[4];
5306 RGBA outputColors1[4];
5307 RGBA outputColors2[4];
5308 RGBA outputColors3[4];
5309 map<string, string> fragments1;
5310 map<string, string> fragments2;
5311 map<string, string> fragments3;
5313 const char typesAndConstants1[] =
5314 "%c_f32_p2 = OpConstant %f32 0.2\n"
5315 "%c_f32_p4 = OpConstant %f32 0.4\n"
5316 "%c_f32_p5 = OpConstant %f32 0.5\n"
5317 "%c_f32_p8 = OpConstant %f32 0.8\n";
5319 // vec4 test_code(vec4 param) {
5320 // vec4 result = param;
5321 // for (int i = 0; i < 4; ++i) {
5324 // case 0: operand = .2; break;
5325 // case 1: operand = .5; break;
5326 // case 2: operand = .4; break;
5327 // case 3: operand = .0; break;
5328 // default: break; // unreachable
5330 // result[i] += operand;
5334 const char function1[] =
5335 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5336 "%param1 = OpFunctionParameter %v4f32\n"
5338 "%iptr = OpVariable %fp_i32 Function\n"
5339 "%result = OpVariable %fp_v4f32 Function\n"
5340 " OpStore %iptr %c_i32_0\n"
5341 " OpStore %result %param1\n"
5345 "%ival = OpLoad %i32 %iptr\n"
5346 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5347 " OpLoopMerge %exit %phi None\n"
5348 " OpBranchConditional %lt_4 %entry %exit\n"
5350 "%entry = OpLabel\n"
5351 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5352 "%val = OpLoad %f32 %loc\n"
5353 " OpSelectionMerge %phi None\n"
5354 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5356 "%case0 = OpLabel\n"
5358 "%case1 = OpLabel\n"
5360 "%case2 = OpLabel\n"
5362 "%case3 = OpLabel\n"
5365 "%default = OpLabel\n"
5369 "%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
5370 "%add = OpFAdd %f32 %val %operand\n"
5371 " OpStore %loc %add\n"
5372 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5373 " OpStore %iptr %ival_next\n"
5377 "%ret = OpLoad %v4f32 %result\n"
5378 " OpReturnValue %ret\n"
5382 fragments1["pre_main"] = typesAndConstants1;
5383 fragments1["testfun"] = function1;
5385 getHalfColorsFullAlpha(inputColors);
5387 outputColors1[0] = RGBA(178, 255, 229, 255);
5388 outputColors1[1] = RGBA(178, 127, 102, 255);
5389 outputColors1[2] = RGBA(51, 255, 102, 255);
5390 outputColors1[3] = RGBA(51, 127, 229, 255);
5392 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
5394 const char typesAndConstants2[] =
5395 "%c_f32_p2 = OpConstant %f32 0.2\n";
5397 // Add .4 to the second element of the given parameter.
5398 const char function2[] =
5399 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5400 "%param = OpFunctionParameter %v4f32\n"
5401 "%entry = OpLabel\n"
5402 "%result = OpVariable %fp_v4f32 Function\n"
5403 " OpStore %result %param\n"
5404 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
5405 "%val = OpLoad %f32 %loc\n"
5409 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
5410 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
5411 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
5412 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
5413 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
5414 " OpLoopMerge %exit %phi None\n"
5415 " OpBranchConditional %still_loop %phi %exit\n"
5418 " OpStore %loc %accum\n"
5419 "%ret = OpLoad %v4f32 %result\n"
5420 " OpReturnValue %ret\n"
5424 fragments2["pre_main"] = typesAndConstants2;
5425 fragments2["testfun"] = function2;
5427 outputColors2[0] = RGBA(127, 229, 127, 255);
5428 outputColors2[1] = RGBA(127, 102, 0, 255);
5429 outputColors2[2] = RGBA(0, 229, 0, 255);
5430 outputColors2[3] = RGBA(0, 102, 127, 255);
5432 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
5434 const char typesAndConstants3[] =
5435 "%true = OpConstantTrue %bool\n"
5436 "%false = OpConstantFalse %bool\n"
5437 "%c_f32_p2 = OpConstant %f32 0.2\n";
5439 // Swap the second and the third element of the given parameter.
5440 const char function3[] =
5441 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5442 "%param = OpFunctionParameter %v4f32\n"
5443 "%entry = OpLabel\n"
5444 "%result = OpVariable %fp_v4f32 Function\n"
5445 " OpStore %result %param\n"
5446 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
5447 "%a_init = OpLoad %f32 %a_loc\n"
5448 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
5449 "%b_init = OpLoad %f32 %b_loc\n"
5453 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
5454 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
5455 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
5456 " OpLoopMerge %exit %phi None\n"
5457 " OpBranchConditional %still_loop %phi %exit\n"
5460 " OpStore %a_loc %a_next\n"
5461 " OpStore %b_loc %b_next\n"
5462 "%ret = OpLoad %v4f32 %result\n"
5463 " OpReturnValue %ret\n"
5467 fragments3["pre_main"] = typesAndConstants3;
5468 fragments3["testfun"] = function3;
5470 outputColors3[0] = RGBA(127, 127, 127, 255);
5471 outputColors3[1] = RGBA(127, 0, 0, 255);
5472 outputColors3[2] = RGBA(0, 0, 127, 255);
5473 outputColors3[3] = RGBA(0, 127, 0, 255);
5475 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
5477 return group.release();
5480 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
5482 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
5483 RGBA inputColors[4];
5484 RGBA outputColors[4];
5486 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
5487 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
5488 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
5489 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
5490 const char constantsAndTypes[] =
5491 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
5492 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5493 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
5494 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
5495 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n"
5498 const char function[] =
5499 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5500 "%param = OpFunctionParameter %v4f32\n"
5501 "%label = OpLabel\n"
5502 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
5503 "%var2 = OpVariable %fp_f32 Function\n"
5504 "%red = OpCompositeExtract %f32 %param 0\n"
5505 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
5506 " OpStore %var2 %plus_red\n"
5507 "%val1 = OpLoad %f32 %var1\n"
5508 "%val2 = OpLoad %f32 %var2\n"
5509 "%mul = OpFMul %f32 %val1 %val2\n"
5510 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
5511 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
5512 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
5513 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
5514 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
5515 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
5516 " OpReturnValue %ret\n"
5519 struct CaseNameDecoration
5526 CaseNameDecoration tests[] = {
5527 {"multiplication", "OpDecorate %mul NoContraction"},
5528 {"addition", "OpDecorate %add NoContraction"},
5529 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
5532 getHalfColorsFullAlpha(inputColors);
5534 for (deUint8 idx = 0; idx < 4; ++idx)
5536 inputColors[idx].setRed(0);
5537 outputColors[idx] = RGBA(0, 0, 0, 255);
5540 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
5542 map<string, string> fragments;
5544 fragments["decoration"] = tests[testNdx].decoration;
5545 fragments["pre_main"] = constantsAndTypes;
5546 fragments["testfun"] = function;
5548 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
5551 return group.release();
5554 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
5556 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
5559 const char constantsAndTypes[] =
5560 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
5561 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
5562 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
5563 "%fp_stype = OpTypePointer Function %stype\n";
5565 const char function[] =
5566 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5567 "%param1 = OpFunctionParameter %v4f32\n"
5569 "%v1 = OpVariable %fp_v4f32 Function\n"
5570 "%v2 = OpVariable %fp_a2f32 Function\n"
5571 "%v3 = OpVariable %fp_f32 Function\n"
5572 "%v = OpVariable %fp_stype Function\n"
5573 "%vv = OpVariable %fp_stype Function\n"
5574 "%vvv = OpVariable %fp_f32 Function\n"
5576 " OpStore %v1 %c_v4f32_1_1_1_1\n"
5577 " OpStore %v2 %c_a2f32_1\n"
5578 " OpStore %v3 %c_f32_1\n"
5580 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5581 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
5582 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5583 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
5584 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
5585 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
5587 " OpStore %p_v4f32 %v1_v ${access_type}\n"
5588 " OpStore %p_a2f32 %v2_v ${access_type}\n"
5589 " OpStore %p_f32 %v3_v ${access_type}\n"
5591 " OpCopyMemory %vv %v ${access_type}\n"
5592 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
5594 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
5595 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
5596 "%v_f32_3 = OpLoad %f32 %vvv\n"
5598 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
5599 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
5600 " OpReturnValue %ret2\n"
5603 struct NameMemoryAccess
5610 NameMemoryAccess tests[] =
5613 { "volatile", "Volatile" },
5614 { "aligned", "Aligned 1" },
5615 { "volatile_aligned", "Volatile|Aligned 1" },
5616 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
5617 { "volatile_nontemporal", "Volatile|Nontemporal" },
5618 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
5621 getHalfColorsFullAlpha(colors);
5623 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
5625 map<string, string> fragments;
5626 map<string, string> memoryAccess;
5627 memoryAccess["access_type"] = tests[testNdx].accessType;
5629 fragments["pre_main"] = constantsAndTypes;
5630 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
5631 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
5633 return memoryAccessTests.release();
5635 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
5637 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
5638 RGBA defaultColors[4];
5639 map<string, string> fragments;
5640 getDefaultColors(defaultColors);
5642 // First, simple cases that don't do anything with the OpUndef result.
5643 struct NameCodePair { string name, decl, type; };
5644 const NameCodePair tests[] =
5646 {"bool", "", "%bool"},
5647 {"vec2uint32", "%type = OpTypeVector %u32 2", "%type"},
5648 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown", "%type"},
5649 {"sampler", "%type = OpTypeSampler", "%type"},
5650 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img", "%type"},
5651 {"pointer", "", "%fp_i32"},
5652 {"runtimearray", "%type = OpTypeRuntimeArray %f32", "%type"},
5653 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100", "%type"},
5654 {"struct", "%type = OpTypeStruct %f32 %i32 %u32", "%type"}};
5655 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5657 fragments["undef_type"] = tests[testNdx].type;
5658 fragments["testfun"] = StringTemplate(
5659 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5660 "%param1 = OpFunctionParameter %v4f32\n"
5661 "%label_testfun = OpLabel\n"
5662 "%undef = OpUndef ${undef_type}\n"
5663 "OpReturnValue %param1\n"
5664 "OpFunctionEnd\n").specialize(fragments);
5665 fragments["pre_main"] = tests[testNdx].decl;
5666 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
5670 fragments["testfun"] =
5671 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5672 "%param1 = OpFunctionParameter %v4f32\n"
5673 "%label_testfun = OpLabel\n"
5674 "%undef = OpUndef %f32\n"
5675 "%zero = OpFMul %f32 %undef %c_f32_0\n"
5676 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
5677 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
5678 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5679 "%b = OpFAdd %f32 %a %actually_zero\n"
5680 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
5681 "OpReturnValue %ret\n"
5684 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
5686 fragments["testfun"] =
5687 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5688 "%param1 = OpFunctionParameter %v4f32\n"
5689 "%label_testfun = OpLabel\n"
5690 "%undef = OpUndef %i32\n"
5691 "%zero = OpIMul %i32 %undef %c_i32_0\n"
5692 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
5693 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
5694 "OpReturnValue %ret\n"
5697 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
5699 fragments["testfun"] =
5700 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5701 "%param1 = OpFunctionParameter %v4f32\n"
5702 "%label_testfun = OpLabel\n"
5703 "%undef = OpUndef %u32\n"
5704 "%zero = OpIMul %u32 %undef %c_i32_0\n"
5705 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
5706 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
5707 "OpReturnValue %ret\n"
5710 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
5712 fragments["testfun"] =
5713 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5714 "%param1 = OpFunctionParameter %v4f32\n"
5715 "%label_testfun = OpLabel\n"
5716 "%undef = OpUndef %v4f32\n"
5717 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
5718 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
5719 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
5720 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
5721 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
5722 "%is_nan_0 = OpIsNan %bool %zero_0\n"
5723 "%is_nan_1 = OpIsNan %bool %zero_1\n"
5724 "%is_nan_2 = OpIsNan %bool %zero_2\n"
5725 "%is_nan_3 = OpIsNan %bool %zero_3\n"
5726 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
5727 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
5728 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
5729 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
5730 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5731 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
5732 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
5733 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
5734 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
5735 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
5736 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
5737 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
5738 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
5739 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
5740 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
5741 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
5742 "OpReturnValue %ret\n"
5745 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
5747 fragments["pre_main"] =
5748 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
5749 fragments["testfun"] =
5750 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5751 "%param1 = OpFunctionParameter %v4f32\n"
5752 "%label_testfun = OpLabel\n"
5753 "%undef = OpUndef %m2x2f32\n"
5754 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
5755 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
5756 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
5757 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
5758 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
5759 "%is_nan_0 = OpIsNan %bool %zero_0\n"
5760 "%is_nan_1 = OpIsNan %bool %zero_1\n"
5761 "%is_nan_2 = OpIsNan %bool %zero_2\n"
5762 "%is_nan_3 = OpIsNan %bool %zero_3\n"
5763 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
5764 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
5765 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
5766 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
5767 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5768 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
5769 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
5770 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
5771 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
5772 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
5773 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
5774 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
5775 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
5776 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
5777 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
5778 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
5779 "OpReturnValue %ret\n"
5782 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
5784 return opUndefTests.release();
5787 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
5789 const RGBA inputColors[4] =
5792 RGBA(0, 0, 255, 255),
5793 RGBA(0, 255, 0, 255),
5794 RGBA(0, 255, 255, 255)
5797 const RGBA expectedColors[4] =
5799 RGBA(255, 0, 0, 255),
5800 RGBA(255, 0, 0, 255),
5801 RGBA(255, 0, 0, 255),
5802 RGBA(255, 0, 0, 255)
5805 const struct SingleFP16Possibility
5808 const char* constant; // Value to assign to %test_constant.
5810 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
5816 -constructNormalizedFloat(1, 0x300000),
5817 "%cond = OpFOrdEqual %bool %c %test_constant\n"
5822 constructNormalizedFloat(7, 0x000000),
5823 "%cond = OpFOrdEqual %bool %c %test_constant\n"
5825 // SPIR-V requires that OpQuantizeToF16 flushes
5826 // any numbers that would end up denormalized in F16 to zero.
5830 std::ldexp(1.5f, -140),
5831 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5836 -std::ldexp(1.5f, -140),
5837 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5842 std::ldexp(1.0f, -16),
5843 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5844 }, // too small positive
5846 "negative_too_small",
5848 -std::ldexp(1.0f, -32),
5849 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5850 }, // too small negative
5854 -std::ldexp(1.0f, 128),
5856 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
5857 "%inf = OpIsInf %bool %c\n"
5858 "%cond = OpLogicalAnd %bool %gz %inf\n"
5863 std::ldexp(1.0f, 128),
5865 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
5866 "%inf = OpIsInf %bool %c\n"
5867 "%cond = OpLogicalAnd %bool %gz %inf\n"
5870 "round_to_negative_inf",
5872 -std::ldexp(1.0f, 32),
5874 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
5875 "%inf = OpIsInf %bool %c\n"
5876 "%cond = OpLogicalAnd %bool %gz %inf\n"
5881 std::ldexp(1.0f, 16),
5883 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
5884 "%inf = OpIsInf %bool %c\n"
5885 "%cond = OpLogicalAnd %bool %gz %inf\n"
5890 std::numeric_limits<float>::quiet_NaN(),
5892 // Test for any NaN value, as NaNs are not preserved
5893 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
5894 "%cond = OpIsNan %bool %direct_quant\n"
5899 std::numeric_limits<float>::quiet_NaN(),
5901 // Test for any NaN value, as NaNs are not preserved
5902 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
5903 "%cond = OpIsNan %bool %direct_quant\n"
5906 const char* constants =
5907 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
5909 StringTemplate function (
5910 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5911 "%param1 = OpFunctionParameter %v4f32\n"
5912 "%label_testfun = OpLabel\n"
5913 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5914 "%b = OpFAdd %f32 %test_constant %a\n"
5915 "%c = OpQuantizeToF16 %f32 %b\n"
5917 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
5918 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
5919 " OpReturnValue %retval\n"
5923 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
5924 const char* specConstants =
5925 "%test_constant = OpSpecConstant %f32 0.\n"
5926 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
5928 StringTemplate specConstantFunction(
5929 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5930 "%param1 = OpFunctionParameter %v4f32\n"
5931 "%label_testfun = OpLabel\n"
5933 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
5934 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
5935 " OpReturnValue %retval\n"
5939 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
5941 map<string, string> codeSpecialization;
5942 map<string, string> fragments;
5943 codeSpecialization["condition"] = tests[idx].condition;
5944 fragments["testfun"] = function.specialize(codeSpecialization);
5945 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
5946 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
5949 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
5951 map<string, string> codeSpecialization;
5952 map<string, string> fragments;
5953 vector<deInt32> passConstants;
5954 deInt32 specConstant;
5956 codeSpecialization["condition"] = tests[idx].condition;
5957 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
5958 fragments["decoration"] = specDecorations;
5959 fragments["pre_main"] = specConstants;
5961 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
5962 passConstants.push_back(specConstant);
5964 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
5968 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
5970 RGBA inputColors[4] = {
5972 RGBA(0, 0, 255, 255),
5973 RGBA(0, 255, 0, 255),
5974 RGBA(0, 255, 255, 255)
5977 RGBA expectedColors[4] =
5979 RGBA(255, 0, 0, 255),
5980 RGBA(255, 0, 0, 255),
5981 RGBA(255, 0, 0, 255),
5982 RGBA(255, 0, 0, 255)
5985 struct DualFP16Possibility
5990 const char* possibleOutput1;
5991 const char* possibleOutput2;
5994 "positive_round_up_or_round_down",
5996 constructNormalizedFloat(8, 0x300300),
6001 "negative_round_up_or_round_down",
6003 -constructNormalizedFloat(-7, 0x600800),
6010 constructNormalizedFloat(2, 0x01e000),
6015 "carry_to_exponent",
6017 constructNormalizedFloat(1, 0xffe000),
6022 StringTemplate constants (
6023 "%input_const = OpConstant %f32 ${input}\n"
6024 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6025 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6028 StringTemplate specConstants (
6029 "%input_const = OpSpecConstant %f32 0.\n"
6030 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6031 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6034 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
6036 const char* function =
6037 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6038 "%param1 = OpFunctionParameter %v4f32\n"
6039 "%label_testfun = OpLabel\n"
6040 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6041 // For the purposes of this test we assume that 0.f will always get
6042 // faithfully passed through the pipeline stages.
6043 "%b = OpFAdd %f32 %input_const %a\n"
6044 "%c = OpQuantizeToF16 %f32 %b\n"
6045 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
6046 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
6047 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
6048 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6049 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
6050 " OpReturnValue %retval\n"
6053 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6054 map<string, string> fragments;
6055 map<string, string> constantSpecialization;
6057 constantSpecialization["input"] = tests[idx].input;
6058 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6059 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6060 fragments["testfun"] = function;
6061 fragments["pre_main"] = constants.specialize(constantSpecialization);
6062 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6065 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6066 map<string, string> fragments;
6067 map<string, string> constantSpecialization;
6068 vector<deInt32> passConstants;
6069 deInt32 specConstant;
6071 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6072 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6073 fragments["testfun"] = function;
6074 fragments["decoration"] = specDecorations;
6075 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
6077 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
6078 passConstants.push_back(specConstant);
6080 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6084 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
6086 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
6087 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
6088 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
6089 return opQuantizeTests.release();
6092 struct ShaderPermutation
6094 deUint8 vertexPermutation;
6095 deUint8 geometryPermutation;
6096 deUint8 tesscPermutation;
6097 deUint8 tessePermutation;
6098 deUint8 fragmentPermutation;
6101 ShaderPermutation getShaderPermutation(deUint8 inputValue)
6103 ShaderPermutation permutation =
6105 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
6106 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
6107 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
6108 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
6109 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
6114 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
6116 RGBA defaultColors[4];
6117 RGBA invertedColors[4];
6118 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
6120 const ShaderElement combinedPipeline[] =
6122 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
6123 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
6124 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
6125 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
6126 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
6129 getDefaultColors(defaultColors);
6130 getInvertedDefaultColors(invertedColors);
6131 addFunctionCaseWithPrograms<InstanceContext>(
6132 moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline,
6133 createInstanceContext(combinedPipeline, map<string, string>()));
6135 const char* numbers[] =
6140 for (deInt8 idx = 0; idx < 32; ++idx)
6142 ShaderPermutation permutation = getShaderPermutation(idx);
6143 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
6144 const ShaderElement pipeline[] =
6146 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
6147 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
6148 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
6149 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
6150 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
6153 // If there are an even number of swaps, then it should be no-op.
6154 // If there are an odd number, the color should be flipped.
6155 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
6157 addFunctionCaseWithPrograms<InstanceContext>(
6158 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
6159 createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
6163 addFunctionCaseWithPrograms<InstanceContext>(
6164 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
6165 createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
6168 return moduleTests.release();
6171 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
6173 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
6174 RGBA defaultColors[4];
6175 getDefaultColors(defaultColors);
6176 map<string, string> fragments;
6177 fragments["pre_main"] =
6178 "%c_f32_5 = OpConstant %f32 5.\n";
6180 // A loop with a single block. The Continue Target is the loop block
6181 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
6182 // -- the "continue construct" forms the entire loop.
6183 fragments["testfun"] =
6184 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6185 "%param1 = OpFunctionParameter %v4f32\n"
6187 "%entry = OpLabel\n"
6188 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6191 ";adds and subtracts 1.0 to %val in alternate iterations\n"
6193 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
6194 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
6195 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
6196 "%val = OpFAdd %f32 %val1 %delta\n"
6197 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
6198 "%count__ = OpISub %i32 %count %c_i32_1\n"
6199 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6200 "OpLoopMerge %exit %loop None\n"
6201 "OpBranchConditional %again %loop %exit\n"
6204 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
6205 "OpReturnValue %result\n"
6209 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
6211 // Body comprised of multiple basic blocks.
6212 const StringTemplate multiBlock(
6213 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6214 "%param1 = OpFunctionParameter %v4f32\n"
6216 "%entry = OpLabel\n"
6217 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6220 ";adds and subtracts 1.0 to %val in alternate iterations\n"
6222 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
6223 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
6224 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
6225 // There are several possibilities for the Continue Target below. Each
6226 // will be specialized into a separate test case.
6227 "OpLoopMerge %exit ${continue_target} None\n"
6231 ";delta_next = (delta > 0) ? -1 : 1;\n"
6232 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
6233 "OpSelectionMerge %gather DontFlatten\n"
6234 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
6237 "OpBranch %gather\n"
6240 "OpBranch %gather\n"
6242 "%gather = OpLabel\n"
6243 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
6244 "%val = OpFAdd %f32 %val1 %delta\n"
6245 "%count__ = OpISub %i32 %count %c_i32_1\n"
6246 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6247 "OpBranchConditional %again %loop %exit\n"
6250 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
6251 "OpReturnValue %result\n"
6255 map<string, string> continue_target;
6257 // The Continue Target is the loop block itself.
6258 continue_target["continue_target"] = "%loop";
6259 fragments["testfun"] = multiBlock.specialize(continue_target);
6260 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
6262 // The Continue Target is at the end of the loop.
6263 continue_target["continue_target"] = "%gather";
6264 fragments["testfun"] = multiBlock.specialize(continue_target);
6265 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
6267 // A loop with continue statement.
6268 fragments["testfun"] =
6269 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6270 "%param1 = OpFunctionParameter %v4f32\n"
6272 "%entry = OpLabel\n"
6273 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6276 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
6278 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6279 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
6280 "OpLoopMerge %exit %continue None\n"
6284 ";skip if %count==2\n"
6285 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
6286 "OpSelectionMerge %continue DontFlatten\n"
6287 "OpBranchConditional %eq2 %continue %body\n"
6290 "%fcount = OpConvertSToF %f32 %count\n"
6291 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6292 "OpBranch %continue\n"
6294 "%continue = OpLabel\n"
6295 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
6296 "%count__ = OpISub %i32 %count %c_i32_1\n"
6297 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6298 "OpBranchConditional %again %loop %exit\n"
6301 "%same = OpFSub %f32 %val %c_f32_8\n"
6302 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6303 "OpReturnValue %result\n"
6305 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
6307 // A loop with break.
6308 fragments["testfun"] =
6309 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6310 "%param1 = OpFunctionParameter %v4f32\n"
6312 "%entry = OpLabel\n"
6313 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6314 "%dot = OpDot %f32 %param1 %param1\n"
6315 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6316 "%zero = OpConvertFToU %u32 %div\n"
6317 "%two = OpIAdd %i32 %zero %c_i32_2\n"
6318 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6321 ";adds 4 and 3 to %val0 (exits early)\n"
6323 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6324 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
6325 "OpLoopMerge %exit %continue None\n"
6329 ";end loop if %count==%two\n"
6330 "%above2 = OpSGreaterThan %bool %count %two\n"
6331 "OpSelectionMerge %continue DontFlatten\n"
6332 "OpBranchConditional %above2 %body %exit\n"
6335 "%fcount = OpConvertSToF %f32 %count\n"
6336 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6337 "OpBranch %continue\n"
6339 "%continue = OpLabel\n"
6340 "%count__ = OpISub %i32 %count %c_i32_1\n"
6341 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6342 "OpBranchConditional %again %loop %exit\n"
6345 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
6346 "%same = OpFSub %f32 %val_post %c_f32_7\n"
6347 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6348 "OpReturnValue %result\n"
6350 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
6352 // A loop with return.
6353 fragments["testfun"] =
6354 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6355 "%param1 = OpFunctionParameter %v4f32\n"
6357 "%entry = OpLabel\n"
6358 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6359 "%dot = OpDot %f32 %param1 %param1\n"
6360 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6361 "%zero = OpConvertFToU %u32 %div\n"
6362 "%two = OpIAdd %i32 %zero %c_i32_2\n"
6363 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6366 ";returns early without modifying %param1\n"
6368 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6369 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
6370 "OpLoopMerge %exit %continue None\n"
6374 ";return if %count==%two\n"
6375 "%above2 = OpSGreaterThan %bool %count %two\n"
6376 "OpSelectionMerge %continue DontFlatten\n"
6377 "OpBranchConditional %above2 %body %early_exit\n"
6379 "%early_exit = OpLabel\n"
6380 "OpReturnValue %param1\n"
6383 "%fcount = OpConvertSToF %f32 %count\n"
6384 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6385 "OpBranch %continue\n"
6387 "%continue = OpLabel\n"
6388 "%count__ = OpISub %i32 %count %c_i32_1\n"
6389 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6390 "OpBranchConditional %again %loop %exit\n"
6393 ";should never get here, so return an incorrect result\n"
6394 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
6395 "OpReturnValue %result\n"
6397 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
6399 return testGroup.release();
6402 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
6403 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
6405 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
6406 map<string, string> fragments;
6408 // A barrier inside a function body.
6409 fragments["pre_main"] =
6410 "%Workgroup = OpConstant %i32 2\n"
6411 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
6412 fragments["testfun"] =
6413 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6414 "%param1 = OpFunctionParameter %v4f32\n"
6415 "%label_testfun = OpLabel\n"
6416 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6417 "OpReturnValue %param1\n"
6419 addTessCtrlTest(testGroup.get(), "in_function", fragments);
6421 // Common setup code for the following tests.
6422 fragments["pre_main"] =
6423 "%Workgroup = OpConstant %i32 2\n"
6424 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
6425 "%c_f32_5 = OpConstant %f32 5.\n";
6426 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
6427 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6428 "%param1 = OpFunctionParameter %v4f32\n"
6429 "%entry = OpLabel\n"
6430 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6431 "%dot = OpDot %f32 %param1 %param1\n"
6432 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6433 "%zero = OpConvertFToU %u32 %div\n";
6435 // Barriers inside OpSwitch branches.
6436 fragments["testfun"] =
6438 "OpSelectionMerge %switch_exit None\n"
6439 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
6441 "%case1 = OpLabel\n"
6442 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6443 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6444 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6445 "OpBranch %switch_exit\n"
6447 "%switch_default = OpLabel\n"
6448 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6449 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6450 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6451 "OpBranch %switch_exit\n"
6453 "%case0 = OpLabel\n"
6454 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6455 "OpBranch %switch_exit\n"
6457 "%switch_exit = OpLabel\n"
6458 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
6459 "OpReturnValue %ret\n"
6461 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
6463 // Barriers inside if-then-else.
6464 fragments["testfun"] =
6466 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
6467 "OpSelectionMerge %exit DontFlatten\n"
6468 "OpBranchConditional %eq0 %then %else\n"
6471 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6472 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6473 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6477 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6481 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
6482 "OpReturnValue %ret\n"
6484 addTessCtrlTest(testGroup.get(), "in_if", fragments);
6486 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
6487 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
6488 fragments["testfun"] =
6490 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
6491 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
6492 "OpSelectionMerge %exit DontFlatten\n"
6493 "OpBranchConditional %thread0 %then %else\n"
6496 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6500 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
6504 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
6505 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6506 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
6507 "OpReturnValue %ret\n"
6509 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
6511 // A barrier inside a loop.
6512 fragments["pre_main"] =
6513 "%Workgroup = OpConstant %i32 2\n"
6514 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
6515 "%c_f32_10 = OpConstant %f32 10.\n";
6516 fragments["testfun"] =
6517 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6518 "%param1 = OpFunctionParameter %v4f32\n"
6519 "%entry = OpLabel\n"
6520 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6523 ";adds 4, 3, 2, and 1 to %val0\n"
6525 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
6526 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
6527 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6528 "%fcount = OpConvertSToF %f32 %count\n"
6529 "%val = OpFAdd %f32 %val1 %fcount\n"
6530 "%count__ = OpISub %i32 %count %c_i32_1\n"
6531 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6532 "OpLoopMerge %exit %loop None\n"
6533 "OpBranchConditional %again %loop %exit\n"
6536 "%same = OpFSub %f32 %val %c_f32_10\n"
6537 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6538 "OpReturnValue %ret\n"
6540 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
6542 return testGroup.release();
6545 // Test for the OpFRem instruction.
6546 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
6548 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
6549 map<string, string> fragments;
6550 RGBA inputColors[4];
6551 RGBA outputColors[4];
6553 fragments["pre_main"] =
6554 "%c_f32_3 = OpConstant %f32 3.0\n"
6555 "%c_f32_n3 = OpConstant %f32 -3.0\n"
6556 "%c_f32_4 = OpConstant %f32 4.0\n"
6557 "%c_f32_p75 = OpConstant %f32 0.75\n"
6558 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
6559 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
6560 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
6562 // The test does the following.
6563 // vec4 result = (param1 * 8.0) - 4.0;
6564 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
6565 fragments["testfun"] =
6566 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6567 "%param1 = OpFunctionParameter %v4f32\n"
6568 "%label_testfun = OpLabel\n"
6569 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
6570 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
6571 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
6572 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
6573 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
6574 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
6575 "OpReturnValue %xy_0_1\n"
6579 inputColors[0] = RGBA(16, 16, 0, 255);
6580 inputColors[1] = RGBA(232, 232, 0, 255);
6581 inputColors[2] = RGBA(232, 16, 0, 255);
6582 inputColors[3] = RGBA(16, 232, 0, 255);
6584 outputColors[0] = RGBA(64, 64, 0, 255);
6585 outputColors[1] = RGBA(255, 255, 0, 255);
6586 outputColors[2] = RGBA(255, 64, 0, 255);
6587 outputColors[3] = RGBA(64, 255, 0, 255);
6589 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
6590 return testGroup.release();
6593 // Test for the OpSRem instruction.
6594 tcu::TestCaseGroup* createOpSRemGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
6596 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "srem", "OpSRem"));
6597 map<string, string> fragments;
6599 fragments["pre_main"] =
6600 "%c_f32_255 = OpConstant %f32 255.0\n"
6601 "%c_i32_128 = OpConstant %i32 128\n"
6602 "%c_i32_255 = OpConstant %i32 255\n"
6603 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
6604 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
6605 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
6607 // The test does the following.
6608 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
6609 // ivec4 result = ivec4(srem(ints.x, ints.y), srem(ints.y, ints.z), srem(ints.z, ints.x), 255);
6610 // return float(result + 128) / 255.0;
6611 fragments["testfun"] =
6612 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6613 "%param1 = OpFunctionParameter %v4f32\n"
6614 "%label_testfun = OpLabel\n"
6615 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
6616 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
6617 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
6618 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
6619 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
6620 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
6621 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
6622 "%x_out = OpSRem %i32 %x_in %y_in\n"
6623 "%y_out = OpSRem %i32 %y_in %z_in\n"
6624 "%z_out = OpSRem %i32 %z_in %x_in\n"
6625 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
6626 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
6627 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
6628 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
6629 "OpReturnValue %float_out\n"
6632 const struct CaseParams
6635 const char* failMessageTemplate; // customized status message
6636 qpTestResult failResult; // override status on failure
6637 int operands[4][3]; // four (x, y, z) vectors of operands
6638 int results[4][3]; // four (x, y, z) vectors of results
6644 QP_TEST_RESULT_FAIL,
6645 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
6646 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
6650 "Inconsistent results, but within specification: ${reason}",
6651 negFailResult, // negative operands, not required by the spec
6652 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
6653 { { 5, 12, -2 }, { 0, -5, 2 }, { 3, 8, -6 }, { 25, -60, 0 } }, // results
6656 // If either operand is negative the result is undefined. Some implementations may still return correct values.
6658 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
6660 const CaseParams& params = cases[caseNdx];
6661 RGBA inputColors[4];
6662 RGBA outputColors[4];
6664 for (int i = 0; i < 4; ++i)
6666 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
6667 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
6670 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
6673 return testGroup.release();
6676 // Test for the OpSMod instruction.
6677 tcu::TestCaseGroup* createOpSModGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
6679 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "smod", "OpSMod"));
6680 map<string, string> fragments;
6682 fragments["pre_main"] =
6683 "%c_f32_255 = OpConstant %f32 255.0\n"
6684 "%c_i32_128 = OpConstant %i32 128\n"
6685 "%c_i32_255 = OpConstant %i32 255\n"
6686 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
6687 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
6688 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
6690 // The test does the following.
6691 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
6692 // ivec4 result = ivec4(smod(ints.x, ints.y), smod(ints.y, ints.z), smod(ints.z, ints.x), 255);
6693 // return float(result + 128) / 255.0;
6694 fragments["testfun"] =
6695 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6696 "%param1 = OpFunctionParameter %v4f32\n"
6697 "%label_testfun = OpLabel\n"
6698 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
6699 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
6700 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
6701 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
6702 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
6703 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
6704 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
6705 "%x_out = OpSMod %i32 %x_in %y_in\n"
6706 "%y_out = OpSMod %i32 %y_in %z_in\n"
6707 "%z_out = OpSMod %i32 %z_in %x_in\n"
6708 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
6709 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
6710 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
6711 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
6712 "OpReturnValue %float_out\n"
6715 const struct CaseParams
6718 const char* failMessageTemplate; // customized status message
6719 qpTestResult failResult; // override status on failure
6720 int operands[4][3]; // four (x, y, z) vectors of operands
6721 int results[4][3]; // four (x, y, z) vectors of results
6727 QP_TEST_RESULT_FAIL,
6728 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
6729 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
6733 "Inconsistent results, but within specification: ${reason}",
6734 negFailResult, // negative operands, not required by the spec
6735 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
6736 { { 5, -5, 3 }, { 0, 2, -3 }, { 3, -73, 69 }, { -35, 40, 0 } }, // results
6739 // If either operand is negative the result is undefined. Some implementations may still return correct values.
6741 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
6743 const CaseParams& params = cases[caseNdx];
6744 RGBA inputColors[4];
6745 RGBA outputColors[4];
6747 for (int i = 0; i < 4; ++i)
6749 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
6750 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
6753 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
6756 return testGroup.release();
6761 INTEGER_TYPE_SIGNED_16,
6762 INTEGER_TYPE_SIGNED_32,
6763 INTEGER_TYPE_SIGNED_64,
6765 INTEGER_TYPE_UNSIGNED_16,
6766 INTEGER_TYPE_UNSIGNED_32,
6767 INTEGER_TYPE_UNSIGNED_64,
6770 const string getBitWidthStr (IntegerType type)
6774 case INTEGER_TYPE_SIGNED_16:
6775 case INTEGER_TYPE_UNSIGNED_16: return "16";
6777 case INTEGER_TYPE_SIGNED_32:
6778 case INTEGER_TYPE_UNSIGNED_32: return "32";
6780 case INTEGER_TYPE_SIGNED_64:
6781 case INTEGER_TYPE_UNSIGNED_64: return "64";
6783 default: DE_ASSERT(false);
6788 const string getByteWidthStr (IntegerType type)
6792 case INTEGER_TYPE_SIGNED_16:
6793 case INTEGER_TYPE_UNSIGNED_16: return "2";
6795 case INTEGER_TYPE_SIGNED_32:
6796 case INTEGER_TYPE_UNSIGNED_32: return "4";
6798 case INTEGER_TYPE_SIGNED_64:
6799 case INTEGER_TYPE_UNSIGNED_64: return "8";
6801 default: DE_ASSERT(false);
6806 bool isSigned (IntegerType type)
6808 return (type <= INTEGER_TYPE_SIGNED_64);
6811 const string getTypeName (IntegerType type)
6813 string prefix = isSigned(type) ? "" : "u";
6814 return prefix + "int" + getBitWidthStr(type);
6817 const string getTestName (IntegerType from, IntegerType to)
6819 return getTypeName(from) + "_to_" + getTypeName(to);
6822 const string getAsmTypeDeclaration (IntegerType type)
6824 string sign = isSigned(type) ? " 1" : " 0";
6825 return "OpTypeInt " + getBitWidthStr(type) + sign;
6828 template<typename T>
6829 BufferSp getSpecializedBuffer (deInt64 number)
6831 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
6834 BufferSp getBuffer (IntegerType type, deInt64 number)
6838 case INTEGER_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
6839 case INTEGER_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
6840 case INTEGER_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
6842 case INTEGER_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
6843 case INTEGER_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
6844 case INTEGER_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
6846 default: DE_ASSERT(false);
6847 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
6851 bool usesInt16 (IntegerType from, IntegerType to)
6853 return (from == INTEGER_TYPE_SIGNED_16 || from == INTEGER_TYPE_UNSIGNED_16
6854 || to == INTEGER_TYPE_SIGNED_16 || to == INTEGER_TYPE_UNSIGNED_16);
6857 bool usesInt64 (IntegerType from, IntegerType to)
6859 return (from == INTEGER_TYPE_SIGNED_64 || from == INTEGER_TYPE_UNSIGNED_64
6860 || to == INTEGER_TYPE_SIGNED_64 || to == INTEGER_TYPE_UNSIGNED_64);
6863 ComputeTestFeatures getConversionUsedFeatures (IntegerType from, IntegerType to)
6865 if (usesInt16(from, to))
6867 if (usesInt64(from, to))
6869 return COMPUTE_TEST_USES_INT16_INT64;
6873 return COMPUTE_TEST_USES_INT16;
6878 return COMPUTE_TEST_USES_INT64;
6884 ConvertCase (IntegerType from, IntegerType to, deInt64 number)
6887 , m_features (getConversionUsedFeatures(from, to))
6888 , m_name (getTestName(from, to))
6889 , m_inputBuffer (getBuffer(from, number))
6890 , m_outputBuffer (getBuffer(to, number))
6892 m_asmTypes["inputType"] = getAsmTypeDeclaration(from);
6893 m_asmTypes["outputType"] = getAsmTypeDeclaration(to);
6895 if (m_features == COMPUTE_TEST_USES_INT16)
6897 m_asmTypes["int_capabilities"] = "OpCapability Int16\n";
6899 else if (m_features == COMPUTE_TEST_USES_INT64)
6901 m_asmTypes["int_capabilities"] = "OpCapability Int64\n";
6903 else if (m_features == COMPUTE_TEST_USES_INT16_INT64)
6905 m_asmTypes["int_capabilities"] = string("OpCapability Int16\n") +
6906 "OpCapability Int64\n";
6914 IntegerType m_fromType;
6915 IntegerType m_toType;
6916 ComputeTestFeatures m_features;
6918 map<string, string> m_asmTypes;
6919 BufferSp m_inputBuffer;
6920 BufferSp m_outputBuffer;
6923 const string getConvertCaseShaderStr (const string& instruction, const ConvertCase& convertCase)
6925 map<string, string> params = convertCase.m_asmTypes;
6927 params["instruction"] = instruction;
6929 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
6930 params["outDecorator"] = getByteWidthStr(convertCase.m_toType);
6932 const StringTemplate shader (
6933 "OpCapability Shader\n"
6934 "${int_capabilities}"
6935 "OpMemoryModel Logical GLSL450\n"
6936 "OpEntryPoint GLCompute %main \"main\" %id\n"
6937 "OpExecutionMode %main LocalSize 1 1 1\n"
6938 "OpSource GLSL 430\n"
6939 "OpName %main \"main\"\n"
6940 "OpName %id \"gl_GlobalInvocationID\"\n"
6942 "OpDecorate %id BuiltIn GlobalInvocationId\n"
6943 "OpDecorate %indata DescriptorSet 0\n"
6944 "OpDecorate %indata Binding 0\n"
6945 "OpDecorate %outdata DescriptorSet 0\n"
6946 "OpDecorate %outdata Binding 1\n"
6947 "OpDecorate %in_arr ArrayStride ${inDecorator}\n"
6948 "OpDecorate %out_arr ArrayStride ${outDecorator}\n"
6949 "OpDecorate %in_buf BufferBlock\n"
6950 "OpDecorate %out_buf BufferBlock\n"
6951 "OpMemberDecorate %in_buf 0 Offset 0\n"
6952 "OpMemberDecorate %out_buf 0 Offset 0\n"
6954 "%void = OpTypeVoid\n"
6955 "%voidf = OpTypeFunction %void\n"
6956 "%u32 = OpTypeInt 32 0\n"
6957 "%i32 = OpTypeInt 32 1\n"
6958 "%uvec3 = OpTypeVector %u32 3\n"
6959 "%uvec3ptr = OpTypePointer Input %uvec3\n"
6961 "%in_type = ${inputType}\n"
6962 "%out_type = ${outputType}\n"
6964 "%in_ptr = OpTypePointer Uniform %in_type\n"
6965 "%out_ptr = OpTypePointer Uniform %out_type\n"
6966 "%in_arr = OpTypeRuntimeArray %in_type\n"
6967 "%out_arr = OpTypeRuntimeArray %out_type\n"
6968 "%in_buf = OpTypeStruct %in_arr\n"
6969 "%out_buf = OpTypeStruct %out_arr\n"
6970 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
6971 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
6972 "%indata = OpVariable %in_bufptr Uniform\n"
6973 "%outdata = OpVariable %out_bufptr Uniform\n"
6974 "%inputptr = OpTypePointer Input %in_type\n"
6975 "%id = OpVariable %uvec3ptr Input\n"
6977 "%zero = OpConstant %i32 0\n"
6979 "%main = OpFunction %void None %voidf\n"
6980 "%label = OpLabel\n"
6981 "%idval = OpLoad %uvec3 %id\n"
6982 "%x = OpCompositeExtract %u32 %idval 0\n"
6983 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
6984 "%outloc = OpAccessChain %out_ptr %outdata %zero %x\n"
6985 "%inval = OpLoad %in_type %inloc\n"
6986 "%conv = ${instruction} %out_type %inval\n"
6987 " OpStore %outloc %conv\n"
6992 return shader.specialize(params);
6995 void createSConvertCases (vector<ConvertCase>& testCases)
6997 // Convert int to int
6998 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_32, 14669));
6999 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_64, 3341));
7001 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_SIGNED_64, 973610259));
7003 // Convert int to unsigned int
7004 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_32, 9288));
7005 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_64, 15460));
7007 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_UNSIGNED_64, 346213461));
7010 // Test for the OpSConvert instruction.
7011 tcu::TestCaseGroup* createSConvertTests (tcu::TestContext& testCtx)
7013 const string instruction ("OpSConvert");
7014 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "sconvert", "OpSConvert"));
7015 vector<ConvertCase> testCases;
7016 createSConvertCases(testCases);
7018 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7020 ComputeShaderSpec spec;
7022 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7023 spec.inputs.push_back(test->m_inputBuffer);
7024 spec.outputs.push_back(test->m_outputBuffer);
7025 spec.numWorkGroups = IVec3(1, 1, 1);
7027 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpSConvert.", spec, test->m_features));
7030 return group.release();
7033 void createUConvertCases (vector<ConvertCase>& testCases)
7035 // Convert unsigned int to unsigned int
7036 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_32, 60653));
7037 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_64, 17991));
7039 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_UNSIGNED_64, 904256275));
7041 // Convert unsigned int to int
7042 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_32, 38002));
7043 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_64, 64921));
7045 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_SIGNED_64, 4294956295ll));
7048 // Test for the OpUConvert instruction.
7049 tcu::TestCaseGroup* createUConvertTests (tcu::TestContext& testCtx)
7051 const string instruction ("OpUConvert");
7052 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "uconvert", "OpUConvert"));
7053 vector<ConvertCase> testCases;
7054 createUConvertCases(testCases);
7056 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7058 ComputeShaderSpec spec;
7060 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7061 spec.inputs.push_back(test->m_inputBuffer);
7062 spec.outputs.push_back(test->m_outputBuffer);
7063 spec.numWorkGroups = IVec3(1, 1, 1);
7065 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpUConvert.", spec, test->m_features));
7067 return group.release();
7070 const string getNumberTypeName (const NumberType type)
7072 if (type == NUMBERTYPE_INT32)
7076 else if (type == NUMBERTYPE_UINT32)
7080 else if (type == NUMBERTYPE_FLOAT32)
7091 deInt32 getInt(de::Random& rnd)
7093 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
7096 const string repeatString (const string& str, int times)
7099 for (int i = 0; i < times; ++i)
7106 const string getRandomConstantString (const NumberType type, de::Random& rnd)
7108 if (type == NUMBERTYPE_INT32)
7110 return numberToString<deInt32>(getInt(rnd));
7112 else if (type == NUMBERTYPE_UINT32)
7114 return numberToString<deUint32>(rnd.getUint32());
7116 else if (type == NUMBERTYPE_FLOAT32)
7118 return numberToString<float>(rnd.getFloat());
7127 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7129 map<string, string> params;
7132 for (int width = 2; width <= 4; ++width)
7134 string randomConst = numberToString(getInt(rnd));
7135 string widthStr = numberToString(width);
7136 int index = rnd.getInt(0, width-1);
7138 params["type"] = "vec";
7139 params["name"] = params["type"] + "_" + widthStr;
7140 params["compositeType"] = "%composite = OpTypeVector %custom " + widthStr +"\n";
7141 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7142 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7143 params["indexes"] = numberToString(index);
7144 testCases.push_back(params);
7148 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7150 const int limit = 10;
7151 map<string, string> params;
7153 for (int width = 2; width <= limit; ++width)
7155 string randomConst = numberToString(getInt(rnd));
7156 string widthStr = numberToString(width);
7157 int index = rnd.getInt(0, width-1);
7159 params["type"] = "array";
7160 params["name"] = params["type"] + "_" + widthStr;
7161 params["compositeType"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
7162 + "%composite = OpTypeArray %custom %arraywidth\n";
7164 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7165 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7166 params["indexes"] = numberToString(index);
7167 testCases.push_back(params);
7171 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7173 const int limit = 10;
7174 map<string, string> params;
7176 for (int width = 2; width <= limit; ++width)
7178 string randomConst = numberToString(getInt(rnd));
7179 int index = rnd.getInt(0, width-1);
7181 params["type"] = "struct";
7182 params["name"] = params["type"] + "_" + numberToString(width);
7183 params["compositeType"] = "%composite = OpTypeStruct" + repeatString(" %custom", width) + "\n";
7184 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7185 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7186 params["indexes"] = numberToString(index);
7187 testCases.push_back(params);
7191 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7193 map<string, string> params;
7196 for (int width = 2; width <= 4; ++width)
7198 string widthStr = numberToString(width);
7200 for (int column = 2 ; column <= 4; ++column)
7202 int index_0 = rnd.getInt(0, column-1);
7203 int index_1 = rnd.getInt(0, width-1);
7204 string columnStr = numberToString(column);
7206 params["type"] = "matrix";
7207 params["name"] = params["type"] + "_" + widthStr + "x" + columnStr;
7208 params["compositeType"] = string("%vectype = OpTypeVector %custom " + widthStr + "\n")
7209 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
7211 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n"
7212 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
7214 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
7215 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
7216 testCases.push_back(params);
7221 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7223 createVectorCompositeCases(testCases, rnd, type);
7224 createArrayCompositeCases(testCases, rnd, type);
7225 createStructCompositeCases(testCases, rnd, type);
7226 // Matrix only supports float types
7227 if (type == NUMBERTYPE_FLOAT32)
7229 createMatrixCompositeCases(testCases, rnd, type);
7233 const string getAssemblyTypeDeclaration (const NumberType type)
7237 case NUMBERTYPE_INT32: return "OpTypeInt 32 1";
7238 case NUMBERTYPE_UINT32: return "OpTypeInt 32 0";
7239 case NUMBERTYPE_FLOAT32: return "OpTypeFloat 32";
7240 default: DE_ASSERT(false); return "";
7244 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
7246 map<string, string> parameters(params);
7248 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7250 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
7252 return StringTemplate (
7253 "OpCapability Shader\n"
7254 "OpCapability Matrix\n"
7255 "OpMemoryModel Logical GLSL450\n"
7256 "OpEntryPoint GLCompute %main \"main\" %id\n"
7257 "OpExecutionMode %main LocalSize 1 1 1\n"
7259 "OpSource GLSL 430\n"
7260 "OpName %main \"main\"\n"
7261 "OpName %id \"gl_GlobalInvocationID\"\n"
7264 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7265 "OpDecorate %buf BufferBlock\n"
7266 "OpDecorate %indata DescriptorSet 0\n"
7267 "OpDecorate %indata Binding 0\n"
7268 "OpDecorate %outdata DescriptorSet 0\n"
7269 "OpDecorate %outdata Binding 1\n"
7270 "OpDecorate %customarr ArrayStride 4\n"
7271 "${compositeDecorator}"
7272 "OpMemberDecorate %buf 0 Offset 0\n"
7275 "%void = OpTypeVoid\n"
7276 "%voidf = OpTypeFunction %void\n"
7277 "%u32 = OpTypeInt 32 0\n"
7278 "%i32 = OpTypeInt 32 1\n"
7279 "%uvec3 = OpTypeVector %u32 3\n"
7280 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7283 "%custom = ${typeDeclaration}\n"
7289 // Inherited from custom
7290 "%customptr = OpTypePointer Uniform %custom\n"
7291 "%customarr = OpTypeRuntimeArray %custom\n"
7292 "%buf = OpTypeStruct %customarr\n"
7293 "%bufptr = OpTypePointer Uniform %buf\n"
7295 "%indata = OpVariable %bufptr Uniform\n"
7296 "%outdata = OpVariable %bufptr Uniform\n"
7298 "%id = OpVariable %uvec3ptr Input\n"
7299 "%zero = OpConstant %i32 0\n"
7301 "%main = OpFunction %void None %voidf\n"
7302 "%label = OpLabel\n"
7303 "%idval = OpLoad %uvec3 %id\n"
7304 "%x = OpCompositeExtract %u32 %idval 0\n"
7306 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
7307 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
7308 // Read the input value
7309 "%inval = OpLoad %custom %inloc\n"
7310 // Create the composite and fill it
7311 "${compositeConstruct}"
7312 // Insert the input value to a place
7313 "%instance2 = OpCompositeInsert %composite %inval %instance ${indexes}\n"
7314 // Read back the value from the position
7315 "%out_val = OpCompositeExtract %custom %instance2 ${indexes}\n"
7316 // Store it in the output position
7317 " OpStore %outloc %out_val\n"
7320 ).specialize(parameters);
7323 template<typename T>
7324 BufferSp createCompositeBuffer(T number)
7326 return BufferSp(new Buffer<T>(vector<T>(1, number)));
7329 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
7331 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
7332 de::Random rnd (deStringHash(group->getName()));
7334 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7336 NumberType numberType = NumberType(type);
7337 const string typeName = getNumberTypeName(numberType);
7338 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
7339 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7340 vector<map<string, string> > testCases;
7342 createCompositeCases(testCases, rnd, numberType);
7344 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7346 ComputeShaderSpec spec;
7348 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
7352 case NUMBERTYPE_INT32:
7354 deInt32 number = getInt(rnd);
7355 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7356 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7359 case NUMBERTYPE_UINT32:
7361 deUint32 number = rnd.getUint32();
7362 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7363 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7366 case NUMBERTYPE_FLOAT32:
7368 float number = rnd.getFloat();
7369 spec.inputs.push_back(createCompositeBuffer<float>(number));
7370 spec.outputs.push_back(createCompositeBuffer<float>(number));
7377 spec.numWorkGroups = IVec3(1, 1, 1);
7378 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
7380 group->addChild(subGroup.release());
7382 return group.release();
7385 struct AssemblyStructInfo
7387 AssemblyStructInfo (const deUint32 comp, const deUint32 idx)
7392 deUint32 components;
7396 const string specializeInBoundsShaderTemplate (const NumberType type, const AssemblyStructInfo& structInfo, const map<string, string>& params)
7398 // Create the full index string
7399 string fullIndex = numberToString(structInfo.index) + " " + params.at("indexes");
7400 // Convert it to list of indexes
7401 vector<string> indexes = de::splitString(fullIndex, ' ');
7403 map<string, string> parameters (params);
7404 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7405 parameters["structType"] = repeatString(" %composite", structInfo.components);
7406 parameters["structConstruct"] = repeatString(" %instance", structInfo.components);
7407 parameters["insertIndexes"] = fullIndex;
7409 // In matrix cases the last two index is the CompositeExtract indexes
7410 const deUint32 extractIndexes = (parameters["type"] == "matrix") ? 2 : 1;
7412 // Construct the extractIndex
7413 for (vector<string>::const_iterator index = indexes.end() - extractIndexes; index != indexes.end(); ++index)
7415 parameters["extractIndexes"] += " " + *index;
7418 // Remove the last 1 or 2 element depends on matrix case or not
7419 indexes.erase(indexes.end() - extractIndexes, indexes.end());
7422 // Generate AccessChain index expressions (except for the last one, because we use ptr to the composite)
7423 for (vector<string>::const_iterator index = indexes.begin(); index != indexes.end(); ++index)
7425 string indexId = "%index_" + numberToString(id++);
7426 parameters["accessChainConstDeclaration"] += indexId + " = OpConstant %u32 " + *index + "\n";
7427 parameters["accessChainIndexes"] += " " + indexId;
7430 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
7432 return StringTemplate (
7433 "OpCapability Shader\n"
7434 "OpCapability Matrix\n"
7435 "OpMemoryModel Logical GLSL450\n"
7436 "OpEntryPoint GLCompute %main \"main\" %id\n"
7437 "OpExecutionMode %main LocalSize 1 1 1\n"
7439 "OpSource GLSL 430\n"
7440 "OpName %main \"main\"\n"
7441 "OpName %id \"gl_GlobalInvocationID\"\n"
7443 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7444 "OpDecorate %buf BufferBlock\n"
7445 "OpDecorate %indata DescriptorSet 0\n"
7446 "OpDecorate %indata Binding 0\n"
7447 "OpDecorate %outdata DescriptorSet 0\n"
7448 "OpDecorate %outdata Binding 1\n"
7449 "OpDecorate %customarr ArrayStride 4\n"
7450 "${compositeDecorator}"
7451 "OpMemberDecorate %buf 0 Offset 0\n"
7453 "%void = OpTypeVoid\n"
7454 "%voidf = OpTypeFunction %void\n"
7455 "%u32 = OpTypeInt 32 0\n"
7456 "%uvec3 = OpTypeVector %u32 3\n"
7457 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7459 "%custom = ${typeDeclaration}\n"
7462 // Inherited from composite
7463 "%composite_p = OpTypePointer Function %composite\n"
7464 "%struct_t = OpTypeStruct${structType}\n"
7465 "%struct_p = OpTypePointer Function %struct_t\n"
7468 "${accessChainConstDeclaration}"
7469 // Inherited from custom
7470 "%customptr = OpTypePointer Uniform %custom\n"
7471 "%customarr = OpTypeRuntimeArray %custom\n"
7472 "%buf = OpTypeStruct %customarr\n"
7473 "%bufptr = OpTypePointer Uniform %buf\n"
7474 "%indata = OpVariable %bufptr Uniform\n"
7475 "%outdata = OpVariable %bufptr Uniform\n"
7477 "%id = OpVariable %uvec3ptr Input\n"
7478 "%zero = OpConstant %u32 0\n"
7479 "%main = OpFunction %void None %voidf\n"
7480 "%label = OpLabel\n"
7481 "%struct_v = OpVariable %struct_p Function\n"
7482 "%idval = OpLoad %uvec3 %id\n"
7483 "%x = OpCompositeExtract %u32 %idval 0\n"
7484 // Create the input/output type
7485 "%inloc = OpInBoundsAccessChain %customptr %indata %zero %x\n"
7486 "%outloc = OpInBoundsAccessChain %customptr %outdata %zero %x\n"
7487 // Read the input value
7488 "%inval = OpLoad %custom %inloc\n"
7489 // Create the composite and fill it
7490 "${compositeConstruct}"
7491 // Create the struct and fill it with the composite
7492 "%struct = OpCompositeConstruct %struct_t${structConstruct}\n"
7494 "%comp_obj = OpCompositeInsert %struct_t %inval %struct ${insertIndexes}\n"
7496 " OpStore %struct_v %comp_obj\n"
7497 // Get deepest possible composite pointer
7498 "%inner_ptr = OpInBoundsAccessChain %composite_p %struct_v${accessChainIndexes}\n"
7499 "%read_obj = OpLoad %composite %inner_ptr\n"
7500 // Read back the stored value
7501 "%read_val = OpCompositeExtract %custom %read_obj${extractIndexes}\n"
7502 " OpStore %outloc %read_val\n"
7504 " OpFunctionEnd\n").specialize(parameters);
7507 tcu::TestCaseGroup* createOpInBoundsAccessChainGroup (tcu::TestContext& testCtx)
7509 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opinboundsaccesschain", "Test the OpInBoundsAccessChain instruction"));
7510 de::Random rnd (deStringHash(group->getName()));
7512 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7514 NumberType numberType = NumberType(type);
7515 const string typeName = getNumberTypeName(numberType);
7516 const string description = "Test the OpInBoundsAccessChain instruction with " + typeName + "s";
7517 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7519 vector<map<string, string> > testCases;
7520 createCompositeCases(testCases, rnd, numberType);
7522 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7524 ComputeShaderSpec spec;
7526 // Number of components inside of a struct
7527 deUint32 structComponents = rnd.getInt(2, 8);
7528 // Component index value
7529 deUint32 structIndex = rnd.getInt(0, structComponents - 1);
7530 AssemblyStructInfo structInfo(structComponents, structIndex);
7532 spec.assembly = specializeInBoundsShaderTemplate(numberType, structInfo, *test);
7536 case NUMBERTYPE_INT32:
7538 deInt32 number = getInt(rnd);
7539 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7540 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7543 case NUMBERTYPE_UINT32:
7545 deUint32 number = rnd.getUint32();
7546 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7547 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7550 case NUMBERTYPE_FLOAT32:
7552 float number = rnd.getFloat();
7553 spec.inputs.push_back(createCompositeBuffer<float>(number));
7554 spec.outputs.push_back(createCompositeBuffer<float>(number));
7560 spec.numWorkGroups = IVec3(1, 1, 1);
7561 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpInBoundsAccessChain test", spec));
7563 group->addChild(subGroup.release());
7565 return group.release();
7568 // If the params missing, uninitialized case
7569 const string specializeDefaultOutputShaderTemplate (const NumberType type, const map<string, string>& params = map<string, string>())
7571 map<string, string> parameters(params);
7573 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7575 // Declare the const value, and use it in the initializer
7576 if (params.find("constValue") != params.end())
7578 parameters["constDeclaration"] = "%const = OpConstant %in_type " + params.at("constValue") + "\n";
7579 parameters["variableInitializer"] = "%const";
7581 // Uninitialized case
7584 parameters["constDeclaration"] = "";
7585 parameters["variableInitializer"] = "";
7588 return StringTemplate(
7589 "OpCapability Shader\n"
7590 "OpMemoryModel Logical GLSL450\n"
7591 "OpEntryPoint GLCompute %main \"main\" %id\n"
7592 "OpExecutionMode %main LocalSize 1 1 1\n"
7593 "OpSource GLSL 430\n"
7594 "OpName %main \"main\"\n"
7595 "OpName %id \"gl_GlobalInvocationID\"\n"
7597 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7598 "OpDecorate %indata DescriptorSet 0\n"
7599 "OpDecorate %indata Binding 0\n"
7600 "OpDecorate %outdata DescriptorSet 0\n"
7601 "OpDecorate %outdata Binding 1\n"
7602 "OpDecorate %in_arr ArrayStride 4\n"
7603 "OpDecorate %in_buf BufferBlock\n"
7604 "OpMemberDecorate %in_buf 0 Offset 0\n"
7606 "%void = OpTypeVoid\n"
7607 "%voidf = OpTypeFunction %void\n"
7608 "%u32 = OpTypeInt 32 0\n"
7609 "%i32 = OpTypeInt 32 1\n"
7610 "%uvec3 = OpTypeVector %u32 3\n"
7611 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7613 "%in_type = ${typeDeclaration}\n"
7614 // "%const = OpConstant %in_type ${constValue}\n"
7615 "${constDeclaration}\n"
7617 "%in_ptr = OpTypePointer Uniform %in_type\n"
7618 "%in_arr = OpTypeRuntimeArray %in_type\n"
7619 "%in_buf = OpTypeStruct %in_arr\n"
7620 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
7621 "%indata = OpVariable %in_bufptr Uniform\n"
7622 "%outdata = OpVariable %in_bufptr Uniform\n"
7623 "%id = OpVariable %uvec3ptr Input\n"
7624 "%var_ptr = OpTypePointer Function %in_type\n"
7626 "%zero = OpConstant %i32 0\n"
7628 "%main = OpFunction %void None %voidf\n"
7629 "%label = OpLabel\n"
7630 "%out_var = OpVariable %var_ptr Function ${variableInitializer}\n"
7631 "%idval = OpLoad %uvec3 %id\n"
7632 "%x = OpCompositeExtract %u32 %idval 0\n"
7633 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
7634 "%outloc = OpAccessChain %in_ptr %outdata %zero %x\n"
7636 "%outval = OpLoad %in_type %out_var\n"
7637 " OpStore %outloc %outval\n"
7640 ).specialize(parameters);
7643 bool compareFloats (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
7645 DE_ASSERT(outputAllocs.size() != 0);
7646 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
7648 // Use custom epsilon because of the float->string conversion
7649 const float epsilon = 0.00001f;
7651 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
7654 memcpy(&expected, expectedOutputs[outputNdx]->data(), expectedOutputs[outputNdx]->getNumBytes());
7657 memcpy(&actual, outputAllocs[outputNdx]->getHostPtr(), expectedOutputs[outputNdx]->getNumBytes());
7659 // Test with epsilon
7660 if (fabs(expected - actual) > epsilon)
7662 log << TestLog::Message << "Error: The actual and expected values not matching."
7663 << " Expected: " << expected << " Actual: " << actual << " Epsilon: " << epsilon << TestLog::EndMessage;
7670 // Checks if the driver crash with uninitialized cases
7671 bool passthruVerify (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
7673 DE_ASSERT(outputAllocs.size() != 0);
7674 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
7676 // Copy and discard the result.
7677 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
7679 size_t width = expectedOutputs[outputNdx]->getNumBytes();
7681 vector<char> data(width);
7682 memcpy(&data[0], outputAllocs[outputNdx]->getHostPtr(), width);
7687 tcu::TestCaseGroup* createShaderDefaultOutputGroup (tcu::TestContext& testCtx)
7689 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "shader_default_output", "Test shader default output."));
7690 de::Random rnd (deStringHash(group->getName()));
7692 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7694 NumberType numberType = NumberType(type);
7695 const string typeName = getNumberTypeName(numberType);
7696 const string description = "Test the OpVariable initializer with " + typeName + ".";
7697 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7699 // 2 similar subcases (initialized and uninitialized)
7700 for (int subCase = 0; subCase < 2; ++subCase)
7702 ComputeShaderSpec spec;
7703 spec.numWorkGroups = IVec3(1, 1, 1);
7705 map<string, string> params;
7709 case NUMBERTYPE_INT32:
7711 deInt32 number = getInt(rnd);
7712 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7713 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7714 params["constValue"] = numberToString(number);
7717 case NUMBERTYPE_UINT32:
7719 deUint32 number = rnd.getUint32();
7720 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7721 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7722 params["constValue"] = numberToString(number);
7725 case NUMBERTYPE_FLOAT32:
7727 float number = rnd.getFloat();
7728 spec.inputs.push_back(createCompositeBuffer<float>(number));
7729 spec.outputs.push_back(createCompositeBuffer<float>(number));
7730 spec.verifyIO = &compareFloats;
7731 params["constValue"] = numberToString(number);
7738 // Initialized subcase
7741 spec.assembly = specializeDefaultOutputShaderTemplate(numberType, params);
7742 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "initialized", "OpVariable initializer tests.", spec));
7744 // Uninitialized subcase
7747 spec.assembly = specializeDefaultOutputShaderTemplate(numberType);
7748 spec.verifyIO = &passthruVerify;
7749 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "uninitialized", "OpVariable initializer tests.", spec));
7752 group->addChild(subGroup.release());
7754 return group.release();
7757 tcu::TestCaseGroup* createOpNopTests (tcu::TestContext& testCtx)
7759 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
7760 RGBA defaultColors[4];
7761 map<string, string> opNopFragments;
7763 getDefaultColors(defaultColors);
7765 opNopFragments["testfun"] =
7766 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7767 "%param1 = OpFunctionParameter %v4f32\n"
7768 "%label_testfun = OpLabel\n"
7777 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7778 "%b = OpFAdd %f32 %a %a\n"
7780 "%c = OpFSub %f32 %b %a\n"
7781 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
7784 "OpReturnValue %ret\n"
7787 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, testGroup.get());
7789 return testGroup.release();
7792 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
7794 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
7795 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
7796 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
7798 computeTests->addChild(createOpNopGroup(testCtx));
7799 computeTests->addChild(createOpFUnordGroup(testCtx));
7800 computeTests->addChild(createOpAtomicGroup(testCtx, false));
7801 computeTests->addChild(createOpAtomicGroup(testCtx, true)); // Using new StorageBuffer decoration
7802 computeTests->addChild(createOpLineGroup(testCtx));
7803 computeTests->addChild(createOpNoLineGroup(testCtx));
7804 computeTests->addChild(createOpConstantNullGroup(testCtx));
7805 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
7806 computeTests->addChild(createOpConstantUsageGroup(testCtx));
7807 computeTests->addChild(createSpecConstantGroup(testCtx));
7808 computeTests->addChild(createOpSourceGroup(testCtx));
7809 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
7810 computeTests->addChild(createDecorationGroupGroup(testCtx));
7811 computeTests->addChild(createOpPhiGroup(testCtx));
7812 computeTests->addChild(createLoopControlGroup(testCtx));
7813 computeTests->addChild(createFunctionControlGroup(testCtx));
7814 computeTests->addChild(createSelectionControlGroup(testCtx));
7815 computeTests->addChild(createBlockOrderGroup(testCtx));
7816 computeTests->addChild(createMultipleShaderGroup(testCtx));
7817 computeTests->addChild(createMemoryAccessGroup(testCtx));
7818 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
7819 computeTests->addChild(createOpCopyObjectGroup(testCtx));
7820 computeTests->addChild(createNoContractionGroup(testCtx));
7821 computeTests->addChild(createOpUndefGroup(testCtx));
7822 computeTests->addChild(createOpUnreachableGroup(testCtx));
7823 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
7824 computeTests ->addChild(createOpFRemGroup(testCtx));
7825 computeTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_PASS));
7826 computeTests->addChild(createOpSRemComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
7827 computeTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_PASS));
7828 computeTests->addChild(createOpSModComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
7829 computeTests->addChild(createSConvertTests(testCtx));
7830 computeTests->addChild(createUConvertTests(testCtx));
7831 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
7832 computeTests->addChild(createOpInBoundsAccessChainGroup(testCtx));
7833 computeTests->addChild(createShaderDefaultOutputGroup(testCtx));
7834 computeTests->addChild(createOpNMinGroup(testCtx));
7835 computeTests->addChild(createOpNMaxGroup(testCtx));
7836 computeTests->addChild(createOpNClampGroup(testCtx));
7838 de::MovePtr<tcu::TestCaseGroup> computeAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
7840 computeAndroidTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7841 computeAndroidTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7843 computeTests->addChild(computeAndroidTests.release());
7846 computeTests->addChild(create16BitStorageComputeGroup(testCtx));
7847 computeTests->addChild(createUboMatrixPaddingComputeGroup(testCtx));
7848 computeTests->addChild(createConditionalBranchComputeGroup(testCtx));
7849 computeTests->addChild(createIndexingComputeGroup(testCtx));
7850 computeTests->addChild(createVariablePointersComputeGroup(testCtx));
7851 graphicsTests->addChild(createOpNopTests(testCtx));
7852 graphicsTests->addChild(createOpSourceTests(testCtx));
7853 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
7854 graphicsTests->addChild(createOpLineTests(testCtx));
7855 graphicsTests->addChild(createOpNoLineTests(testCtx));
7856 graphicsTests->addChild(createOpConstantNullTests(testCtx));
7857 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
7858 graphicsTests->addChild(createMemoryAccessTests(testCtx));
7859 graphicsTests->addChild(createOpUndefTests(testCtx));
7860 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
7861 graphicsTests->addChild(createModuleTests(testCtx));
7862 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
7863 graphicsTests->addChild(createOpPhiTests(testCtx));
7864 graphicsTests->addChild(createNoContractionTests(testCtx));
7865 graphicsTests->addChild(createOpQuantizeTests(testCtx));
7866 graphicsTests->addChild(createLoopTests(testCtx));
7867 graphicsTests->addChild(createSpecConstantTests(testCtx));
7868 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
7869 graphicsTests->addChild(createBarrierTests(testCtx));
7870 graphicsTests->addChild(createDecorationGroupTests(testCtx));
7871 graphicsTests->addChild(createFRemTests(testCtx));
7872 graphicsTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
7873 graphicsTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
7876 de::MovePtr<tcu::TestCaseGroup> graphicsAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
7878 graphicsAndroidTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7879 graphicsAndroidTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7881 graphicsTests->addChild(graphicsAndroidTests.release());
7884 graphicsTests->addChild(create16BitStorageGraphicsGroup(testCtx));
7885 graphicsTests->addChild(createUboMatrixPaddingGraphicsGroup(testCtx));
7886 graphicsTests->addChild(createConditionalBranchGraphicsGroup(testCtx));
7887 graphicsTests->addChild(createIndexingGraphicsGroup(testCtx));
7888 graphicsTests->addChild(createVariablePointersGraphicsGroup(testCtx));
7890 instructionTests->addChild(computeTests.release());
7891 instructionTests->addChild(graphicsTests.release());
7893 return instructionTests.release();