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 string generateConstantDefinitions (int count)
2529 std::stringstream r;
2530 for (int i = 0; i < count; i++)
2531 r << "%cf" << (i * 10 + 5) << " = OpConstant %f32 " <<(i * 10 + 5) << ".0\n";
2532 return r.str() + string("\n");
2535 string generateSwitchCases (int count)
2537 std::stringstream r;
2538 for (int i = 0; i < count; i++)
2539 r << " " << i << " %case" << i;
2540 return r.str() + string("\n");
2543 string generateSwitchTargets (int count)
2545 std::stringstream r;
2546 for (int i = 0; i < count; i++)
2547 r << "%case" << i << " = OpLabel\n OpBranch %phi\n";
2548 return r.str() + string("\n");
2551 string generateOpPhiParams (int count)
2553 std::stringstream r;
2554 for (int i = 0; i < count; i++)
2555 r << " %cf" << (i * 10 + 5) << " %case" << i;
2556 return r.str() + string("\n");
2559 string generateIntWidth (int value)
2561 std::stringstream r;
2566 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
2568 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
2569 ComputeShaderSpec spec1;
2570 ComputeShaderSpec spec2;
2571 ComputeShaderSpec spec3;
2572 ComputeShaderSpec spec4;
2573 de::Random rnd (deStringHash(group->getName()));
2574 const int numElements = 100;
2575 vector<float> inputFloats (numElements, 0);
2576 vector<float> outputFloats1 (numElements, 0);
2577 vector<float> outputFloats2 (numElements, 0);
2578 vector<float> outputFloats3 (numElements, 0);
2579 vector<float> outputFloats4 (numElements, 0);
2580 const int test4Width = 1024;
2582 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
2584 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2585 floorAll(inputFloats);
2587 for (size_t ndx = 0; ndx < numElements; ++ndx)
2591 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
2592 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
2593 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
2596 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
2597 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
2599 int index4 = (int)deFloor(deAbs((float)ndx * inputFloats[ndx]));
2600 outputFloats4[ndx] = (float)(index4 % test4Width) * 10.0f + 5.0f;
2604 string(getComputeAsmShaderPreamble()) +
2606 "OpSource GLSL 430\n"
2607 "OpName %main \"main\"\n"
2608 "OpName %id \"gl_GlobalInvocationID\"\n"
2610 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2612 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2614 "%id = OpVariable %uvec3ptr Input\n"
2615 "%zero = OpConstant %i32 0\n"
2616 "%three = OpConstant %u32 3\n"
2617 "%constf5p5 = OpConstant %f32 5.5\n"
2618 "%constf20p5 = OpConstant %f32 20.5\n"
2619 "%constf1p75 = OpConstant %f32 1.75\n"
2620 "%constf8p5 = OpConstant %f32 8.5\n"
2621 "%constf6p5 = OpConstant %f32 6.5\n"
2623 "%main = OpFunction %void None %voidf\n"
2624 "%entry = OpLabel\n"
2625 "%idval = OpLoad %uvec3 %id\n"
2626 "%x = OpCompositeExtract %u32 %idval 0\n"
2627 "%selector = OpUMod %u32 %x %three\n"
2628 " OpSelectionMerge %phi None\n"
2629 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
2631 // Case 1 before OpPhi.
2632 "%case1 = OpLabel\n"
2635 "%default = OpLabel\n"
2639 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
2640 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2641 "%inval = OpLoad %f32 %inloc\n"
2642 "%add = OpFAdd %f32 %inval %operand\n"
2643 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2644 " OpStore %outloc %add\n"
2647 // Case 0 after OpPhi.
2648 "%case0 = OpLabel\n"
2652 // Case 2 after OpPhi.
2653 "%case2 = OpLabel\n"
2657 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2658 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
2659 spec1.numWorkGroups = IVec3(numElements, 1, 1);
2661 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
2664 string(getComputeAsmShaderPreamble()) +
2666 "OpName %main \"main\"\n"
2667 "OpName %id \"gl_GlobalInvocationID\"\n"
2669 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2671 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2673 "%id = OpVariable %uvec3ptr Input\n"
2674 "%zero = OpConstant %i32 0\n"
2675 "%one = OpConstant %i32 1\n"
2676 "%three = OpConstant %i32 3\n"
2677 "%constf6p5 = OpConstant %f32 6.5\n"
2679 "%main = OpFunction %void None %voidf\n"
2680 "%entry = OpLabel\n"
2681 "%idval = OpLoad %uvec3 %id\n"
2682 "%x = OpCompositeExtract %u32 %idval 0\n"
2683 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2684 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2685 "%inval = OpLoad %f32 %inloc\n"
2689 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
2690 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
2691 "%step_next = OpIAdd %i32 %step %one\n"
2692 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
2693 "%still_loop = OpSLessThan %bool %step %three\n"
2694 " OpLoopMerge %exit %phi None\n"
2695 " OpBranchConditional %still_loop %phi %exit\n"
2698 " OpStore %outloc %accum\n"
2701 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2702 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
2703 spec2.numWorkGroups = IVec3(numElements, 1, 1);
2705 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
2708 string(getComputeAsmShaderPreamble()) +
2710 "OpName %main \"main\"\n"
2711 "OpName %id \"gl_GlobalInvocationID\"\n"
2713 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2715 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2717 "%f32ptr_f = OpTypePointer Function %f32\n"
2718 "%id = OpVariable %uvec3ptr Input\n"
2719 "%true = OpConstantTrue %bool\n"
2720 "%false = OpConstantFalse %bool\n"
2721 "%zero = OpConstant %i32 0\n"
2722 "%constf8p5 = OpConstant %f32 8.5\n"
2724 "%main = OpFunction %void None %voidf\n"
2725 "%entry = OpLabel\n"
2726 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
2727 "%idval = OpLoad %uvec3 %id\n"
2728 "%x = OpCompositeExtract %u32 %idval 0\n"
2729 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2730 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2731 "%a_init = OpLoad %f32 %inloc\n"
2732 "%b_init = OpLoad %f32 %b\n"
2736 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
2737 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
2738 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
2739 " OpLoopMerge %exit %phi None\n"
2740 " OpBranchConditional %still_loop %phi %exit\n"
2743 "%sub = OpFSub %f32 %a_next %b_next\n"
2744 " OpStore %outloc %sub\n"
2747 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2748 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
2749 spec3.numWorkGroups = IVec3(numElements, 1, 1);
2751 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
2754 "OpCapability Shader\n"
2755 "%ext = OpExtInstImport \"GLSL.std.450\"\n"
2756 "OpMemoryModel Logical GLSL450\n"
2757 "OpEntryPoint GLCompute %main \"main\" %id\n"
2758 "OpExecutionMode %main LocalSize 1 1 1\n"
2760 "OpSource GLSL 430\n"
2761 "OpName %main \"main\"\n"
2762 "OpName %id \"gl_GlobalInvocationID\"\n"
2764 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2766 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
2768 "%id = OpVariable %uvec3ptr Input\n"
2769 "%zero = OpConstant %i32 0\n"
2770 "%cimod = OpConstant %u32 " + generateIntWidth(test4Width) + "\n"
2772 + generateConstantDefinitions(test4Width) +
2774 "%main = OpFunction %void None %voidf\n"
2775 "%entry = OpLabel\n"
2776 "%idval = OpLoad %uvec3 %id\n"
2777 "%x = OpCompositeExtract %u32 %idval 0\n"
2778 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2779 "%inval = OpLoad %f32 %inloc\n"
2780 "%xf = OpConvertUToF %f32 %x\n"
2781 "%xm = OpFMul %f32 %xf %inval\n"
2782 "%xa = OpExtInst %f32 %ext FAbs %xm\n"
2783 "%xi = OpConvertFToU %u32 %xa\n"
2784 "%selector = OpUMod %u32 %xi %cimod\n"
2785 " OpSelectionMerge %phi None\n"
2786 " OpSwitch %selector %default "
2788 + generateSwitchCases(test4Width) +
2790 "%default = OpLabel\n"
2793 + generateSwitchTargets(test4Width) +
2796 "%result = OpPhi %f32"
2798 + generateOpPhiParams(test4Width) +
2800 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2801 " OpStore %outloc %result\n"
2805 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2806 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
2807 spec4.numWorkGroups = IVec3(numElements, 1, 1);
2809 group->addChild(new SpvAsmComputeShaderCase(testCtx, "wide", "OpPhi with a lot of parameters", spec4));
2811 return group.release();
2814 // Assembly code used for testing block order is based on GLSL source code:
2818 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2819 // float elements[];
2821 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2822 // float elements[];
2826 // uint x = gl_GlobalInvocationID.x;
2827 // output_data.elements[x] = input_data.elements[x];
2828 // if (x > uint(50)) {
2829 // switch (x % uint(3)) {
2830 // case 0: output_data.elements[x] += 1.5f; break;
2831 // case 1: output_data.elements[x] += 42.f; break;
2832 // case 2: output_data.elements[x] -= 27.f; break;
2836 // output_data.elements[x] = -input_data.elements[x];
2839 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
2841 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
2842 ComputeShaderSpec spec;
2843 de::Random rnd (deStringHash(group->getName()));
2844 const int numElements = 100;
2845 vector<float> inputFloats (numElements, 0);
2846 vector<float> outputFloats (numElements, 0);
2848 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2850 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2851 floorAll(inputFloats);
2853 for (size_t ndx = 0; ndx <= 50; ++ndx)
2854 outputFloats[ndx] = -inputFloats[ndx];
2856 for (size_t ndx = 51; ndx < numElements; ++ndx)
2860 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
2861 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
2862 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
2868 string(getComputeAsmShaderPreamble()) +
2870 "OpSource GLSL 430\n"
2871 "OpName %main \"main\"\n"
2872 "OpName %id \"gl_GlobalInvocationID\"\n"
2874 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2876 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
2878 "%u32ptr = OpTypePointer Function %u32\n"
2879 "%u32ptr_input = OpTypePointer Input %u32\n"
2881 + string(getComputeAsmInputOutputBuffer()) +
2883 "%id = OpVariable %uvec3ptr Input\n"
2884 "%zero = OpConstant %i32 0\n"
2885 "%const3 = OpConstant %u32 3\n"
2886 "%const50 = OpConstant %u32 50\n"
2887 "%constf1p5 = OpConstant %f32 1.5\n"
2888 "%constf27 = OpConstant %f32 27.0\n"
2889 "%constf42 = OpConstant %f32 42.0\n"
2891 "%main = OpFunction %void None %voidf\n"
2894 "%entry = OpLabel\n"
2896 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
2897 "%xvar = OpVariable %u32ptr Function\n"
2898 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
2899 "%x = OpLoad %u32 %xptr\n"
2900 " OpStore %xvar %x\n"
2902 "%cmp = OpUGreaterThan %bool %x %const50\n"
2903 " OpSelectionMerge %if_merge None\n"
2904 " OpBranchConditional %cmp %if_true %if_false\n"
2906 // False branch for if-statement: placed in the middle of switch cases and before true branch.
2907 "%if_false = OpLabel\n"
2908 "%x_f = OpLoad %u32 %xvar\n"
2909 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
2910 "%inval_f = OpLoad %f32 %inloc_f\n"
2911 "%negate = OpFNegate %f32 %inval_f\n"
2912 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
2913 " OpStore %outloc_f %negate\n"
2914 " OpBranch %if_merge\n"
2916 // Merge block for if-statement: placed in the middle of true and false branch.
2917 "%if_merge = OpLabel\n"
2920 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
2921 "%if_true = OpLabel\n"
2922 "%xval_t = OpLoad %u32 %xvar\n"
2923 "%mod = OpUMod %u32 %xval_t %const3\n"
2924 " OpSelectionMerge %switch_merge None\n"
2925 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
2927 // Merge block for switch-statement: placed before the case
2928 // bodies. But it must follow OpSwitch which dominates it.
2929 "%switch_merge = OpLabel\n"
2930 " OpBranch %if_merge\n"
2932 // Case 1 for switch-statement: placed before case 0.
2933 // It must follow the OpSwitch that dominates it.
2934 "%case1 = OpLabel\n"
2935 "%x_1 = OpLoad %u32 %xvar\n"
2936 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
2937 "%inval_1 = OpLoad %f32 %inloc_1\n"
2938 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
2939 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
2940 " OpStore %outloc_1 %addf42\n"
2941 " OpBranch %switch_merge\n"
2943 // Case 2 for switch-statement.
2944 "%case2 = OpLabel\n"
2945 "%x_2 = OpLoad %u32 %xvar\n"
2946 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
2947 "%inval_2 = OpLoad %f32 %inloc_2\n"
2948 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
2949 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
2950 " OpStore %outloc_2 %subf27\n"
2951 " OpBranch %switch_merge\n"
2953 // Default case for switch-statement: placed in the middle of normal cases.
2954 "%default = OpLabel\n"
2955 " OpBranch %switch_merge\n"
2957 // Case 0 for switch-statement: out of order.
2958 "%case0 = OpLabel\n"
2959 "%x_0 = OpLoad %u32 %xvar\n"
2960 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
2961 "%inval_0 = OpLoad %f32 %inloc_0\n"
2962 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
2963 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
2964 " OpStore %outloc_0 %addf1p5\n"
2965 " OpBranch %switch_merge\n"
2968 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2969 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2970 spec.numWorkGroups = IVec3(numElements, 1, 1);
2972 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
2974 return group.release();
2977 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
2979 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
2980 ComputeShaderSpec spec1;
2981 ComputeShaderSpec spec2;
2982 de::Random rnd (deStringHash(group->getName()));
2983 const int numElements = 100;
2984 vector<float> inputFloats (numElements, 0);
2985 vector<float> outputFloats1 (numElements, 0);
2986 vector<float> outputFloats2 (numElements, 0);
2987 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
2989 for (size_t ndx = 0; ndx < numElements; ++ndx)
2991 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
2992 outputFloats2[ndx] = -inputFloats[ndx];
2995 const string assembly(
2996 "OpCapability Shader\n"
2997 "OpCapability ClipDistance\n"
2998 "OpMemoryModel Logical GLSL450\n"
2999 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
3000 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
3001 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
3002 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
3003 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
3004 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
3006 "OpName %comp_main1 \"entrypoint1\"\n"
3007 "OpName %comp_main2 \"entrypoint2\"\n"
3008 "OpName %vert_main \"entrypoint2\"\n"
3009 "OpName %id \"gl_GlobalInvocationID\"\n"
3010 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
3011 "OpName %vertexIndex \"gl_VertexIndex\"\n"
3012 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
3013 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
3014 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
3015 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
3017 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3018 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
3019 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
3020 "OpDecorate %vert_builtin_st Block\n"
3021 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
3022 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
3023 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
3025 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3027 "%zero = OpConstant %i32 0\n"
3028 "%one = OpConstant %u32 1\n"
3029 "%c_f32_1 = OpConstant %f32 1\n"
3031 "%i32inputptr = OpTypePointer Input %i32\n"
3032 "%vec4 = OpTypeVector %f32 4\n"
3033 "%vec4ptr = OpTypePointer Output %vec4\n"
3034 "%f32arr1 = OpTypeArray %f32 %one\n"
3035 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
3036 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
3037 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
3039 "%id = OpVariable %uvec3ptr Input\n"
3040 "%vertexIndex = OpVariable %i32inputptr Input\n"
3041 "%instanceIndex = OpVariable %i32inputptr Input\n"
3042 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
3044 // gl_Position = vec4(1.);
3045 "%vert_main = OpFunction %void None %voidf\n"
3046 "%vert_entry = OpLabel\n"
3047 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
3048 " OpStore %position %c_vec4_1\n"
3053 "%comp_main1 = OpFunction %void None %voidf\n"
3054 "%comp1_entry = OpLabel\n"
3055 "%idval1 = OpLoad %uvec3 %id\n"
3056 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
3057 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
3058 "%inval1 = OpLoad %f32 %inloc1\n"
3059 "%add = OpFAdd %f32 %inval1 %inval1\n"
3060 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
3061 " OpStore %outloc1 %add\n"
3066 "%comp_main2 = OpFunction %void None %voidf\n"
3067 "%comp2_entry = OpLabel\n"
3068 "%idval2 = OpLoad %uvec3 %id\n"
3069 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
3070 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
3071 "%inval2 = OpLoad %f32 %inloc2\n"
3072 "%neg = OpFNegate %f32 %inval2\n"
3073 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
3074 " OpStore %outloc2 %neg\n"
3076 " OpFunctionEnd\n");
3078 spec1.assembly = assembly;
3079 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3080 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
3081 spec1.numWorkGroups = IVec3(numElements, 1, 1);
3082 spec1.entryPoint = "entrypoint1";
3084 spec2.assembly = assembly;
3085 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3086 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
3087 spec2.numWorkGroups = IVec3(numElements, 1, 1);
3088 spec2.entryPoint = "entrypoint2";
3090 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
3091 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
3093 return group.release();
3096 inline std::string makeLongUTF8String (size_t num4ByteChars)
3098 // An example of a longest valid UTF-8 character. Be explicit about the
3099 // character type because Microsoft compilers can otherwise interpret the
3100 // character string as being over wide (16-bit) characters. Ideally, we
3101 // would just use a C++11 UTF-8 string literal, but we want to support older
3102 // Microsoft compilers.
3103 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
3104 std::string longString;
3105 longString.reserve(num4ByteChars * 4);
3106 for (size_t count = 0; count < num4ByteChars; count++)
3108 longString += earthAfrica;
3113 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
3115 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
3116 vector<CaseParameter> cases;
3117 de::Random rnd (deStringHash(group->getName()));
3118 const int numElements = 100;
3119 vector<float> positiveFloats (numElements, 0);
3120 vector<float> negativeFloats (numElements, 0);
3121 const StringTemplate shaderTemplate (
3122 "OpCapability Shader\n"
3123 "OpMemoryModel Logical GLSL450\n"
3125 "OpEntryPoint GLCompute %main \"main\" %id\n"
3126 "OpExecutionMode %main LocalSize 1 1 1\n"
3130 "OpName %main \"main\"\n"
3131 "OpName %id \"gl_GlobalInvocationID\"\n"
3133 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3135 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3137 "%id = OpVariable %uvec3ptr Input\n"
3138 "%zero = OpConstant %i32 0\n"
3140 "%main = OpFunction %void None %voidf\n"
3141 "%label = OpLabel\n"
3142 "%idval = OpLoad %uvec3 %id\n"
3143 "%x = OpCompositeExtract %u32 %idval 0\n"
3144 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3145 "%inval = OpLoad %f32 %inloc\n"
3146 "%neg = OpFNegate %f32 %inval\n"
3147 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3148 " OpStore %outloc %neg\n"
3150 " OpFunctionEnd\n");
3152 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
3153 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
3154 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
3155 "OpSource GLSL 430 %fname"));
3156 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
3157 "OpSource GLSL 430 %fname"));
3158 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
3159 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
3160 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
3161 "OpSource GLSL 430 %fname \"\""));
3162 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
3163 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
3164 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
3165 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
3166 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
3167 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
3168 "OpSourceContinued \"id main() {}\""));
3169 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
3170 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3171 "OpSourceContinued \"\""));
3172 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
3173 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3174 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
3175 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
3176 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
3177 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
3178 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
3179 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
3180 "OpSourceContinued \"void\"\n"
3181 "OpSourceContinued \"main()\"\n"
3182 "OpSourceContinued \"{}\""));
3183 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
3184 "OpSource GLSL 430 %fname \"\"\n"
3185 "OpSourceContinued \"#version 430\nvoid main() {}\""));
3187 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3189 for (size_t ndx = 0; ndx < numElements; ++ndx)
3190 negativeFloats[ndx] = -positiveFloats[ndx];
3192 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3194 map<string, string> specializations;
3195 ComputeShaderSpec spec;
3197 specializations["SOURCE"] = cases[caseNdx].param;
3198 spec.assembly = shaderTemplate.specialize(specializations);
3199 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3200 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3201 spec.numWorkGroups = IVec3(numElements, 1, 1);
3203 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3206 return group.release();
3209 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
3211 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
3212 vector<CaseParameter> cases;
3213 de::Random rnd (deStringHash(group->getName()));
3214 const int numElements = 100;
3215 vector<float> inputFloats (numElements, 0);
3216 vector<float> outputFloats (numElements, 0);
3217 const StringTemplate shaderTemplate (
3218 string(getComputeAsmShaderPreamble()) +
3220 "OpSourceExtension \"${EXTENSION}\"\n"
3222 "OpName %main \"main\"\n"
3223 "OpName %id \"gl_GlobalInvocationID\"\n"
3225 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3227 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3229 "%id = OpVariable %uvec3ptr Input\n"
3230 "%zero = OpConstant %i32 0\n"
3232 "%main = OpFunction %void None %voidf\n"
3233 "%label = OpLabel\n"
3234 "%idval = OpLoad %uvec3 %id\n"
3235 "%x = OpCompositeExtract %u32 %idval 0\n"
3236 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3237 "%inval = OpLoad %f32 %inloc\n"
3238 "%neg = OpFNegate %f32 %inval\n"
3239 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3240 " OpStore %outloc %neg\n"
3242 " OpFunctionEnd\n");
3244 cases.push_back(CaseParameter("empty_extension", ""));
3245 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
3246 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
3247 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
3248 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
3250 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
3252 for (size_t ndx = 0; ndx < numElements; ++ndx)
3253 outputFloats[ndx] = -inputFloats[ndx];
3255 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3257 map<string, string> specializations;
3258 ComputeShaderSpec spec;
3260 specializations["EXTENSION"] = cases[caseNdx].param;
3261 spec.assembly = shaderTemplate.specialize(specializations);
3262 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3263 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3264 spec.numWorkGroups = IVec3(numElements, 1, 1);
3266 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3269 return group.release();
3272 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
3273 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
3275 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
3276 vector<CaseParameter> cases;
3277 de::Random rnd (deStringHash(group->getName()));
3278 const int numElements = 100;
3279 vector<float> positiveFloats (numElements, 0);
3280 vector<float> negativeFloats (numElements, 0);
3281 const StringTemplate shaderTemplate (
3282 string(getComputeAsmShaderPreamble()) +
3284 "OpSource GLSL 430\n"
3285 "OpName %main \"main\"\n"
3286 "OpName %id \"gl_GlobalInvocationID\"\n"
3288 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3290 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3291 "%uvec2 = OpTypeVector %u32 2\n"
3292 "%bvec3 = OpTypeVector %bool 3\n"
3293 "%fvec4 = OpTypeVector %f32 4\n"
3294 "%fmat33 = OpTypeMatrix %fvec3 3\n"
3295 "%const100 = OpConstant %u32 100\n"
3296 "%uarr100 = OpTypeArray %i32 %const100\n"
3297 "%struct = OpTypeStruct %f32 %i32 %u32\n"
3298 "%pointer = OpTypePointer Function %i32\n"
3299 + string(getComputeAsmInputOutputBuffer()) +
3301 "%null = OpConstantNull ${TYPE}\n"
3303 "%id = OpVariable %uvec3ptr Input\n"
3304 "%zero = OpConstant %i32 0\n"
3306 "%main = OpFunction %void None %voidf\n"
3307 "%label = OpLabel\n"
3308 "%idval = OpLoad %uvec3 %id\n"
3309 "%x = OpCompositeExtract %u32 %idval 0\n"
3310 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3311 "%inval = OpLoad %f32 %inloc\n"
3312 "%neg = OpFNegate %f32 %inval\n"
3313 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3314 " OpStore %outloc %neg\n"
3316 " OpFunctionEnd\n");
3318 cases.push_back(CaseParameter("bool", "%bool"));
3319 cases.push_back(CaseParameter("sint32", "%i32"));
3320 cases.push_back(CaseParameter("uint32", "%u32"));
3321 cases.push_back(CaseParameter("float32", "%f32"));
3322 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
3323 cases.push_back(CaseParameter("vec3bool", "%bvec3"));
3324 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
3325 cases.push_back(CaseParameter("matrix", "%fmat33"));
3326 cases.push_back(CaseParameter("array", "%uarr100"));
3327 cases.push_back(CaseParameter("struct", "%struct"));
3328 cases.push_back(CaseParameter("pointer", "%pointer"));
3330 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3332 for (size_t ndx = 0; ndx < numElements; ++ndx)
3333 negativeFloats[ndx] = -positiveFloats[ndx];
3335 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3337 map<string, string> specializations;
3338 ComputeShaderSpec spec;
3340 specializations["TYPE"] = cases[caseNdx].param;
3341 spec.assembly = shaderTemplate.specialize(specializations);
3342 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3343 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3344 spec.numWorkGroups = IVec3(numElements, 1, 1);
3346 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3349 return group.release();
3352 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
3353 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
3355 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
3356 vector<CaseParameter> cases;
3357 de::Random rnd (deStringHash(group->getName()));
3358 const int numElements = 100;
3359 vector<float> positiveFloats (numElements, 0);
3360 vector<float> negativeFloats (numElements, 0);
3361 const StringTemplate shaderTemplate (
3362 string(getComputeAsmShaderPreamble()) +
3364 "OpSource GLSL 430\n"
3365 "OpName %main \"main\"\n"
3366 "OpName %id \"gl_GlobalInvocationID\"\n"
3368 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3370 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3372 "%id = OpVariable %uvec3ptr Input\n"
3373 "%zero = OpConstant %i32 0\n"
3377 "%main = OpFunction %void None %voidf\n"
3378 "%label = OpLabel\n"
3379 "%idval = OpLoad %uvec3 %id\n"
3380 "%x = OpCompositeExtract %u32 %idval 0\n"
3381 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3382 "%inval = OpLoad %f32 %inloc\n"
3383 "%neg = OpFNegate %f32 %inval\n"
3384 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3385 " OpStore %outloc %neg\n"
3387 " OpFunctionEnd\n");
3389 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
3390 "%const = OpConstantComposite %uvec3 %five %zero %five"));
3391 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
3392 "%ten = OpConstant %f32 10.\n"
3393 "%fzero = OpConstant %f32 0.\n"
3394 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
3395 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
3396 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
3397 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
3398 "%fzero = OpConstant %f32 0.\n"
3399 "%one = OpConstant %f32 1.\n"
3400 "%point5 = OpConstant %f32 0.5\n"
3401 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
3402 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
3403 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
3404 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
3405 "%st2 = OpTypeStruct %i32 %i32\n"
3406 "%struct = OpTypeStruct %st1 %st2\n"
3407 "%point5 = OpConstant %f32 0.5\n"
3408 "%one = OpConstant %u32 1\n"
3409 "%ten = OpConstant %i32 10\n"
3410 "%st1val = OpConstantComposite %st1 %one %point5\n"
3411 "%st2val = OpConstantComposite %st2 %ten %ten\n"
3412 "%const = OpConstantComposite %struct %st1val %st2val"));
3414 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3416 for (size_t ndx = 0; ndx < numElements; ++ndx)
3417 negativeFloats[ndx] = -positiveFloats[ndx];
3419 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3421 map<string, string> specializations;
3422 ComputeShaderSpec spec;
3424 specializations["CONSTANT"] = cases[caseNdx].param;
3425 spec.assembly = shaderTemplate.specialize(specializations);
3426 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3427 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3428 spec.numWorkGroups = IVec3(numElements, 1, 1);
3430 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3433 return group.release();
3436 // Creates a floating point number with the given exponent, and significand
3437 // bits set. It can only create normalized numbers. Only the least significant
3438 // 24 bits of the significand will be examined. The final bit of the
3439 // significand will also be ignored. This allows alignment to be written
3440 // similarly to C99 hex-floats.
3441 // For example if you wanted to write 0x1.7f34p-12 you would call
3442 // constructNormalizedFloat(-12, 0x7f3400)
3443 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
3447 for (deInt32 idx = 0; idx < 23; ++idx)
3449 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
3453 return std::ldexp(f, exponent);
3456 // Compare instruction for the OpQuantizeF16 compute exact case.
3457 // Returns true if the output is what is expected from the test case.
3458 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
3460 if (outputAllocs.size() != 1)
3463 // We really just need this for size because we cannot compare Nans.
3464 const BufferSp& expectedOutput = expectedOutputs[0];
3465 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
3467 if (expectedOutput->getNumBytes() != 4*sizeof(float)) {
3471 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
3472 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
3477 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
3478 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
3483 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
3484 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
3489 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
3490 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
3497 // Checks that every output from a test-case is a float NaN.
3498 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
3500 if (outputAllocs.size() != 1)
3503 // We really just need this for size because we cannot compare Nans.
3504 const BufferSp& expectedOutput = expectedOutputs[0];
3505 const float* output_as_float = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
3507 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
3509 if (!deFloatIsNaN(output_as_float[idx]))
3518 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
3519 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
3521 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
3523 const std::string shader (
3524 string(getComputeAsmShaderPreamble()) +
3526 "OpSource GLSL 430\n"
3527 "OpName %main \"main\"\n"
3528 "OpName %id \"gl_GlobalInvocationID\"\n"
3530 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3532 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3534 "%id = OpVariable %uvec3ptr Input\n"
3535 "%zero = OpConstant %i32 0\n"
3537 "%main = OpFunction %void None %voidf\n"
3538 "%label = OpLabel\n"
3539 "%idval = OpLoad %uvec3 %id\n"
3540 "%x = OpCompositeExtract %u32 %idval 0\n"
3541 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3542 "%inval = OpLoad %f32 %inloc\n"
3543 "%quant = OpQuantizeToF16 %f32 %inval\n"
3544 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3545 " OpStore %outloc %quant\n"
3547 " OpFunctionEnd\n");
3550 ComputeShaderSpec spec;
3551 const deUint32 numElements = 100;
3552 vector<float> infinities;
3553 vector<float> results;
3555 infinities.reserve(numElements);
3556 results.reserve(numElements);
3558 for (size_t idx = 0; idx < numElements; ++idx)
3563 infinities.push_back(std::numeric_limits<float>::infinity());
3564 results.push_back(std::numeric_limits<float>::infinity());
3567 infinities.push_back(-std::numeric_limits<float>::infinity());
3568 results.push_back(-std::numeric_limits<float>::infinity());
3571 infinities.push_back(std::ldexp(1.0f, 16));
3572 results.push_back(std::numeric_limits<float>::infinity());
3575 infinities.push_back(std::ldexp(-1.0f, 32));
3576 results.push_back(-std::numeric_limits<float>::infinity());
3581 spec.assembly = shader;
3582 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
3583 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
3584 spec.numWorkGroups = IVec3(numElements, 1, 1);
3586 group->addChild(new SpvAsmComputeShaderCase(
3587 testCtx, "infinities", "Check that infinities propagated and created", spec));
3591 ComputeShaderSpec spec;
3593 const deUint32 numElements = 100;
3595 nans.reserve(numElements);
3597 for (size_t idx = 0; idx < numElements; ++idx)
3601 nans.push_back(std::numeric_limits<float>::quiet_NaN());
3605 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
3609 spec.assembly = shader;
3610 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
3611 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
3612 spec.numWorkGroups = IVec3(numElements, 1, 1);
3613 spec.verifyIO = &compareNan;
3615 group->addChild(new SpvAsmComputeShaderCase(
3616 testCtx, "propagated_nans", "Check that nans are propagated", spec));
3620 ComputeShaderSpec spec;
3621 vector<float> small;
3622 vector<float> zeros;
3623 const deUint32 numElements = 100;
3625 small.reserve(numElements);
3626 zeros.reserve(numElements);
3628 for (size_t idx = 0; idx < numElements; ++idx)
3633 small.push_back(0.f);
3634 zeros.push_back(0.f);
3637 small.push_back(-0.f);
3638 zeros.push_back(-0.f);
3641 small.push_back(std::ldexp(1.0f, -16));
3642 zeros.push_back(0.f);
3645 small.push_back(std::ldexp(-1.0f, -32));
3646 zeros.push_back(-0.f);
3649 small.push_back(std::ldexp(1.0f, -127));
3650 zeros.push_back(0.f);
3653 small.push_back(-std::ldexp(1.0f, -128));
3654 zeros.push_back(-0.f);
3659 spec.assembly = shader;
3660 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
3661 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
3662 spec.numWorkGroups = IVec3(numElements, 1, 1);
3664 group->addChild(new SpvAsmComputeShaderCase(
3665 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
3669 ComputeShaderSpec spec;
3670 vector<float> exact;
3671 const deUint32 numElements = 200;
3673 exact.reserve(numElements);
3675 for (size_t idx = 0; idx < numElements; ++idx)
3676 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
3678 spec.assembly = shader;
3679 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
3680 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
3681 spec.numWorkGroups = IVec3(numElements, 1, 1);
3683 group->addChild(new SpvAsmComputeShaderCase(
3684 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
3688 ComputeShaderSpec spec;
3689 vector<float> inputs;
3690 const deUint32 numElements = 4;
3692 inputs.push_back(constructNormalizedFloat(8, 0x300300));
3693 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
3694 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
3695 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
3697 spec.assembly = shader;
3698 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
3699 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3700 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
3701 spec.numWorkGroups = IVec3(numElements, 1, 1);
3703 group->addChild(new SpvAsmComputeShaderCase(
3704 testCtx, "rounded", "Check that are rounded when needed", spec));
3707 return group.release();
3710 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
3712 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
3714 const std::string shader (
3715 string(getComputeAsmShaderPreamble()) +
3717 "OpName %main \"main\"\n"
3718 "OpName %id \"gl_GlobalInvocationID\"\n"
3720 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3722 "OpDecorate %sc_0 SpecId 0\n"
3723 "OpDecorate %sc_1 SpecId 1\n"
3724 "OpDecorate %sc_2 SpecId 2\n"
3725 "OpDecorate %sc_3 SpecId 3\n"
3726 "OpDecorate %sc_4 SpecId 4\n"
3727 "OpDecorate %sc_5 SpecId 5\n"
3729 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
3731 "%id = OpVariable %uvec3ptr Input\n"
3732 "%zero = OpConstant %i32 0\n"
3733 "%c_u32_6 = OpConstant %u32 6\n"
3735 "%sc_0 = OpSpecConstant %f32 0.\n"
3736 "%sc_1 = OpSpecConstant %f32 0.\n"
3737 "%sc_2 = OpSpecConstant %f32 0.\n"
3738 "%sc_3 = OpSpecConstant %f32 0.\n"
3739 "%sc_4 = OpSpecConstant %f32 0.\n"
3740 "%sc_5 = OpSpecConstant %f32 0.\n"
3742 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
3743 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
3744 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
3745 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
3746 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
3747 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
3749 "%main = OpFunction %void None %voidf\n"
3750 "%label = OpLabel\n"
3751 "%idval = OpLoad %uvec3 %id\n"
3752 "%x = OpCompositeExtract %u32 %idval 0\n"
3753 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3754 "%selector = OpUMod %u32 %x %c_u32_6\n"
3755 " OpSelectionMerge %exit None\n"
3756 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
3758 "%case0 = OpLabel\n"
3759 " OpStore %outloc %sc_0_quant\n"
3762 "%case1 = OpLabel\n"
3763 " OpStore %outloc %sc_1_quant\n"
3766 "%case2 = OpLabel\n"
3767 " OpStore %outloc %sc_2_quant\n"
3770 "%case3 = OpLabel\n"
3771 " OpStore %outloc %sc_3_quant\n"
3774 "%case4 = OpLabel\n"
3775 " OpStore %outloc %sc_4_quant\n"
3778 "%case5 = OpLabel\n"
3779 " OpStore %outloc %sc_5_quant\n"
3785 " OpFunctionEnd\n");
3788 ComputeShaderSpec spec;
3789 const deUint8 numCases = 4;
3790 vector<float> inputs (numCases, 0.f);
3791 vector<float> outputs;
3793 spec.assembly = shader;
3794 spec.numWorkGroups = IVec3(numCases, 1, 1);
3796 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
3797 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
3798 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
3799 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
3801 outputs.push_back(std::numeric_limits<float>::infinity());
3802 outputs.push_back(-std::numeric_limits<float>::infinity());
3803 outputs.push_back(std::numeric_limits<float>::infinity());
3804 outputs.push_back(-std::numeric_limits<float>::infinity());
3806 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3807 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3809 group->addChild(new SpvAsmComputeShaderCase(
3810 testCtx, "infinities", "Check that infinities propagated and created", spec));
3814 ComputeShaderSpec spec;
3815 const deUint8 numCases = 2;
3816 vector<float> inputs (numCases, 0.f);
3817 vector<float> outputs;
3819 spec.assembly = shader;
3820 spec.numWorkGroups = IVec3(numCases, 1, 1);
3821 spec.verifyIO = &compareNan;
3823 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
3824 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
3826 for (deUint8 idx = 0; idx < numCases; ++idx)
3827 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
3829 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3830 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3832 group->addChild(new SpvAsmComputeShaderCase(
3833 testCtx, "propagated_nans", "Check that nans are propagated", spec));
3837 ComputeShaderSpec spec;
3838 const deUint8 numCases = 6;
3839 vector<float> inputs (numCases, 0.f);
3840 vector<float> outputs;
3842 spec.assembly = shader;
3843 spec.numWorkGroups = IVec3(numCases, 1, 1);
3845 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
3846 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
3847 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
3848 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
3849 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
3850 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
3852 outputs.push_back(0.f);
3853 outputs.push_back(-0.f);
3854 outputs.push_back(0.f);
3855 outputs.push_back(-0.f);
3856 outputs.push_back(0.f);
3857 outputs.push_back(-0.f);
3859 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3860 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3862 group->addChild(new SpvAsmComputeShaderCase(
3863 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
3867 ComputeShaderSpec spec;
3868 const deUint8 numCases = 6;
3869 vector<float> inputs (numCases, 0.f);
3870 vector<float> outputs;
3872 spec.assembly = shader;
3873 spec.numWorkGroups = IVec3(numCases, 1, 1);
3875 for (deUint8 idx = 0; idx < 6; ++idx)
3877 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
3878 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
3879 outputs.push_back(f);
3882 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3883 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3885 group->addChild(new SpvAsmComputeShaderCase(
3886 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
3890 ComputeShaderSpec spec;
3891 const deUint8 numCases = 4;
3892 vector<float> inputs (numCases, 0.f);
3893 vector<float> outputs;
3895 spec.assembly = shader;
3896 spec.numWorkGroups = IVec3(numCases, 1, 1);
3897 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
3899 outputs.push_back(constructNormalizedFloat(8, 0x300300));
3900 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
3901 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
3902 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
3904 for (deUint8 idx = 0; idx < numCases; ++idx)
3905 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
3907 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
3908 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
3910 group->addChild(new SpvAsmComputeShaderCase(
3911 testCtx, "rounded", "Check that are rounded when needed", spec));
3914 return group.release();
3917 // Checks that constant null/composite values can be used in computation.
3918 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
3920 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
3921 ComputeShaderSpec spec;
3922 de::Random rnd (deStringHash(group->getName()));
3923 const int numElements = 100;
3924 vector<float> positiveFloats (numElements, 0);
3925 vector<float> negativeFloats (numElements, 0);
3927 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3929 for (size_t ndx = 0; ndx < numElements; ++ndx)
3930 negativeFloats[ndx] = -positiveFloats[ndx];
3933 "OpCapability Shader\n"
3934 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
3935 "OpMemoryModel Logical GLSL450\n"
3936 "OpEntryPoint GLCompute %main \"main\" %id\n"
3937 "OpExecutionMode %main LocalSize 1 1 1\n"
3939 "OpSource GLSL 430\n"
3940 "OpName %main \"main\"\n"
3941 "OpName %id \"gl_GlobalInvocationID\"\n"
3943 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3945 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
3947 "%fmat = OpTypeMatrix %fvec3 3\n"
3948 "%ten = OpConstant %u32 10\n"
3949 "%f32arr10 = OpTypeArray %f32 %ten\n"
3950 "%fst = OpTypeStruct %f32 %f32\n"
3952 + string(getComputeAsmInputOutputBuffer()) +
3954 "%id = OpVariable %uvec3ptr Input\n"
3955 "%zero = OpConstant %i32 0\n"
3957 // Create a bunch of null values
3958 "%unull = OpConstantNull %u32\n"
3959 "%fnull = OpConstantNull %f32\n"
3960 "%vnull = OpConstantNull %fvec3\n"
3961 "%mnull = OpConstantNull %fmat\n"
3962 "%anull = OpConstantNull %f32arr10\n"
3963 "%snull = OpConstantComposite %fst %fnull %fnull\n"
3965 "%main = OpFunction %void None %voidf\n"
3966 "%label = OpLabel\n"
3967 "%idval = OpLoad %uvec3 %id\n"
3968 "%x = OpCompositeExtract %u32 %idval 0\n"
3969 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3970 "%inval = OpLoad %f32 %inloc\n"
3971 "%neg = OpFNegate %f32 %inval\n"
3973 // Get the abs() of (a certain element of) those null values
3974 "%unull_cov = OpConvertUToF %f32 %unull\n"
3975 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
3976 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
3977 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
3978 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
3979 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
3980 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
3981 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
3982 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
3983 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
3984 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
3987 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
3988 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
3989 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
3990 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
3991 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
3992 "%final = OpFAdd %f32 %add5 %snull_abs\n"
3994 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3995 " OpStore %outloc %final\n" // write to output
3998 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3999 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4000 spec.numWorkGroups = IVec3(numElements, 1, 1);
4002 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
4004 return group.release();
4007 // Assembly code used for testing loop control is based on GLSL source code:
4010 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4011 // float elements[];
4013 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4014 // float elements[];
4018 // uint x = gl_GlobalInvocationID.x;
4019 // output_data.elements[x] = input_data.elements[x];
4020 // for (uint i = 0; i < 4; ++i)
4021 // output_data.elements[x] += 1.f;
4023 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
4025 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
4026 vector<CaseParameter> cases;
4027 de::Random rnd (deStringHash(group->getName()));
4028 const int numElements = 100;
4029 vector<float> inputFloats (numElements, 0);
4030 vector<float> outputFloats (numElements, 0);
4031 const StringTemplate shaderTemplate (
4032 string(getComputeAsmShaderPreamble()) +
4034 "OpSource GLSL 430\n"
4035 "OpName %main \"main\"\n"
4036 "OpName %id \"gl_GlobalInvocationID\"\n"
4038 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4040 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4042 "%u32ptr = OpTypePointer Function %u32\n"
4044 "%id = OpVariable %uvec3ptr Input\n"
4045 "%zero = OpConstant %i32 0\n"
4046 "%uzero = OpConstant %u32 0\n"
4047 "%one = OpConstant %i32 1\n"
4048 "%constf1 = OpConstant %f32 1.0\n"
4049 "%four = OpConstant %u32 4\n"
4051 "%main = OpFunction %void None %voidf\n"
4052 "%entry = OpLabel\n"
4053 "%i = OpVariable %u32ptr Function\n"
4054 " OpStore %i %uzero\n"
4056 "%idval = OpLoad %uvec3 %id\n"
4057 "%x = OpCompositeExtract %u32 %idval 0\n"
4058 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4059 "%inval = OpLoad %f32 %inloc\n"
4060 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4061 " OpStore %outloc %inval\n"
4062 " OpBranch %loop_entry\n"
4064 "%loop_entry = OpLabel\n"
4065 "%i_val = OpLoad %u32 %i\n"
4066 "%cmp_lt = OpULessThan %bool %i_val %four\n"
4067 " OpLoopMerge %loop_merge %loop_body ${CONTROL}\n"
4068 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
4069 "%loop_body = OpLabel\n"
4070 "%outval = OpLoad %f32 %outloc\n"
4071 "%addf1 = OpFAdd %f32 %outval %constf1\n"
4072 " OpStore %outloc %addf1\n"
4073 "%new_i = OpIAdd %u32 %i_val %one\n"
4074 " OpStore %i %new_i\n"
4075 " OpBranch %loop_entry\n"
4076 "%loop_merge = OpLabel\n"
4078 " OpFunctionEnd\n");
4080 cases.push_back(CaseParameter("none", "None"));
4081 cases.push_back(CaseParameter("unroll", "Unroll"));
4082 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
4083 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
4085 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4087 for (size_t ndx = 0; ndx < numElements; ++ndx)
4088 outputFloats[ndx] = inputFloats[ndx] + 4.f;
4090 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4092 map<string, string> specializations;
4093 ComputeShaderSpec spec;
4095 specializations["CONTROL"] = cases[caseNdx].param;
4096 spec.assembly = shaderTemplate.specialize(specializations);
4097 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4098 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4099 spec.numWorkGroups = IVec3(numElements, 1, 1);
4101 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4104 return group.release();
4107 // Assembly code used for testing selection control is based on GLSL source code:
4110 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4111 // float elements[];
4113 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4114 // float elements[];
4118 // uint x = gl_GlobalInvocationID.x;
4119 // float val = input_data.elements[x];
4121 // output_data.elements[x] = val + 1.f;
4123 // output_data.elements[x] = val - 1.f;
4125 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
4127 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
4128 vector<CaseParameter> cases;
4129 de::Random rnd (deStringHash(group->getName()));
4130 const int numElements = 100;
4131 vector<float> inputFloats (numElements, 0);
4132 vector<float> outputFloats (numElements, 0);
4133 const StringTemplate shaderTemplate (
4134 string(getComputeAsmShaderPreamble()) +
4136 "OpSource GLSL 430\n"
4137 "OpName %main \"main\"\n"
4138 "OpName %id \"gl_GlobalInvocationID\"\n"
4140 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4142 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4144 "%id = OpVariable %uvec3ptr Input\n"
4145 "%zero = OpConstant %i32 0\n"
4146 "%constf1 = OpConstant %f32 1.0\n"
4147 "%constf10 = OpConstant %f32 10.0\n"
4149 "%main = OpFunction %void None %voidf\n"
4150 "%entry = OpLabel\n"
4151 "%idval = OpLoad %uvec3 %id\n"
4152 "%x = OpCompositeExtract %u32 %idval 0\n"
4153 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4154 "%inval = OpLoad %f32 %inloc\n"
4155 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4156 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
4158 " OpSelectionMerge %if_end ${CONTROL}\n"
4159 " OpBranchConditional %cmp_gt %if_true %if_false\n"
4160 "%if_true = OpLabel\n"
4161 "%addf1 = OpFAdd %f32 %inval %constf1\n"
4162 " OpStore %outloc %addf1\n"
4163 " OpBranch %if_end\n"
4164 "%if_false = OpLabel\n"
4165 "%subf1 = OpFSub %f32 %inval %constf1\n"
4166 " OpStore %outloc %subf1\n"
4167 " OpBranch %if_end\n"
4168 "%if_end = OpLabel\n"
4170 " OpFunctionEnd\n");
4172 cases.push_back(CaseParameter("none", "None"));
4173 cases.push_back(CaseParameter("flatten", "Flatten"));
4174 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
4175 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
4177 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4179 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4180 floorAll(inputFloats);
4182 for (size_t ndx = 0; ndx < numElements; ++ndx)
4183 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
4185 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4187 map<string, string> specializations;
4188 ComputeShaderSpec spec;
4190 specializations["CONTROL"] = cases[caseNdx].param;
4191 spec.assembly = shaderTemplate.specialize(specializations);
4192 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4193 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4194 spec.numWorkGroups = IVec3(numElements, 1, 1);
4196 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4199 return group.release();
4202 // Assembly code used for testing function control is based on GLSL source code:
4206 // layout(std140, set = 0, binding = 0) readonly buffer Input {
4207 // float elements[];
4209 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
4210 // float elements[];
4213 // float const10() { return 10.f; }
4216 // uint x = gl_GlobalInvocationID.x;
4217 // output_data.elements[x] = input_data.elements[x] + const10();
4219 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
4221 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
4222 vector<CaseParameter> cases;
4223 de::Random rnd (deStringHash(group->getName()));
4224 const int numElements = 100;
4225 vector<float> inputFloats (numElements, 0);
4226 vector<float> outputFloats (numElements, 0);
4227 const StringTemplate shaderTemplate (
4228 string(getComputeAsmShaderPreamble()) +
4230 "OpSource GLSL 430\n"
4231 "OpName %main \"main\"\n"
4232 "OpName %func_const10 \"const10(\"\n"
4233 "OpName %id \"gl_GlobalInvocationID\"\n"
4235 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4237 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4239 "%f32f = OpTypeFunction %f32\n"
4240 "%id = OpVariable %uvec3ptr Input\n"
4241 "%zero = OpConstant %i32 0\n"
4242 "%constf10 = OpConstant %f32 10.0\n"
4244 "%main = OpFunction %void None %voidf\n"
4245 "%entry = OpLabel\n"
4246 "%idval = OpLoad %uvec3 %id\n"
4247 "%x = OpCompositeExtract %u32 %idval 0\n"
4248 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4249 "%inval = OpLoad %f32 %inloc\n"
4250 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
4251 "%fadd = OpFAdd %f32 %inval %ret_10\n"
4252 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4253 " OpStore %outloc %fadd\n"
4257 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
4258 "%label = OpLabel\n"
4259 " OpReturnValue %constf10\n"
4260 " OpFunctionEnd\n");
4262 cases.push_back(CaseParameter("none", "None"));
4263 cases.push_back(CaseParameter("inline", "Inline"));
4264 cases.push_back(CaseParameter("dont_inline", "DontInline"));
4265 cases.push_back(CaseParameter("pure", "Pure"));
4266 cases.push_back(CaseParameter("const", "Const"));
4267 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
4268 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
4269 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
4270 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
4272 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4274 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
4275 floorAll(inputFloats);
4277 for (size_t ndx = 0; ndx < numElements; ++ndx)
4278 outputFloats[ndx] = inputFloats[ndx] + 10.f;
4280 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4282 map<string, string> specializations;
4283 ComputeShaderSpec spec;
4285 specializations["CONTROL"] = cases[caseNdx].param;
4286 spec.assembly = shaderTemplate.specialize(specializations);
4287 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4288 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4289 spec.numWorkGroups = IVec3(numElements, 1, 1);
4291 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4294 return group.release();
4297 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
4299 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
4300 vector<CaseParameter> cases;
4301 de::Random rnd (deStringHash(group->getName()));
4302 const int numElements = 100;
4303 vector<float> inputFloats (numElements, 0);
4304 vector<float> outputFloats (numElements, 0);
4305 const StringTemplate shaderTemplate (
4306 string(getComputeAsmShaderPreamble()) +
4308 "OpSource GLSL 430\n"
4309 "OpName %main \"main\"\n"
4310 "OpName %id \"gl_GlobalInvocationID\"\n"
4312 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4314 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
4316 "%f32ptr_f = OpTypePointer Function %f32\n"
4318 "%id = OpVariable %uvec3ptr Input\n"
4319 "%zero = OpConstant %i32 0\n"
4320 "%four = OpConstant %i32 4\n"
4322 "%main = OpFunction %void None %voidf\n"
4323 "%label = OpLabel\n"
4324 "%copy = OpVariable %f32ptr_f Function\n"
4325 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
4326 "%x = OpCompositeExtract %u32 %idval 0\n"
4327 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4328 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4329 " OpCopyMemory %copy %inloc ${ACCESS}\n"
4330 "%val1 = OpLoad %f32 %copy\n"
4331 "%val2 = OpLoad %f32 %inloc\n"
4332 "%add = OpFAdd %f32 %val1 %val2\n"
4333 " OpStore %outloc %add ${ACCESS}\n"
4335 " OpFunctionEnd\n");
4337 cases.push_back(CaseParameter("null", ""));
4338 cases.push_back(CaseParameter("none", "None"));
4339 cases.push_back(CaseParameter("volatile", "Volatile"));
4340 cases.push_back(CaseParameter("aligned", "Aligned 4"));
4341 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
4342 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
4343 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
4345 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
4347 for (size_t ndx = 0; ndx < numElements; ++ndx)
4348 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
4350 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4352 map<string, string> specializations;
4353 ComputeShaderSpec spec;
4355 specializations["ACCESS"] = cases[caseNdx].param;
4356 spec.assembly = shaderTemplate.specialize(specializations);
4357 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
4358 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
4359 spec.numWorkGroups = IVec3(numElements, 1, 1);
4361 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4364 return group.release();
4367 // Checks that we can get undefined values for various types, without exercising a computation with it.
4368 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
4370 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
4371 vector<CaseParameter> cases;
4372 de::Random rnd (deStringHash(group->getName()));
4373 const int numElements = 100;
4374 vector<float> positiveFloats (numElements, 0);
4375 vector<float> negativeFloats (numElements, 0);
4376 const StringTemplate shaderTemplate (
4377 string(getComputeAsmShaderPreamble()) +
4379 "OpSource GLSL 430\n"
4380 "OpName %main \"main\"\n"
4381 "OpName %id \"gl_GlobalInvocationID\"\n"
4383 "OpDecorate %id BuiltIn GlobalInvocationId\n"
4385 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) +
4386 "%uvec2 = OpTypeVector %u32 2\n"
4387 "%fvec4 = OpTypeVector %f32 4\n"
4388 "%fmat33 = OpTypeMatrix %fvec3 3\n"
4389 "%image = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
4390 "%sampler = OpTypeSampler\n"
4391 "%simage = OpTypeSampledImage %image\n"
4392 "%const100 = OpConstant %u32 100\n"
4393 "%uarr100 = OpTypeArray %i32 %const100\n"
4394 "%struct = OpTypeStruct %f32 %i32 %u32\n"
4395 "%pointer = OpTypePointer Function %i32\n"
4396 + string(getComputeAsmInputOutputBuffer()) +
4398 "%id = OpVariable %uvec3ptr Input\n"
4399 "%zero = OpConstant %i32 0\n"
4401 "%main = OpFunction %void None %voidf\n"
4402 "%label = OpLabel\n"
4404 "%undef = OpUndef ${TYPE}\n"
4406 "%idval = OpLoad %uvec3 %id\n"
4407 "%x = OpCompositeExtract %u32 %idval 0\n"
4409 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
4410 "%inval = OpLoad %f32 %inloc\n"
4411 "%neg = OpFNegate %f32 %inval\n"
4412 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
4413 " OpStore %outloc %neg\n"
4415 " OpFunctionEnd\n");
4417 cases.push_back(CaseParameter("bool", "%bool"));
4418 cases.push_back(CaseParameter("sint32", "%i32"));
4419 cases.push_back(CaseParameter("uint32", "%u32"));
4420 cases.push_back(CaseParameter("float32", "%f32"));
4421 cases.push_back(CaseParameter("vec4float32", "%fvec4"));
4422 cases.push_back(CaseParameter("vec2uint32", "%uvec2"));
4423 cases.push_back(CaseParameter("matrix", "%fmat33"));
4424 cases.push_back(CaseParameter("image", "%image"));
4425 cases.push_back(CaseParameter("sampler", "%sampler"));
4426 cases.push_back(CaseParameter("sampledimage", "%simage"));
4427 cases.push_back(CaseParameter("array", "%uarr100"));
4428 cases.push_back(CaseParameter("runtimearray", "%f32arr"));
4429 cases.push_back(CaseParameter("struct", "%struct"));
4430 cases.push_back(CaseParameter("pointer", "%pointer"));
4432 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
4434 for (size_t ndx = 0; ndx < numElements; ++ndx)
4435 negativeFloats[ndx] = -positiveFloats[ndx];
4437 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
4439 map<string, string> specializations;
4440 ComputeShaderSpec spec;
4442 specializations["TYPE"] = cases[caseNdx].param;
4443 spec.assembly = shaderTemplate.specialize(specializations);
4444 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
4445 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
4446 spec.numWorkGroups = IVec3(numElements, 1, 1);
4448 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
4451 return group.release();
4456 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
4458 struct NameCodePair { string name, code; };
4459 RGBA defaultColors[4];
4460 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
4461 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
4462 map<string, string> fragments = passthruFragments();
4463 const NameCodePair tests[] =
4465 {"unknown", "OpSource Unknown 321"},
4466 {"essl", "OpSource ESSL 310"},
4467 {"glsl", "OpSource GLSL 450"},
4468 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
4469 {"opencl_c", "OpSource OpenCL_C 120"},
4470 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
4471 {"file", opsourceGLSLWithFile},
4472 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
4473 // Longest possible source string: SPIR-V limits instructions to 65535
4474 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
4475 // contain 65530 UTF8 characters (one word each) plus one last word
4476 // containing 3 ASCII characters and \0.
4477 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
4480 getDefaultColors(defaultColors);
4481 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
4483 fragments["debug"] = tests[testNdx].code;
4484 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
4487 return opSourceTests.release();
4490 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
4492 struct NameCodePair { string name, code; };
4493 RGBA defaultColors[4];
4494 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
4495 map<string, string> fragments = passthruFragments();
4496 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
4497 const NameCodePair tests[] =
4499 {"empty", opsource + "OpSourceContinued \"\""},
4500 {"short", opsource + "OpSourceContinued \"abcde\""},
4501 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
4502 // Longest possible source string: SPIR-V limits instructions to 65535
4503 // words, of which the first one is OpSourceContinued/length; the rest
4504 // will contain 65533 UTF8 characters (one word each) plus one last word
4505 // containing 3 ASCII characters and \0.
4506 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
4509 getDefaultColors(defaultColors);
4510 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
4512 fragments["debug"] = tests[testNdx].code;
4513 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
4516 return opSourceTests.release();
4519 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
4521 RGBA defaultColors[4];
4522 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
4523 map<string, string> fragments;
4524 getDefaultColors(defaultColors);
4525 fragments["debug"] =
4526 "%name = OpString \"name\"\n";
4528 fragments["pre_main"] =
4531 "OpLine %name 1 1\n"
4533 "OpLine %name 1 1\n"
4534 "OpLine %name 1 1\n"
4535 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
4537 "OpLine %name 1 1\n"
4539 "OpLine %name 1 1\n"
4540 "OpLine %name 1 1\n"
4541 "%second_param1 = OpFunctionParameter %v4f32\n"
4544 "%label_secondfunction = OpLabel\n"
4546 "OpReturnValue %second_param1\n"
4551 fragments["testfun"] =
4552 // A %test_code function that returns its argument unchanged.
4555 "OpLine %name 1 1\n"
4556 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4558 "%param1 = OpFunctionParameter %v4f32\n"
4561 "%label_testfun = OpLabel\n"
4563 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
4564 "OpReturnValue %val1\n"
4566 "OpLine %name 1 1\n"
4569 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
4571 return opLineTests.release();
4575 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
4577 RGBA defaultColors[4];
4578 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
4579 map<string, string> fragments;
4580 std::vector<std::pair<std::string, std::string> > problemStrings;
4582 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
4583 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
4584 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
4585 getDefaultColors(defaultColors);
4587 fragments["debug"] =
4588 "%other_name = OpString \"other_name\"\n";
4590 fragments["pre_main"] =
4591 "OpLine %file_name 32 0\n"
4592 "OpLine %file_name 32 32\n"
4593 "OpLine %file_name 32 40\n"
4594 "OpLine %other_name 32 40\n"
4595 "OpLine %other_name 0 100\n"
4596 "OpLine %other_name 0 4294967295\n"
4597 "OpLine %other_name 4294967295 0\n"
4598 "OpLine %other_name 32 40\n"
4599 "OpLine %file_name 0 0\n"
4600 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
4601 "OpLine %file_name 1 0\n"
4602 "%second_param1 = OpFunctionParameter %v4f32\n"
4603 "OpLine %file_name 1 3\n"
4604 "OpLine %file_name 1 2\n"
4605 "%label_secondfunction = OpLabel\n"
4606 "OpLine %file_name 0 2\n"
4607 "OpReturnValue %second_param1\n"
4609 "OpLine %file_name 0 2\n"
4610 "OpLine %file_name 0 2\n";
4612 fragments["testfun"] =
4613 // A %test_code function that returns its argument unchanged.
4614 "OpLine %file_name 1 0\n"
4615 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4616 "OpLine %file_name 16 330\n"
4617 "%param1 = OpFunctionParameter %v4f32\n"
4618 "OpLine %file_name 14 442\n"
4619 "%label_testfun = OpLabel\n"
4620 "OpLine %file_name 11 1024\n"
4621 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
4622 "OpLine %file_name 2 97\n"
4623 "OpReturnValue %val1\n"
4625 "OpLine %file_name 5 32\n";
4627 for (size_t i = 0; i < problemStrings.size(); ++i)
4629 map<string, string> testFragments = fragments;
4630 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
4631 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
4634 return opLineTests.release();
4637 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
4639 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
4643 const char functionStart[] =
4644 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4645 "%param1 = OpFunctionParameter %v4f32\n"
4648 const char functionEnd[] =
4649 "OpReturnValue %transformed_param\n"
4652 struct NameConstantsCode
4659 NameConstantsCode tests[] =
4663 "%cnull = OpConstantNull %v4f32\n",
4664 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
4668 "%cnull = OpConstantNull %f32\n",
4669 "%vp = OpVariable %fp_v4f32 Function\n"
4670 "%v = OpLoad %v4f32 %vp\n"
4671 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
4672 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
4673 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
4674 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
4675 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
4679 "%cnull = OpConstantNull %bool\n",
4680 "%v = OpVariable %fp_v4f32 Function\n"
4681 " OpStore %v %param1\n"
4682 " OpSelectionMerge %false_label None\n"
4683 " OpBranchConditional %cnull %true_label %false_label\n"
4684 "%true_label = OpLabel\n"
4685 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
4686 " OpBranch %false_label\n"
4687 "%false_label = OpLabel\n"
4688 "%transformed_param = OpLoad %v4f32 %v\n"
4692 "%cnull = OpConstantNull %i32\n",
4693 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
4694 "%b = OpIEqual %bool %cnull %c_i32_0\n"
4695 " OpSelectionMerge %false_label None\n"
4696 " OpBranchConditional %b %true_label %false_label\n"
4697 "%true_label = OpLabel\n"
4698 " OpStore %v %param1\n"
4699 " OpBranch %false_label\n"
4700 "%false_label = OpLabel\n"
4701 "%transformed_param = OpLoad %v4f32 %v\n"
4705 "%stype = OpTypeStruct %f32 %v4f32\n"
4706 "%fp_stype = OpTypePointer Function %stype\n"
4707 "%cnull = OpConstantNull %stype\n",
4708 "%v = OpVariable %fp_stype Function %cnull\n"
4709 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
4710 "%f_val = OpLoad %v4f32 %f\n"
4711 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
4715 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
4716 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
4717 "%cnull = OpConstantNull %a4_v4f32\n",
4718 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
4719 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
4720 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
4721 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
4722 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
4723 "%f_val = OpLoad %v4f32 %f\n"
4724 "%f1_val = OpLoad %v4f32 %f1\n"
4725 "%f2_val = OpLoad %v4f32 %f2\n"
4726 "%f3_val = OpLoad %v4f32 %f3\n"
4727 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
4728 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
4729 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
4730 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
4734 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
4735 "%cnull = OpConstantNull %mat4x4_f32\n",
4736 // Our null matrix * any vector should result in a zero vector.
4737 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
4738 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
4742 getHalfColorsFullAlpha(colors);
4744 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
4746 map<string, string> fragments;
4747 fragments["pre_main"] = tests[testNdx].constants;
4748 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
4749 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
4751 return opConstantNullTests.release();
4753 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
4755 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
4756 RGBA inputColors[4];
4757 RGBA outputColors[4];
4760 const char functionStart[] =
4761 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4762 "%param1 = OpFunctionParameter %v4f32\n"
4765 const char functionEnd[] =
4766 "OpReturnValue %transformed_param\n"
4769 struct NameConstantsCode
4776 NameConstantsCode tests[] =
4781 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
4782 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
4787 "%stype = OpTypeStruct %v4f32 %f32\n"
4788 "%fp_stype = OpTypePointer Function %stype\n"
4789 "%f32_n_1 = OpConstant %f32 -1.0\n"
4790 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
4791 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
4792 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
4794 "%v = OpVariable %fp_stype Function %cval\n"
4795 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
4796 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
4797 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
4798 "%f32_val = OpLoad %f32 %f32_ptr\n"
4799 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
4800 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
4801 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
4804 // [1|0|0|0.5] [x] = x + 0.5
4805 // [0|1|0|0.5] [y] = y + 0.5
4806 // [0|0|1|0.5] [z] = z + 0.5
4807 // [0|0|0|1 ] [1] = 1
4810 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
4811 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
4812 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
4813 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
4814 "%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"
4815 "%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",
4817 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
4822 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4823 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
4824 "%f32_n_1 = OpConstant %f32 -1.0\n"
4825 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
4826 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
4828 "%v = OpVariable %fp_a4f32 Function %carr\n"
4829 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
4830 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
4831 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
4832 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
4833 "%f_val = OpLoad %f32 %f\n"
4834 "%f1_val = OpLoad %f32 %f1\n"
4835 "%f2_val = OpLoad %f32 %f2\n"
4836 "%f3_val = OpLoad %f32 %f3\n"
4837 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
4838 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
4839 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
4840 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
4841 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
4848 // [ 1.0, 1.0, 1.0, 1.0]
4852 // [ 0.0, 0.5, 0.0, 0.0]
4856 // [ 1.0, 1.0, 1.0, 1.0]
4859 "array_of_struct_of_array",
4861 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4862 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
4863 "%stype = OpTypeStruct %f32 %a4f32\n"
4864 "%a3stype = OpTypeArray %stype %c_u32_3\n"
4865 "%fp_a3stype = OpTypePointer Function %a3stype\n"
4866 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
4867 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
4868 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
4869 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
4870 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
4872 "%v = OpVariable %fp_a3stype Function %carr\n"
4873 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
4874 "%f_l = OpLoad %f32 %f\n"
4875 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
4876 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
4880 getHalfColorsFullAlpha(inputColors);
4881 outputColors[0] = RGBA(255, 255, 255, 255);
4882 outputColors[1] = RGBA(255, 127, 127, 255);
4883 outputColors[2] = RGBA(127, 255, 127, 255);
4884 outputColors[3] = RGBA(127, 127, 255, 255);
4886 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
4888 map<string, string> fragments;
4889 fragments["pre_main"] = tests[testNdx].constants;
4890 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
4891 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
4893 return opConstantCompositeTests.release();
4896 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
4898 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
4899 RGBA inputColors[4];
4900 RGBA outputColors[4];
4901 map<string, string> fragments;
4903 // vec4 test_code(vec4 param) {
4904 // vec4 result = param;
4905 // for (int i = 0; i < 4; ++i) {
4906 // if (i == 0) result[i] = 0.;
4907 // else result[i] = 1. - result[i];
4911 const char function[] =
4912 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
4913 "%param1 = OpFunctionParameter %v4f32\n"
4915 "%iptr = OpVariable %fp_i32 Function\n"
4916 "%result = OpVariable %fp_v4f32 Function\n"
4917 " OpStore %iptr %c_i32_0\n"
4918 " OpStore %result %param1\n"
4921 // Loop entry block.
4923 "%ival = OpLoad %i32 %iptr\n"
4924 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
4925 " OpLoopMerge %exit %if_entry None\n"
4926 " OpBranchConditional %lt_4 %if_entry %exit\n"
4928 // Merge block for loop.
4930 "%ret = OpLoad %v4f32 %result\n"
4931 " OpReturnValue %ret\n"
4933 // If-statement entry block.
4934 "%if_entry = OpLabel\n"
4935 "%loc = OpAccessChain %fp_f32 %result %ival\n"
4936 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
4937 " OpSelectionMerge %if_exit None\n"
4938 " OpBranchConditional %eq_0 %if_true %if_false\n"
4940 // False branch for if-statement.
4941 "%if_false = OpLabel\n"
4942 "%val = OpLoad %f32 %loc\n"
4943 "%sub = OpFSub %f32 %c_f32_1 %val\n"
4944 " OpStore %loc %sub\n"
4945 " OpBranch %if_exit\n"
4947 // Merge block for if-statement.
4948 "%if_exit = OpLabel\n"
4949 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
4950 " OpStore %iptr %ival_next\n"
4953 // True branch for if-statement.
4954 "%if_true = OpLabel\n"
4955 " OpStore %loc %c_f32_0\n"
4956 " OpBranch %if_exit\n"
4960 fragments["testfun"] = function;
4962 inputColors[0] = RGBA(127, 127, 127, 0);
4963 inputColors[1] = RGBA(127, 0, 0, 0);
4964 inputColors[2] = RGBA(0, 127, 0, 0);
4965 inputColors[3] = RGBA(0, 0, 127, 0);
4967 outputColors[0] = RGBA(0, 128, 128, 255);
4968 outputColors[1] = RGBA(0, 255, 255, 255);
4969 outputColors[2] = RGBA(0, 128, 255, 255);
4970 outputColors[3] = RGBA(0, 255, 128, 255);
4972 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
4974 return group.release();
4977 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
4979 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
4980 RGBA inputColors[4];
4981 RGBA outputColors[4];
4982 map<string, string> fragments;
4984 const char typesAndConstants[] =
4985 "%c_f32_p2 = OpConstant %f32 0.2\n"
4986 "%c_f32_p4 = OpConstant %f32 0.4\n"
4987 "%c_f32_p6 = OpConstant %f32 0.6\n"
4988 "%c_f32_p8 = OpConstant %f32 0.8\n";
4990 // vec4 test_code(vec4 param) {
4991 // vec4 result = param;
4992 // for (int i = 0; i < 4; ++i) {
4994 // case 0: result[i] += .2; break;
4995 // case 1: result[i] += .6; break;
4996 // case 2: result[i] += .4; break;
4997 // case 3: result[i] += .8; break;
4998 // default: break; // unreachable
5003 const char function[] =
5004 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5005 "%param1 = OpFunctionParameter %v4f32\n"
5007 "%iptr = OpVariable %fp_i32 Function\n"
5008 "%result = OpVariable %fp_v4f32 Function\n"
5009 " OpStore %iptr %c_i32_0\n"
5010 " OpStore %result %param1\n"
5013 // Loop entry block.
5015 "%ival = OpLoad %i32 %iptr\n"
5016 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5017 " OpLoopMerge %exit %switch_exit None\n"
5018 " OpBranchConditional %lt_4 %switch_entry %exit\n"
5020 // Merge block for loop.
5022 "%ret = OpLoad %v4f32 %result\n"
5023 " OpReturnValue %ret\n"
5025 // Switch-statement entry block.
5026 "%switch_entry = OpLabel\n"
5027 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5028 "%val = OpLoad %f32 %loc\n"
5029 " OpSelectionMerge %switch_exit None\n"
5030 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5032 "%case2 = OpLabel\n"
5033 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
5034 " OpStore %loc %addp4\n"
5035 " OpBranch %switch_exit\n"
5037 "%switch_default = OpLabel\n"
5040 "%case3 = OpLabel\n"
5041 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
5042 " OpStore %loc %addp8\n"
5043 " OpBranch %switch_exit\n"
5045 "%case0 = OpLabel\n"
5046 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
5047 " OpStore %loc %addp2\n"
5048 " OpBranch %switch_exit\n"
5050 // Merge block for switch-statement.
5051 "%switch_exit = OpLabel\n"
5052 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5053 " OpStore %iptr %ival_next\n"
5056 "%case1 = OpLabel\n"
5057 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
5058 " OpStore %loc %addp6\n"
5059 " OpBranch %switch_exit\n"
5063 fragments["pre_main"] = typesAndConstants;
5064 fragments["testfun"] = function;
5066 inputColors[0] = RGBA(127, 27, 127, 51);
5067 inputColors[1] = RGBA(127, 0, 0, 51);
5068 inputColors[2] = RGBA(0, 27, 0, 51);
5069 inputColors[3] = RGBA(0, 0, 127, 51);
5071 outputColors[0] = RGBA(178, 180, 229, 255);
5072 outputColors[1] = RGBA(178, 153, 102, 255);
5073 outputColors[2] = RGBA(51, 180, 102, 255);
5074 outputColors[3] = RGBA(51, 153, 229, 255);
5076 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5078 return group.release();
5081 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
5083 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
5084 RGBA inputColors[4];
5085 RGBA outputColors[4];
5086 map<string, string> fragments;
5088 const char decorations[] =
5089 "OpDecorate %array_group ArrayStride 4\n"
5090 "OpDecorate %struct_member_group Offset 0\n"
5091 "%array_group = OpDecorationGroup\n"
5092 "%struct_member_group = OpDecorationGroup\n"
5094 "OpDecorate %group1 RelaxedPrecision\n"
5095 "OpDecorate %group3 RelaxedPrecision\n"
5096 "OpDecorate %group3 Invariant\n"
5097 "OpDecorate %group3 Restrict\n"
5098 "%group0 = OpDecorationGroup\n"
5099 "%group1 = OpDecorationGroup\n"
5100 "%group3 = OpDecorationGroup\n";
5102 const char typesAndConstants[] =
5103 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
5104 "%struct1 = OpTypeStruct %a3f32\n"
5105 "%struct2 = OpTypeStruct %a3f32\n"
5106 "%fp_struct1 = OpTypePointer Function %struct1\n"
5107 "%fp_struct2 = OpTypePointer Function %struct2\n"
5108 "%c_f32_2 = OpConstant %f32 2.\n"
5109 "%c_f32_n2 = OpConstant %f32 -2.\n"
5111 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
5112 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
5113 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
5114 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
5116 const char function[] =
5117 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5118 "%param = OpFunctionParameter %v4f32\n"
5119 "%entry = OpLabel\n"
5120 "%result = OpVariable %fp_v4f32 Function\n"
5121 "%v_struct1 = OpVariable %fp_struct1 Function\n"
5122 "%v_struct2 = OpVariable %fp_struct2 Function\n"
5123 " OpStore %result %param\n"
5124 " OpStore %v_struct1 %c_struct1\n"
5125 " OpStore %v_struct2 %c_struct2\n"
5126 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
5127 "%val1 = OpLoad %f32 %ptr1\n"
5128 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
5129 "%val2 = OpLoad %f32 %ptr2\n"
5130 "%addvalues = OpFAdd %f32 %val1 %val2\n"
5131 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
5132 "%val = OpLoad %f32 %ptr\n"
5133 "%addresult = OpFAdd %f32 %addvalues %val\n"
5134 " OpStore %ptr %addresult\n"
5135 "%ret = OpLoad %v4f32 %result\n"
5136 " OpReturnValue %ret\n"
5139 struct CaseNameDecoration
5145 CaseNameDecoration tests[] =
5148 "same_decoration_group_on_multiple_types",
5149 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
5152 "empty_decoration_group",
5153 "OpGroupDecorate %group0 %a3f32\n"
5154 "OpGroupDecorate %group0 %result\n"
5157 "one_element_decoration_group",
5158 "OpGroupDecorate %array_group %a3f32\n"
5161 "multiple_elements_decoration_group",
5162 "OpGroupDecorate %group3 %v_struct1\n"
5165 "multiple_decoration_groups_on_same_variable",
5166 "OpGroupDecorate %group0 %v_struct2\n"
5167 "OpGroupDecorate %group1 %v_struct2\n"
5168 "OpGroupDecorate %group3 %v_struct2\n"
5171 "same_decoration_group_multiple_times",
5172 "OpGroupDecorate %group1 %addvalues\n"
5173 "OpGroupDecorate %group1 %addvalues\n"
5174 "OpGroupDecorate %group1 %addvalues\n"
5179 getHalfColorsFullAlpha(inputColors);
5180 getHalfColorsFullAlpha(outputColors);
5182 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
5184 fragments["decoration"] = decorations + tests[idx].decoration;
5185 fragments["pre_main"] = typesAndConstants;
5186 fragments["testfun"] = function;
5188 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
5191 return group.release();
5194 struct SpecConstantTwoIntGraphicsCase
5196 const char* caseName;
5197 const char* scDefinition0;
5198 const char* scDefinition1;
5199 const char* scResultType;
5200 const char* scOperation;
5201 deInt32 scActualValue0;
5202 deInt32 scActualValue1;
5203 const char* resultOperation;
5204 RGBA expectedColors[4];
5206 SpecConstantTwoIntGraphicsCase (const char* name,
5207 const char* definition0,
5208 const char* definition1,
5209 const char* resultType,
5210 const char* operation,
5213 const char* resultOp,
5214 const RGBA (&output)[4])
5216 , scDefinition0 (definition0)
5217 , scDefinition1 (definition1)
5218 , scResultType (resultType)
5219 , scOperation (operation)
5220 , scActualValue0 (value0)
5221 , scActualValue1 (value1)
5222 , resultOperation (resultOp)
5224 expectedColors[0] = output[0];
5225 expectedColors[1] = output[1];
5226 expectedColors[2] = output[2];
5227 expectedColors[3] = output[3];
5231 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
5233 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
5234 vector<SpecConstantTwoIntGraphicsCase> cases;
5235 RGBA inputColors[4];
5236 RGBA outputColors0[4];
5237 RGBA outputColors1[4];
5238 RGBA outputColors2[4];
5240 const char decorations1[] =
5241 "OpDecorate %sc_0 SpecId 0\n"
5242 "OpDecorate %sc_1 SpecId 1\n";
5244 const char typesAndConstants1[] =
5245 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
5246 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
5247 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
5249 const char function1[] =
5250 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5251 "%param = OpFunctionParameter %v4f32\n"
5252 "%label = OpLabel\n"
5253 "%result = OpVariable %fp_v4f32 Function\n"
5254 " OpStore %result %param\n"
5255 "%gen = ${GEN_RESULT}\n"
5256 "%index = OpIAdd %i32 %gen %c_i32_1\n"
5257 "%loc = OpAccessChain %fp_f32 %result %index\n"
5258 "%val = OpLoad %f32 %loc\n"
5259 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
5260 " OpStore %loc %add\n"
5261 "%ret = OpLoad %v4f32 %result\n"
5262 " OpReturnValue %ret\n"
5265 inputColors[0] = RGBA(127, 127, 127, 255);
5266 inputColors[1] = RGBA(127, 0, 0, 255);
5267 inputColors[2] = RGBA(0, 127, 0, 255);
5268 inputColors[3] = RGBA(0, 0, 127, 255);
5270 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
5271 outputColors0[0] = RGBA(255, 127, 127, 255);
5272 outputColors0[1] = RGBA(255, 0, 0, 255);
5273 outputColors0[2] = RGBA(128, 127, 0, 255);
5274 outputColors0[3] = RGBA(128, 0, 127, 255);
5276 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
5277 outputColors1[0] = RGBA(127, 255, 127, 255);
5278 outputColors1[1] = RGBA(127, 128, 0, 255);
5279 outputColors1[2] = RGBA(0, 255, 0, 255);
5280 outputColors1[3] = RGBA(0, 128, 127, 255);
5282 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
5283 outputColors2[0] = RGBA(127, 127, 255, 255);
5284 outputColors2[1] = RGBA(127, 0, 128, 255);
5285 outputColors2[2] = RGBA(0, 127, 128, 255);
5286 outputColors2[3] = RGBA(0, 0, 255, 255);
5288 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
5289 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
5290 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
5292 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
5293 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
5294 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
5295 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
5296 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
5297 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
5298 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
5299 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
5300 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
5301 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
5302 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
5303 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
5304 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
5305 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
5306 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
5307 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
5308 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
5309 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
5310 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
5311 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
5312 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
5313 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
5314 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
5315 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
5316 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
5317 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
5318 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
5319 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
5320 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
5321 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
5322 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
5323 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
5324 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
5326 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
5328 map<string, string> specializations;
5329 map<string, string> fragments;
5330 vector<deInt32> specConstants;
5332 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
5333 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
5334 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
5335 specializations["SC_OP"] = cases[caseNdx].scOperation;
5336 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
5338 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
5339 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
5340 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
5342 specConstants.push_back(cases[caseNdx].scActualValue0);
5343 specConstants.push_back(cases[caseNdx].scActualValue1);
5345 createTestsForAllStages(cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants, group.get());
5348 const char decorations2[] =
5349 "OpDecorate %sc_0 SpecId 0\n"
5350 "OpDecorate %sc_1 SpecId 1\n"
5351 "OpDecorate %sc_2 SpecId 2\n";
5353 const char typesAndConstants2[] =
5354 "%v3i32 = OpTypeVector %i32 3\n"
5355 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
5356 "%vec3_undef = OpUndef %v3i32\n"
5358 "%sc_0 = OpSpecConstant %i32 0\n"
5359 "%sc_1 = OpSpecConstant %i32 0\n"
5360 "%sc_2 = OpSpecConstant %i32 0\n"
5361 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
5362 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
5363 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
5364 "%sc_vec3_0_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %vec3_undef 0 0xFFFFFFFF 2\n" // (sc_0, ???, 0)
5365 "%sc_vec3_1_s = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_1 %vec3_undef 0xFFFFFFFF 1 0\n" // (???, sc_1, 0)
5366 "%sc_vec3_2_s = OpSpecConstantOp %v3i32 VectorShuffle %vec3_undef %sc_vec3_2 5 0xFFFFFFFF 5\n" // (sc_2, ???, sc_2)
5367 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0_s %sc_vec3_1_s 1 0 4\n" // (0, sc_0, sc_1)
5368 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2_s 5 1 2\n" // (sc_2, sc_0, sc_1)
5369 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
5370 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
5371 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
5372 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
5373 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
5375 const char function2[] =
5376 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5377 "%param = OpFunctionParameter %v4f32\n"
5378 "%label = OpLabel\n"
5379 "%result = OpVariable %fp_v4f32 Function\n"
5380 " OpStore %result %param\n"
5381 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
5382 "%val = OpLoad %f32 %loc\n"
5383 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
5384 " OpStore %loc %add\n"
5385 "%ret = OpLoad %v4f32 %result\n"
5386 " OpReturnValue %ret\n"
5389 map<string, string> fragments;
5390 vector<deInt32> specConstants;
5392 fragments["decoration"] = decorations2;
5393 fragments["pre_main"] = typesAndConstants2;
5394 fragments["testfun"] = function2;
5396 specConstants.push_back(56789);
5397 specConstants.push_back(-2);
5398 specConstants.push_back(56788);
5400 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
5402 return group.release();
5405 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
5407 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
5408 RGBA inputColors[4];
5409 RGBA outputColors1[4];
5410 RGBA outputColors2[4];
5411 RGBA outputColors3[4];
5412 map<string, string> fragments1;
5413 map<string, string> fragments2;
5414 map<string, string> fragments3;
5416 const char typesAndConstants1[] =
5417 "%c_f32_p2 = OpConstant %f32 0.2\n"
5418 "%c_f32_p4 = OpConstant %f32 0.4\n"
5419 "%c_f32_p5 = OpConstant %f32 0.5\n"
5420 "%c_f32_p8 = OpConstant %f32 0.8\n";
5422 // vec4 test_code(vec4 param) {
5423 // vec4 result = param;
5424 // for (int i = 0; i < 4; ++i) {
5427 // case 0: operand = .2; break;
5428 // case 1: operand = .5; break;
5429 // case 2: operand = .4; break;
5430 // case 3: operand = .0; break;
5431 // default: break; // unreachable
5433 // result[i] += operand;
5437 const char function1[] =
5438 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5439 "%param1 = OpFunctionParameter %v4f32\n"
5441 "%iptr = OpVariable %fp_i32 Function\n"
5442 "%result = OpVariable %fp_v4f32 Function\n"
5443 " OpStore %iptr %c_i32_0\n"
5444 " OpStore %result %param1\n"
5448 "%ival = OpLoad %i32 %iptr\n"
5449 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5450 " OpLoopMerge %exit %phi None\n"
5451 " OpBranchConditional %lt_4 %entry %exit\n"
5453 "%entry = OpLabel\n"
5454 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5455 "%val = OpLoad %f32 %loc\n"
5456 " OpSelectionMerge %phi None\n"
5457 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5459 "%case0 = OpLabel\n"
5461 "%case1 = OpLabel\n"
5463 "%case2 = OpLabel\n"
5465 "%case3 = OpLabel\n"
5468 "%default = OpLabel\n"
5472 "%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
5473 "%add = OpFAdd %f32 %val %operand\n"
5474 " OpStore %loc %add\n"
5475 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5476 " OpStore %iptr %ival_next\n"
5480 "%ret = OpLoad %v4f32 %result\n"
5481 " OpReturnValue %ret\n"
5485 fragments1["pre_main"] = typesAndConstants1;
5486 fragments1["testfun"] = function1;
5488 getHalfColorsFullAlpha(inputColors);
5490 outputColors1[0] = RGBA(178, 255, 229, 255);
5491 outputColors1[1] = RGBA(178, 127, 102, 255);
5492 outputColors1[2] = RGBA(51, 255, 102, 255);
5493 outputColors1[3] = RGBA(51, 127, 229, 255);
5495 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
5497 const char typesAndConstants2[] =
5498 "%c_f32_p2 = OpConstant %f32 0.2\n";
5500 // Add .4 to the second element of the given parameter.
5501 const char function2[] =
5502 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5503 "%param = OpFunctionParameter %v4f32\n"
5504 "%entry = OpLabel\n"
5505 "%result = OpVariable %fp_v4f32 Function\n"
5506 " OpStore %result %param\n"
5507 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
5508 "%val = OpLoad %f32 %loc\n"
5512 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
5513 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
5514 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
5515 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
5516 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
5517 " OpLoopMerge %exit %phi None\n"
5518 " OpBranchConditional %still_loop %phi %exit\n"
5521 " OpStore %loc %accum\n"
5522 "%ret = OpLoad %v4f32 %result\n"
5523 " OpReturnValue %ret\n"
5527 fragments2["pre_main"] = typesAndConstants2;
5528 fragments2["testfun"] = function2;
5530 outputColors2[0] = RGBA(127, 229, 127, 255);
5531 outputColors2[1] = RGBA(127, 102, 0, 255);
5532 outputColors2[2] = RGBA(0, 229, 0, 255);
5533 outputColors2[3] = RGBA(0, 102, 127, 255);
5535 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
5537 const char typesAndConstants3[] =
5538 "%true = OpConstantTrue %bool\n"
5539 "%false = OpConstantFalse %bool\n"
5540 "%c_f32_p2 = OpConstant %f32 0.2\n";
5542 // Swap the second and the third element of the given parameter.
5543 const char function3[] =
5544 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5545 "%param = OpFunctionParameter %v4f32\n"
5546 "%entry = OpLabel\n"
5547 "%result = OpVariable %fp_v4f32 Function\n"
5548 " OpStore %result %param\n"
5549 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
5550 "%a_init = OpLoad %f32 %a_loc\n"
5551 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
5552 "%b_init = OpLoad %f32 %b_loc\n"
5556 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
5557 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
5558 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
5559 " OpLoopMerge %exit %phi None\n"
5560 " OpBranchConditional %still_loop %phi %exit\n"
5563 " OpStore %a_loc %a_next\n"
5564 " OpStore %b_loc %b_next\n"
5565 "%ret = OpLoad %v4f32 %result\n"
5566 " OpReturnValue %ret\n"
5570 fragments3["pre_main"] = typesAndConstants3;
5571 fragments3["testfun"] = function3;
5573 outputColors3[0] = RGBA(127, 127, 127, 255);
5574 outputColors3[1] = RGBA(127, 0, 0, 255);
5575 outputColors3[2] = RGBA(0, 0, 127, 255);
5576 outputColors3[3] = RGBA(0, 127, 0, 255);
5578 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
5580 return group.release();
5583 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
5585 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
5586 RGBA inputColors[4];
5587 RGBA outputColors[4];
5589 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
5590 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
5591 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
5592 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
5593 const char constantsAndTypes[] =
5594 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
5595 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5596 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
5597 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
5598 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n"
5601 const char function[] =
5602 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5603 "%param = OpFunctionParameter %v4f32\n"
5604 "%label = OpLabel\n"
5605 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
5606 "%var2 = OpVariable %fp_f32 Function\n"
5607 "%red = OpCompositeExtract %f32 %param 0\n"
5608 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
5609 " OpStore %var2 %plus_red\n"
5610 "%val1 = OpLoad %f32 %var1\n"
5611 "%val2 = OpLoad %f32 %var2\n"
5612 "%mul = OpFMul %f32 %val1 %val2\n"
5613 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
5614 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
5615 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
5616 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
5617 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
5618 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
5619 " OpReturnValue %ret\n"
5622 struct CaseNameDecoration
5629 CaseNameDecoration tests[] = {
5630 {"multiplication", "OpDecorate %mul NoContraction"},
5631 {"addition", "OpDecorate %add NoContraction"},
5632 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
5635 getHalfColorsFullAlpha(inputColors);
5637 for (deUint8 idx = 0; idx < 4; ++idx)
5639 inputColors[idx].setRed(0);
5640 outputColors[idx] = RGBA(0, 0, 0, 255);
5643 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
5645 map<string, string> fragments;
5647 fragments["decoration"] = tests[testNdx].decoration;
5648 fragments["pre_main"] = constantsAndTypes;
5649 fragments["testfun"] = function;
5651 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
5654 return group.release();
5657 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
5659 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
5662 const char constantsAndTypes[] =
5663 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
5664 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
5665 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
5666 "%fp_stype = OpTypePointer Function %stype\n";
5668 const char function[] =
5669 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5670 "%param1 = OpFunctionParameter %v4f32\n"
5672 "%v1 = OpVariable %fp_v4f32 Function\n"
5673 "%v2 = OpVariable %fp_a2f32 Function\n"
5674 "%v3 = OpVariable %fp_f32 Function\n"
5675 "%v = OpVariable %fp_stype Function\n"
5676 "%vv = OpVariable %fp_stype Function\n"
5677 "%vvv = OpVariable %fp_f32 Function\n"
5679 " OpStore %v1 %c_v4f32_1_1_1_1\n"
5680 " OpStore %v2 %c_a2f32_1\n"
5681 " OpStore %v3 %c_f32_1\n"
5683 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5684 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
5685 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5686 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
5687 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
5688 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
5690 " OpStore %p_v4f32 %v1_v ${access_type}\n"
5691 " OpStore %p_a2f32 %v2_v ${access_type}\n"
5692 " OpStore %p_f32 %v3_v ${access_type}\n"
5694 " OpCopyMemory %vv %v ${access_type}\n"
5695 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
5697 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
5698 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
5699 "%v_f32_3 = OpLoad %f32 %vvv\n"
5701 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
5702 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
5703 " OpReturnValue %ret2\n"
5706 struct NameMemoryAccess
5713 NameMemoryAccess tests[] =
5716 { "volatile", "Volatile" },
5717 { "aligned", "Aligned 1" },
5718 { "volatile_aligned", "Volatile|Aligned 1" },
5719 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
5720 { "volatile_nontemporal", "Volatile|Nontemporal" },
5721 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
5724 getHalfColorsFullAlpha(colors);
5726 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
5728 map<string, string> fragments;
5729 map<string, string> memoryAccess;
5730 memoryAccess["access_type"] = tests[testNdx].accessType;
5732 fragments["pre_main"] = constantsAndTypes;
5733 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
5734 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
5736 return memoryAccessTests.release();
5738 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
5740 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
5741 RGBA defaultColors[4];
5742 map<string, string> fragments;
5743 getDefaultColors(defaultColors);
5745 // First, simple cases that don't do anything with the OpUndef result.
5746 struct NameCodePair { string name, decl, type; };
5747 const NameCodePair tests[] =
5749 {"bool", "", "%bool"},
5750 {"vec2uint32", "%type = OpTypeVector %u32 2", "%type"},
5751 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown", "%type"},
5752 {"sampler", "%type = OpTypeSampler", "%type"},
5753 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img", "%type"},
5754 {"pointer", "", "%fp_i32"},
5755 {"runtimearray", "%type = OpTypeRuntimeArray %f32", "%type"},
5756 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100", "%type"},
5757 {"struct", "%type = OpTypeStruct %f32 %i32 %u32", "%type"}};
5758 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5760 fragments["undef_type"] = tests[testNdx].type;
5761 fragments["testfun"] = StringTemplate(
5762 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5763 "%param1 = OpFunctionParameter %v4f32\n"
5764 "%label_testfun = OpLabel\n"
5765 "%undef = OpUndef ${undef_type}\n"
5766 "OpReturnValue %param1\n"
5767 "OpFunctionEnd\n").specialize(fragments);
5768 fragments["pre_main"] = tests[testNdx].decl;
5769 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
5773 fragments["testfun"] =
5774 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5775 "%param1 = OpFunctionParameter %v4f32\n"
5776 "%label_testfun = OpLabel\n"
5777 "%undef = OpUndef %f32\n"
5778 "%zero = OpFMul %f32 %undef %c_f32_0\n"
5779 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
5780 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
5781 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5782 "%b = OpFAdd %f32 %a %actually_zero\n"
5783 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
5784 "OpReturnValue %ret\n"
5787 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
5789 fragments["testfun"] =
5790 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5791 "%param1 = OpFunctionParameter %v4f32\n"
5792 "%label_testfun = OpLabel\n"
5793 "%undef = OpUndef %i32\n"
5794 "%zero = OpIMul %i32 %undef %c_i32_0\n"
5795 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
5796 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
5797 "OpReturnValue %ret\n"
5800 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
5802 fragments["testfun"] =
5803 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5804 "%param1 = OpFunctionParameter %v4f32\n"
5805 "%label_testfun = OpLabel\n"
5806 "%undef = OpUndef %u32\n"
5807 "%zero = OpIMul %u32 %undef %c_i32_0\n"
5808 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
5809 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
5810 "OpReturnValue %ret\n"
5813 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
5815 fragments["testfun"] =
5816 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5817 "%param1 = OpFunctionParameter %v4f32\n"
5818 "%label_testfun = OpLabel\n"
5819 "%undef = OpUndef %v4f32\n"
5820 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
5821 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
5822 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
5823 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
5824 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
5825 "%is_nan_0 = OpIsNan %bool %zero_0\n"
5826 "%is_nan_1 = OpIsNan %bool %zero_1\n"
5827 "%is_nan_2 = OpIsNan %bool %zero_2\n"
5828 "%is_nan_3 = OpIsNan %bool %zero_3\n"
5829 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
5830 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
5831 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
5832 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
5833 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5834 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
5835 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
5836 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
5837 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
5838 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
5839 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
5840 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
5841 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
5842 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
5843 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
5844 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
5845 "OpReturnValue %ret\n"
5848 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
5850 fragments["pre_main"] =
5851 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
5852 fragments["testfun"] =
5853 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5854 "%param1 = OpFunctionParameter %v4f32\n"
5855 "%label_testfun = OpLabel\n"
5856 "%undef = OpUndef %m2x2f32\n"
5857 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
5858 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
5859 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
5860 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
5861 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
5862 "%is_nan_0 = OpIsNan %bool %zero_0\n"
5863 "%is_nan_1 = OpIsNan %bool %zero_1\n"
5864 "%is_nan_2 = OpIsNan %bool %zero_2\n"
5865 "%is_nan_3 = OpIsNan %bool %zero_3\n"
5866 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
5867 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
5868 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
5869 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
5870 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
5871 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
5872 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
5873 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
5874 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
5875 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
5876 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
5877 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
5878 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
5879 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
5880 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
5881 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
5882 "OpReturnValue %ret\n"
5885 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
5887 return opUndefTests.release();
5890 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
5892 const RGBA inputColors[4] =
5895 RGBA(0, 0, 255, 255),
5896 RGBA(0, 255, 0, 255),
5897 RGBA(0, 255, 255, 255)
5900 const RGBA expectedColors[4] =
5902 RGBA(255, 0, 0, 255),
5903 RGBA(255, 0, 0, 255),
5904 RGBA(255, 0, 0, 255),
5905 RGBA(255, 0, 0, 255)
5908 const struct SingleFP16Possibility
5911 const char* constant; // Value to assign to %test_constant.
5913 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
5919 -constructNormalizedFloat(1, 0x300000),
5920 "%cond = OpFOrdEqual %bool %c %test_constant\n"
5925 constructNormalizedFloat(7, 0x000000),
5926 "%cond = OpFOrdEqual %bool %c %test_constant\n"
5928 // SPIR-V requires that OpQuantizeToF16 flushes
5929 // any numbers that would end up denormalized in F16 to zero.
5933 std::ldexp(1.5f, -140),
5934 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5939 -std::ldexp(1.5f, -140),
5940 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5945 std::ldexp(1.0f, -16),
5946 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5947 }, // too small positive
5949 "negative_too_small",
5951 -std::ldexp(1.0f, -32),
5952 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
5953 }, // too small negative
5957 -std::ldexp(1.0f, 128),
5959 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
5960 "%inf = OpIsInf %bool %c\n"
5961 "%cond = OpLogicalAnd %bool %gz %inf\n"
5966 std::ldexp(1.0f, 128),
5968 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
5969 "%inf = OpIsInf %bool %c\n"
5970 "%cond = OpLogicalAnd %bool %gz %inf\n"
5973 "round_to_negative_inf",
5975 -std::ldexp(1.0f, 32),
5977 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
5978 "%inf = OpIsInf %bool %c\n"
5979 "%cond = OpLogicalAnd %bool %gz %inf\n"
5984 std::ldexp(1.0f, 16),
5986 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
5987 "%inf = OpIsInf %bool %c\n"
5988 "%cond = OpLogicalAnd %bool %gz %inf\n"
5993 std::numeric_limits<float>::quiet_NaN(),
5995 // Test for any NaN value, as NaNs are not preserved
5996 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
5997 "%cond = OpIsNan %bool %direct_quant\n"
6002 std::numeric_limits<float>::quiet_NaN(),
6004 // Test for any NaN value, as NaNs are not preserved
6005 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6006 "%cond = OpIsNan %bool %direct_quant\n"
6009 const char* constants =
6010 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
6012 StringTemplate function (
6013 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6014 "%param1 = OpFunctionParameter %v4f32\n"
6015 "%label_testfun = OpLabel\n"
6016 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6017 "%b = OpFAdd %f32 %test_constant %a\n"
6018 "%c = OpQuantizeToF16 %f32 %b\n"
6020 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6021 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
6022 " OpReturnValue %retval\n"
6026 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
6027 const char* specConstants =
6028 "%test_constant = OpSpecConstant %f32 0.\n"
6029 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
6031 StringTemplate specConstantFunction(
6032 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6033 "%param1 = OpFunctionParameter %v4f32\n"
6034 "%label_testfun = OpLabel\n"
6036 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6037 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
6038 " OpReturnValue %retval\n"
6042 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6044 map<string, string> codeSpecialization;
6045 map<string, string> fragments;
6046 codeSpecialization["condition"] = tests[idx].condition;
6047 fragments["testfun"] = function.specialize(codeSpecialization);
6048 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
6049 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6052 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6054 map<string, string> codeSpecialization;
6055 map<string, string> fragments;
6056 vector<deInt32> passConstants;
6057 deInt32 specConstant;
6059 codeSpecialization["condition"] = tests[idx].condition;
6060 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
6061 fragments["decoration"] = specDecorations;
6062 fragments["pre_main"] = specConstants;
6064 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
6065 passConstants.push_back(specConstant);
6067 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6071 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
6073 RGBA inputColors[4] = {
6075 RGBA(0, 0, 255, 255),
6076 RGBA(0, 255, 0, 255),
6077 RGBA(0, 255, 255, 255)
6080 RGBA expectedColors[4] =
6082 RGBA(255, 0, 0, 255),
6083 RGBA(255, 0, 0, 255),
6084 RGBA(255, 0, 0, 255),
6085 RGBA(255, 0, 0, 255)
6088 struct DualFP16Possibility
6093 const char* possibleOutput1;
6094 const char* possibleOutput2;
6097 "positive_round_up_or_round_down",
6099 constructNormalizedFloat(8, 0x300300),
6104 "negative_round_up_or_round_down",
6106 -constructNormalizedFloat(-7, 0x600800),
6113 constructNormalizedFloat(2, 0x01e000),
6118 "carry_to_exponent",
6120 constructNormalizedFloat(1, 0xffe000),
6125 StringTemplate constants (
6126 "%input_const = OpConstant %f32 ${input}\n"
6127 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6128 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6131 StringTemplate specConstants (
6132 "%input_const = OpSpecConstant %f32 0.\n"
6133 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6134 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6137 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
6139 const char* function =
6140 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6141 "%param1 = OpFunctionParameter %v4f32\n"
6142 "%label_testfun = OpLabel\n"
6143 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6144 // For the purposes of this test we assume that 0.f will always get
6145 // faithfully passed through the pipeline stages.
6146 "%b = OpFAdd %f32 %input_const %a\n"
6147 "%c = OpQuantizeToF16 %f32 %b\n"
6148 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
6149 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
6150 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
6151 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6152 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
6153 " OpReturnValue %retval\n"
6156 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6157 map<string, string> fragments;
6158 map<string, string> constantSpecialization;
6160 constantSpecialization["input"] = tests[idx].input;
6161 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6162 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6163 fragments["testfun"] = function;
6164 fragments["pre_main"] = constants.specialize(constantSpecialization);
6165 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6168 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6169 map<string, string> fragments;
6170 map<string, string> constantSpecialization;
6171 vector<deInt32> passConstants;
6172 deInt32 specConstant;
6174 constantSpecialization["output1"] = tests[idx].possibleOutput1;
6175 constantSpecialization["output2"] = tests[idx].possibleOutput2;
6176 fragments["testfun"] = function;
6177 fragments["decoration"] = specDecorations;
6178 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
6180 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
6181 passConstants.push_back(specConstant);
6183 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6187 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
6189 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
6190 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
6191 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
6192 return opQuantizeTests.release();
6195 struct ShaderPermutation
6197 deUint8 vertexPermutation;
6198 deUint8 geometryPermutation;
6199 deUint8 tesscPermutation;
6200 deUint8 tessePermutation;
6201 deUint8 fragmentPermutation;
6204 ShaderPermutation getShaderPermutation(deUint8 inputValue)
6206 ShaderPermutation permutation =
6208 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
6209 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
6210 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
6211 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
6212 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
6217 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
6219 RGBA defaultColors[4];
6220 RGBA invertedColors[4];
6221 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
6223 const ShaderElement combinedPipeline[] =
6225 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
6226 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
6227 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
6228 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
6229 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
6232 getDefaultColors(defaultColors);
6233 getInvertedDefaultColors(invertedColors);
6234 addFunctionCaseWithPrograms<InstanceContext>(
6235 moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline,
6236 createInstanceContext(combinedPipeline, map<string, string>()));
6238 const char* numbers[] =
6243 for (deInt8 idx = 0; idx < 32; ++idx)
6245 ShaderPermutation permutation = getShaderPermutation(idx);
6246 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
6247 const ShaderElement pipeline[] =
6249 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
6250 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
6251 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
6252 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
6253 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
6256 // If there are an even number of swaps, then it should be no-op.
6257 // If there are an odd number, the color should be flipped.
6258 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
6260 addFunctionCaseWithPrograms<InstanceContext>(
6261 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
6262 createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
6266 addFunctionCaseWithPrograms<InstanceContext>(
6267 moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline,
6268 createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
6271 return moduleTests.release();
6274 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
6276 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
6277 RGBA defaultColors[4];
6278 getDefaultColors(defaultColors);
6279 map<string, string> fragments;
6280 fragments["pre_main"] =
6281 "%c_f32_5 = OpConstant %f32 5.\n";
6283 // A loop with a single block. The Continue Target is the loop block
6284 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
6285 // -- the "continue construct" forms the entire loop.
6286 fragments["testfun"] =
6287 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6288 "%param1 = OpFunctionParameter %v4f32\n"
6290 "%entry = OpLabel\n"
6291 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6294 ";adds and subtracts 1.0 to %val in alternate iterations\n"
6296 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
6297 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
6298 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
6299 "%val = OpFAdd %f32 %val1 %delta\n"
6300 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
6301 "%count__ = OpISub %i32 %count %c_i32_1\n"
6302 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6303 "OpLoopMerge %exit %loop None\n"
6304 "OpBranchConditional %again %loop %exit\n"
6307 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
6308 "OpReturnValue %result\n"
6312 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
6314 // Body comprised of multiple basic blocks.
6315 const StringTemplate multiBlock(
6316 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6317 "%param1 = OpFunctionParameter %v4f32\n"
6319 "%entry = OpLabel\n"
6320 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6323 ";adds and subtracts 1.0 to %val in alternate iterations\n"
6325 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
6326 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
6327 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
6328 // There are several possibilities for the Continue Target below. Each
6329 // will be specialized into a separate test case.
6330 "OpLoopMerge %exit ${continue_target} None\n"
6334 ";delta_next = (delta > 0) ? -1 : 1;\n"
6335 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
6336 "OpSelectionMerge %gather DontFlatten\n"
6337 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
6340 "OpBranch %gather\n"
6343 "OpBranch %gather\n"
6345 "%gather = OpLabel\n"
6346 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
6347 "%val = OpFAdd %f32 %val1 %delta\n"
6348 "%count__ = OpISub %i32 %count %c_i32_1\n"
6349 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6350 "OpBranchConditional %again %loop %exit\n"
6353 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
6354 "OpReturnValue %result\n"
6358 map<string, string> continue_target;
6360 // The Continue Target is the loop block itself.
6361 continue_target["continue_target"] = "%loop";
6362 fragments["testfun"] = multiBlock.specialize(continue_target);
6363 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
6365 // The Continue Target is at the end of the loop.
6366 continue_target["continue_target"] = "%gather";
6367 fragments["testfun"] = multiBlock.specialize(continue_target);
6368 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
6370 // A loop with continue statement.
6371 fragments["testfun"] =
6372 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6373 "%param1 = OpFunctionParameter %v4f32\n"
6375 "%entry = OpLabel\n"
6376 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6379 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
6381 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6382 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
6383 "OpLoopMerge %exit %continue None\n"
6387 ";skip if %count==2\n"
6388 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
6389 "OpSelectionMerge %continue DontFlatten\n"
6390 "OpBranchConditional %eq2 %continue %body\n"
6393 "%fcount = OpConvertSToF %f32 %count\n"
6394 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6395 "OpBranch %continue\n"
6397 "%continue = OpLabel\n"
6398 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
6399 "%count__ = OpISub %i32 %count %c_i32_1\n"
6400 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6401 "OpBranchConditional %again %loop %exit\n"
6404 "%same = OpFSub %f32 %val %c_f32_8\n"
6405 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6406 "OpReturnValue %result\n"
6408 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
6410 // A loop with break.
6411 fragments["testfun"] =
6412 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6413 "%param1 = OpFunctionParameter %v4f32\n"
6415 "%entry = OpLabel\n"
6416 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6417 "%dot = OpDot %f32 %param1 %param1\n"
6418 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6419 "%zero = OpConvertFToU %u32 %div\n"
6420 "%two = OpIAdd %i32 %zero %c_i32_2\n"
6421 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6424 ";adds 4 and 3 to %val0 (exits early)\n"
6426 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6427 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
6428 "OpLoopMerge %exit %continue None\n"
6432 ";end loop if %count==%two\n"
6433 "%above2 = OpSGreaterThan %bool %count %two\n"
6434 "OpSelectionMerge %continue DontFlatten\n"
6435 "OpBranchConditional %above2 %body %exit\n"
6438 "%fcount = OpConvertSToF %f32 %count\n"
6439 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6440 "OpBranch %continue\n"
6442 "%continue = OpLabel\n"
6443 "%count__ = OpISub %i32 %count %c_i32_1\n"
6444 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6445 "OpBranchConditional %again %loop %exit\n"
6448 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
6449 "%same = OpFSub %f32 %val_post %c_f32_7\n"
6450 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6451 "OpReturnValue %result\n"
6453 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
6455 // A loop with return.
6456 fragments["testfun"] =
6457 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6458 "%param1 = OpFunctionParameter %v4f32\n"
6460 "%entry = OpLabel\n"
6461 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6462 "%dot = OpDot %f32 %param1 %param1\n"
6463 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6464 "%zero = OpConvertFToU %u32 %div\n"
6465 "%two = OpIAdd %i32 %zero %c_i32_2\n"
6466 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6469 ";returns early without modifying %param1\n"
6471 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
6472 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
6473 "OpLoopMerge %exit %continue None\n"
6477 ";return if %count==%two\n"
6478 "%above2 = OpSGreaterThan %bool %count %two\n"
6479 "OpSelectionMerge %continue DontFlatten\n"
6480 "OpBranchConditional %above2 %body %early_exit\n"
6482 "%early_exit = OpLabel\n"
6483 "OpReturnValue %param1\n"
6486 "%fcount = OpConvertSToF %f32 %count\n"
6487 "%val2 = OpFAdd %f32 %val1 %fcount\n"
6488 "OpBranch %continue\n"
6490 "%continue = OpLabel\n"
6491 "%count__ = OpISub %i32 %count %c_i32_1\n"
6492 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6493 "OpBranchConditional %again %loop %exit\n"
6496 ";should never get here, so return an incorrect result\n"
6497 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
6498 "OpReturnValue %result\n"
6500 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
6502 return testGroup.release();
6505 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
6506 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
6508 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
6509 map<string, string> fragments;
6511 // A barrier inside a function body.
6512 fragments["pre_main"] =
6513 "%Workgroup = OpConstant %i32 2\n"
6514 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
6515 fragments["testfun"] =
6516 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6517 "%param1 = OpFunctionParameter %v4f32\n"
6518 "%label_testfun = OpLabel\n"
6519 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6520 "OpReturnValue %param1\n"
6522 addTessCtrlTest(testGroup.get(), "in_function", fragments);
6524 // Common setup code for the following tests.
6525 fragments["pre_main"] =
6526 "%Workgroup = OpConstant %i32 2\n"
6527 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
6528 "%c_f32_5 = OpConstant %f32 5.\n";
6529 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
6530 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6531 "%param1 = OpFunctionParameter %v4f32\n"
6532 "%entry = OpLabel\n"
6533 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
6534 "%dot = OpDot %f32 %param1 %param1\n"
6535 "%div = OpFDiv %f32 %dot %c_f32_5\n"
6536 "%zero = OpConvertFToU %u32 %div\n";
6538 // Barriers inside OpSwitch branches.
6539 fragments["testfun"] =
6541 "OpSelectionMerge %switch_exit None\n"
6542 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
6544 "%case1 = OpLabel\n"
6545 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6546 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6547 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6548 "OpBranch %switch_exit\n"
6550 "%switch_default = OpLabel\n"
6551 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6552 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6553 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6554 "OpBranch %switch_exit\n"
6556 "%case0 = OpLabel\n"
6557 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6558 "OpBranch %switch_exit\n"
6560 "%switch_exit = OpLabel\n"
6561 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
6562 "OpReturnValue %ret\n"
6564 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
6566 // Barriers inside if-then-else.
6567 fragments["testfun"] =
6569 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
6570 "OpSelectionMerge %exit DontFlatten\n"
6571 "OpBranchConditional %eq0 %then %else\n"
6574 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
6575 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6576 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
6580 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6584 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
6585 "OpReturnValue %ret\n"
6587 addTessCtrlTest(testGroup.get(), "in_if", fragments);
6589 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
6590 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
6591 fragments["testfun"] =
6593 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
6594 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
6595 "OpSelectionMerge %exit DontFlatten\n"
6596 "OpBranchConditional %thread0 %then %else\n"
6599 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6603 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
6607 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
6608 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6609 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
6610 "OpReturnValue %ret\n"
6612 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
6614 // A barrier inside a loop.
6615 fragments["pre_main"] =
6616 "%Workgroup = OpConstant %i32 2\n"
6617 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
6618 "%c_f32_10 = OpConstant %f32 10.\n";
6619 fragments["testfun"] =
6620 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6621 "%param1 = OpFunctionParameter %v4f32\n"
6622 "%entry = OpLabel\n"
6623 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6626 ";adds 4, 3, 2, and 1 to %val0\n"
6628 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
6629 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
6630 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
6631 "%fcount = OpConvertSToF %f32 %count\n"
6632 "%val = OpFAdd %f32 %val1 %fcount\n"
6633 "%count__ = OpISub %i32 %count %c_i32_1\n"
6634 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
6635 "OpLoopMerge %exit %loop None\n"
6636 "OpBranchConditional %again %loop %exit\n"
6639 "%same = OpFSub %f32 %val %c_f32_10\n"
6640 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
6641 "OpReturnValue %ret\n"
6643 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
6645 return testGroup.release();
6648 // Test for the OpFRem instruction.
6649 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
6651 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
6652 map<string, string> fragments;
6653 RGBA inputColors[4];
6654 RGBA outputColors[4];
6656 fragments["pre_main"] =
6657 "%c_f32_3 = OpConstant %f32 3.0\n"
6658 "%c_f32_n3 = OpConstant %f32 -3.0\n"
6659 "%c_f32_4 = OpConstant %f32 4.0\n"
6660 "%c_f32_p75 = OpConstant %f32 0.75\n"
6661 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
6662 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
6663 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
6665 // The test does the following.
6666 // vec4 result = (param1 * 8.0) - 4.0;
6667 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
6668 fragments["testfun"] =
6669 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6670 "%param1 = OpFunctionParameter %v4f32\n"
6671 "%label_testfun = OpLabel\n"
6672 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
6673 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
6674 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
6675 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
6676 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
6677 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
6678 "OpReturnValue %xy_0_1\n"
6682 inputColors[0] = RGBA(16, 16, 0, 255);
6683 inputColors[1] = RGBA(232, 232, 0, 255);
6684 inputColors[2] = RGBA(232, 16, 0, 255);
6685 inputColors[3] = RGBA(16, 232, 0, 255);
6687 outputColors[0] = RGBA(64, 64, 0, 255);
6688 outputColors[1] = RGBA(255, 255, 0, 255);
6689 outputColors[2] = RGBA(255, 64, 0, 255);
6690 outputColors[3] = RGBA(64, 255, 0, 255);
6692 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
6693 return testGroup.release();
6696 // Test for the OpSRem instruction.
6697 tcu::TestCaseGroup* createOpSRemGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
6699 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "srem", "OpSRem"));
6700 map<string, string> fragments;
6702 fragments["pre_main"] =
6703 "%c_f32_255 = OpConstant %f32 255.0\n"
6704 "%c_i32_128 = OpConstant %i32 128\n"
6705 "%c_i32_255 = OpConstant %i32 255\n"
6706 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
6707 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
6708 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
6710 // The test does the following.
6711 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
6712 // ivec4 result = ivec4(srem(ints.x, ints.y), srem(ints.y, ints.z), srem(ints.z, ints.x), 255);
6713 // return float(result + 128) / 255.0;
6714 fragments["testfun"] =
6715 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6716 "%param1 = OpFunctionParameter %v4f32\n"
6717 "%label_testfun = OpLabel\n"
6718 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
6719 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
6720 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
6721 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
6722 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
6723 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
6724 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
6725 "%x_out = OpSRem %i32 %x_in %y_in\n"
6726 "%y_out = OpSRem %i32 %y_in %z_in\n"
6727 "%z_out = OpSRem %i32 %z_in %x_in\n"
6728 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
6729 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
6730 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
6731 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
6732 "OpReturnValue %float_out\n"
6735 const struct CaseParams
6738 const char* failMessageTemplate; // customized status message
6739 qpTestResult failResult; // override status on failure
6740 int operands[4][3]; // four (x, y, z) vectors of operands
6741 int results[4][3]; // four (x, y, z) vectors of results
6747 QP_TEST_RESULT_FAIL,
6748 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
6749 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
6753 "Inconsistent results, but within specification: ${reason}",
6754 negFailResult, // negative operands, not required by the spec
6755 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
6756 { { 5, 12, -2 }, { 0, -5, 2 }, { 3, 8, -6 }, { 25, -60, 0 } }, // results
6759 // If either operand is negative the result is undefined. Some implementations may still return correct values.
6761 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
6763 const CaseParams& params = cases[caseNdx];
6764 RGBA inputColors[4];
6765 RGBA outputColors[4];
6767 for (int i = 0; i < 4; ++i)
6769 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
6770 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
6773 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
6776 return testGroup.release();
6779 // Test for the OpSMod instruction.
6780 tcu::TestCaseGroup* createOpSModGraphicsTests(tcu::TestContext& testCtx, qpTestResult negFailResult)
6782 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "smod", "OpSMod"));
6783 map<string, string> fragments;
6785 fragments["pre_main"] =
6786 "%c_f32_255 = OpConstant %f32 255.0\n"
6787 "%c_i32_128 = OpConstant %i32 128\n"
6788 "%c_i32_255 = OpConstant %i32 255\n"
6789 "%c_v4f32_255 = OpConstantComposite %v4f32 %c_f32_255 %c_f32_255 %c_f32_255 %c_f32_255 \n"
6790 "%c_v4f32_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 \n"
6791 "%c_v4i32_128 = OpConstantComposite %v4i32 %c_i32_128 %c_i32_128 %c_i32_128 %c_i32_128 \n";
6793 // The test does the following.
6794 // ivec4 ints = int(param1 * 255.0 + 0.5) - 128;
6795 // ivec4 result = ivec4(smod(ints.x, ints.y), smod(ints.y, ints.z), smod(ints.z, ints.x), 255);
6796 // return float(result + 128) / 255.0;
6797 fragments["testfun"] =
6798 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6799 "%param1 = OpFunctionParameter %v4f32\n"
6800 "%label_testfun = OpLabel\n"
6801 "%div255 = OpFMul %v4f32 %param1 %c_v4f32_255\n"
6802 "%add0_5 = OpFAdd %v4f32 %div255 %c_v4f32_0_5\n"
6803 "%uints_in = OpConvertFToS %v4i32 %add0_5\n"
6804 "%ints_in = OpISub %v4i32 %uints_in %c_v4i32_128\n"
6805 "%x_in = OpCompositeExtract %i32 %ints_in 0\n"
6806 "%y_in = OpCompositeExtract %i32 %ints_in 1\n"
6807 "%z_in = OpCompositeExtract %i32 %ints_in 2\n"
6808 "%x_out = OpSMod %i32 %x_in %y_in\n"
6809 "%y_out = OpSMod %i32 %y_in %z_in\n"
6810 "%z_out = OpSMod %i32 %z_in %x_in\n"
6811 "%ints_out = OpCompositeConstruct %v4i32 %x_out %y_out %z_out %c_i32_255\n"
6812 "%ints_offset = OpIAdd %v4i32 %ints_out %c_v4i32_128\n"
6813 "%f_ints_offset = OpConvertSToF %v4f32 %ints_offset\n"
6814 "%float_out = OpFDiv %v4f32 %f_ints_offset %c_v4f32_255\n"
6815 "OpReturnValue %float_out\n"
6818 const struct CaseParams
6821 const char* failMessageTemplate; // customized status message
6822 qpTestResult failResult; // override status on failure
6823 int operands[4][3]; // four (x, y, z) vectors of operands
6824 int results[4][3]; // four (x, y, z) vectors of results
6830 QP_TEST_RESULT_FAIL,
6831 { { 5, 12, 17 }, { 5, 5, 7 }, { 75, 8, 81 }, { 25, 60, 100 } }, // operands
6832 { { 5, 12, 2 }, { 0, 5, 2 }, { 3, 8, 6 }, { 25, 60, 0 } }, // results
6836 "Inconsistent results, but within specification: ${reason}",
6837 negFailResult, // negative operands, not required by the spec
6838 { { 5, 12, -17 }, { -5, -5, 7 }, { 75, 8, -81 }, { 25, -60, 100 } }, // operands
6839 { { 5, -5, 3 }, { 0, 2, -3 }, { 3, -73, 69 }, { -35, 40, 0 } }, // results
6842 // If either operand is negative the result is undefined. Some implementations may still return correct values.
6844 for (int caseNdx = 0; caseNdx < DE_LENGTH_OF_ARRAY(cases); ++caseNdx)
6846 const CaseParams& params = cases[caseNdx];
6847 RGBA inputColors[4];
6848 RGBA outputColors[4];
6850 for (int i = 0; i < 4; ++i)
6852 inputColors [i] = RGBA(params.operands[i][0] + 128, params.operands[i][1] + 128, params.operands[i][2] + 128, 255);
6853 outputColors[i] = RGBA(params.results [i][0] + 128, params.results [i][1] + 128, params.results [i][2] + 128, 255);
6856 createTestsForAllStages(params.name, inputColors, outputColors, fragments, testGroup.get(), params.failResult, params.failMessageTemplate);
6859 return testGroup.release();
6864 INTEGER_TYPE_SIGNED_16,
6865 INTEGER_TYPE_SIGNED_32,
6866 INTEGER_TYPE_SIGNED_64,
6868 INTEGER_TYPE_UNSIGNED_16,
6869 INTEGER_TYPE_UNSIGNED_32,
6870 INTEGER_TYPE_UNSIGNED_64,
6873 const string getBitWidthStr (IntegerType type)
6877 case INTEGER_TYPE_SIGNED_16:
6878 case INTEGER_TYPE_UNSIGNED_16: return "16";
6880 case INTEGER_TYPE_SIGNED_32:
6881 case INTEGER_TYPE_UNSIGNED_32: return "32";
6883 case INTEGER_TYPE_SIGNED_64:
6884 case INTEGER_TYPE_UNSIGNED_64: return "64";
6886 default: DE_ASSERT(false);
6891 const string getByteWidthStr (IntegerType type)
6895 case INTEGER_TYPE_SIGNED_16:
6896 case INTEGER_TYPE_UNSIGNED_16: return "2";
6898 case INTEGER_TYPE_SIGNED_32:
6899 case INTEGER_TYPE_UNSIGNED_32: return "4";
6901 case INTEGER_TYPE_SIGNED_64:
6902 case INTEGER_TYPE_UNSIGNED_64: return "8";
6904 default: DE_ASSERT(false);
6909 bool isSigned (IntegerType type)
6911 return (type <= INTEGER_TYPE_SIGNED_64);
6914 const string getTypeName (IntegerType type)
6916 string prefix = isSigned(type) ? "" : "u";
6917 return prefix + "int" + getBitWidthStr(type);
6920 const string getTestName (IntegerType from, IntegerType to)
6922 return getTypeName(from) + "_to_" + getTypeName(to);
6925 const string getAsmTypeDeclaration (IntegerType type)
6927 string sign = isSigned(type) ? " 1" : " 0";
6928 return "OpTypeInt " + getBitWidthStr(type) + sign;
6931 template<typename T>
6932 BufferSp getSpecializedBuffer (deInt64 number)
6934 return BufferSp(new Buffer<T>(vector<T>(1, (T)number)));
6937 BufferSp getBuffer (IntegerType type, deInt64 number)
6941 case INTEGER_TYPE_SIGNED_16: return getSpecializedBuffer<deInt16>(number);
6942 case INTEGER_TYPE_SIGNED_32: return getSpecializedBuffer<deInt32>(number);
6943 case INTEGER_TYPE_SIGNED_64: return getSpecializedBuffer<deInt64>(number);
6945 case INTEGER_TYPE_UNSIGNED_16: return getSpecializedBuffer<deUint16>(number);
6946 case INTEGER_TYPE_UNSIGNED_32: return getSpecializedBuffer<deUint32>(number);
6947 case INTEGER_TYPE_UNSIGNED_64: return getSpecializedBuffer<deUint64>(number);
6949 default: DE_ASSERT(false);
6950 return BufferSp(new Buffer<deInt32>(vector<deInt32>(1, 0)));
6954 bool usesInt16 (IntegerType from, IntegerType to)
6956 return (from == INTEGER_TYPE_SIGNED_16 || from == INTEGER_TYPE_UNSIGNED_16
6957 || to == INTEGER_TYPE_SIGNED_16 || to == INTEGER_TYPE_UNSIGNED_16);
6960 bool usesInt64 (IntegerType from, IntegerType to)
6962 return (from == INTEGER_TYPE_SIGNED_64 || from == INTEGER_TYPE_UNSIGNED_64
6963 || to == INTEGER_TYPE_SIGNED_64 || to == INTEGER_TYPE_UNSIGNED_64);
6966 ComputeTestFeatures getConversionUsedFeatures (IntegerType from, IntegerType to)
6968 if (usesInt16(from, to))
6970 if (usesInt64(from, to))
6972 return COMPUTE_TEST_USES_INT16_INT64;
6976 return COMPUTE_TEST_USES_INT16;
6981 return COMPUTE_TEST_USES_INT64;
6987 ConvertCase (IntegerType from, IntegerType to, deInt64 number)
6990 , m_features (getConversionUsedFeatures(from, to))
6991 , m_name (getTestName(from, to))
6992 , m_inputBuffer (getBuffer(from, number))
6993 , m_outputBuffer (getBuffer(to, number))
6995 m_asmTypes["inputType"] = getAsmTypeDeclaration(from);
6996 m_asmTypes["outputType"] = getAsmTypeDeclaration(to);
6998 if (m_features == COMPUTE_TEST_USES_INT16)
7000 m_asmTypes["int_capabilities"] = "OpCapability Int16\n";
7002 else if (m_features == COMPUTE_TEST_USES_INT64)
7004 m_asmTypes["int_capabilities"] = "OpCapability Int64\n";
7006 else if (m_features == COMPUTE_TEST_USES_INT16_INT64)
7008 m_asmTypes["int_capabilities"] = string("OpCapability Int16\n") +
7009 "OpCapability Int64\n";
7017 IntegerType m_fromType;
7018 IntegerType m_toType;
7019 ComputeTestFeatures m_features;
7021 map<string, string> m_asmTypes;
7022 BufferSp m_inputBuffer;
7023 BufferSp m_outputBuffer;
7026 const string getConvertCaseShaderStr (const string& instruction, const ConvertCase& convertCase)
7028 map<string, string> params = convertCase.m_asmTypes;
7030 params["instruction"] = instruction;
7032 params["inDecorator"] = getByteWidthStr(convertCase.m_fromType);
7033 params["outDecorator"] = getByteWidthStr(convertCase.m_toType);
7035 const StringTemplate shader (
7036 "OpCapability Shader\n"
7037 "${int_capabilities}"
7038 "OpMemoryModel Logical GLSL450\n"
7039 "OpEntryPoint GLCompute %main \"main\" %id\n"
7040 "OpExecutionMode %main LocalSize 1 1 1\n"
7041 "OpSource GLSL 430\n"
7042 "OpName %main \"main\"\n"
7043 "OpName %id \"gl_GlobalInvocationID\"\n"
7045 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7046 "OpDecorate %indata DescriptorSet 0\n"
7047 "OpDecorate %indata Binding 0\n"
7048 "OpDecorate %outdata DescriptorSet 0\n"
7049 "OpDecorate %outdata Binding 1\n"
7050 "OpDecorate %in_arr ArrayStride ${inDecorator}\n"
7051 "OpDecorate %out_arr ArrayStride ${outDecorator}\n"
7052 "OpDecorate %in_buf BufferBlock\n"
7053 "OpDecorate %out_buf BufferBlock\n"
7054 "OpMemberDecorate %in_buf 0 Offset 0\n"
7055 "OpMemberDecorate %out_buf 0 Offset 0\n"
7057 "%void = OpTypeVoid\n"
7058 "%voidf = OpTypeFunction %void\n"
7059 "%u32 = OpTypeInt 32 0\n"
7060 "%i32 = OpTypeInt 32 1\n"
7061 "%uvec3 = OpTypeVector %u32 3\n"
7062 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7064 "%in_type = ${inputType}\n"
7065 "%out_type = ${outputType}\n"
7067 "%in_ptr = OpTypePointer Uniform %in_type\n"
7068 "%out_ptr = OpTypePointer Uniform %out_type\n"
7069 "%in_arr = OpTypeRuntimeArray %in_type\n"
7070 "%out_arr = OpTypeRuntimeArray %out_type\n"
7071 "%in_buf = OpTypeStruct %in_arr\n"
7072 "%out_buf = OpTypeStruct %out_arr\n"
7073 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
7074 "%out_bufptr = OpTypePointer Uniform %out_buf\n"
7075 "%indata = OpVariable %in_bufptr Uniform\n"
7076 "%outdata = OpVariable %out_bufptr Uniform\n"
7077 "%inputptr = OpTypePointer Input %in_type\n"
7078 "%id = OpVariable %uvec3ptr Input\n"
7080 "%zero = OpConstant %i32 0\n"
7082 "%main = OpFunction %void None %voidf\n"
7083 "%label = OpLabel\n"
7084 "%idval = OpLoad %uvec3 %id\n"
7085 "%x = OpCompositeExtract %u32 %idval 0\n"
7086 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
7087 "%outloc = OpAccessChain %out_ptr %outdata %zero %x\n"
7088 "%inval = OpLoad %in_type %inloc\n"
7089 "%conv = ${instruction} %out_type %inval\n"
7090 " OpStore %outloc %conv\n"
7095 return shader.specialize(params);
7098 void createSConvertCases (vector<ConvertCase>& testCases)
7100 // Convert int to int
7101 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_32, 14669));
7102 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_SIGNED_64, 3341));
7104 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_SIGNED_64, 973610259));
7106 // Convert int to unsigned int
7107 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_32, 9288));
7108 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_16, INTEGER_TYPE_UNSIGNED_64, 15460));
7110 testCases.push_back(ConvertCase(INTEGER_TYPE_SIGNED_32, INTEGER_TYPE_UNSIGNED_64, 346213461));
7113 // Test for the OpSConvert instruction.
7114 tcu::TestCaseGroup* createSConvertTests (tcu::TestContext& testCtx)
7116 const string instruction ("OpSConvert");
7117 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "sconvert", "OpSConvert"));
7118 vector<ConvertCase> testCases;
7119 createSConvertCases(testCases);
7121 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7123 ComputeShaderSpec spec;
7125 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7126 spec.inputs.push_back(test->m_inputBuffer);
7127 spec.outputs.push_back(test->m_outputBuffer);
7128 spec.numWorkGroups = IVec3(1, 1, 1);
7130 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpSConvert.", spec, test->m_features));
7133 return group.release();
7136 void createUConvertCases (vector<ConvertCase>& testCases)
7138 // Convert unsigned int to unsigned int
7139 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_32, 60653));
7140 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_UNSIGNED_64, 17991));
7142 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_UNSIGNED_64, 904256275));
7144 // Convert unsigned int to int
7145 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_32, 38002));
7146 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_16, INTEGER_TYPE_SIGNED_64, 64921));
7148 testCases.push_back(ConvertCase(INTEGER_TYPE_UNSIGNED_32, INTEGER_TYPE_SIGNED_64, 4294956295ll));
7151 // Test for the OpUConvert instruction.
7152 tcu::TestCaseGroup* createUConvertTests (tcu::TestContext& testCtx)
7154 const string instruction ("OpUConvert");
7155 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "uconvert", "OpUConvert"));
7156 vector<ConvertCase> testCases;
7157 createUConvertCases(testCases);
7159 for (vector<ConvertCase>::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7161 ComputeShaderSpec spec;
7163 spec.assembly = getConvertCaseShaderStr(instruction, *test);
7164 spec.inputs.push_back(test->m_inputBuffer);
7165 spec.outputs.push_back(test->m_outputBuffer);
7166 spec.numWorkGroups = IVec3(1, 1, 1);
7168 group->addChild(new SpvAsmComputeShaderCase(testCtx, test->m_name.c_str(), "Convert integers with OpUConvert.", spec, test->m_features));
7170 return group.release();
7173 const string getNumberTypeName (const NumberType type)
7175 if (type == NUMBERTYPE_INT32)
7179 else if (type == NUMBERTYPE_UINT32)
7183 else if (type == NUMBERTYPE_FLOAT32)
7194 deInt32 getInt(de::Random& rnd)
7196 return rnd.getInt(std::numeric_limits<int>::min(), std::numeric_limits<int>::max());
7199 const string repeatString (const string& str, int times)
7202 for (int i = 0; i < times; ++i)
7209 const string getRandomConstantString (const NumberType type, de::Random& rnd)
7211 if (type == NUMBERTYPE_INT32)
7213 return numberToString<deInt32>(getInt(rnd));
7215 else if (type == NUMBERTYPE_UINT32)
7217 return numberToString<deUint32>(rnd.getUint32());
7219 else if (type == NUMBERTYPE_FLOAT32)
7221 return numberToString<float>(rnd.getFloat());
7230 void createVectorCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7232 map<string, string> params;
7235 for (int width = 2; width <= 4; ++width)
7237 string randomConst = numberToString(getInt(rnd));
7238 string widthStr = numberToString(width);
7239 int index = rnd.getInt(0, width-1);
7241 params["type"] = "vec";
7242 params["name"] = params["type"] + "_" + widthStr;
7243 params["compositeType"] = "%composite = OpTypeVector %custom " + widthStr +"\n";
7244 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7245 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7246 params["indexes"] = numberToString(index);
7247 testCases.push_back(params);
7251 void createArrayCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7253 const int limit = 10;
7254 map<string, string> params;
7256 for (int width = 2; width <= limit; ++width)
7258 string randomConst = numberToString(getInt(rnd));
7259 string widthStr = numberToString(width);
7260 int index = rnd.getInt(0, width-1);
7262 params["type"] = "array";
7263 params["name"] = params["type"] + "_" + widthStr;
7264 params["compositeType"] = string("%arraywidth = OpConstant %u32 " + widthStr + "\n")
7265 + "%composite = OpTypeArray %custom %arraywidth\n";
7267 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7268 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7269 params["indexes"] = numberToString(index);
7270 testCases.push_back(params);
7274 void createStructCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7276 const int limit = 10;
7277 map<string, string> params;
7279 for (int width = 2; width <= limit; ++width)
7281 string randomConst = numberToString(getInt(rnd));
7282 int index = rnd.getInt(0, width-1);
7284 params["type"] = "struct";
7285 params["name"] = params["type"] + "_" + numberToString(width);
7286 params["compositeType"] = "%composite = OpTypeStruct" + repeatString(" %custom", width) + "\n";
7287 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n";
7288 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %filler", width) + "\n";
7289 params["indexes"] = numberToString(index);
7290 testCases.push_back(params);
7294 void createMatrixCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7296 map<string, string> params;
7299 for (int width = 2; width <= 4; ++width)
7301 string widthStr = numberToString(width);
7303 for (int column = 2 ; column <= 4; ++column)
7305 int index_0 = rnd.getInt(0, column-1);
7306 int index_1 = rnd.getInt(0, width-1);
7307 string columnStr = numberToString(column);
7309 params["type"] = "matrix";
7310 params["name"] = params["type"] + "_" + widthStr + "x" + columnStr;
7311 params["compositeType"] = string("%vectype = OpTypeVector %custom " + widthStr + "\n")
7312 + "%composite = OpTypeMatrix %vectype " + columnStr + "\n";
7314 params["filler"] = string("%filler = OpConstant %custom ") + getRandomConstantString(type, rnd) + "\n"
7315 + "%fillerVec = OpConstantComposite %vectype" + repeatString(" %filler", width) + "\n";
7317 params["compositeConstruct"] = "%instance = OpCompositeConstruct %composite" + repeatString(" %fillerVec", column) + "\n";
7318 params["indexes"] = numberToString(index_0) + " " + numberToString(index_1);
7319 testCases.push_back(params);
7324 void createCompositeCases (vector<map<string, string> >& testCases, de::Random& rnd, const NumberType type)
7326 createVectorCompositeCases(testCases, rnd, type);
7327 createArrayCompositeCases(testCases, rnd, type);
7328 createStructCompositeCases(testCases, rnd, type);
7329 // Matrix only supports float types
7330 if (type == NUMBERTYPE_FLOAT32)
7332 createMatrixCompositeCases(testCases, rnd, type);
7336 const string getAssemblyTypeDeclaration (const NumberType type)
7340 case NUMBERTYPE_INT32: return "OpTypeInt 32 1";
7341 case NUMBERTYPE_UINT32: return "OpTypeInt 32 0";
7342 case NUMBERTYPE_FLOAT32: return "OpTypeFloat 32";
7343 default: DE_ASSERT(false); return "";
7347 const string specializeCompositeInsertShaderTemplate (const NumberType type, const map<string, string>& params)
7349 map<string, string> parameters(params);
7351 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7353 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
7355 return StringTemplate (
7356 "OpCapability Shader\n"
7357 "OpCapability Matrix\n"
7358 "OpMemoryModel Logical GLSL450\n"
7359 "OpEntryPoint GLCompute %main \"main\" %id\n"
7360 "OpExecutionMode %main LocalSize 1 1 1\n"
7362 "OpSource GLSL 430\n"
7363 "OpName %main \"main\"\n"
7364 "OpName %id \"gl_GlobalInvocationID\"\n"
7367 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7368 "OpDecorate %buf BufferBlock\n"
7369 "OpDecorate %indata DescriptorSet 0\n"
7370 "OpDecorate %indata Binding 0\n"
7371 "OpDecorate %outdata DescriptorSet 0\n"
7372 "OpDecorate %outdata Binding 1\n"
7373 "OpDecorate %customarr ArrayStride 4\n"
7374 "${compositeDecorator}"
7375 "OpMemberDecorate %buf 0 Offset 0\n"
7378 "%void = OpTypeVoid\n"
7379 "%voidf = OpTypeFunction %void\n"
7380 "%u32 = OpTypeInt 32 0\n"
7381 "%i32 = OpTypeInt 32 1\n"
7382 "%uvec3 = OpTypeVector %u32 3\n"
7383 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7386 "%custom = ${typeDeclaration}\n"
7392 // Inherited from custom
7393 "%customptr = OpTypePointer Uniform %custom\n"
7394 "%customarr = OpTypeRuntimeArray %custom\n"
7395 "%buf = OpTypeStruct %customarr\n"
7396 "%bufptr = OpTypePointer Uniform %buf\n"
7398 "%indata = OpVariable %bufptr Uniform\n"
7399 "%outdata = OpVariable %bufptr Uniform\n"
7401 "%id = OpVariable %uvec3ptr Input\n"
7402 "%zero = OpConstant %i32 0\n"
7404 "%main = OpFunction %void None %voidf\n"
7405 "%label = OpLabel\n"
7406 "%idval = OpLoad %uvec3 %id\n"
7407 "%x = OpCompositeExtract %u32 %idval 0\n"
7409 "%inloc = OpAccessChain %customptr %indata %zero %x\n"
7410 "%outloc = OpAccessChain %customptr %outdata %zero %x\n"
7411 // Read the input value
7412 "%inval = OpLoad %custom %inloc\n"
7413 // Create the composite and fill it
7414 "${compositeConstruct}"
7415 // Insert the input value to a place
7416 "%instance2 = OpCompositeInsert %composite %inval %instance ${indexes}\n"
7417 // Read back the value from the position
7418 "%out_val = OpCompositeExtract %custom %instance2 ${indexes}\n"
7419 // Store it in the output position
7420 " OpStore %outloc %out_val\n"
7423 ).specialize(parameters);
7426 template<typename T>
7427 BufferSp createCompositeBuffer(T number)
7429 return BufferSp(new Buffer<T>(vector<T>(1, number)));
7432 tcu::TestCaseGroup* createOpCompositeInsertGroup (tcu::TestContext& testCtx)
7434 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcompositeinsert", "Test the OpCompositeInsert instruction"));
7435 de::Random rnd (deStringHash(group->getName()));
7437 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7439 NumberType numberType = NumberType(type);
7440 const string typeName = getNumberTypeName(numberType);
7441 const string description = "Test the OpCompositeInsert instruction with " + typeName + "s";
7442 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7443 vector<map<string, string> > testCases;
7445 createCompositeCases(testCases, rnd, numberType);
7447 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7449 ComputeShaderSpec spec;
7451 spec.assembly = specializeCompositeInsertShaderTemplate(numberType, *test);
7455 case NUMBERTYPE_INT32:
7457 deInt32 number = getInt(rnd);
7458 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7459 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7462 case NUMBERTYPE_UINT32:
7464 deUint32 number = rnd.getUint32();
7465 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7466 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7469 case NUMBERTYPE_FLOAT32:
7471 float number = rnd.getFloat();
7472 spec.inputs.push_back(createCompositeBuffer<float>(number));
7473 spec.outputs.push_back(createCompositeBuffer<float>(number));
7480 spec.numWorkGroups = IVec3(1, 1, 1);
7481 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpCompositeInsert test", spec));
7483 group->addChild(subGroup.release());
7485 return group.release();
7488 struct AssemblyStructInfo
7490 AssemblyStructInfo (const deUint32 comp, const deUint32 idx)
7495 deUint32 components;
7499 const string specializeInBoundsShaderTemplate (const NumberType type, const AssemblyStructInfo& structInfo, const map<string, string>& params)
7501 // Create the full index string
7502 string fullIndex = numberToString(structInfo.index) + " " + params.at("indexes");
7503 // Convert it to list of indexes
7504 vector<string> indexes = de::splitString(fullIndex, ' ');
7506 map<string, string> parameters (params);
7507 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7508 parameters["structType"] = repeatString(" %composite", structInfo.components);
7509 parameters["structConstruct"] = repeatString(" %instance", structInfo.components);
7510 parameters["insertIndexes"] = fullIndex;
7512 // In matrix cases the last two index is the CompositeExtract indexes
7513 const deUint32 extractIndexes = (parameters["type"] == "matrix") ? 2 : 1;
7515 // Construct the extractIndex
7516 for (vector<string>::const_iterator index = indexes.end() - extractIndexes; index != indexes.end(); ++index)
7518 parameters["extractIndexes"] += " " + *index;
7521 // Remove the last 1 or 2 element depends on matrix case or not
7522 indexes.erase(indexes.end() - extractIndexes, indexes.end());
7525 // Generate AccessChain index expressions (except for the last one, because we use ptr to the composite)
7526 for (vector<string>::const_iterator index = indexes.begin(); index != indexes.end(); ++index)
7528 string indexId = "%index_" + numberToString(id++);
7529 parameters["accessChainConstDeclaration"] += indexId + " = OpConstant %u32 " + *index + "\n";
7530 parameters["accessChainIndexes"] += " " + indexId;
7533 parameters["compositeDecorator"] = (parameters["type"] == "array") ? "OpDecorate %composite ArrayStride 4\n" : "";
7535 return StringTemplate (
7536 "OpCapability Shader\n"
7537 "OpCapability Matrix\n"
7538 "OpMemoryModel Logical GLSL450\n"
7539 "OpEntryPoint GLCompute %main \"main\" %id\n"
7540 "OpExecutionMode %main LocalSize 1 1 1\n"
7542 "OpSource GLSL 430\n"
7543 "OpName %main \"main\"\n"
7544 "OpName %id \"gl_GlobalInvocationID\"\n"
7546 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7547 "OpDecorate %buf BufferBlock\n"
7548 "OpDecorate %indata DescriptorSet 0\n"
7549 "OpDecorate %indata Binding 0\n"
7550 "OpDecorate %outdata DescriptorSet 0\n"
7551 "OpDecorate %outdata Binding 1\n"
7552 "OpDecorate %customarr ArrayStride 4\n"
7553 "${compositeDecorator}"
7554 "OpMemberDecorate %buf 0 Offset 0\n"
7556 "%void = OpTypeVoid\n"
7557 "%voidf = OpTypeFunction %void\n"
7558 "%u32 = OpTypeInt 32 0\n"
7559 "%uvec3 = OpTypeVector %u32 3\n"
7560 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7562 "%custom = ${typeDeclaration}\n"
7565 // Inherited from composite
7566 "%composite_p = OpTypePointer Function %composite\n"
7567 "%struct_t = OpTypeStruct${structType}\n"
7568 "%struct_p = OpTypePointer Function %struct_t\n"
7571 "${accessChainConstDeclaration}"
7572 // Inherited from custom
7573 "%customptr = OpTypePointer Uniform %custom\n"
7574 "%customarr = OpTypeRuntimeArray %custom\n"
7575 "%buf = OpTypeStruct %customarr\n"
7576 "%bufptr = OpTypePointer Uniform %buf\n"
7577 "%indata = OpVariable %bufptr Uniform\n"
7578 "%outdata = OpVariable %bufptr Uniform\n"
7580 "%id = OpVariable %uvec3ptr Input\n"
7581 "%zero = OpConstant %u32 0\n"
7582 "%main = OpFunction %void None %voidf\n"
7583 "%label = OpLabel\n"
7584 "%struct_v = OpVariable %struct_p Function\n"
7585 "%idval = OpLoad %uvec3 %id\n"
7586 "%x = OpCompositeExtract %u32 %idval 0\n"
7587 // Create the input/output type
7588 "%inloc = OpInBoundsAccessChain %customptr %indata %zero %x\n"
7589 "%outloc = OpInBoundsAccessChain %customptr %outdata %zero %x\n"
7590 // Read the input value
7591 "%inval = OpLoad %custom %inloc\n"
7592 // Create the composite and fill it
7593 "${compositeConstruct}"
7594 // Create the struct and fill it with the composite
7595 "%struct = OpCompositeConstruct %struct_t${structConstruct}\n"
7597 "%comp_obj = OpCompositeInsert %struct_t %inval %struct ${insertIndexes}\n"
7599 " OpStore %struct_v %comp_obj\n"
7600 // Get deepest possible composite pointer
7601 "%inner_ptr = OpInBoundsAccessChain %composite_p %struct_v${accessChainIndexes}\n"
7602 "%read_obj = OpLoad %composite %inner_ptr\n"
7603 // Read back the stored value
7604 "%read_val = OpCompositeExtract %custom %read_obj${extractIndexes}\n"
7605 " OpStore %outloc %read_val\n"
7607 " OpFunctionEnd\n").specialize(parameters);
7610 tcu::TestCaseGroup* createOpInBoundsAccessChainGroup (tcu::TestContext& testCtx)
7612 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opinboundsaccesschain", "Test the OpInBoundsAccessChain instruction"));
7613 de::Random rnd (deStringHash(group->getName()));
7615 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7617 NumberType numberType = NumberType(type);
7618 const string typeName = getNumberTypeName(numberType);
7619 const string description = "Test the OpInBoundsAccessChain instruction with " + typeName + "s";
7620 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7622 vector<map<string, string> > testCases;
7623 createCompositeCases(testCases, rnd, numberType);
7625 for (vector<map<string, string> >::const_iterator test = testCases.begin(); test != testCases.end(); ++test)
7627 ComputeShaderSpec spec;
7629 // Number of components inside of a struct
7630 deUint32 structComponents = rnd.getInt(2, 8);
7631 // Component index value
7632 deUint32 structIndex = rnd.getInt(0, structComponents - 1);
7633 AssemblyStructInfo structInfo(structComponents, structIndex);
7635 spec.assembly = specializeInBoundsShaderTemplate(numberType, structInfo, *test);
7639 case NUMBERTYPE_INT32:
7641 deInt32 number = getInt(rnd);
7642 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7643 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7646 case NUMBERTYPE_UINT32:
7648 deUint32 number = rnd.getUint32();
7649 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7650 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7653 case NUMBERTYPE_FLOAT32:
7655 float number = rnd.getFloat();
7656 spec.inputs.push_back(createCompositeBuffer<float>(number));
7657 spec.outputs.push_back(createCompositeBuffer<float>(number));
7663 spec.numWorkGroups = IVec3(1, 1, 1);
7664 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, test->at("name").c_str(), "OpInBoundsAccessChain test", spec));
7666 group->addChild(subGroup.release());
7668 return group.release();
7671 // If the params missing, uninitialized case
7672 const string specializeDefaultOutputShaderTemplate (const NumberType type, const map<string, string>& params = map<string, string>())
7674 map<string, string> parameters(params);
7676 parameters["typeDeclaration"] = getAssemblyTypeDeclaration(type);
7678 // Declare the const value, and use it in the initializer
7679 if (params.find("constValue") != params.end())
7681 parameters["constDeclaration"] = "%const = OpConstant %in_type " + params.at("constValue") + "\n";
7682 parameters["variableInitializer"] = "%const";
7684 // Uninitialized case
7687 parameters["constDeclaration"] = "";
7688 parameters["variableInitializer"] = "";
7691 return StringTemplate(
7692 "OpCapability Shader\n"
7693 "OpMemoryModel Logical GLSL450\n"
7694 "OpEntryPoint GLCompute %main \"main\" %id\n"
7695 "OpExecutionMode %main LocalSize 1 1 1\n"
7696 "OpSource GLSL 430\n"
7697 "OpName %main \"main\"\n"
7698 "OpName %id \"gl_GlobalInvocationID\"\n"
7700 "OpDecorate %id BuiltIn GlobalInvocationId\n"
7701 "OpDecorate %indata DescriptorSet 0\n"
7702 "OpDecorate %indata Binding 0\n"
7703 "OpDecorate %outdata DescriptorSet 0\n"
7704 "OpDecorate %outdata Binding 1\n"
7705 "OpDecorate %in_arr ArrayStride 4\n"
7706 "OpDecorate %in_buf BufferBlock\n"
7707 "OpMemberDecorate %in_buf 0 Offset 0\n"
7709 "%void = OpTypeVoid\n"
7710 "%voidf = OpTypeFunction %void\n"
7711 "%u32 = OpTypeInt 32 0\n"
7712 "%i32 = OpTypeInt 32 1\n"
7713 "%uvec3 = OpTypeVector %u32 3\n"
7714 "%uvec3ptr = OpTypePointer Input %uvec3\n"
7716 "%in_type = ${typeDeclaration}\n"
7717 // "%const = OpConstant %in_type ${constValue}\n"
7718 "${constDeclaration}\n"
7720 "%in_ptr = OpTypePointer Uniform %in_type\n"
7721 "%in_arr = OpTypeRuntimeArray %in_type\n"
7722 "%in_buf = OpTypeStruct %in_arr\n"
7723 "%in_bufptr = OpTypePointer Uniform %in_buf\n"
7724 "%indata = OpVariable %in_bufptr Uniform\n"
7725 "%outdata = OpVariable %in_bufptr Uniform\n"
7726 "%id = OpVariable %uvec3ptr Input\n"
7727 "%var_ptr = OpTypePointer Function %in_type\n"
7729 "%zero = OpConstant %i32 0\n"
7731 "%main = OpFunction %void None %voidf\n"
7732 "%label = OpLabel\n"
7733 "%out_var = OpVariable %var_ptr Function ${variableInitializer}\n"
7734 "%idval = OpLoad %uvec3 %id\n"
7735 "%x = OpCompositeExtract %u32 %idval 0\n"
7736 "%inloc = OpAccessChain %in_ptr %indata %zero %x\n"
7737 "%outloc = OpAccessChain %in_ptr %outdata %zero %x\n"
7739 "%outval = OpLoad %in_type %out_var\n"
7740 " OpStore %outloc %outval\n"
7743 ).specialize(parameters);
7746 bool compareFloats (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog& log)
7748 DE_ASSERT(outputAllocs.size() != 0);
7749 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
7751 // Use custom epsilon because of the float->string conversion
7752 const float epsilon = 0.00001f;
7754 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
7757 memcpy(&expected, expectedOutputs[outputNdx]->data(), expectedOutputs[outputNdx]->getNumBytes());
7760 memcpy(&actual, outputAllocs[outputNdx]->getHostPtr(), expectedOutputs[outputNdx]->getNumBytes());
7762 // Test with epsilon
7763 if (fabs(expected - actual) > epsilon)
7765 log << TestLog::Message << "Error: The actual and expected values not matching."
7766 << " Expected: " << expected << " Actual: " << actual << " Epsilon: " << epsilon << TestLog::EndMessage;
7773 // Checks if the driver crash with uninitialized cases
7774 bool passthruVerify (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs, TestLog&)
7776 DE_ASSERT(outputAllocs.size() != 0);
7777 DE_ASSERT(outputAllocs.size() == expectedOutputs.size());
7779 // Copy and discard the result.
7780 for (size_t outputNdx = 0; outputNdx < outputAllocs.size(); ++outputNdx)
7782 size_t width = expectedOutputs[outputNdx]->getNumBytes();
7784 vector<char> data(width);
7785 memcpy(&data[0], outputAllocs[outputNdx]->getHostPtr(), width);
7790 tcu::TestCaseGroup* createShaderDefaultOutputGroup (tcu::TestContext& testCtx)
7792 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "shader_default_output", "Test shader default output."));
7793 de::Random rnd (deStringHash(group->getName()));
7795 for (int type = NUMBERTYPE_INT32; type != NUMBERTYPE_END32; ++type)
7797 NumberType numberType = NumberType(type);
7798 const string typeName = getNumberTypeName(numberType);
7799 const string description = "Test the OpVariable initializer with " + typeName + ".";
7800 de::MovePtr<tcu::TestCaseGroup> subGroup (new tcu::TestCaseGroup(testCtx, typeName.c_str(), description.c_str()));
7802 // 2 similar subcases (initialized and uninitialized)
7803 for (int subCase = 0; subCase < 2; ++subCase)
7805 ComputeShaderSpec spec;
7806 spec.numWorkGroups = IVec3(1, 1, 1);
7808 map<string, string> params;
7812 case NUMBERTYPE_INT32:
7814 deInt32 number = getInt(rnd);
7815 spec.inputs.push_back(createCompositeBuffer<deInt32>(number));
7816 spec.outputs.push_back(createCompositeBuffer<deInt32>(number));
7817 params["constValue"] = numberToString(number);
7820 case NUMBERTYPE_UINT32:
7822 deUint32 number = rnd.getUint32();
7823 spec.inputs.push_back(createCompositeBuffer<deUint32>(number));
7824 spec.outputs.push_back(createCompositeBuffer<deUint32>(number));
7825 params["constValue"] = numberToString(number);
7828 case NUMBERTYPE_FLOAT32:
7830 float number = rnd.getFloat();
7831 spec.inputs.push_back(createCompositeBuffer<float>(number));
7832 spec.outputs.push_back(createCompositeBuffer<float>(number));
7833 spec.verifyIO = &compareFloats;
7834 params["constValue"] = numberToString(number);
7841 // Initialized subcase
7844 spec.assembly = specializeDefaultOutputShaderTemplate(numberType, params);
7845 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "initialized", "OpVariable initializer tests.", spec));
7847 // Uninitialized subcase
7850 spec.assembly = specializeDefaultOutputShaderTemplate(numberType);
7851 spec.verifyIO = &passthruVerify;
7852 subGroup->addChild(new SpvAsmComputeShaderCase(testCtx, "uninitialized", "OpVariable initializer tests.", spec));
7855 group->addChild(subGroup.release());
7857 return group.release();
7860 tcu::TestCaseGroup* createOpNopTests (tcu::TestContext& testCtx)
7862 de::MovePtr<tcu::TestCaseGroup> testGroup (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
7863 RGBA defaultColors[4];
7864 map<string, string> opNopFragments;
7866 getDefaultColors(defaultColors);
7868 opNopFragments["testfun"] =
7869 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7870 "%param1 = OpFunctionParameter %v4f32\n"
7871 "%label_testfun = OpLabel\n"
7880 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7881 "%b = OpFAdd %f32 %a %a\n"
7883 "%c = OpFSub %f32 %b %a\n"
7884 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
7887 "OpReturnValue %ret\n"
7890 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, testGroup.get());
7892 return testGroup.release();
7895 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
7897 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
7898 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
7899 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
7901 computeTests->addChild(createOpNopGroup(testCtx));
7902 computeTests->addChild(createOpFUnordGroup(testCtx));
7903 computeTests->addChild(createOpAtomicGroup(testCtx, false));
7904 computeTests->addChild(createOpAtomicGroup(testCtx, true)); // Using new StorageBuffer decoration
7905 computeTests->addChild(createOpLineGroup(testCtx));
7906 computeTests->addChild(createOpNoLineGroup(testCtx));
7907 computeTests->addChild(createOpConstantNullGroup(testCtx));
7908 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
7909 computeTests->addChild(createOpConstantUsageGroup(testCtx));
7910 computeTests->addChild(createSpecConstantGroup(testCtx));
7911 computeTests->addChild(createOpSourceGroup(testCtx));
7912 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
7913 computeTests->addChild(createDecorationGroupGroup(testCtx));
7914 computeTests->addChild(createOpPhiGroup(testCtx));
7915 computeTests->addChild(createLoopControlGroup(testCtx));
7916 computeTests->addChild(createFunctionControlGroup(testCtx));
7917 computeTests->addChild(createSelectionControlGroup(testCtx));
7918 computeTests->addChild(createBlockOrderGroup(testCtx));
7919 computeTests->addChild(createMultipleShaderGroup(testCtx));
7920 computeTests->addChild(createMemoryAccessGroup(testCtx));
7921 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
7922 computeTests->addChild(createOpCopyObjectGroup(testCtx));
7923 computeTests->addChild(createNoContractionGroup(testCtx));
7924 computeTests->addChild(createOpUndefGroup(testCtx));
7925 computeTests->addChild(createOpUnreachableGroup(testCtx));
7926 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
7927 computeTests ->addChild(createOpFRemGroup(testCtx));
7928 computeTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_PASS));
7929 computeTests->addChild(createOpSRemComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
7930 computeTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_PASS));
7931 computeTests->addChild(createOpSModComputeGroup64(testCtx, QP_TEST_RESULT_PASS));
7932 computeTests->addChild(createSConvertTests(testCtx));
7933 computeTests->addChild(createUConvertTests(testCtx));
7934 computeTests->addChild(createOpCompositeInsertGroup(testCtx));
7935 computeTests->addChild(createOpInBoundsAccessChainGroup(testCtx));
7936 computeTests->addChild(createShaderDefaultOutputGroup(testCtx));
7937 computeTests->addChild(createOpNMinGroup(testCtx));
7938 computeTests->addChild(createOpNMaxGroup(testCtx));
7939 computeTests->addChild(createOpNClampGroup(testCtx));
7941 de::MovePtr<tcu::TestCaseGroup> computeAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
7943 computeAndroidTests->addChild(createOpSRemComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7944 computeAndroidTests->addChild(createOpSModComputeGroup(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7946 computeTests->addChild(computeAndroidTests.release());
7949 computeTests->addChild(create16BitStorageComputeGroup(testCtx));
7950 computeTests->addChild(createUboMatrixPaddingComputeGroup(testCtx));
7951 computeTests->addChild(createConditionalBranchComputeGroup(testCtx));
7952 computeTests->addChild(createIndexingComputeGroup(testCtx));
7953 computeTests->addChild(createVariablePointersComputeGroup(testCtx));
7954 graphicsTests->addChild(createOpNopTests(testCtx));
7955 graphicsTests->addChild(createOpSourceTests(testCtx));
7956 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
7957 graphicsTests->addChild(createOpLineTests(testCtx));
7958 graphicsTests->addChild(createOpNoLineTests(testCtx));
7959 graphicsTests->addChild(createOpConstantNullTests(testCtx));
7960 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
7961 graphicsTests->addChild(createMemoryAccessTests(testCtx));
7962 graphicsTests->addChild(createOpUndefTests(testCtx));
7963 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
7964 graphicsTests->addChild(createModuleTests(testCtx));
7965 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
7966 graphicsTests->addChild(createOpPhiTests(testCtx));
7967 graphicsTests->addChild(createNoContractionTests(testCtx));
7968 graphicsTests->addChild(createOpQuantizeTests(testCtx));
7969 graphicsTests->addChild(createLoopTests(testCtx));
7970 graphicsTests->addChild(createSpecConstantTests(testCtx));
7971 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
7972 graphicsTests->addChild(createBarrierTests(testCtx));
7973 graphicsTests->addChild(createDecorationGroupTests(testCtx));
7974 graphicsTests->addChild(createFRemTests(testCtx));
7975 graphicsTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
7976 graphicsTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_PASS));
7979 de::MovePtr<tcu::TestCaseGroup> graphicsAndroidTests (new tcu::TestCaseGroup(testCtx, "android", "Android CTS Tests"));
7981 graphicsAndroidTests->addChild(createOpSRemGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7982 graphicsAndroidTests->addChild(createOpSModGraphicsTests(testCtx, QP_TEST_RESULT_QUALITY_WARNING));
7984 graphicsTests->addChild(graphicsAndroidTests.release());
7987 graphicsTests->addChild(create16BitStorageGraphicsGroup(testCtx));
7988 graphicsTests->addChild(createUboMatrixPaddingGraphicsGroup(testCtx));
7989 graphicsTests->addChild(createConditionalBranchGraphicsGroup(testCtx));
7990 graphicsTests->addChild(createIndexingGraphicsGroup(testCtx));
7991 graphicsTests->addChild(createVariablePointersGraphicsGroup(testCtx));
7993 instructionTests->addChild(computeTests.release());
7994 instructionTests->addChild(graphicsTests.release());
7996 return instructionTests.release();