1 /*-------------------------------------------------------------------------
2 * Vulkan Conformance Tests
3 * ------------------------
4 *
5 * Copyright (c) 2015 Google Inc.
6 *
7 * Licensed under the Apache License, Version 2.0 (the "License");
8 * you may not use this file except in compliance with the License.
9 * You may obtain a copy of the License at
10 *
11 * http://www.apache.org/licenses/LICENSE-2.0
12 *
13 * Unless required by applicable law or agreed to in writing, software
14 * distributed under the License is distributed on an "AS IS" BASIS,
15 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16 * See the License for the specific language governing permissions and
17 * limitations under the License.
18 *
19 *//*!
20 * \file
21 * \brief SPIR-V Assembly Tests for Instructions (special opcode/operand)
22 *//*--------------------------------------------------------------------*/
23
24 #include "vktSpvAsmInstructionTests.hpp"
25
26 #include "tcuCommandLine.hpp"
27 #include "tcuFormatUtil.hpp"
28 #include "tcuRGBA.hpp"
29 #include "tcuStringTemplate.hpp"
30 #include "tcuTestLog.hpp"
31 #include "tcuVectorUtil.hpp"
32
33 #include "vkDefs.hpp"
34 #include "vkDeviceUtil.hpp"
35 #include "vkMemUtil.hpp"
36 #include "vkPlatform.hpp"
37 #include "vkPrograms.hpp"
38 #include "vkQueryUtil.hpp"
39 #include "vkRef.hpp"
40 #include "vkRefUtil.hpp"
41 #include "vkStrUtil.hpp"
42 #include "vkTypeUtil.hpp"
43
44 #include "deRandom.hpp"
45 #include "deStringUtil.hpp"
46 #include "deUniquePtr.hpp"
47 #include "tcuStringTemplate.hpp"
48
49 #include <cmath>
50 #include "vktSpvAsmComputeShaderCase.hpp"
51 #include "vktSpvAsmComputeShaderTestUtil.hpp"
52 #include "vktTestCaseUtil.hpp"
53
54 #include <cmath>
55 #include <limits>
56 #include <map>
57 #include <string>
58 #include <sstream>
59
60 namespace vkt
61 {
62 namespace SpirVAssembly
63 {
64
65 namespace
66 {
67
68 using namespace vk;
69 using std::map;
70 using std::string;
71 using std::vector;
72 using tcu::IVec3;
73 using tcu::IVec4;
74 using tcu::RGBA;
75 using tcu::TestLog;
76 using tcu::TestStatus;
77 using tcu::Vec4;
78 using de::UniquePtr;
79 using tcu::StringTemplate;
80 using tcu::Vec4;
81
82 typedef Unique<VkShaderModule> ModuleHandleUp;
83 typedef de::SharedPtr<ModuleHandleUp> ModuleHandleSp;
84
85 template<typename T> T randomScalar (de::Random& rnd, T minValue, T maxValue);
randomScalar(de::Random & rnd,float minValue,float maxValue)86 template<> inline float randomScalar (de::Random& rnd, float minValue, float maxValue) { return rnd.getFloat(minValue, maxValue); }
randomScalar(de::Random & rnd,deInt32 minValue,deInt32 maxValue)87 template<> inline deInt32 randomScalar (de::Random& rnd, deInt32 minValue, deInt32 maxValue) { return rnd.getInt(minValue, maxValue); }
88
89 template<typename T>
fillRandomScalars(de::Random & rnd,T minValue,T maxValue,void * dst,int numValues,int offset=0)90 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0)
91 {
92 T* const typedPtr = (T*)dst;
93 for (int ndx = 0; ndx < numValues; ndx++)
94 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue);
95 }
96
floorAll(vector<float> & values)97 static void floorAll (vector<float>& values)
98 {
99 for (size_t i = 0; i < values.size(); i++)
100 values[i] = deFloatFloor(values[i]);
101 }
102
floorAll(vector<Vec4> & values)103 static void floorAll (vector<Vec4>& values)
104 {
105 for (size_t i = 0; i < values.size(); i++)
106 values[i] = floor(values[i]);
107 }
108
109 struct CaseParameter
110 {
111 const char* name;
112 string param;
113
CaseParametervkt::SpirVAssembly::__anon889ef7250111::CaseParameter114 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {}
115 };
116
117 // Assembly code used for testing OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code:
118 //
119 // #version 430
120 //
121 // layout(std140, set = 0, binding = 0) readonly buffer Input {
122 // float elements[];
123 // } input_data;
124 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
125 // float elements[];
126 // } output_data;
127 //
128 // layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
129 //
130 // void main() {
131 // uint x = gl_GlobalInvocationID.x;
132 // output_data.elements[x] = -input_data.elements[x];
133 // }
134
135 static const char* const s_ShaderPreamble =
136 "OpCapability Shader\n"
137 "OpMemoryModel Logical GLSL450\n"
138 "OpEntryPoint GLCompute %main \"main\" %id\n"
139 "OpExecutionMode %main LocalSize 1 1 1\n";
140
141 static const char* const s_CommonTypes =
142 "%bool = OpTypeBool\n"
143 "%void = OpTypeVoid\n"
144 "%voidf = OpTypeFunction %void\n"
145 "%u32 = OpTypeInt 32 0\n"
146 "%i32 = OpTypeInt 32 1\n"
147 "%f32 = OpTypeFloat 32\n"
148 "%uvec3 = OpTypeVector %u32 3\n"
149 "%fvec3 = OpTypeVector %f32 3\n"
150 "%uvec3ptr = OpTypePointer Input %uvec3\n"
151 "%f32ptr = OpTypePointer Uniform %f32\n"
152 "%f32arr = OpTypeRuntimeArray %f32\n";
153
154 // Declares two uniform variables (indata, outdata) of type "struct { float[] }". Depends on type "f32arr" (for "float[]").
155 static const char* const s_InputOutputBuffer =
156 "%buf = OpTypeStruct %f32arr\n"
157 "%bufptr = OpTypePointer Uniform %buf\n"
158 "%indata = OpVariable %bufptr Uniform\n"
159 "%outdata = OpVariable %bufptr Uniform\n";
160
161 // Declares buffer type and layout for uniform variables indata and outdata. Both of them are SSBO bounded to descriptor set 0.
162 // indata is at binding point 0, while outdata is at 1.
163 static const char* const s_InputOutputBufferTraits =
164 "OpDecorate %buf BufferBlock\n"
165 "OpDecorate %indata DescriptorSet 0\n"
166 "OpDecorate %indata Binding 0\n"
167 "OpDecorate %outdata DescriptorSet 0\n"
168 "OpDecorate %outdata Binding 1\n"
169 "OpDecorate %f32arr ArrayStride 4\n"
170 "OpMemberDecorate %buf 0 Offset 0\n";
171
createOpNopGroup(tcu::TestContext & testCtx)172 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
173 {
174 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
175 ComputeShaderSpec spec;
176 de::Random rnd (deStringHash(group->getName()));
177 const int numElements = 100;
178 vector<float> positiveFloats (numElements, 0);
179 vector<float> negativeFloats (numElements, 0);
180
181 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
182
183 for (size_t ndx = 0; ndx < numElements; ++ndx)
184 negativeFloats[ndx] = -positiveFloats[ndx];
185
186 spec.assembly =
187 string(s_ShaderPreamble) +
188
189 "OpSource GLSL 430\n"
190 "OpName %main \"main\"\n"
191 "OpName %id \"gl_GlobalInvocationID\"\n"
192
193 "OpDecorate %id BuiltIn GlobalInvocationId\n"
194
195 + string(s_InputOutputBufferTraits) + string(s_CommonTypes)
196
197 + string(s_InputOutputBuffer) +
198
199 "%id = OpVariable %uvec3ptr Input\n"
200 "%zero = OpConstant %i32 0\n"
201
202 "%main = OpFunction %void None %voidf\n"
203 "%label = OpLabel\n"
204 "%idval = OpLoad %uvec3 %id\n"
205 "%x = OpCompositeExtract %u32 %idval 0\n"
206
207 " OpNop\n" // Inside a function body
208
209 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
210 "%inval = OpLoad %f32 %inloc\n"
211 "%neg = OpFNegate %f32 %inval\n"
212 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
213 " OpStore %outloc %neg\n"
214 " OpReturn\n"
215 " OpFunctionEnd\n";
216 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
217 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
218 spec.numWorkGroups = IVec3(numElements, 1, 1);
219
220 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
221
222 return group.release();
223 }
224
createOpLineGroup(tcu::TestContext & testCtx)225 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
226 {
227 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
228 ComputeShaderSpec spec;
229 de::Random rnd (deStringHash(group->getName()));
230 const int numElements = 100;
231 vector<float> positiveFloats (numElements, 0);
232 vector<float> negativeFloats (numElements, 0);
233
234 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
235
236 for (size_t ndx = 0; ndx < numElements; ++ndx)
237 negativeFloats[ndx] = -positiveFloats[ndx];
238
239 spec.assembly =
240 string(s_ShaderPreamble) +
241
242 "%fname1 = OpString \"negateInputs.comp\"\n"
243 "%fname2 = OpString \"negateInputs\"\n"
244
245 "OpSource GLSL 430\n"
246 "OpName %main \"main\"\n"
247 "OpName %id \"gl_GlobalInvocationID\"\n"
248
249 "OpDecorate %id BuiltIn GlobalInvocationId\n"
250
251 + string(s_InputOutputBufferTraits) +
252
253 "OpLine %fname1 0 0\n" // At the earliest possible position
254
255 + string(s_CommonTypes) + string(s_InputOutputBuffer) +
256
257 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
258 "OpLine %fname2 1 0\n" // Different filenames
259 "OpLine %fname1 1000 100000\n"
260
261 "%id = OpVariable %uvec3ptr Input\n"
262 "%zero = OpConstant %i32 0\n"
263
264 "OpLine %fname1 1 1\n" // Before a function
265
266 "%main = OpFunction %void None %voidf\n"
267 "%label = OpLabel\n"
268
269 "OpLine %fname1 1 1\n" // In a function
270
271 "%idval = OpLoad %uvec3 %id\n"
272 "%x = OpCompositeExtract %u32 %idval 0\n"
273 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
274 "%inval = OpLoad %f32 %inloc\n"
275 "%neg = OpFNegate %f32 %inval\n"
276 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
277 " OpStore %outloc %neg\n"
278 " OpReturn\n"
279 " OpFunctionEnd\n";
280 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
281 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
282 spec.numWorkGroups = IVec3(numElements, 1, 1);
283
284 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
285
286 return group.release();
287 }
288
createOpNoLineGroup(tcu::TestContext & testCtx)289 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
290 {
291 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
292 ComputeShaderSpec spec;
293 de::Random rnd (deStringHash(group->getName()));
294 const int numElements = 100;
295 vector<float> positiveFloats (numElements, 0);
296 vector<float> negativeFloats (numElements, 0);
297
298 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
299
300 for (size_t ndx = 0; ndx < numElements; ++ndx)
301 negativeFloats[ndx] = -positiveFloats[ndx];
302
303 spec.assembly =
304 string(s_ShaderPreamble) +
305
306 "%fname = OpString \"negateInputs.comp\"\n"
307
308 "OpSource GLSL 430\n"
309 "OpName %main \"main\"\n"
310 "OpName %id \"gl_GlobalInvocationID\"\n"
311
312 "OpDecorate %id BuiltIn GlobalInvocationId\n"
313
314 + string(s_InputOutputBufferTraits) +
315
316 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
317
318 + string(s_CommonTypes) + string(s_InputOutputBuffer) +
319
320 "OpLine %fname 0 1\n"
321 "OpNoLine\n" // Immediately following a preceding OpLine
322
323 "OpLine %fname 1000 1\n"
324
325 "%id = OpVariable %uvec3ptr Input\n"
326 "%zero = OpConstant %i32 0\n"
327
328 "OpNoLine\n" // Contents after the previous OpLine
329
330 "%main = OpFunction %void None %voidf\n"
331 "%label = OpLabel\n"
332 "%idval = OpLoad %uvec3 %id\n"
333 "%x = OpCompositeExtract %u32 %idval 0\n"
334
335 "OpNoLine\n" // Multiple OpNoLine
336 "OpNoLine\n"
337 "OpNoLine\n"
338
339 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
340 "%inval = OpLoad %f32 %inloc\n"
341 "%neg = OpFNegate %f32 %inval\n"
342 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
343 " OpStore %outloc %neg\n"
344 " OpReturn\n"
345 " OpFunctionEnd\n";
346 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
347 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
348 spec.numWorkGroups = IVec3(numElements, 1, 1);
349
350 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
351
352 return group.release();
353 }
354
355 // Compare instruction for the contraction compute case.
356 // Returns true if the output is what is expected from the test case.
compareNoContractCase(const std::vector<BufferSp> &,const vector<AllocationSp> & outputAllocs,const std::vector<BufferSp> & expectedOutputs)357 bool compareNoContractCase(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs)
358 {
359 if (outputAllocs.size() != 1)
360 return false;
361
362 // We really just need this for size because we are not comparing the exact values.
363 const BufferSp& expectedOutput = expectedOutputs[0];
364 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
365
366 for(size_t i = 0; i < expectedOutput->getNumBytes() / sizeof(float); ++i) {
367 if (outputAsFloat[i] != 0.f &&
368 outputAsFloat[i] != -ldexp(1, -24)) {
369 return false;
370 }
371 }
372
373 return true;
374 }
375
createNoContractionGroup(tcu::TestContext & testCtx)376 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
377 {
378 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
379 vector<CaseParameter> cases;
380 const int numElements = 100;
381 vector<float> inputFloats1 (numElements, 0);
382 vector<float> inputFloats2 (numElements, 0);
383 vector<float> outputFloats (numElements, 0);
384 const StringTemplate shaderTemplate (
385 string(s_ShaderPreamble) +
386
387 "OpName %main \"main\"\n"
388 "OpName %id \"gl_GlobalInvocationID\"\n"
389
390 "OpDecorate %id BuiltIn GlobalInvocationId\n"
391
392 "${DECORATION}\n"
393
394 "OpDecorate %buf BufferBlock\n"
395 "OpDecorate %indata1 DescriptorSet 0\n"
396 "OpDecorate %indata1 Binding 0\n"
397 "OpDecorate %indata2 DescriptorSet 0\n"
398 "OpDecorate %indata2 Binding 1\n"
399 "OpDecorate %outdata DescriptorSet 0\n"
400 "OpDecorate %outdata Binding 2\n"
401 "OpDecorate %f32arr ArrayStride 4\n"
402 "OpMemberDecorate %buf 0 Offset 0\n"
403
404 + string(s_CommonTypes) +
405
406 "%buf = OpTypeStruct %f32arr\n"
407 "%bufptr = OpTypePointer Uniform %buf\n"
408 "%indata1 = OpVariable %bufptr Uniform\n"
409 "%indata2 = OpVariable %bufptr Uniform\n"
410 "%outdata = OpVariable %bufptr Uniform\n"
411
412 "%id = OpVariable %uvec3ptr Input\n"
413 "%zero = OpConstant %i32 0\n"
414 "%c_f_m1 = OpConstant %f32 -1.\n"
415
416 "%main = OpFunction %void None %voidf\n"
417 "%label = OpLabel\n"
418 "%idval = OpLoad %uvec3 %id\n"
419 "%x = OpCompositeExtract %u32 %idval 0\n"
420 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
421 "%inval1 = OpLoad %f32 %inloc1\n"
422 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
423 "%inval2 = OpLoad %f32 %inloc2\n"
424 "%mul = OpFMul %f32 %inval1 %inval2\n"
425 "%add = OpFAdd %f32 %mul %c_f_m1\n"
426 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
427 " OpStore %outloc %add\n"
428 " OpReturn\n"
429 " OpFunctionEnd\n");
430
431 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
432 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
433 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
434
435 for (size_t ndx = 0; ndx < numElements; ++ndx)
436 {
437 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
438 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
439 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
440 // conducted separately and the result is rounded to 1, or 0x1.fffffcp-1
441 // So the final result will be 0.f or 0x1p-24.
442 // If the operation is combined into a precise fused multiply-add, then the result would be
443 // 2^-46 (0xa8800000).
444 outputFloats[ndx] = 0.f;
445 }
446
447 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
448 {
449 map<string, string> specializations;
450 ComputeShaderSpec spec;
451
452 specializations["DECORATION"] = cases[caseNdx].param;
453 spec.assembly = shaderTemplate.specialize(specializations);
454 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
455 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
456 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
457 spec.numWorkGroups = IVec3(numElements, 1, 1);
458 // Check against the two possible answers based on rounding mode.
459 spec.verifyIO = &compareNoContractCase;
460
461 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
462 }
463 return group.release();
464 }
465
compareFRem(const std::vector<BufferSp> &,const vector<AllocationSp> & outputAllocs,const std::vector<BufferSp> & expectedOutputs)466 bool compareFRem(const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs)
467 {
468 if (outputAllocs.size() != 1)
469 return false;
470
471 const BufferSp& expectedOutput = expectedOutputs[0];
472 const float *expectedOutputAsFloat = static_cast<const float*>(expectedOutput->data());
473 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
474
475 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
476 {
477 const float f0 = expectedOutputAsFloat[idx];
478 const float f1 = outputAsFloat[idx];
479 // \todo relative error needs to be fairly high because FRem may be implemented as
480 // (roughly) frac(a/b)*b, so LSB errors can be magnified. But this should be fine for now.
481 if (deFloatAbs((f1 - f0) / f0) > 0.02)
482 return false;
483 }
484
485 return true;
486 }
487
createOpFRemGroup(tcu::TestContext & testCtx)488 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
489 {
490 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
491 ComputeShaderSpec spec;
492 de::Random rnd (deStringHash(group->getName()));
493 const int numElements = 200;
494 vector<float> inputFloats1 (numElements, 0);
495 vector<float> inputFloats2 (numElements, 0);
496 vector<float> outputFloats (numElements, 0);
497
498 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
499 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
500
501 for (size_t ndx = 0; ndx < numElements; ++ndx)
502 {
503 // Guard against divisors near zero.
504 if (std::fabs(inputFloats2[ndx]) < 1e-3)
505 inputFloats2[ndx] = 8.f;
506
507 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
508 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
509 }
510
511 spec.assembly =
512 string(s_ShaderPreamble) +
513
514 "OpName %main \"main\"\n"
515 "OpName %id \"gl_GlobalInvocationID\"\n"
516
517 "OpDecorate %id BuiltIn GlobalInvocationId\n"
518
519 "OpDecorate %buf BufferBlock\n"
520 "OpDecorate %indata1 DescriptorSet 0\n"
521 "OpDecorate %indata1 Binding 0\n"
522 "OpDecorate %indata2 DescriptorSet 0\n"
523 "OpDecorate %indata2 Binding 1\n"
524 "OpDecorate %outdata DescriptorSet 0\n"
525 "OpDecorate %outdata Binding 2\n"
526 "OpDecorate %f32arr ArrayStride 4\n"
527 "OpMemberDecorate %buf 0 Offset 0\n"
528
529 + string(s_CommonTypes) +
530
531 "%buf = OpTypeStruct %f32arr\n"
532 "%bufptr = OpTypePointer Uniform %buf\n"
533 "%indata1 = OpVariable %bufptr Uniform\n"
534 "%indata2 = OpVariable %bufptr Uniform\n"
535 "%outdata = OpVariable %bufptr Uniform\n"
536
537 "%id = OpVariable %uvec3ptr Input\n"
538 "%zero = OpConstant %i32 0\n"
539
540 "%main = OpFunction %void None %voidf\n"
541 "%label = OpLabel\n"
542 "%idval = OpLoad %uvec3 %id\n"
543 "%x = OpCompositeExtract %u32 %idval 0\n"
544 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
545 "%inval1 = OpLoad %f32 %inloc1\n"
546 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
547 "%inval2 = OpLoad %f32 %inloc2\n"
548 "%rem = OpFRem %f32 %inval1 %inval2\n"
549 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
550 " OpStore %outloc %rem\n"
551 " OpReturn\n"
552 " OpFunctionEnd\n";
553
554 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
555 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
556 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
557 spec.numWorkGroups = IVec3(numElements, 1, 1);
558 spec.verifyIO = &compareFRem;
559
560 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
561
562 return group.release();
563 }
564
565 // Copy contents in the input buffer to the output buffer.
createOpCopyMemoryGroup(tcu::TestContext & testCtx)566 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
567 {
568 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
569 de::Random rnd (deStringHash(group->getName()));
570 const int numElements = 100;
571
572 // 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.
573 ComputeShaderSpec spec1;
574 vector<Vec4> inputFloats1 (numElements);
575 vector<Vec4> outputFloats1 (numElements);
576
577 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
578
579 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
580 floorAll(inputFloats1);
581
582 for (size_t ndx = 0; ndx < numElements; ++ndx)
583 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
584
585 spec1.assembly =
586 string(s_ShaderPreamble) +
587
588 "OpName %main \"main\"\n"
589 "OpName %id \"gl_GlobalInvocationID\"\n"
590
591 "OpDecorate %id BuiltIn GlobalInvocationId\n"
592 "OpDecorate %vec4arr ArrayStride 16\n"
593
594 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
595
596 "%vec4 = OpTypeVector %f32 4\n"
597 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
598 "%vec4ptr_f = OpTypePointer Function %vec4\n"
599 "%vec4arr = OpTypeRuntimeArray %vec4\n"
600 "%buf = OpTypeStruct %vec4arr\n"
601 "%bufptr = OpTypePointer Uniform %buf\n"
602 "%indata = OpVariable %bufptr Uniform\n"
603 "%outdata = OpVariable %bufptr Uniform\n"
604
605 "%id = OpVariable %uvec3ptr Input\n"
606 "%zero = OpConstant %i32 0\n"
607 "%c_f_0 = OpConstant %f32 0.\n"
608 "%c_f_0_5 = OpConstant %f32 0.5\n"
609 "%c_f_1_5 = OpConstant %f32 1.5\n"
610 "%c_f_2_5 = OpConstant %f32 2.5\n"
611 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
612
613 "%main = OpFunction %void None %voidf\n"
614 "%label = OpLabel\n"
615 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
616 "%idval = OpLoad %uvec3 %id\n"
617 "%x = OpCompositeExtract %u32 %idval 0\n"
618 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
619 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
620 " OpCopyMemory %v_vec4 %inloc\n"
621 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
622 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
623 " OpStore %outloc %add\n"
624 " OpReturn\n"
625 " OpFunctionEnd\n";
626
627 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
628 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
629 spec1.numWorkGroups = IVec3(numElements, 1, 1);
630
631 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
632
633 // The following case copies a float[100] variable from the input buffer to the output buffer.
634 ComputeShaderSpec spec2;
635 vector<float> inputFloats2 (numElements);
636 vector<float> outputFloats2 (numElements);
637
638 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
639
640 for (size_t ndx = 0; ndx < numElements; ++ndx)
641 outputFloats2[ndx] = inputFloats2[ndx];
642
643 spec2.assembly =
644 string(s_ShaderPreamble) +
645
646 "OpName %main \"main\"\n"
647 "OpName %id \"gl_GlobalInvocationID\"\n"
648
649 "OpDecorate %id BuiltIn GlobalInvocationId\n"
650 "OpDecorate %f32arr100 ArrayStride 4\n"
651
652 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
653
654 "%hundred = OpConstant %u32 100\n"
655 "%f32arr100 = OpTypeArray %f32 %hundred\n"
656 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
657 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
658 "%buf = OpTypeStruct %f32arr100\n"
659 "%bufptr = OpTypePointer Uniform %buf\n"
660 "%indata = OpVariable %bufptr Uniform\n"
661 "%outdata = OpVariable %bufptr Uniform\n"
662
663 "%id = OpVariable %uvec3ptr Input\n"
664 "%zero = OpConstant %i32 0\n"
665
666 "%main = OpFunction %void None %voidf\n"
667 "%label = OpLabel\n"
668 "%var = OpVariable %f32arr100ptr_f Function\n"
669 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
670 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
671 " OpCopyMemory %var %inarr\n"
672 " OpCopyMemory %outarr %var\n"
673 " OpReturn\n"
674 " OpFunctionEnd\n";
675
676 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
677 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
678 spec2.numWorkGroups = IVec3(1, 1, 1);
679
680 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
681
682 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
683 ComputeShaderSpec spec3;
684 vector<float> inputFloats3 (16);
685 vector<float> outputFloats3 (16);
686
687 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
688
689 for (size_t ndx = 0; ndx < 16; ++ndx)
690 outputFloats3[ndx] = inputFloats3[ndx];
691
692 spec3.assembly =
693 string(s_ShaderPreamble) +
694
695 "OpName %main \"main\"\n"
696 "OpName %id \"gl_GlobalInvocationID\"\n"
697
698 "OpDecorate %id BuiltIn GlobalInvocationId\n"
699 "OpMemberDecorate %buf 0 Offset 0\n"
700 "OpMemberDecorate %buf 1 Offset 16\n"
701 "OpMemberDecorate %buf 2 Offset 32\n"
702 "OpMemberDecorate %buf 3 Offset 48\n"
703
704 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
705
706 "%vec4 = OpTypeVector %f32 4\n"
707 "%buf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
708 "%bufptr = OpTypePointer Uniform %buf\n"
709 "%indata = OpVariable %bufptr Uniform\n"
710 "%outdata = OpVariable %bufptr Uniform\n"
711 "%vec4stptr = OpTypePointer Function %buf\n"
712
713 "%id = OpVariable %uvec3ptr Input\n"
714 "%zero = OpConstant %i32 0\n"
715
716 "%main = OpFunction %void None %voidf\n"
717 "%label = OpLabel\n"
718 "%var = OpVariable %vec4stptr Function\n"
719 " OpCopyMemory %var %indata\n"
720 " OpCopyMemory %outdata %var\n"
721 " OpReturn\n"
722 " OpFunctionEnd\n";
723
724 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
725 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
726 spec3.numWorkGroups = IVec3(1, 1, 1);
727
728 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
729
730 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
731 ComputeShaderSpec spec4;
732 vector<float> inputFloats4 (numElements);
733 vector<float> outputFloats4 (numElements);
734
735 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
736
737 for (size_t ndx = 0; ndx < numElements; ++ndx)
738 outputFloats4[ndx] = -inputFloats4[ndx];
739
740 spec4.assembly =
741 string(s_ShaderPreamble) +
742
743 "OpName %main \"main\"\n"
744 "OpName %id \"gl_GlobalInvocationID\"\n"
745
746 "OpDecorate %id BuiltIn GlobalInvocationId\n"
747
748 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
749
750 "%f32ptr_f = OpTypePointer Function %f32\n"
751 "%id = OpVariable %uvec3ptr Input\n"
752 "%zero = OpConstant %i32 0\n"
753
754 "%main = OpFunction %void None %voidf\n"
755 "%label = OpLabel\n"
756 "%var = OpVariable %f32ptr_f Function\n"
757 "%idval = OpLoad %uvec3 %id\n"
758 "%x = OpCompositeExtract %u32 %idval 0\n"
759 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
760 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
761 " OpCopyMemory %var %inloc\n"
762 "%val = OpLoad %f32 %var\n"
763 "%neg = OpFNegate %f32 %val\n"
764 " OpStore %outloc %neg\n"
765 " OpReturn\n"
766 " OpFunctionEnd\n";
767
768 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
769 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
770 spec4.numWorkGroups = IVec3(numElements, 1, 1);
771
772 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
773
774 return group.release();
775 }
776
createOpCopyObjectGroup(tcu::TestContext & testCtx)777 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
778 {
779 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
780 ComputeShaderSpec spec;
781 de::Random rnd (deStringHash(group->getName()));
782 const int numElements = 100;
783 vector<float> inputFloats (numElements, 0);
784 vector<float> outputFloats (numElements, 0);
785
786 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
787
788 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
789 floorAll(inputFloats);
790
791 for (size_t ndx = 0; ndx < numElements; ++ndx)
792 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
793
794 spec.assembly =
795 string(s_ShaderPreamble) +
796
797 "OpName %main \"main\"\n"
798 "OpName %id \"gl_GlobalInvocationID\"\n"
799
800 "OpDecorate %id BuiltIn GlobalInvocationId\n"
801
802 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
803
804 "%fmat = OpTypeMatrix %fvec3 3\n"
805 "%three = OpConstant %u32 3\n"
806 "%farr = OpTypeArray %f32 %three\n"
807 "%fst = OpTypeStruct %f32 %f32\n"
808
809 + string(s_InputOutputBuffer) +
810
811 "%id = OpVariable %uvec3ptr Input\n"
812 "%zero = OpConstant %i32 0\n"
813 "%c_f = OpConstant %f32 1.5\n"
814 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
815 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
816 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
817 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
818
819 "%main = OpFunction %void None %voidf\n"
820 "%label = OpLabel\n"
821 "%c_f_copy = OpCopyObject %f32 %c_f\n"
822 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
823 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
824 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
825 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
826 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
827 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
828 "%farr_elem = OpCompositeExtract %f32 %c_farr_copy 2\n"
829 "%fst_elem = OpCompositeExtract %f32 %c_fst_copy 1\n"
830 // Add up. 1.5 * 5 = 7.5.
831 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
832 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
833 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
834 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
835
836 "%idval = OpLoad %uvec3 %id\n"
837 "%x = OpCompositeExtract %u32 %idval 0\n"
838 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
839 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
840 "%inval = OpLoad %f32 %inloc\n"
841 "%add = OpFAdd %f32 %add4 %inval\n"
842 " OpStore %outloc %add\n"
843 " OpReturn\n"
844 " OpFunctionEnd\n";
845 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
846 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
847 spec.numWorkGroups = IVec3(numElements, 1, 1);
848
849 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
850
851 return group.release();
852 }
853 // Assembly code used for testing OpUnreachable is based on GLSL source code:
854 //
855 // #version 430
856 //
857 // layout(std140, set = 0, binding = 0) readonly buffer Input {
858 // float elements[];
859 // } input_data;
860 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
861 // float elements[];
862 // } output_data;
863 //
864 // void not_called_func() {
865 // // place OpUnreachable here
866 // }
867 //
868 // uint modulo4(uint val) {
869 // switch (val % uint(4)) {
870 // case 0: return 3;
871 // case 1: return 2;
872 // case 2: return 1;
873 // case 3: return 0;
874 // default: return 100; // place OpUnreachable here
875 // }
876 // }
877 //
878 // uint const5() {
879 // return 5;
880 // // place OpUnreachable here
881 // }
882 //
883 // void main() {
884 // uint x = gl_GlobalInvocationID.x;
885 // if (const5() > modulo4(1000)) {
886 // output_data.elements[x] = -input_data.elements[x];
887 // } else {
888 // // place OpUnreachable here
889 // output_data.elements[x] = input_data.elements[x];
890 // }
891 // }
892
createOpUnreachableGroup(tcu::TestContext & testCtx)893 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
894 {
895 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
896 ComputeShaderSpec spec;
897 de::Random rnd (deStringHash(group->getName()));
898 const int numElements = 100;
899 vector<float> positiveFloats (numElements, 0);
900 vector<float> negativeFloats (numElements, 0);
901
902 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
903
904 for (size_t ndx = 0; ndx < numElements; ++ndx)
905 negativeFloats[ndx] = -positiveFloats[ndx];
906
907 spec.assembly =
908 string(s_ShaderPreamble) +
909
910 "OpSource GLSL 430\n"
911 "OpName %main \"main\"\n"
912 "OpName %func_not_called_func \"not_called_func(\"\n"
913 "OpName %func_modulo4 \"modulo4(u1;\"\n"
914 "OpName %func_const5 \"const5(\"\n"
915 "OpName %id \"gl_GlobalInvocationID\"\n"
916
917 "OpDecorate %id BuiltIn GlobalInvocationId\n"
918
919 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
920
921 "%u32ptr = OpTypePointer Function %u32\n"
922 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
923 "%unitf = OpTypeFunction %u32\n"
924
925 "%id = OpVariable %uvec3ptr Input\n"
926 "%zero = OpConstant %u32 0\n"
927 "%one = OpConstant %u32 1\n"
928 "%two = OpConstant %u32 2\n"
929 "%three = OpConstant %u32 3\n"
930 "%four = OpConstant %u32 4\n"
931 "%five = OpConstant %u32 5\n"
932 "%hundred = OpConstant %u32 100\n"
933 "%thousand = OpConstant %u32 1000\n"
934
935 + string(s_InputOutputBuffer) +
936
937 // Main()
938 "%main = OpFunction %void None %voidf\n"
939 "%main_entry = OpLabel\n"
940 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
941 "%idval = OpLoad %uvec3 %id\n"
942 "%x = OpCompositeExtract %u32 %idval 0\n"
943 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
944 "%inval = OpLoad %f32 %inloc\n"
945 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
946 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
947 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
948 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
949 " OpSelectionMerge %if_end None\n"
950 " OpBranchConditional %cmp_gt %if_true %if_false\n"
951 "%if_true = OpLabel\n"
952 "%negate = OpFNegate %f32 %inval\n"
953 " OpStore %outloc %negate\n"
954 " OpBranch %if_end\n"
955 "%if_false = OpLabel\n"
956 " OpUnreachable\n" // Unreachable else branch for if statement
957 "%if_end = OpLabel\n"
958 " OpReturn\n"
959 " OpFunctionEnd\n"
960
961 // not_called_function()
962 "%func_not_called_func = OpFunction %void None %voidf\n"
963 "%not_called_func_entry = OpLabel\n"
964 " OpUnreachable\n" // Unreachable entry block in not called static function
965 " OpFunctionEnd\n"
966
967 // modulo4()
968 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
969 "%valptr = OpFunctionParameter %u32ptr\n"
970 "%modulo4_entry = OpLabel\n"
971 "%val = OpLoad %u32 %valptr\n"
972 "%modulo = OpUMod %u32 %val %four\n"
973 " OpSelectionMerge %switch_merge None\n"
974 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
975 "%case0 = OpLabel\n"
976 " OpReturnValue %three\n"
977 "%case1 = OpLabel\n"
978 " OpReturnValue %two\n"
979 "%case2 = OpLabel\n"
980 " OpReturnValue %one\n"
981 "%case3 = OpLabel\n"
982 " OpReturnValue %zero\n"
983 "%default = OpLabel\n"
984 " OpUnreachable\n" // Unreachable default case for switch statement
985 "%switch_merge = OpLabel\n"
986 " OpUnreachable\n" // Unreachable merge block for switch statement
987 " OpFunctionEnd\n"
988
989 // const5()
990 "%func_const5 = OpFunction %u32 None %unitf\n"
991 "%const5_entry = OpLabel\n"
992 " OpReturnValue %five\n"
993 "%unreachable = OpLabel\n"
994 " OpUnreachable\n" // Unreachable block in function
995 " OpFunctionEnd\n";
996 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
997 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
998 spec.numWorkGroups = IVec3(numElements, 1, 1);
999
1000 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
1001
1002 return group.release();
1003 }
1004
1005 // Assembly code used for testing decoration group is based on GLSL source code:
1006 //
1007 // #version 430
1008 //
1009 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
1010 // float elements[];
1011 // } input_data0;
1012 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
1013 // float elements[];
1014 // } input_data1;
1015 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
1016 // float elements[];
1017 // } input_data2;
1018 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
1019 // float elements[];
1020 // } input_data3;
1021 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
1022 // float elements[];
1023 // } input_data4;
1024 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
1025 // float elements[];
1026 // } output_data;
1027 //
1028 // void main() {
1029 // uint x = gl_GlobalInvocationID.x;
1030 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
1031 // }
createDecorationGroupGroup(tcu::TestContext & testCtx)1032 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
1033 {
1034 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
1035 ComputeShaderSpec spec;
1036 de::Random rnd (deStringHash(group->getName()));
1037 const int numElements = 100;
1038 vector<float> inputFloats0 (numElements, 0);
1039 vector<float> inputFloats1 (numElements, 0);
1040 vector<float> inputFloats2 (numElements, 0);
1041 vector<float> inputFloats3 (numElements, 0);
1042 vector<float> inputFloats4 (numElements, 0);
1043 vector<float> outputFloats (numElements, 0);
1044
1045 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
1046 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
1047 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
1048 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
1049 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
1050
1051 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1052 floorAll(inputFloats0);
1053 floorAll(inputFloats1);
1054 floorAll(inputFloats2);
1055 floorAll(inputFloats3);
1056 floorAll(inputFloats4);
1057
1058 for (size_t ndx = 0; ndx < numElements; ++ndx)
1059 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
1060
1061 spec.assembly =
1062 string(s_ShaderPreamble) +
1063
1064 "OpSource GLSL 430\n"
1065 "OpName %main \"main\"\n"
1066 "OpName %id \"gl_GlobalInvocationID\"\n"
1067
1068 // Not using group decoration on variable.
1069 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1070 // Not using group decoration on type.
1071 "OpDecorate %f32arr ArrayStride 4\n"
1072
1073 "OpDecorate %groups BufferBlock\n"
1074 "OpDecorate %groupm Offset 0\n"
1075 "%groups = OpDecorationGroup\n"
1076 "%groupm = OpDecorationGroup\n"
1077
1078 // Group decoration on multiple structs.
1079 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
1080 // Group decoration on multiple struct members.
1081 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
1082
1083 "OpDecorate %group1 DescriptorSet 0\n"
1084 "OpDecorate %group3 DescriptorSet 0\n"
1085 "OpDecorate %group3 NonWritable\n"
1086 "OpDecorate %group3 Restrict\n"
1087 "%group0 = OpDecorationGroup\n"
1088 "%group1 = OpDecorationGroup\n"
1089 "%group3 = OpDecorationGroup\n"
1090
1091 // Applying the same decoration group multiple times.
1092 "OpGroupDecorate %group1 %outdata\n"
1093 "OpGroupDecorate %group1 %outdata\n"
1094 "OpGroupDecorate %group1 %outdata\n"
1095 "OpDecorate %outdata DescriptorSet 0\n"
1096 "OpDecorate %outdata Binding 5\n"
1097 // Applying decoration group containing nothing.
1098 "OpGroupDecorate %group0 %indata0\n"
1099 "OpDecorate %indata0 DescriptorSet 0\n"
1100 "OpDecorate %indata0 Binding 0\n"
1101 // Applying decoration group containing one decoration.
1102 "OpGroupDecorate %group1 %indata1\n"
1103 "OpDecorate %indata1 Binding 1\n"
1104 // Applying decoration group containing multiple decorations.
1105 "OpGroupDecorate %group3 %indata2 %indata3\n"
1106 "OpDecorate %indata2 Binding 2\n"
1107 "OpDecorate %indata3 Binding 3\n"
1108 // Applying multiple decoration groups (with overlapping).
1109 "OpGroupDecorate %group0 %indata4\n"
1110 "OpGroupDecorate %group1 %indata4\n"
1111 "OpGroupDecorate %group3 %indata4\n"
1112 "OpDecorate %indata4 Binding 4\n"
1113
1114 + string(s_CommonTypes) +
1115
1116 "%id = OpVariable %uvec3ptr Input\n"
1117 "%zero = OpConstant %i32 0\n"
1118
1119 "%outbuf = OpTypeStruct %f32arr\n"
1120 "%outbufptr = OpTypePointer Uniform %outbuf\n"
1121 "%outdata = OpVariable %outbufptr Uniform\n"
1122 "%inbuf0 = OpTypeStruct %f32arr\n"
1123 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
1124 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
1125 "%inbuf1 = OpTypeStruct %f32arr\n"
1126 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
1127 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
1128 "%inbuf2 = OpTypeStruct %f32arr\n"
1129 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
1130 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
1131 "%inbuf3 = OpTypeStruct %f32arr\n"
1132 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
1133 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
1134 "%inbuf4 = OpTypeStruct %f32arr\n"
1135 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
1136 "%indata4 = OpVariable %inbufptr Uniform\n"
1137
1138 "%main = OpFunction %void None %voidf\n"
1139 "%label = OpLabel\n"
1140 "%idval = OpLoad %uvec3 %id\n"
1141 "%x = OpCompositeExtract %u32 %idval 0\n"
1142 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
1143 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1144 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1145 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1146 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
1147 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1148 "%inval0 = OpLoad %f32 %inloc0\n"
1149 "%inval1 = OpLoad %f32 %inloc1\n"
1150 "%inval2 = OpLoad %f32 %inloc2\n"
1151 "%inval3 = OpLoad %f32 %inloc3\n"
1152 "%inval4 = OpLoad %f32 %inloc4\n"
1153 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
1154 "%add1 = OpFAdd %f32 %add0 %inval2\n"
1155 "%add2 = OpFAdd %f32 %add1 %inval3\n"
1156 "%add = OpFAdd %f32 %add2 %inval4\n"
1157 " OpStore %outloc %add\n"
1158 " OpReturn\n"
1159 " OpFunctionEnd\n";
1160 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
1161 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1162 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1163 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1164 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
1165 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1166 spec.numWorkGroups = IVec3(numElements, 1, 1);
1167
1168 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
1169
1170 return group.release();
1171 }
1172
1173 struct SpecConstantTwoIntCase
1174 {
1175 const char* caseName;
1176 const char* scDefinition0;
1177 const char* scDefinition1;
1178 const char* scResultType;
1179 const char* scOperation;
1180 deInt32 scActualValue0;
1181 deInt32 scActualValue1;
1182 const char* resultOperation;
1183 vector<deInt32> expectedOutput;
1184
SpecConstantTwoIntCasevkt::SpirVAssembly::__anon889ef7250111::SpecConstantTwoIntCase1185 SpecConstantTwoIntCase (const char* name,
1186 const char* definition0,
1187 const char* definition1,
1188 const char* resultType,
1189 const char* operation,
1190 deInt32 value0,
1191 deInt32 value1,
1192 const char* resultOp,
1193 const vector<deInt32>& output)
1194 : caseName (name)
1195 , scDefinition0 (definition0)
1196 , scDefinition1 (definition1)
1197 , scResultType (resultType)
1198 , scOperation (operation)
1199 , scActualValue0 (value0)
1200 , scActualValue1 (value1)
1201 , resultOperation (resultOp)
1202 , expectedOutput (output) {}
1203 };
1204
createSpecConstantGroup(tcu::TestContext & testCtx)1205 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
1206 {
1207 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
1208 vector<SpecConstantTwoIntCase> cases;
1209 de::Random rnd (deStringHash(group->getName()));
1210 const int numElements = 100;
1211 vector<deInt32> inputInts (numElements, 0);
1212 vector<deInt32> outputInts1 (numElements, 0);
1213 vector<deInt32> outputInts2 (numElements, 0);
1214 vector<deInt32> outputInts3 (numElements, 0);
1215 vector<deInt32> outputInts4 (numElements, 0);
1216 const StringTemplate shaderTemplate (
1217 string(s_ShaderPreamble) +
1218
1219 "OpName %main \"main\"\n"
1220 "OpName %id \"gl_GlobalInvocationID\"\n"
1221
1222 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1223 "OpDecorate %sc_0 SpecId 0\n"
1224 "OpDecorate %sc_1 SpecId 1\n"
1225 "OpDecorate %i32arr ArrayStride 4\n"
1226
1227 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1228
1229 "%i32ptr = OpTypePointer Uniform %i32\n"
1230 "%i32arr = OpTypeRuntimeArray %i32\n"
1231 "%boolptr = OpTypePointer Uniform %bool\n"
1232 "%boolarr = OpTypeRuntimeArray %bool\n"
1233 "%buf = OpTypeStruct %i32arr\n"
1234 "%bufptr = OpTypePointer Uniform %buf\n"
1235 "%indata = OpVariable %bufptr Uniform\n"
1236 "%outdata = OpVariable %bufptr Uniform\n"
1237
1238 "%id = OpVariable %uvec3ptr Input\n"
1239 "%zero = OpConstant %i32 0\n"
1240
1241 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
1242 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
1243 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
1244
1245 "%main = OpFunction %void None %voidf\n"
1246 "%label = OpLabel\n"
1247 "%idval = OpLoad %uvec3 %id\n"
1248 "%x = OpCompositeExtract %u32 %idval 0\n"
1249 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
1250 "%inval = OpLoad %i32 %inloc\n"
1251 "%final = ${GEN_RESULT}\n"
1252 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1253 " OpStore %outloc %final\n"
1254 " OpReturn\n"
1255 " OpFunctionEnd\n");
1256
1257 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
1258
1259 for (size_t ndx = 0; ndx < numElements; ++ndx)
1260 {
1261 outputInts1[ndx] = inputInts[ndx] + 42;
1262 outputInts2[ndx] = inputInts[ndx];
1263 outputInts3[ndx] = inputInts[ndx] - 11200;
1264 outputInts4[ndx] = inputInts[ndx] + 1;
1265 }
1266
1267 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
1268 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
1269 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
1270
1271 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
1272 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
1273 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
1274 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
1275 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
1276 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
1277 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
1278 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
1279 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
1280 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
1281 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
1282 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
1283 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
1284 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
1285 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
1286 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
1287 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
1288 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
1289 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
1290 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
1291 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
1292 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
1293 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
1294 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
1295 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
1296 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
1297 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
1298 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
1299 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
1300 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
1301 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
1302 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
1303
1304 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
1305 {
1306 map<string, string> specializations;
1307 ComputeShaderSpec spec;
1308
1309 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
1310 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
1311 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
1312 specializations["SC_OP"] = cases[caseNdx].scOperation;
1313 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
1314
1315 spec.assembly = shaderTemplate.specialize(specializations);
1316 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
1317 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
1318 spec.numWorkGroups = IVec3(numElements, 1, 1);
1319 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
1320 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
1321
1322 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec));
1323 }
1324
1325 ComputeShaderSpec spec;
1326
1327 spec.assembly =
1328 string(s_ShaderPreamble) +
1329
1330 "OpName %main \"main\"\n"
1331 "OpName %id \"gl_GlobalInvocationID\"\n"
1332
1333 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1334 "OpDecorate %sc_0 SpecId 0\n"
1335 "OpDecorate %sc_1 SpecId 1\n"
1336 "OpDecorate %sc_2 SpecId 2\n"
1337 "OpDecorate %i32arr ArrayStride 4\n"
1338
1339 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1340
1341 "%ivec3 = OpTypeVector %i32 3\n"
1342 "%i32ptr = OpTypePointer Uniform %i32\n"
1343 "%i32arr = OpTypeRuntimeArray %i32\n"
1344 "%boolptr = OpTypePointer Uniform %bool\n"
1345 "%boolarr = OpTypeRuntimeArray %bool\n"
1346 "%buf = OpTypeStruct %i32arr\n"
1347 "%bufptr = OpTypePointer Uniform %buf\n"
1348 "%indata = OpVariable %bufptr Uniform\n"
1349 "%outdata = OpVariable %bufptr Uniform\n"
1350
1351 "%id = OpVariable %uvec3ptr Input\n"
1352 "%zero = OpConstant %i32 0\n"
1353 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
1354
1355 "%sc_0 = OpSpecConstant %i32 0\n"
1356 "%sc_1 = OpSpecConstant %i32 0\n"
1357 "%sc_2 = OpSpecConstant %i32 0\n"
1358 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
1359 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
1360 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
1361 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %sc_vec3_1 1 0 4\n" // (0, sc_0, sc_1)
1362 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2 5 1 2\n" // (sc_2, sc_0, sc_1)
1363 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
1364 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
1365 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
1366 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
1367 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
1368
1369 "%main = OpFunction %void None %voidf\n"
1370 "%label = OpLabel\n"
1371 "%idval = OpLoad %uvec3 %id\n"
1372 "%x = OpCompositeExtract %u32 %idval 0\n"
1373 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
1374 "%inval = OpLoad %i32 %inloc\n"
1375 "%final = OpIAdd %i32 %inval %sc_final\n"
1376 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1377 " OpStore %outloc %final\n"
1378 " OpReturn\n"
1379 " OpFunctionEnd\n";
1380 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
1381 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
1382 spec.numWorkGroups = IVec3(numElements, 1, 1);
1383 spec.specConstants.push_back(123);
1384 spec.specConstants.push_back(56);
1385 spec.specConstants.push_back(-77);
1386
1387 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
1388
1389 return group.release();
1390 }
1391
createOpPhiGroup(tcu::TestContext & testCtx)1392 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
1393 {
1394 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
1395 ComputeShaderSpec spec1;
1396 ComputeShaderSpec spec2;
1397 ComputeShaderSpec spec3;
1398 de::Random rnd (deStringHash(group->getName()));
1399 const int numElements = 100;
1400 vector<float> inputFloats (numElements, 0);
1401 vector<float> outputFloats1 (numElements, 0);
1402 vector<float> outputFloats2 (numElements, 0);
1403 vector<float> outputFloats3 (numElements, 0);
1404
1405 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
1406
1407 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1408 floorAll(inputFloats);
1409
1410 for (size_t ndx = 0; ndx < numElements; ++ndx)
1411 {
1412 switch (ndx % 3)
1413 {
1414 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
1415 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
1416 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
1417 default: break;
1418 }
1419 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
1420 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
1421 }
1422
1423 spec1.assembly =
1424 string(s_ShaderPreamble) +
1425
1426 "OpSource GLSL 430\n"
1427 "OpName %main \"main\"\n"
1428 "OpName %id \"gl_GlobalInvocationID\"\n"
1429
1430 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1431
1432 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1433
1434 "%id = OpVariable %uvec3ptr Input\n"
1435 "%zero = OpConstant %i32 0\n"
1436 "%three = OpConstant %u32 3\n"
1437 "%constf5p5 = OpConstant %f32 5.5\n"
1438 "%constf20p5 = OpConstant %f32 20.5\n"
1439 "%constf1p75 = OpConstant %f32 1.75\n"
1440 "%constf8p5 = OpConstant %f32 8.5\n"
1441 "%constf6p5 = OpConstant %f32 6.5\n"
1442
1443 "%main = OpFunction %void None %voidf\n"
1444 "%entry = OpLabel\n"
1445 "%idval = OpLoad %uvec3 %id\n"
1446 "%x = OpCompositeExtract %u32 %idval 0\n"
1447 "%selector = OpUMod %u32 %x %three\n"
1448 " OpSelectionMerge %phi None\n"
1449 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
1450
1451 // Case 1 before OpPhi.
1452 "%case1 = OpLabel\n"
1453 " OpBranch %phi\n"
1454
1455 "%default = OpLabel\n"
1456 " OpUnreachable\n"
1457
1458 "%phi = OpLabel\n"
1459 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
1460 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1461 "%inval = OpLoad %f32 %inloc\n"
1462 "%add = OpFAdd %f32 %inval %operand\n"
1463 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1464 " OpStore %outloc %add\n"
1465 " OpReturn\n"
1466
1467 // Case 0 after OpPhi.
1468 "%case0 = OpLabel\n"
1469 " OpBranch %phi\n"
1470
1471
1472 // Case 2 after OpPhi.
1473 "%case2 = OpLabel\n"
1474 " OpBranch %phi\n"
1475
1476 " OpFunctionEnd\n";
1477 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1478 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
1479 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1480
1481 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
1482
1483 spec2.assembly =
1484 string(s_ShaderPreamble) +
1485
1486 "OpName %main \"main\"\n"
1487 "OpName %id \"gl_GlobalInvocationID\"\n"
1488
1489 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1490
1491 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1492
1493 "%id = OpVariable %uvec3ptr Input\n"
1494 "%zero = OpConstant %i32 0\n"
1495 "%one = OpConstant %i32 1\n"
1496 "%three = OpConstant %i32 3\n"
1497 "%constf6p5 = OpConstant %f32 6.5\n"
1498
1499 "%main = OpFunction %void None %voidf\n"
1500 "%entry = OpLabel\n"
1501 "%idval = OpLoad %uvec3 %id\n"
1502 "%x = OpCompositeExtract %u32 %idval 0\n"
1503 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1504 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1505 "%inval = OpLoad %f32 %inloc\n"
1506 " OpBranch %phi\n"
1507
1508 "%phi = OpLabel\n"
1509 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
1510 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
1511 "%step_next = OpIAdd %i32 %step %one\n"
1512 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
1513 "%still_loop = OpSLessThan %bool %step %three\n"
1514 " OpLoopMerge %exit %phi None\n"
1515 " OpBranchConditional %still_loop %phi %exit\n"
1516
1517 "%exit = OpLabel\n"
1518 " OpStore %outloc %accum\n"
1519 " OpReturn\n"
1520 " OpFunctionEnd\n";
1521 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1522 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1523 spec2.numWorkGroups = IVec3(numElements, 1, 1);
1524
1525 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
1526
1527 spec3.assembly =
1528 string(s_ShaderPreamble) +
1529
1530 "OpName %main \"main\"\n"
1531 "OpName %id \"gl_GlobalInvocationID\"\n"
1532
1533 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1534
1535 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1536
1537 "%f32ptr_f = OpTypePointer Function %f32\n"
1538 "%id = OpVariable %uvec3ptr Input\n"
1539 "%true = OpConstantTrue %bool\n"
1540 "%false = OpConstantFalse %bool\n"
1541 "%zero = OpConstant %i32 0\n"
1542 "%constf8p5 = OpConstant %f32 8.5\n"
1543
1544 "%main = OpFunction %void None %voidf\n"
1545 "%entry = OpLabel\n"
1546 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
1547 "%idval = OpLoad %uvec3 %id\n"
1548 "%x = OpCompositeExtract %u32 %idval 0\n"
1549 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1550 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1551 "%a_init = OpLoad %f32 %inloc\n"
1552 "%b_init = OpLoad %f32 %b\n"
1553 " OpBranch %phi\n"
1554
1555 "%phi = OpLabel\n"
1556 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
1557 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
1558 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
1559 " OpLoopMerge %exit %phi None\n"
1560 " OpBranchConditional %still_loop %phi %exit\n"
1561
1562 "%exit = OpLabel\n"
1563 "%sub = OpFSub %f32 %a_next %b_next\n"
1564 " OpStore %outloc %sub\n"
1565 " OpReturn\n"
1566 " OpFunctionEnd\n";
1567 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1568 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
1569 spec3.numWorkGroups = IVec3(numElements, 1, 1);
1570
1571 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
1572
1573 return group.release();
1574 }
1575
1576 // Assembly code used for testing block order is based on GLSL source code:
1577 //
1578 // #version 430
1579 //
1580 // layout(std140, set = 0, binding = 0) readonly buffer Input {
1581 // float elements[];
1582 // } input_data;
1583 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
1584 // float elements[];
1585 // } output_data;
1586 //
1587 // void main() {
1588 // uint x = gl_GlobalInvocationID.x;
1589 // output_data.elements[x] = input_data.elements[x];
1590 // if (x > uint(50)) {
1591 // switch (x % uint(3)) {
1592 // case 0: output_data.elements[x] += 1.5f; break;
1593 // case 1: output_data.elements[x] += 42.f; break;
1594 // case 2: output_data.elements[x] -= 27.f; break;
1595 // default: break;
1596 // }
1597 // } else {
1598 // output_data.elements[x] = -input_data.elements[x];
1599 // }
1600 // }
createBlockOrderGroup(tcu::TestContext & testCtx)1601 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
1602 {
1603 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
1604 ComputeShaderSpec spec;
1605 de::Random rnd (deStringHash(group->getName()));
1606 const int numElements = 100;
1607 vector<float> inputFloats (numElements, 0);
1608 vector<float> outputFloats (numElements, 0);
1609
1610 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
1611
1612 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
1613 floorAll(inputFloats);
1614
1615 for (size_t ndx = 0; ndx <= 50; ++ndx)
1616 outputFloats[ndx] = -inputFloats[ndx];
1617
1618 for (size_t ndx = 51; ndx < numElements; ++ndx)
1619 {
1620 switch (ndx % 3)
1621 {
1622 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
1623 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
1624 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
1625 default: break;
1626 }
1627 }
1628
1629 spec.assembly =
1630 string(s_ShaderPreamble) +
1631
1632 "OpSource GLSL 430\n"
1633 "OpName %main \"main\"\n"
1634 "OpName %id \"gl_GlobalInvocationID\"\n"
1635
1636 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1637
1638 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1639
1640 "%u32ptr = OpTypePointer Function %u32\n"
1641 "%u32ptr_input = OpTypePointer Input %u32\n"
1642
1643 + string(s_InputOutputBuffer) +
1644
1645 "%id = OpVariable %uvec3ptr Input\n"
1646 "%zero = OpConstant %i32 0\n"
1647 "%const3 = OpConstant %u32 3\n"
1648 "%const50 = OpConstant %u32 50\n"
1649 "%constf1p5 = OpConstant %f32 1.5\n"
1650 "%constf27 = OpConstant %f32 27.0\n"
1651 "%constf42 = OpConstant %f32 42.0\n"
1652
1653 "%main = OpFunction %void None %voidf\n"
1654
1655 // entry block.
1656 "%entry = OpLabel\n"
1657
1658 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
1659 "%xvar = OpVariable %u32ptr Function\n"
1660 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
1661 "%x = OpLoad %u32 %xptr\n"
1662 " OpStore %xvar %x\n"
1663
1664 "%cmp = OpUGreaterThan %bool %x %const50\n"
1665 " OpSelectionMerge %if_merge None\n"
1666 " OpBranchConditional %cmp %if_true %if_false\n"
1667
1668 // Merge block for switch-statement: placed at the beginning.
1669 "%switch_merge = OpLabel\n"
1670 " OpBranch %if_merge\n"
1671
1672 // Case 1 for switch-statement.
1673 "%case1 = OpLabel\n"
1674 "%x_1 = OpLoad %u32 %xvar\n"
1675 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
1676 "%inval_1 = OpLoad %f32 %inloc_1\n"
1677 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
1678 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
1679 " OpStore %outloc_1 %addf42\n"
1680 " OpBranch %switch_merge\n"
1681
1682 // False branch for if-statement: placed in the middle of switch cases and before true branch.
1683 "%if_false = OpLabel\n"
1684 "%x_f = OpLoad %u32 %xvar\n"
1685 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
1686 "%inval_f = OpLoad %f32 %inloc_f\n"
1687 "%negate = OpFNegate %f32 %inval_f\n"
1688 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
1689 " OpStore %outloc_f %negate\n"
1690 " OpBranch %if_merge\n"
1691
1692 // Merge block for if-statement: placed in the middle of true and false branch.
1693 "%if_merge = OpLabel\n"
1694 " OpReturn\n"
1695
1696 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
1697 "%if_true = OpLabel\n"
1698 "%xval_t = OpLoad %u32 %xvar\n"
1699 "%mod = OpUMod %u32 %xval_t %const3\n"
1700 " OpSelectionMerge %switch_merge None\n"
1701 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
1702
1703 // Case 2 for switch-statement.
1704 "%case2 = OpLabel\n"
1705 "%x_2 = OpLoad %u32 %xvar\n"
1706 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
1707 "%inval_2 = OpLoad %f32 %inloc_2\n"
1708 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
1709 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
1710 " OpStore %outloc_2 %subf27\n"
1711 " OpBranch %switch_merge\n"
1712
1713 // Default case for switch-statement: placed in the middle of normal cases.
1714 "%default = OpLabel\n"
1715 " OpBranch %switch_merge\n"
1716
1717 // Case 0 for switch-statement: out of order.
1718 "%case0 = OpLabel\n"
1719 "%x_0 = OpLoad %u32 %xvar\n"
1720 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
1721 "%inval_0 = OpLoad %f32 %inloc_0\n"
1722 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
1723 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
1724 " OpStore %outloc_0 %addf1p5\n"
1725 " OpBranch %switch_merge\n"
1726
1727 " OpFunctionEnd\n";
1728 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1729 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1730 spec.numWorkGroups = IVec3(numElements, 1, 1);
1731
1732 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
1733
1734 return group.release();
1735 }
1736
createMultipleShaderGroup(tcu::TestContext & testCtx)1737 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
1738 {
1739 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
1740 ComputeShaderSpec spec1;
1741 ComputeShaderSpec spec2;
1742 de::Random rnd (deStringHash(group->getName()));
1743 const int numElements = 100;
1744 vector<float> inputFloats (numElements, 0);
1745 vector<float> outputFloats1 (numElements, 0);
1746 vector<float> outputFloats2 (numElements, 0);
1747 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
1748
1749 for (size_t ndx = 0; ndx < numElements; ++ndx)
1750 {
1751 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
1752 outputFloats2[ndx] = -inputFloats[ndx];
1753 }
1754
1755 const string assembly(
1756 "OpCapability Shader\n"
1757 "OpCapability ClipDistance\n"
1758 "OpMemoryModel Logical GLSL450\n"
1759 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
1760 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
1761 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
1762 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
1763 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
1764 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
1765
1766 "OpName %comp_main1 \"entrypoint1\"\n"
1767 "OpName %comp_main2 \"entrypoint2\"\n"
1768 "OpName %vert_main \"entrypoint2\"\n"
1769 "OpName %id \"gl_GlobalInvocationID\"\n"
1770 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
1771 "OpName %vertexIndex \"gl_VertexIndex\"\n"
1772 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
1773 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
1774 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
1775 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
1776
1777 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1778 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
1779 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
1780 "OpDecorate %vert_builtin_st Block\n"
1781 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
1782 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
1783 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
1784
1785 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1786
1787 "%zero = OpConstant %i32 0\n"
1788 "%one = OpConstant %u32 1\n"
1789 "%c_f32_1 = OpConstant %f32 1\n"
1790
1791 "%i32ptr = OpTypePointer Input %i32\n"
1792 "%vec4 = OpTypeVector %f32 4\n"
1793 "%vec4ptr = OpTypePointer Output %vec4\n"
1794 "%f32arr1 = OpTypeArray %f32 %one\n"
1795 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
1796 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
1797 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
1798
1799 "%id = OpVariable %uvec3ptr Input\n"
1800 "%vertexIndex = OpVariable %i32ptr Input\n"
1801 "%instanceIndex = OpVariable %i32ptr Input\n"
1802 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
1803
1804 // gl_Position = vec4(1.);
1805 "%vert_main = OpFunction %void None %voidf\n"
1806 "%vert_entry = OpLabel\n"
1807 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
1808 " OpStore %position %c_vec4_1\n"
1809 " OpReturn\n"
1810 " OpFunctionEnd\n"
1811
1812 // Double inputs.
1813 "%comp_main1 = OpFunction %void None %voidf\n"
1814 "%comp1_entry = OpLabel\n"
1815 "%idval1 = OpLoad %uvec3 %id\n"
1816 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
1817 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
1818 "%inval1 = OpLoad %f32 %inloc1\n"
1819 "%add = OpFAdd %f32 %inval1 %inval1\n"
1820 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
1821 " OpStore %outloc1 %add\n"
1822 " OpReturn\n"
1823 " OpFunctionEnd\n"
1824
1825 // Negate inputs.
1826 "%comp_main2 = OpFunction %void None %voidf\n"
1827 "%comp2_entry = OpLabel\n"
1828 "%idval2 = OpLoad %uvec3 %id\n"
1829 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
1830 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
1831 "%inval2 = OpLoad %f32 %inloc2\n"
1832 "%neg = OpFNegate %f32 %inval2\n"
1833 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
1834 " OpStore %outloc2 %neg\n"
1835 " OpReturn\n"
1836 " OpFunctionEnd\n");
1837
1838 spec1.assembly = assembly;
1839 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1840 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
1841 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1842 spec1.entryPoint = "entrypoint1";
1843
1844 spec2.assembly = assembly;
1845 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1846 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1847 spec2.numWorkGroups = IVec3(numElements, 1, 1);
1848 spec2.entryPoint = "entrypoint2";
1849
1850 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
1851 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
1852
1853 return group.release();
1854 }
1855
makeLongUTF8String(size_t num4ByteChars)1856 inline std::string makeLongUTF8String (size_t num4ByteChars)
1857 {
1858 // An example of a longest valid UTF-8 character. Be explicit about the
1859 // character type because Microsoft compilers can otherwise interpret the
1860 // character string as being over wide (16-bit) characters. Ideally, we
1861 // would just use a C++11 UTF-8 string literal, but we want to support older
1862 // Microsoft compilers.
1863 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
1864 std::string longString;
1865 longString.reserve(num4ByteChars * 4);
1866 for (size_t count = 0; count < num4ByteChars; count++)
1867 {
1868 longString += earthAfrica;
1869 }
1870 return longString;
1871 }
1872
createOpSourceGroup(tcu::TestContext & testCtx)1873 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
1874 {
1875 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
1876 vector<CaseParameter> cases;
1877 de::Random rnd (deStringHash(group->getName()));
1878 const int numElements = 100;
1879 vector<float> positiveFloats (numElements, 0);
1880 vector<float> negativeFloats (numElements, 0);
1881 const StringTemplate shaderTemplate (
1882 "OpCapability Shader\n"
1883 "OpMemoryModel Logical GLSL450\n"
1884
1885 "OpEntryPoint GLCompute %main \"main\" %id\n"
1886 "OpExecutionMode %main LocalSize 1 1 1\n"
1887
1888 "${SOURCE}\n"
1889
1890 "OpName %main \"main\"\n"
1891 "OpName %id \"gl_GlobalInvocationID\"\n"
1892
1893 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1894
1895 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1896
1897 "%id = OpVariable %uvec3ptr Input\n"
1898 "%zero = OpConstant %i32 0\n"
1899
1900 "%main = OpFunction %void None %voidf\n"
1901 "%label = OpLabel\n"
1902 "%idval = OpLoad %uvec3 %id\n"
1903 "%x = OpCompositeExtract %u32 %idval 0\n"
1904 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1905 "%inval = OpLoad %f32 %inloc\n"
1906 "%neg = OpFNegate %f32 %inval\n"
1907 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1908 " OpStore %outloc %neg\n"
1909 " OpReturn\n"
1910 " OpFunctionEnd\n");
1911
1912 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
1913 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
1914 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
1915 "OpSource GLSL 430 %fname"));
1916 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
1917 "OpSource GLSL 430 %fname"));
1918 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
1919 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
1920 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
1921 "OpSource GLSL 430 %fname \"\""));
1922 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
1923 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
1924 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
1925 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
1926 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
1927 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
1928 "OpSourceContinued \"id main() {}\""));
1929 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
1930 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
1931 "OpSourceContinued \"\""));
1932 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
1933 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
1934 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
1935 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
1936 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
1937 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
1938 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
1939 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
1940 "OpSourceContinued \"void\"\n"
1941 "OpSourceContinued \"main()\"\n"
1942 "OpSourceContinued \"{}\""));
1943 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
1944 "OpSource GLSL 430 %fname \"\"\n"
1945 "OpSourceContinued \"#version 430\nvoid main() {}\""));
1946
1947 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
1948
1949 for (size_t ndx = 0; ndx < numElements; ++ndx)
1950 negativeFloats[ndx] = -positiveFloats[ndx];
1951
1952 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
1953 {
1954 map<string, string> specializations;
1955 ComputeShaderSpec spec;
1956
1957 specializations["SOURCE"] = cases[caseNdx].param;
1958 spec.assembly = shaderTemplate.specialize(specializations);
1959 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
1960 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
1961 spec.numWorkGroups = IVec3(numElements, 1, 1);
1962
1963 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
1964 }
1965
1966 return group.release();
1967 }
1968
createOpSourceExtensionGroup(tcu::TestContext & testCtx)1969 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
1970 {
1971 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
1972 vector<CaseParameter> cases;
1973 de::Random rnd (deStringHash(group->getName()));
1974 const int numElements = 100;
1975 vector<float> inputFloats (numElements, 0);
1976 vector<float> outputFloats (numElements, 0);
1977 const StringTemplate shaderTemplate (
1978 string(s_ShaderPreamble) +
1979
1980 "OpSourceExtension \"${EXTENSION}\"\n"
1981
1982 "OpName %main \"main\"\n"
1983 "OpName %id \"gl_GlobalInvocationID\"\n"
1984
1985 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1986
1987 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1988
1989 "%id = OpVariable %uvec3ptr Input\n"
1990 "%zero = OpConstant %i32 0\n"
1991
1992 "%main = OpFunction %void None %voidf\n"
1993 "%label = OpLabel\n"
1994 "%idval = OpLoad %uvec3 %id\n"
1995 "%x = OpCompositeExtract %u32 %idval 0\n"
1996 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1997 "%inval = OpLoad %f32 %inloc\n"
1998 "%neg = OpFNegate %f32 %inval\n"
1999 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2000 " OpStore %outloc %neg\n"
2001 " OpReturn\n"
2002 " OpFunctionEnd\n");
2003
2004 cases.push_back(CaseParameter("empty_extension", ""));
2005 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
2006 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
2007 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
2008 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
2009
2010 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
2011
2012 for (size_t ndx = 0; ndx < numElements; ++ndx)
2013 outputFloats[ndx] = -inputFloats[ndx];
2014
2015 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2016 {
2017 map<string, string> specializations;
2018 ComputeShaderSpec spec;
2019
2020 specializations["EXTENSION"] = cases[caseNdx].param;
2021 spec.assembly = shaderTemplate.specialize(specializations);
2022 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2023 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2024 spec.numWorkGroups = IVec3(numElements, 1, 1);
2025
2026 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2027 }
2028
2029 return group.release();
2030 }
2031
2032 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
createOpConstantNullGroup(tcu::TestContext & testCtx)2033 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
2034 {
2035 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
2036 vector<CaseParameter> cases;
2037 de::Random rnd (deStringHash(group->getName()));
2038 const int numElements = 100;
2039 vector<float> positiveFloats (numElements, 0);
2040 vector<float> negativeFloats (numElements, 0);
2041 const StringTemplate shaderTemplate (
2042 string(s_ShaderPreamble) +
2043
2044 "OpSource GLSL 430\n"
2045 "OpName %main \"main\"\n"
2046 "OpName %id \"gl_GlobalInvocationID\"\n"
2047
2048 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2049
2050 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2051
2052 "${TYPE}\n"
2053 "%null = OpConstantNull %type\n"
2054
2055 "%id = OpVariable %uvec3ptr Input\n"
2056 "%zero = OpConstant %i32 0\n"
2057
2058 "%main = OpFunction %void None %voidf\n"
2059 "%label = OpLabel\n"
2060 "%idval = OpLoad %uvec3 %id\n"
2061 "%x = OpCompositeExtract %u32 %idval 0\n"
2062 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2063 "%inval = OpLoad %f32 %inloc\n"
2064 "%neg = OpFNegate %f32 %inval\n"
2065 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2066 " OpStore %outloc %neg\n"
2067 " OpReturn\n"
2068 " OpFunctionEnd\n");
2069
2070 cases.push_back(CaseParameter("bool", "%type = OpTypeBool"));
2071 cases.push_back(CaseParameter("sint32", "%type = OpTypeInt 32 1"));
2072 cases.push_back(CaseParameter("uint32", "%type = OpTypeInt 32 0"));
2073 cases.push_back(CaseParameter("float32", "%type = OpTypeFloat 32"));
2074 cases.push_back(CaseParameter("vec4float32", "%type = OpTypeVector %f32 4"));
2075 cases.push_back(CaseParameter("vec3bool", "%type = OpTypeVector %bool 3"));
2076 cases.push_back(CaseParameter("vec2uint32", "%type = OpTypeVector %u32 2"));
2077 cases.push_back(CaseParameter("matrix", "%type = OpTypeMatrix %fvec3 3"));
2078 cases.push_back(CaseParameter("array", "%100 = OpConstant %u32 100\n"
2079 "%type = OpTypeArray %i32 %100"));
2080 cases.push_back(CaseParameter("struct", "%type = OpTypeStruct %f32 %i32 %u32"));
2081 cases.push_back(CaseParameter("pointer", "%type = OpTypePointer Function %i32"));
2082
2083 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2084
2085 for (size_t ndx = 0; ndx < numElements; ++ndx)
2086 negativeFloats[ndx] = -positiveFloats[ndx];
2087
2088 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2089 {
2090 map<string, string> specializations;
2091 ComputeShaderSpec spec;
2092
2093 specializations["TYPE"] = cases[caseNdx].param;
2094 spec.assembly = shaderTemplate.specialize(specializations);
2095 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2096 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2097 spec.numWorkGroups = IVec3(numElements, 1, 1);
2098
2099 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2100 }
2101
2102 return group.release();
2103 }
2104
2105 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
createOpConstantCompositeGroup(tcu::TestContext & testCtx)2106 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
2107 {
2108 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
2109 vector<CaseParameter> cases;
2110 de::Random rnd (deStringHash(group->getName()));
2111 const int numElements = 100;
2112 vector<float> positiveFloats (numElements, 0);
2113 vector<float> negativeFloats (numElements, 0);
2114 const StringTemplate shaderTemplate (
2115 string(s_ShaderPreamble) +
2116
2117 "OpSource GLSL 430\n"
2118 "OpName %main \"main\"\n"
2119 "OpName %id \"gl_GlobalInvocationID\"\n"
2120
2121 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2122
2123 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2124
2125 "%id = OpVariable %uvec3ptr Input\n"
2126 "%zero = OpConstant %i32 0\n"
2127
2128 "${CONSTANT}\n"
2129
2130 "%main = OpFunction %void None %voidf\n"
2131 "%label = OpLabel\n"
2132 "%idval = OpLoad %uvec3 %id\n"
2133 "%x = OpCompositeExtract %u32 %idval 0\n"
2134 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2135 "%inval = OpLoad %f32 %inloc\n"
2136 "%neg = OpFNegate %f32 %inval\n"
2137 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2138 " OpStore %outloc %neg\n"
2139 " OpReturn\n"
2140 " OpFunctionEnd\n");
2141
2142 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
2143 "%const = OpConstantComposite %uvec3 %five %zero %five"));
2144 cases.push_back(CaseParameter("matrix", "%m3fvec3 = OpTypeMatrix %fvec3 3\n"
2145 "%ten = OpConstant %f32 10.\n"
2146 "%fzero = OpConstant %f32 0.\n"
2147 "%vec = OpConstantComposite %fvec3 %ten %fzero %ten\n"
2148 "%mat = OpConstantComposite %m3fvec3 %vec %vec %vec"));
2149 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
2150 "%struct = OpTypeStruct %i32 %f32 %fvec3 %m2vec3\n"
2151 "%fzero = OpConstant %f32 0.\n"
2152 "%one = OpConstant %f32 1.\n"
2153 "%point5 = OpConstant %f32 0.5\n"
2154 "%vec = OpConstantComposite %fvec3 %one %one %fzero\n"
2155 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
2156 "%const = OpConstantComposite %struct %zero %point5 %vec %mat"));
2157 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
2158 "%st2 = OpTypeStruct %i32 %i32\n"
2159 "%struct = OpTypeStruct %st1 %st2\n"
2160 "%point5 = OpConstant %f32 0.5\n"
2161 "%one = OpConstant %u32 1\n"
2162 "%ten = OpConstant %i32 10\n"
2163 "%st1val = OpConstantComposite %st1 %one %point5\n"
2164 "%st2val = OpConstantComposite %st2 %ten %ten\n"
2165 "%const = OpConstantComposite %struct %st1val %st2val"));
2166
2167 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2168
2169 for (size_t ndx = 0; ndx < numElements; ++ndx)
2170 negativeFloats[ndx] = -positiveFloats[ndx];
2171
2172 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2173 {
2174 map<string, string> specializations;
2175 ComputeShaderSpec spec;
2176
2177 specializations["CONSTANT"] = cases[caseNdx].param;
2178 spec.assembly = shaderTemplate.specialize(specializations);
2179 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2180 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2181 spec.numWorkGroups = IVec3(numElements, 1, 1);
2182
2183 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2184 }
2185
2186 return group.release();
2187 }
2188
2189 // Creates a floating point number with the given exponent, and significand
2190 // bits set. It can only create normalized numbers. Only the least significant
2191 // 24 bits of the significand will be examined. The final bit of the
2192 // significand will also be ignored. This allows alignment to be written
2193 // similarly to C99 hex-floats.
2194 // For example if you wanted to write 0x1.7f34p-12 you would call
2195 // constructNormalizedFloat(-12, 0x7f3400)
constructNormalizedFloat(deInt32 exponent,deUint32 significand)2196 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
2197 {
2198 float f = 1.0f;
2199
2200 for (deInt32 idx = 0; idx < 23; ++idx)
2201 {
2202 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
2203 significand <<= 1;
2204 }
2205
2206 return std::ldexp(f, exponent);
2207 }
2208
2209 // Compare instruction for the OpQuantizeF16 compute exact case.
2210 // Returns true if the output is what is expected from the test case.
compareOpQuantizeF16ComputeExactCase(const std::vector<BufferSp> &,const vector<AllocationSp> & outputAllocs,const std::vector<BufferSp> & expectedOutputs)2211 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs)
2212 {
2213 if (outputAllocs.size() != 1)
2214 return false;
2215
2216 // We really just need this for size because we cannot compare Nans.
2217 const BufferSp& expectedOutput = expectedOutputs[0];
2218 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
2219
2220 if (expectedOutput->getNumBytes() != 4*sizeof(float)) {
2221 return false;
2222 }
2223
2224 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
2225 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
2226 return false;
2227 }
2228 outputAsFloat++;
2229
2230 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
2231 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
2232 return false;
2233 }
2234 outputAsFloat++;
2235
2236 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
2237 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
2238 return false;
2239 }
2240 outputAsFloat++;
2241
2242 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
2243 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
2244 return false;
2245 }
2246
2247 return true;
2248 }
2249
2250 // Checks that every output from a test-case is a float NaN.
compareNan(const std::vector<BufferSp> &,const vector<AllocationSp> & outputAllocs,const std::vector<BufferSp> & expectedOutputs)2251 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs)
2252 {
2253 if (outputAllocs.size() != 1)
2254 return false;
2255
2256 // We really just need this for size because we cannot compare Nans.
2257 const BufferSp& expectedOutput = expectedOutputs[0];
2258 const float* output_as_float = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
2259
2260 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
2261 {
2262 if (!isnan(output_as_float[idx]))
2263 {
2264 return false;
2265 }
2266 }
2267
2268 return true;
2269 }
2270
2271 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
createOpQuantizeToF16Group(tcu::TestContext & testCtx)2272 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
2273 {
2274 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
2275
2276 const std::string shader (
2277 string(s_ShaderPreamble) +
2278
2279 "OpSource GLSL 430\n"
2280 "OpName %main \"main\"\n"
2281 "OpName %id \"gl_GlobalInvocationID\"\n"
2282
2283 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2284
2285 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2286
2287 "%id = OpVariable %uvec3ptr Input\n"
2288 "%zero = OpConstant %i32 0\n"
2289
2290 "%main = OpFunction %void None %voidf\n"
2291 "%label = OpLabel\n"
2292 "%idval = OpLoad %uvec3 %id\n"
2293 "%x = OpCompositeExtract %u32 %idval 0\n"
2294 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2295 "%inval = OpLoad %f32 %inloc\n"
2296 "%quant = OpQuantizeToF16 %f32 %inval\n"
2297 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2298 " OpStore %outloc %quant\n"
2299 " OpReturn\n"
2300 " OpFunctionEnd\n");
2301
2302 {
2303 ComputeShaderSpec spec;
2304 const deUint32 numElements = 100;
2305 vector<float> infinities;
2306 vector<float> results;
2307
2308 infinities.reserve(numElements);
2309 results.reserve(numElements);
2310
2311 for (size_t idx = 0; idx < numElements; ++idx)
2312 {
2313 switch(idx % 4)
2314 {
2315 case 0:
2316 infinities.push_back(std::numeric_limits<float>::infinity());
2317 results.push_back(std::numeric_limits<float>::infinity());
2318 break;
2319 case 1:
2320 infinities.push_back(-std::numeric_limits<float>::infinity());
2321 results.push_back(-std::numeric_limits<float>::infinity());
2322 break;
2323 case 2:
2324 infinities.push_back(std::ldexp(1.0f, 16));
2325 results.push_back(std::numeric_limits<float>::infinity());
2326 break;
2327 case 3:
2328 infinities.push_back(std::ldexp(-1.0f, 32));
2329 results.push_back(-std::numeric_limits<float>::infinity());
2330 break;
2331 }
2332 }
2333
2334 spec.assembly = shader;
2335 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
2336 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
2337 spec.numWorkGroups = IVec3(numElements, 1, 1);
2338
2339 group->addChild(new SpvAsmComputeShaderCase(
2340 testCtx, "infinities", "Check that infinities propagated and created", spec));
2341 }
2342
2343 {
2344 ComputeShaderSpec spec;
2345 vector<float> nans;
2346 const deUint32 numElements = 100;
2347
2348 nans.reserve(numElements);
2349
2350 for (size_t idx = 0; idx < numElements; ++idx)
2351 {
2352 if (idx % 2 == 0)
2353 {
2354 nans.push_back(std::numeric_limits<float>::quiet_NaN());
2355 }
2356 else
2357 {
2358 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
2359 }
2360 }
2361
2362 spec.assembly = shader;
2363 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
2364 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
2365 spec.numWorkGroups = IVec3(numElements, 1, 1);
2366 spec.verifyIO = &compareNan;
2367
2368 group->addChild(new SpvAsmComputeShaderCase(
2369 testCtx, "propagated_nans", "Check that nans are propagated", spec));
2370 }
2371
2372 {
2373 ComputeShaderSpec spec;
2374 vector<float> small;
2375 vector<float> zeros;
2376 const deUint32 numElements = 100;
2377
2378 small.reserve(numElements);
2379 zeros.reserve(numElements);
2380
2381 for (size_t idx = 0; idx < numElements; ++idx)
2382 {
2383 switch(idx % 6)
2384 {
2385 case 0:
2386 small.push_back(0.f);
2387 zeros.push_back(0.f);
2388 break;
2389 case 1:
2390 small.push_back(-0.f);
2391 zeros.push_back(-0.f);
2392 break;
2393 case 2:
2394 small.push_back(std::ldexp(1.0f, -16));
2395 zeros.push_back(0.f);
2396 break;
2397 case 3:
2398 small.push_back(std::ldexp(-1.0f, -32));
2399 zeros.push_back(-0.f);
2400 break;
2401 case 4:
2402 small.push_back(std::ldexp(1.0f, -127));
2403 zeros.push_back(0.f);
2404 break;
2405 case 5:
2406 small.push_back(-std::ldexp(1.0f, -128));
2407 zeros.push_back(-0.f);
2408 break;
2409 }
2410 }
2411
2412 spec.assembly = shader;
2413 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
2414 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
2415 spec.numWorkGroups = IVec3(numElements, 1, 1);
2416
2417 group->addChild(new SpvAsmComputeShaderCase(
2418 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
2419 }
2420
2421 {
2422 ComputeShaderSpec spec;
2423 vector<float> exact;
2424 const deUint32 numElements = 200;
2425
2426 exact.reserve(numElements);
2427
2428 for (size_t idx = 0; idx < numElements; ++idx)
2429 exact.push_back(static_cast<float>(static_cast<int>(idx) - 100));
2430
2431 spec.assembly = shader;
2432 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
2433 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
2434 spec.numWorkGroups = IVec3(numElements, 1, 1);
2435
2436 group->addChild(new SpvAsmComputeShaderCase(
2437 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
2438 }
2439
2440 {
2441 ComputeShaderSpec spec;
2442 vector<float> inputs;
2443 const deUint32 numElements = 4;
2444
2445 inputs.push_back(constructNormalizedFloat(8, 0x300300));
2446 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
2447 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
2448 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
2449
2450 spec.assembly = shader;
2451 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
2452 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2453 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
2454 spec.numWorkGroups = IVec3(numElements, 1, 1);
2455
2456 group->addChild(new SpvAsmComputeShaderCase(
2457 testCtx, "rounded", "Check that are rounded when needed", spec));
2458 }
2459
2460 return group.release();
2461 }
2462
2463 // Performs a bitwise copy of source to the destination type Dest.
2464 template <typename Dest, typename Src>
bitwiseCast(Src source)2465 Dest bitwiseCast(Src source)
2466 {
2467 Dest dest;
2468 DE_STATIC_ASSERT(sizeof(source) == sizeof(dest));
2469 deMemcpy(&dest, &source, sizeof(dest));
2470 return dest;
2471 }
2472
createSpecConstantOpQuantizeToF16Group(tcu::TestContext & testCtx)2473 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
2474 {
2475 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
2476
2477 const std::string shader (
2478 string(s_ShaderPreamble) +
2479
2480 "OpName %main \"main\"\n"
2481 "OpName %id \"gl_GlobalInvocationID\"\n"
2482
2483 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2484
2485 "OpDecorate %sc_0 SpecId 0\n"
2486 "OpDecorate %sc_1 SpecId 1\n"
2487 "OpDecorate %sc_2 SpecId 2\n"
2488 "OpDecorate %sc_3 SpecId 3\n"
2489 "OpDecorate %sc_4 SpecId 4\n"
2490 "OpDecorate %sc_5 SpecId 5\n"
2491
2492 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2493
2494 "%id = OpVariable %uvec3ptr Input\n"
2495 "%zero = OpConstant %i32 0\n"
2496 "%c_u32_6 = OpConstant %u32 6\n"
2497
2498 "%sc_0 = OpSpecConstant %f32 0.\n"
2499 "%sc_1 = OpSpecConstant %f32 0.\n"
2500 "%sc_2 = OpSpecConstant %f32 0.\n"
2501 "%sc_3 = OpSpecConstant %f32 0.\n"
2502 "%sc_4 = OpSpecConstant %f32 0.\n"
2503 "%sc_5 = OpSpecConstant %f32 0.\n"
2504
2505 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
2506 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
2507 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
2508 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
2509 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
2510 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
2511
2512 "%main = OpFunction %void None %voidf\n"
2513 "%label = OpLabel\n"
2514 "%idval = OpLoad %uvec3 %id\n"
2515 "%x = OpCompositeExtract %u32 %idval 0\n"
2516 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2517 "%selector = OpUMod %u32 %x %c_u32_6\n"
2518 " OpSelectionMerge %exit None\n"
2519 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
2520
2521 "%case0 = OpLabel\n"
2522 " OpStore %outloc %sc_0_quant\n"
2523 " OpBranch %exit\n"
2524
2525 "%case1 = OpLabel\n"
2526 " OpStore %outloc %sc_1_quant\n"
2527 " OpBranch %exit\n"
2528
2529 "%case2 = OpLabel\n"
2530 " OpStore %outloc %sc_2_quant\n"
2531 " OpBranch %exit\n"
2532
2533 "%case3 = OpLabel\n"
2534 " OpStore %outloc %sc_3_quant\n"
2535 " OpBranch %exit\n"
2536
2537 "%case4 = OpLabel\n"
2538 " OpStore %outloc %sc_4_quant\n"
2539 " OpBranch %exit\n"
2540
2541 "%case5 = OpLabel\n"
2542 " OpStore %outloc %sc_5_quant\n"
2543 " OpBranch %exit\n"
2544
2545 "%exit = OpLabel\n"
2546 " OpReturn\n"
2547
2548 " OpFunctionEnd\n");
2549
2550 {
2551 ComputeShaderSpec spec;
2552 const deUint8 numCases = 4;
2553 vector<float> inputs (numCases, 0.f);
2554 vector<float> outputs;
2555
2556 spec.assembly = shader;
2557 spec.numWorkGroups = IVec3(numCases, 1, 1);
2558
2559 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
2560 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
2561 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
2562 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
2563
2564 outputs.push_back(std::numeric_limits<float>::infinity());
2565 outputs.push_back(-std::numeric_limits<float>::infinity());
2566 outputs.push_back(std::numeric_limits<float>::infinity());
2567 outputs.push_back(-std::numeric_limits<float>::infinity());
2568
2569 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2570 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2571
2572 group->addChild(new SpvAsmComputeShaderCase(
2573 testCtx, "infinities", "Check that infinities propagated and created", spec));
2574 }
2575
2576 {
2577 ComputeShaderSpec spec;
2578 const deUint8 numCases = 2;
2579 vector<float> inputs (numCases, 0.f);
2580 vector<float> outputs;
2581
2582 spec.assembly = shader;
2583 spec.numWorkGroups = IVec3(numCases, 1, 1);
2584 spec.verifyIO = &compareNan;
2585
2586 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
2587 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
2588
2589 for (deUint8 idx = 0; idx < numCases; ++idx)
2590 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
2591
2592 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2593 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2594
2595 group->addChild(new SpvAsmComputeShaderCase(
2596 testCtx, "propagated_nans", "Check that nans are propagated", spec));
2597 }
2598
2599 {
2600 ComputeShaderSpec spec;
2601 const deUint8 numCases = 6;
2602 vector<float> inputs (numCases, 0.f);
2603 vector<float> outputs;
2604
2605 spec.assembly = shader;
2606 spec.numWorkGroups = IVec3(numCases, 1, 1);
2607
2608 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
2609 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
2610 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
2611 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
2612 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
2613 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
2614
2615 outputs.push_back(0.f);
2616 outputs.push_back(-0.f);
2617 outputs.push_back(0.f);
2618 outputs.push_back(-0.f);
2619 outputs.push_back(0.f);
2620 outputs.push_back(-0.f);
2621
2622 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2623 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2624
2625 group->addChild(new SpvAsmComputeShaderCase(
2626 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
2627 }
2628
2629 {
2630 ComputeShaderSpec spec;
2631 const deUint8 numCases = 6;
2632 vector<float> inputs (numCases, 0.f);
2633 vector<float> outputs;
2634
2635 spec.assembly = shader;
2636 spec.numWorkGroups = IVec3(numCases, 1, 1);
2637
2638 for (deUint8 idx = 0; idx < 6; ++idx)
2639 {
2640 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
2641 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
2642 outputs.push_back(f);
2643 }
2644
2645 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2646 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2647
2648 group->addChild(new SpvAsmComputeShaderCase(
2649 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
2650 }
2651
2652 {
2653 ComputeShaderSpec spec;
2654 const deUint8 numCases = 4;
2655 vector<float> inputs (numCases, 0.f);
2656 vector<float> outputs;
2657
2658 spec.assembly = shader;
2659 spec.numWorkGroups = IVec3(numCases, 1, 1);
2660 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
2661
2662 outputs.push_back(constructNormalizedFloat(8, 0x300300));
2663 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
2664 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
2665 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
2666
2667 for (deUint8 idx = 0; idx < numCases; ++idx)
2668 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
2669
2670 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2671 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2672
2673 group->addChild(new SpvAsmComputeShaderCase(
2674 testCtx, "rounded", "Check that are rounded when needed", spec));
2675 }
2676
2677 return group.release();
2678 }
2679
2680 // Checks that constant null/composite values can be used in computation.
createOpConstantUsageGroup(tcu::TestContext & testCtx)2681 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
2682 {
2683 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
2684 ComputeShaderSpec spec;
2685 de::Random rnd (deStringHash(group->getName()));
2686 const int numElements = 100;
2687 vector<float> positiveFloats (numElements, 0);
2688 vector<float> negativeFloats (numElements, 0);
2689
2690 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2691
2692 for (size_t ndx = 0; ndx < numElements; ++ndx)
2693 negativeFloats[ndx] = -positiveFloats[ndx];
2694
2695 spec.assembly =
2696 "OpCapability Shader\n"
2697 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
2698 "OpMemoryModel Logical GLSL450\n"
2699 "OpEntryPoint GLCompute %main \"main\" %id\n"
2700 "OpExecutionMode %main LocalSize 1 1 1\n"
2701
2702 "OpSource GLSL 430\n"
2703 "OpName %main \"main\"\n"
2704 "OpName %id \"gl_GlobalInvocationID\"\n"
2705
2706 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2707
2708 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
2709
2710 "%fmat = OpTypeMatrix %fvec3 3\n"
2711 "%ten = OpConstant %u32 10\n"
2712 "%f32arr10 = OpTypeArray %f32 %ten\n"
2713 "%fst = OpTypeStruct %f32 %f32\n"
2714
2715 + string(s_InputOutputBuffer) +
2716
2717 "%id = OpVariable %uvec3ptr Input\n"
2718 "%zero = OpConstant %i32 0\n"
2719
2720 // Create a bunch of null values
2721 "%unull = OpConstantNull %u32\n"
2722 "%fnull = OpConstantNull %f32\n"
2723 "%vnull = OpConstantNull %fvec3\n"
2724 "%mnull = OpConstantNull %fmat\n"
2725 "%anull = OpConstantNull %f32arr10\n"
2726 "%snull = OpConstantComposite %fst %fnull %fnull\n"
2727
2728 "%main = OpFunction %void None %voidf\n"
2729 "%label = OpLabel\n"
2730 "%idval = OpLoad %uvec3 %id\n"
2731 "%x = OpCompositeExtract %u32 %idval 0\n"
2732 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2733 "%inval = OpLoad %f32 %inloc\n"
2734 "%neg = OpFNegate %f32 %inval\n"
2735
2736 // Get the abs() of (a certain element of) those null values
2737 "%unull_cov = OpConvertUToF %f32 %unull\n"
2738 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
2739 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
2740 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
2741 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
2742 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
2743 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
2744 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
2745 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
2746 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
2747 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
2748
2749 // Add them all
2750 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
2751 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
2752 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
2753 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
2754 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
2755 "%final = OpFAdd %f32 %add5 %snull_abs\n"
2756
2757 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2758 " OpStore %outloc %final\n" // write to output
2759 " OpReturn\n"
2760 " OpFunctionEnd\n";
2761 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2762 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2763 spec.numWorkGroups = IVec3(numElements, 1, 1);
2764
2765 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
2766
2767 return group.release();
2768 }
2769
2770 // Assembly code used for testing loop control is based on GLSL source code:
2771 // #version 430
2772 //
2773 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2774 // float elements[];
2775 // } input_data;
2776 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2777 // float elements[];
2778 // } output_data;
2779 //
2780 // void main() {
2781 // uint x = gl_GlobalInvocationID.x;
2782 // output_data.elements[x] = input_data.elements[x];
2783 // for (uint i = 0; i < 4; ++i)
2784 // output_data.elements[x] += 1.f;
2785 // }
createLoopControlGroup(tcu::TestContext & testCtx)2786 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
2787 {
2788 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
2789 vector<CaseParameter> cases;
2790 de::Random rnd (deStringHash(group->getName()));
2791 const int numElements = 100;
2792 vector<float> inputFloats (numElements, 0);
2793 vector<float> outputFloats (numElements, 0);
2794 const StringTemplate shaderTemplate (
2795 string(s_ShaderPreamble) +
2796
2797 "OpSource GLSL 430\n"
2798 "OpName %main \"main\"\n"
2799 "OpName %id \"gl_GlobalInvocationID\"\n"
2800
2801 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2802
2803 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2804
2805 "%u32ptr = OpTypePointer Function %u32\n"
2806
2807 "%id = OpVariable %uvec3ptr Input\n"
2808 "%zero = OpConstant %i32 0\n"
2809 "%uzero = OpConstant %u32 0\n"
2810 "%one = OpConstant %i32 1\n"
2811 "%constf1 = OpConstant %f32 1.0\n"
2812 "%four = OpConstant %u32 4\n"
2813
2814 "%main = OpFunction %void None %voidf\n"
2815 "%entry = OpLabel\n"
2816 "%i = OpVariable %u32ptr Function\n"
2817 " OpStore %i %uzero\n"
2818
2819 "%idval = OpLoad %uvec3 %id\n"
2820 "%x = OpCompositeExtract %u32 %idval 0\n"
2821 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2822 "%inval = OpLoad %f32 %inloc\n"
2823 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2824 " OpStore %outloc %inval\n"
2825 " OpBranch %loop_entry\n"
2826
2827 "%loop_entry = OpLabel\n"
2828 "%i_val = OpLoad %u32 %i\n"
2829 "%cmp_lt = OpULessThan %bool %i_val %four\n"
2830 " OpLoopMerge %loop_merge %loop_entry ${CONTROL}\n"
2831 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
2832 "%loop_body = OpLabel\n"
2833 "%outval = OpLoad %f32 %outloc\n"
2834 "%addf1 = OpFAdd %f32 %outval %constf1\n"
2835 " OpStore %outloc %addf1\n"
2836 "%new_i = OpIAdd %u32 %i_val %one\n"
2837 " OpStore %i %new_i\n"
2838 " OpBranch %loop_entry\n"
2839 "%loop_merge = OpLabel\n"
2840 " OpReturn\n"
2841 " OpFunctionEnd\n");
2842
2843 cases.push_back(CaseParameter("none", "None"));
2844 cases.push_back(CaseParameter("unroll", "Unroll"));
2845 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
2846 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
2847
2848 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2849
2850 for (size_t ndx = 0; ndx < numElements; ++ndx)
2851 outputFloats[ndx] = inputFloats[ndx] + 4.f;
2852
2853 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2854 {
2855 map<string, string> specializations;
2856 ComputeShaderSpec spec;
2857
2858 specializations["CONTROL"] = cases[caseNdx].param;
2859 spec.assembly = shaderTemplate.specialize(specializations);
2860 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2861 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2862 spec.numWorkGroups = IVec3(numElements, 1, 1);
2863
2864 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2865 }
2866
2867 return group.release();
2868 }
2869
2870 // Assembly code used for testing selection control is based on GLSL source code:
2871 // #version 430
2872 //
2873 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2874 // float elements[];
2875 // } input_data;
2876 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2877 // float elements[];
2878 // } output_data;
2879 //
2880 // void main() {
2881 // uint x = gl_GlobalInvocationID.x;
2882 // float val = input_data.elements[x];
2883 // if (val > 10.f)
2884 // output_data.elements[x] = val + 1.f;
2885 // else
2886 // output_data.elements[x] = val - 1.f;
2887 // }
createSelectionControlGroup(tcu::TestContext & testCtx)2888 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
2889 {
2890 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
2891 vector<CaseParameter> cases;
2892 de::Random rnd (deStringHash(group->getName()));
2893 const int numElements = 100;
2894 vector<float> inputFloats (numElements, 0);
2895 vector<float> outputFloats (numElements, 0);
2896 const StringTemplate shaderTemplate (
2897 string(s_ShaderPreamble) +
2898
2899 "OpSource GLSL 430\n"
2900 "OpName %main \"main\"\n"
2901 "OpName %id \"gl_GlobalInvocationID\"\n"
2902
2903 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2904
2905 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2906
2907 "%id = OpVariable %uvec3ptr Input\n"
2908 "%zero = OpConstant %i32 0\n"
2909 "%constf1 = OpConstant %f32 1.0\n"
2910 "%constf10 = OpConstant %f32 10.0\n"
2911
2912 "%main = OpFunction %void None %voidf\n"
2913 "%entry = OpLabel\n"
2914 "%idval = OpLoad %uvec3 %id\n"
2915 "%x = OpCompositeExtract %u32 %idval 0\n"
2916 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2917 "%inval = OpLoad %f32 %inloc\n"
2918 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2919 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
2920
2921 " OpSelectionMerge %if_end ${CONTROL}\n"
2922 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2923 "%if_true = OpLabel\n"
2924 "%addf1 = OpFAdd %f32 %inval %constf1\n"
2925 " OpStore %outloc %addf1\n"
2926 " OpBranch %if_end\n"
2927 "%if_false = OpLabel\n"
2928 "%subf1 = OpFSub %f32 %inval %constf1\n"
2929 " OpStore %outloc %subf1\n"
2930 " OpBranch %if_end\n"
2931 "%if_end = OpLabel\n"
2932 " OpReturn\n"
2933 " OpFunctionEnd\n");
2934
2935 cases.push_back(CaseParameter("none", "None"));
2936 cases.push_back(CaseParameter("flatten", "Flatten"));
2937 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
2938 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
2939
2940 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2941
2942 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
2943 floorAll(inputFloats);
2944
2945 for (size_t ndx = 0; ndx < numElements; ++ndx)
2946 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
2947
2948 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2949 {
2950 map<string, string> specializations;
2951 ComputeShaderSpec spec;
2952
2953 specializations["CONTROL"] = cases[caseNdx].param;
2954 spec.assembly = shaderTemplate.specialize(specializations);
2955 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2956 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2957 spec.numWorkGroups = IVec3(numElements, 1, 1);
2958
2959 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2960 }
2961
2962 return group.release();
2963 }
2964
2965 // Assembly code used for testing function control is based on GLSL source code:
2966 //
2967 // #version 430
2968 //
2969 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2970 // float elements[];
2971 // } input_data;
2972 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2973 // float elements[];
2974 // } output_data;
2975 //
2976 // float const10() { return 10.f; }
2977 //
2978 // void main() {
2979 // uint x = gl_GlobalInvocationID.x;
2980 // output_data.elements[x] = input_data.elements[x] + const10();
2981 // }
createFunctionControlGroup(tcu::TestContext & testCtx)2982 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
2983 {
2984 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
2985 vector<CaseParameter> cases;
2986 de::Random rnd (deStringHash(group->getName()));
2987 const int numElements = 100;
2988 vector<float> inputFloats (numElements, 0);
2989 vector<float> outputFloats (numElements, 0);
2990 const StringTemplate shaderTemplate (
2991 string(s_ShaderPreamble) +
2992
2993 "OpSource GLSL 430\n"
2994 "OpName %main \"main\"\n"
2995 "OpName %func_const10 \"const10(\"\n"
2996 "OpName %id \"gl_GlobalInvocationID\"\n"
2997
2998 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2999
3000 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
3001
3002 "%f32f = OpTypeFunction %f32\n"
3003 "%id = OpVariable %uvec3ptr Input\n"
3004 "%zero = OpConstant %i32 0\n"
3005 "%constf10 = OpConstant %f32 10.0\n"
3006
3007 "%main = OpFunction %void None %voidf\n"
3008 "%entry = OpLabel\n"
3009 "%idval = OpLoad %uvec3 %id\n"
3010 "%x = OpCompositeExtract %u32 %idval 0\n"
3011 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3012 "%inval = OpLoad %f32 %inloc\n"
3013 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
3014 "%fadd = OpFAdd %f32 %inval %ret_10\n"
3015 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3016 " OpStore %outloc %fadd\n"
3017 " OpReturn\n"
3018 " OpFunctionEnd\n"
3019
3020 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
3021 "%label = OpLabel\n"
3022 " OpReturnValue %constf10\n"
3023 " OpFunctionEnd\n");
3024
3025 cases.push_back(CaseParameter("none", "None"));
3026 cases.push_back(CaseParameter("inline", "Inline"));
3027 cases.push_back(CaseParameter("dont_inline", "DontInline"));
3028 cases.push_back(CaseParameter("pure", "Pure"));
3029 cases.push_back(CaseParameter("const", "Const"));
3030 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
3031 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
3032 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
3033 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
3034
3035 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3036
3037 // CPU might not use the same rounding mode as the GPU. Use whole numbers to avoid rounding differences.
3038 floorAll(inputFloats);
3039
3040 for (size_t ndx = 0; ndx < numElements; ++ndx)
3041 outputFloats[ndx] = inputFloats[ndx] + 10.f;
3042
3043 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3044 {
3045 map<string, string> specializations;
3046 ComputeShaderSpec spec;
3047
3048 specializations["CONTROL"] = cases[caseNdx].param;
3049 spec.assembly = shaderTemplate.specialize(specializations);
3050 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3051 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3052 spec.numWorkGroups = IVec3(numElements, 1, 1);
3053
3054 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3055 }
3056
3057 return group.release();
3058 }
3059
createMemoryAccessGroup(tcu::TestContext & testCtx)3060 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
3061 {
3062 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
3063 vector<CaseParameter> cases;
3064 de::Random rnd (deStringHash(group->getName()));
3065 const int numElements = 100;
3066 vector<float> inputFloats (numElements, 0);
3067 vector<float> outputFloats (numElements, 0);
3068 const StringTemplate shaderTemplate (
3069 string(s_ShaderPreamble) +
3070
3071 "OpSource GLSL 430\n"
3072 "OpName %main \"main\"\n"
3073 "OpName %id \"gl_GlobalInvocationID\"\n"
3074
3075 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3076
3077 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
3078
3079 "%f32ptr_f = OpTypePointer Function %f32\n"
3080
3081 "%id = OpVariable %uvec3ptr Input\n"
3082 "%zero = OpConstant %i32 0\n"
3083 "%four = OpConstant %i32 4\n"
3084
3085 "%main = OpFunction %void None %voidf\n"
3086 "%label = OpLabel\n"
3087 "%copy = OpVariable %f32ptr_f Function\n"
3088 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
3089 "%x = OpCompositeExtract %u32 %idval 0\n"
3090 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3091 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3092 " OpCopyMemory %copy %inloc ${ACCESS}\n"
3093 "%val1 = OpLoad %f32 %copy\n"
3094 "%val2 = OpLoad %f32 %inloc\n"
3095 "%add = OpFAdd %f32 %val1 %val2\n"
3096 " OpStore %outloc %add ${ACCESS}\n"
3097 " OpReturn\n"
3098 " OpFunctionEnd\n");
3099
3100 cases.push_back(CaseParameter("null", ""));
3101 cases.push_back(CaseParameter("none", "None"));
3102 cases.push_back(CaseParameter("volatile", "Volatile"));
3103 cases.push_back(CaseParameter("aligned", "Aligned 4"));
3104 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
3105 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
3106 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
3107
3108 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3109
3110 for (size_t ndx = 0; ndx < numElements; ++ndx)
3111 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
3112
3113 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3114 {
3115 map<string, string> specializations;
3116 ComputeShaderSpec spec;
3117
3118 specializations["ACCESS"] = cases[caseNdx].param;
3119 spec.assembly = shaderTemplate.specialize(specializations);
3120 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3121 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3122 spec.numWorkGroups = IVec3(numElements, 1, 1);
3123
3124 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3125 }
3126
3127 return group.release();
3128 }
3129
3130 // Checks that we can get undefined values for various types, without exercising a computation with it.
createOpUndefGroup(tcu::TestContext & testCtx)3131 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
3132 {
3133 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
3134 vector<CaseParameter> cases;
3135 de::Random rnd (deStringHash(group->getName()));
3136 const int numElements = 100;
3137 vector<float> positiveFloats (numElements, 0);
3138 vector<float> negativeFloats (numElements, 0);
3139 const StringTemplate shaderTemplate (
3140 string(s_ShaderPreamble) +
3141
3142 "OpSource GLSL 430\n"
3143 "OpName %main \"main\"\n"
3144 "OpName %id \"gl_GlobalInvocationID\"\n"
3145
3146 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3147
3148 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
3149
3150 "${TYPE}\n"
3151
3152 "%id = OpVariable %uvec3ptr Input\n"
3153 "%zero = OpConstant %i32 0\n"
3154
3155 "%main = OpFunction %void None %voidf\n"
3156 "%label = OpLabel\n"
3157
3158 "%undef = OpUndef %type\n"
3159
3160 "%idval = OpLoad %uvec3 %id\n"
3161 "%x = OpCompositeExtract %u32 %idval 0\n"
3162
3163 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3164 "%inval = OpLoad %f32 %inloc\n"
3165 "%neg = OpFNegate %f32 %inval\n"
3166 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3167 " OpStore %outloc %neg\n"
3168 " OpReturn\n"
3169 " OpFunctionEnd\n");
3170
3171 cases.push_back(CaseParameter("bool", "%type = OpTypeBool"));
3172 cases.push_back(CaseParameter("sint32", "%type = OpTypeInt 32 1"));
3173 cases.push_back(CaseParameter("uint32", "%type = OpTypeInt 32 0"));
3174 cases.push_back(CaseParameter("float32", "%type = OpTypeFloat 32"));
3175 cases.push_back(CaseParameter("vec4float32", "%type = OpTypeVector %f32 4"));
3176 cases.push_back(CaseParameter("vec2uint32", "%type = OpTypeVector %u32 2"));
3177 cases.push_back(CaseParameter("matrix", "%type = OpTypeMatrix %fvec3 3"));
3178 cases.push_back(CaseParameter("image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown"));
3179 cases.push_back(CaseParameter("sampler", "%type = OpTypeSampler"));
3180 cases.push_back(CaseParameter("sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n"
3181 "%type = OpTypeSampledImage %img"));
3182 cases.push_back(CaseParameter("array", "%100 = OpConstant %u32 100\n"
3183 "%type = OpTypeArray %i32 %100"));
3184 cases.push_back(CaseParameter("runtimearray", "%type = OpTypeRuntimeArray %f32"));
3185 cases.push_back(CaseParameter("struct", "%type = OpTypeStruct %f32 %i32 %u32"));
3186 cases.push_back(CaseParameter("pointer", "%type = OpTypePointer Function %i32"));
3187
3188 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3189
3190 for (size_t ndx = 0; ndx < numElements; ++ndx)
3191 negativeFloats[ndx] = -positiveFloats[ndx];
3192
3193 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3194 {
3195 map<string, string> specializations;
3196 ComputeShaderSpec spec;
3197
3198 specializations["TYPE"] = cases[caseNdx].param;
3199 spec.assembly = shaderTemplate.specialize(specializations);
3200 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3201 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3202 spec.numWorkGroups = IVec3(numElements, 1, 1);
3203
3204 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3205 }
3206
3207 return group.release();
3208 }
3209 typedef std::pair<std::string, VkShaderStageFlagBits> EntryToStage;
3210 typedef map<string, vector<EntryToStage> > ModuleMap;
3211 typedef map<VkShaderStageFlagBits, vector<deInt32> > StageToSpecConstantMap;
3212
3213 // Context for a specific test instantiation. For example, an instantiation
3214 // may test colors yellow/magenta/cyan/mauve in a tesselation shader
3215 // with an entry point named 'main_to_the_main'
3216 struct InstanceContext
3217 {
3218 // Map of modules to what entry_points we care to use from those modules.
3219 ModuleMap moduleMap;
3220 RGBA inputColors[4];
3221 RGBA outputColors[4];
3222 // Concrete SPIR-V code to test via boilerplate specialization.
3223 map<string, string> testCodeFragments;
3224 StageToSpecConstantMap specConstants;
3225 bool hasTessellation;
3226 VkShaderStageFlagBits requiredStages;
3227
InstanceContextvkt::SpirVAssembly::__anon889ef7250111::InstanceContext3228 InstanceContext (const RGBA (&inputs)[4], const RGBA (&outputs)[4], const map<string, string>& testCodeFragments_, const StageToSpecConstantMap& specConstants_)
3229 : testCodeFragments (testCodeFragments_)
3230 , specConstants (specConstants_)
3231 , hasTessellation (false)
3232 , requiredStages (static_cast<VkShaderStageFlagBits>(0))
3233 {
3234 inputColors[0] = inputs[0];
3235 inputColors[1] = inputs[1];
3236 inputColors[2] = inputs[2];
3237 inputColors[3] = inputs[3];
3238
3239 outputColors[0] = outputs[0];
3240 outputColors[1] = outputs[1];
3241 outputColors[2] = outputs[2];
3242 outputColors[3] = outputs[3];
3243 }
3244
InstanceContextvkt::SpirVAssembly::__anon889ef7250111::InstanceContext3245 InstanceContext (const InstanceContext& other)
3246 : moduleMap (other.moduleMap)
3247 , testCodeFragments (other.testCodeFragments)
3248 , specConstants (other.specConstants)
3249 , hasTessellation (other.hasTessellation)
3250 , requiredStages (other.requiredStages)
3251 {
3252 inputColors[0] = other.inputColors[0];
3253 inputColors[1] = other.inputColors[1];
3254 inputColors[2] = other.inputColors[2];
3255 inputColors[3] = other.inputColors[3];
3256
3257 outputColors[0] = other.outputColors[0];
3258 outputColors[1] = other.outputColors[1];
3259 outputColors[2] = other.outputColors[2];
3260 outputColors[3] = other.outputColors[3];
3261 }
3262 };
3263
3264 // A description of a shader to be used for a single stage of the graphics pipeline.
3265 struct ShaderElement
3266 {
3267 // The module that contains this shader entrypoint.
3268 string moduleName;
3269
3270 // The name of the entrypoint.
3271 string entryName;
3272
3273 // Which shader stage this entry point represents.
3274 VkShaderStageFlagBits stage;
3275
ShaderElementvkt::SpirVAssembly::__anon889ef7250111::ShaderElement3276 ShaderElement (const string& moduleName_, const string& entryPoint_, VkShaderStageFlagBits shaderStage_)
3277 : moduleName(moduleName_)
3278 , entryName(entryPoint_)
3279 , stage(shaderStage_)
3280 {
3281 }
3282 };
3283
getDefaultColors(RGBA (& colors)[4])3284 void getDefaultColors (RGBA (&colors)[4])
3285 {
3286 colors[0] = RGBA::white();
3287 colors[1] = RGBA::red();
3288 colors[2] = RGBA::green();
3289 colors[3] = RGBA::blue();
3290 }
3291
getHalfColorsFullAlpha(RGBA (& colors)[4])3292 void getHalfColorsFullAlpha (RGBA (&colors)[4])
3293 {
3294 colors[0] = RGBA(127, 127, 127, 255);
3295 colors[1] = RGBA(127, 0, 0, 255);
3296 colors[2] = RGBA(0, 127, 0, 255);
3297 colors[3] = RGBA(0, 0, 127, 255);
3298 }
3299
getInvertedDefaultColors(RGBA (& colors)[4])3300 void getInvertedDefaultColors (RGBA (&colors)[4])
3301 {
3302 colors[0] = RGBA(0, 0, 0, 255);
3303 colors[1] = RGBA(0, 255, 255, 255);
3304 colors[2] = RGBA(255, 0, 255, 255);
3305 colors[3] = RGBA(255, 255, 0, 255);
3306 }
3307
3308 // Turns a statically sized array of ShaderElements into an instance-context
3309 // by setting up the mapping of modules to their contained shaders and stages.
3310 // The inputs and expected outputs are given by inputColors and outputColors
3311 template<size_t N>
createInstanceContext(const ShaderElement (& elements)[N],const RGBA (& inputColors)[4],const RGBA (& outputColors)[4],const map<string,string> & testCodeFragments,const StageToSpecConstantMap & specConstants)3312 InstanceContext createInstanceContext (const ShaderElement (&elements)[N], const RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments, const StageToSpecConstantMap& specConstants)
3313 {
3314 InstanceContext ctx (inputColors, outputColors, testCodeFragments, specConstants);
3315 for (size_t i = 0; i < N; ++i)
3316 {
3317 ctx.moduleMap[elements[i].moduleName].push_back(std::make_pair(elements[i].entryName, elements[i].stage));
3318 ctx.requiredStages = static_cast<VkShaderStageFlagBits>(ctx.requiredStages | elements[i].stage);
3319 }
3320 return ctx;
3321 }
3322
3323 template<size_t N>
createInstanceContext(const ShaderElement (& elements)[N],RGBA (& inputColors)[4],const RGBA (& outputColors)[4],const map<string,string> & testCodeFragments)3324 inline InstanceContext createInstanceContext (const ShaderElement (&elements)[N], RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments)
3325 {
3326 return createInstanceContext(elements, inputColors, outputColors, testCodeFragments, StageToSpecConstantMap());
3327 }
3328
3329 // The same as createInstanceContext above, but with default colors.
3330 template<size_t N>
createInstanceContext(const ShaderElement (& elements)[N],const map<string,string> & testCodeFragments)3331 InstanceContext createInstanceContext (const ShaderElement (&elements)[N], const map<string, string>& testCodeFragments)
3332 {
3333 RGBA defaultColors[4];
3334 getDefaultColors(defaultColors);
3335 return createInstanceContext(elements, defaultColors, defaultColors, testCodeFragments);
3336 }
3337
3338 // For the current InstanceContext, constructs the required modules and shader stage create infos.
createPipelineShaderStages(const DeviceInterface & vk,const VkDevice vkDevice,InstanceContext & instance,Context & context,vector<ModuleHandleSp> & modules,vector<VkPipelineShaderStageCreateInfo> & createInfos)3339 void createPipelineShaderStages (const DeviceInterface& vk, const VkDevice vkDevice, InstanceContext& instance, Context& context, vector<ModuleHandleSp>& modules, vector<VkPipelineShaderStageCreateInfo>& createInfos)
3340 {
3341 for (ModuleMap::const_iterator moduleNdx = instance.moduleMap.begin(); moduleNdx != instance.moduleMap.end(); ++moduleNdx)
3342 {
3343 const ModuleHandleSp mod(new Unique<VkShaderModule>(createShaderModule(vk, vkDevice, context.getBinaryCollection().get(moduleNdx->first), 0)));
3344 modules.push_back(ModuleHandleSp(mod));
3345 for (vector<EntryToStage>::const_iterator shaderNdx = moduleNdx->second.begin(); shaderNdx != moduleNdx->second.end(); ++shaderNdx)
3346 {
3347 const EntryToStage& stage = *shaderNdx;
3348 const VkPipelineShaderStageCreateInfo shaderParam =
3349 {
3350 VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, // VkStructureType sType;
3351 DE_NULL, // const void* pNext;
3352 (VkPipelineShaderStageCreateFlags)0,
3353 stage.second, // VkShaderStageFlagBits stage;
3354 **modules.back(), // VkShaderModule module;
3355 stage.first.c_str(), // const char* pName;
3356 (const VkSpecializationInfo*)DE_NULL,
3357 };
3358 createInfos.push_back(shaderParam);
3359 }
3360 }
3361 }
3362
3363 #define SPIRV_ASSEMBLY_TYPES \
3364 "%void = OpTypeVoid\n" \
3365 "%bool = OpTypeBool\n" \
3366 \
3367 "%i32 = OpTypeInt 32 1\n" \
3368 "%u32 = OpTypeInt 32 0\n" \
3369 \
3370 "%f32 = OpTypeFloat 32\n" \
3371 "%v3f32 = OpTypeVector %f32 3\n" \
3372 "%v4f32 = OpTypeVector %f32 4\n" \
3373 "%v4bool = OpTypeVector %bool 4\n" \
3374 \
3375 "%v4f32_function = OpTypeFunction %v4f32 %v4f32\n" \
3376 "%fun = OpTypeFunction %void\n" \
3377 \
3378 "%ip_f32 = OpTypePointer Input %f32\n" \
3379 "%ip_i32 = OpTypePointer Input %i32\n" \
3380 "%ip_v3f32 = OpTypePointer Input %v3f32\n" \
3381 "%ip_v4f32 = OpTypePointer Input %v4f32\n" \
3382 \
3383 "%op_f32 = OpTypePointer Output %f32\n" \
3384 "%op_v4f32 = OpTypePointer Output %v4f32\n" \
3385 \
3386 "%fp_f32 = OpTypePointer Function %f32\n" \
3387 "%fp_i32 = OpTypePointer Function %i32\n" \
3388 "%fp_v4f32 = OpTypePointer Function %v4f32\n"
3389
3390 #define SPIRV_ASSEMBLY_CONSTANTS \
3391 "%c_f32_1 = OpConstant %f32 1.0\n" \
3392 "%c_f32_0 = OpConstant %f32 0.0\n" \
3393 "%c_f32_0_5 = OpConstant %f32 0.5\n" \
3394 "%c_f32_n1 = OpConstant %f32 -1.\n" \
3395 "%c_f32_7 = OpConstant %f32 7.0\n" \
3396 "%c_f32_8 = OpConstant %f32 8.0\n" \
3397 "%c_i32_0 = OpConstant %i32 0\n" \
3398 "%c_i32_1 = OpConstant %i32 1\n" \
3399 "%c_i32_2 = OpConstant %i32 2\n" \
3400 "%c_i32_3 = OpConstant %i32 3\n" \
3401 "%c_i32_4 = OpConstant %i32 4\n" \
3402 "%c_u32_0 = OpConstant %u32 0\n" \
3403 "%c_u32_1 = OpConstant %u32 1\n" \
3404 "%c_u32_2 = OpConstant %u32 2\n" \
3405 "%c_u32_3 = OpConstant %u32 3\n" \
3406 "%c_u32_32 = OpConstant %u32 32\n" \
3407 "%c_u32_4 = OpConstant %u32 4\n" \
3408 "%c_u32_31_bits = OpConstant %u32 0x7FFFFFFF\n" \
3409 "%c_v4f32_1_1_1_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n" \
3410 "%c_v4f32_1_0_0_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_1\n" \
3411 "%c_v4f32_0_5_0_5_0_5_0_5 = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5\n"
3412
3413 #define SPIRV_ASSEMBLY_ARRAYS \
3414 "%a1f32 = OpTypeArray %f32 %c_u32_1\n" \
3415 "%a2f32 = OpTypeArray %f32 %c_u32_2\n" \
3416 "%a3v4f32 = OpTypeArray %v4f32 %c_u32_3\n" \
3417 "%a4f32 = OpTypeArray %f32 %c_u32_4\n" \
3418 "%a32v4f32 = OpTypeArray %v4f32 %c_u32_32\n" \
3419 "%ip_a3v4f32 = OpTypePointer Input %a3v4f32\n" \
3420 "%ip_a32v4f32 = OpTypePointer Input %a32v4f32\n" \
3421 "%op_a2f32 = OpTypePointer Output %a2f32\n" \
3422 "%op_a3v4f32 = OpTypePointer Output %a3v4f32\n" \
3423 "%op_a4f32 = OpTypePointer Output %a4f32\n"
3424
3425 // Creates vertex-shader assembly by specializing a boilerplate StringTemplate
3426 // on fragments, which must (at least) map "testfun" to an OpFunction definition
3427 // for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed
3428 // with "BP_" to avoid collisions with fragments.
3429 //
3430 // It corresponds roughly to this GLSL:
3431 //;
3432 // layout(location = 0) in vec4 position;
3433 // layout(location = 1) in vec4 color;
3434 // layout(location = 1) out highp vec4 vtxColor;
3435 // void main (void) { gl_Position = position; vtxColor = test_func(color); }
makeVertexShaderAssembly(const map<string,string> & fragments)3436 string makeVertexShaderAssembly(const map<string, string>& fragments)
3437 {
3438 // \todo [2015-11-23 awoloszyn] Remove OpName once these have stabalized
3439 static const char vertexShaderBoilerplate[] =
3440 "OpCapability Shader\n"
3441 "OpCapability ClipDistance\n"
3442 "OpCapability CullDistance\n"
3443 "OpMemoryModel Logical GLSL450\n"
3444 "OpEntryPoint Vertex %main \"main\" %BP_stream %BP_position %BP_vtx_color %BP_color %BP_gl_VertexIndex %BP_gl_InstanceIndex\n"
3445 "${debug:opt}\n"
3446 "OpName %main \"main\"\n"
3447 "OpName %BP_gl_PerVertex \"gl_PerVertex\"\n"
3448 "OpMemberName %BP_gl_PerVertex 0 \"gl_Position\"\n"
3449 "OpMemberName %BP_gl_PerVertex 1 \"gl_PointSize\"\n"
3450 "OpMemberName %BP_gl_PerVertex 2 \"gl_ClipDistance\"\n"
3451 "OpMemberName %BP_gl_PerVertex 3 \"gl_CullDistance\"\n"
3452 "OpName %test_code \"testfun(vf4;\"\n"
3453 "OpName %BP_stream \"\"\n"
3454 "OpName %BP_position \"position\"\n"
3455 "OpName %BP_vtx_color \"vtxColor\"\n"
3456 "OpName %BP_color \"color\"\n"
3457 "OpName %BP_gl_VertexIndex \"gl_VertexIndex\"\n"
3458 "OpName %BP_gl_InstanceIndex \"gl_InstanceIndex\"\n"
3459 "OpMemberDecorate %BP_gl_PerVertex 0 BuiltIn Position\n"
3460 "OpMemberDecorate %BP_gl_PerVertex 1 BuiltIn PointSize\n"
3461 "OpMemberDecorate %BP_gl_PerVertex 2 BuiltIn ClipDistance\n"
3462 "OpMemberDecorate %BP_gl_PerVertex 3 BuiltIn CullDistance\n"
3463 "OpDecorate %BP_gl_PerVertex Block\n"
3464 "OpDecorate %BP_position Location 0\n"
3465 "OpDecorate %BP_vtx_color Location 1\n"
3466 "OpDecorate %BP_color Location 1\n"
3467 "OpDecorate %BP_gl_VertexIndex BuiltIn VertexIndex\n"
3468 "OpDecorate %BP_gl_InstanceIndex BuiltIn InstanceIndex\n"
3469 "${decoration:opt}\n"
3470 SPIRV_ASSEMBLY_TYPES
3471 SPIRV_ASSEMBLY_CONSTANTS
3472 SPIRV_ASSEMBLY_ARRAYS
3473 "%BP_gl_PerVertex = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3474 "%BP_op_gl_PerVertex = OpTypePointer Output %BP_gl_PerVertex\n"
3475 "%BP_stream = OpVariable %BP_op_gl_PerVertex Output\n"
3476 "%BP_position = OpVariable %ip_v4f32 Input\n"
3477 "%BP_vtx_color = OpVariable %op_v4f32 Output\n"
3478 "%BP_color = OpVariable %ip_v4f32 Input\n"
3479 "%BP_gl_VertexIndex = OpVariable %ip_i32 Input\n"
3480 "%BP_gl_InstanceIndex = OpVariable %ip_i32 Input\n"
3481 "${pre_main:opt}\n"
3482 "%main = OpFunction %void None %fun\n"
3483 "%BP_label = OpLabel\n"
3484 "%BP_pos = OpLoad %v4f32 %BP_position\n"
3485 "%BP_gl_pos = OpAccessChain %op_v4f32 %BP_stream %c_i32_0\n"
3486 "OpStore %BP_gl_pos %BP_pos\n"
3487 "%BP_col = OpLoad %v4f32 %BP_color\n"
3488 "%BP_col_transformed = OpFunctionCall %v4f32 %test_code %BP_col\n"
3489 "OpStore %BP_vtx_color %BP_col_transformed\n"
3490 "OpReturn\n"
3491 "OpFunctionEnd\n"
3492 "${testfun}\n";
3493 return tcu::StringTemplate(vertexShaderBoilerplate).specialize(fragments);
3494 }
3495
3496 // Creates tess-control-shader assembly by specializing a boilerplate
3497 // StringTemplate on fragments, which must (at least) map "testfun" to an
3498 // OpFunction definition for %test_code that takes and returns a %v4f32.
3499 // Boilerplate IDs are prefixed with "BP_" to avoid collisions with fragments.
3500 //
3501 // It roughly corresponds to the following GLSL.
3502 //
3503 // #version 450
3504 // layout(vertices = 3) out;
3505 // layout(location = 1) in vec4 in_color[];
3506 // layout(location = 1) out vec4 out_color[];
3507 //
3508 // void main() {
3509 // out_color[gl_InvocationID] = testfun(in_color[gl_InvocationID]);
3510 // gl_out[gl_InvocationID].gl_Position = gl_in[gl_InvocationID].gl_Position;
3511 // if (gl_InvocationID == 0) {
3512 // gl_TessLevelOuter[0] = 1.0;
3513 // gl_TessLevelOuter[1] = 1.0;
3514 // gl_TessLevelOuter[2] = 1.0;
3515 // gl_TessLevelInner[0] = 1.0;
3516 // }
3517 // }
makeTessControlShaderAssembly(const map<string,string> & fragments)3518 string makeTessControlShaderAssembly (const map<string, string>& fragments)
3519 {
3520 static const char tessControlShaderBoilerplate[] =
3521 "OpCapability Tessellation\n"
3522 "OpCapability ClipDistance\n"
3523 "OpCapability CullDistance\n"
3524 "OpMemoryModel Logical GLSL450\n"
3525 "OpEntryPoint TessellationControl %BP_main \"main\" %BP_out_color %BP_gl_InvocationID %BP_in_color %BP_gl_out %BP_gl_in %BP_gl_TessLevelOuter %BP_gl_TessLevelInner\n"
3526 "OpExecutionMode %BP_main OutputVertices 3\n"
3527 "${debug:opt}\n"
3528 "OpName %BP_main \"main\"\n"
3529 "OpName %test_code \"testfun(vf4;\"\n"
3530 "OpName %BP_out_color \"out_color\"\n"
3531 "OpName %BP_gl_InvocationID \"gl_InvocationID\"\n"
3532 "OpName %BP_in_color \"in_color\"\n"
3533 "OpName %BP_gl_PerVertex \"gl_PerVertex\"\n"
3534 "OpMemberName %BP_gl_PerVertex 0 \"gl_Position\"\n"
3535 "OpMemberName %BP_gl_PerVertex 1 \"gl_PointSize\"\n"
3536 "OpMemberName %BP_gl_PerVertex 2 \"gl_ClipDistance\"\n"
3537 "OpMemberName %BP_gl_PerVertex 3 \"gl_CullDistance\"\n"
3538 "OpName %BP_gl_out \"gl_out\"\n"
3539 "OpName %BP_gl_PVOut \"gl_PerVertex\"\n"
3540 "OpMemberName %BP_gl_PVOut 0 \"gl_Position\"\n"
3541 "OpMemberName %BP_gl_PVOut 1 \"gl_PointSize\"\n"
3542 "OpMemberName %BP_gl_PVOut 2 \"gl_ClipDistance\"\n"
3543 "OpMemberName %BP_gl_PVOut 3 \"gl_CullDistance\"\n"
3544 "OpName %BP_gl_in \"gl_in\"\n"
3545 "OpName %BP_gl_TessLevelOuter \"gl_TessLevelOuter\"\n"
3546 "OpName %BP_gl_TessLevelInner \"gl_TessLevelInner\"\n"
3547 "OpDecorate %BP_out_color Location 1\n"
3548 "OpDecorate %BP_gl_InvocationID BuiltIn InvocationId\n"
3549 "OpDecorate %BP_in_color Location 1\n"
3550 "OpMemberDecorate %BP_gl_PerVertex 0 BuiltIn Position\n"
3551 "OpMemberDecorate %BP_gl_PerVertex 1 BuiltIn PointSize\n"
3552 "OpMemberDecorate %BP_gl_PerVertex 2 BuiltIn ClipDistance\n"
3553 "OpMemberDecorate %BP_gl_PerVertex 3 BuiltIn CullDistance\n"
3554 "OpDecorate %BP_gl_PerVertex Block\n"
3555 "OpMemberDecorate %BP_gl_PVOut 0 BuiltIn Position\n"
3556 "OpMemberDecorate %BP_gl_PVOut 1 BuiltIn PointSize\n"
3557 "OpMemberDecorate %BP_gl_PVOut 2 BuiltIn ClipDistance\n"
3558 "OpMemberDecorate %BP_gl_PVOut 3 BuiltIn CullDistance\n"
3559 "OpDecorate %BP_gl_PVOut Block\n"
3560 "OpDecorate %BP_gl_TessLevelOuter Patch\n"
3561 "OpDecorate %BP_gl_TessLevelOuter BuiltIn TessLevelOuter\n"
3562 "OpDecorate %BP_gl_TessLevelInner Patch\n"
3563 "OpDecorate %BP_gl_TessLevelInner BuiltIn TessLevelInner\n"
3564 "${decoration:opt}\n"
3565 SPIRV_ASSEMBLY_TYPES
3566 SPIRV_ASSEMBLY_CONSTANTS
3567 SPIRV_ASSEMBLY_ARRAYS
3568 "%BP_out_color = OpVariable %op_a3v4f32 Output\n"
3569 "%BP_gl_InvocationID = OpVariable %ip_i32 Input\n"
3570 "%BP_in_color = OpVariable %ip_a32v4f32 Input\n"
3571 "%BP_gl_PerVertex = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3572 "%BP_a3_gl_PerVertex = OpTypeArray %BP_gl_PerVertex %c_u32_3\n"
3573 "%BP_op_a3_gl_PerVertex = OpTypePointer Output %BP_a3_gl_PerVertex\n"
3574 "%BP_gl_out = OpVariable %BP_op_a3_gl_PerVertex Output\n"
3575 "%BP_gl_PVOut = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3576 "%BP_a32_gl_PVOut = OpTypeArray %BP_gl_PVOut %c_u32_32\n"
3577 "%BP_ip_a32_gl_PVOut = OpTypePointer Input %BP_a32_gl_PVOut\n"
3578 "%BP_gl_in = OpVariable %BP_ip_a32_gl_PVOut Input\n"
3579 "%BP_gl_TessLevelOuter = OpVariable %op_a4f32 Output\n"
3580 "%BP_gl_TessLevelInner = OpVariable %op_a2f32 Output\n"
3581 "${pre_main:opt}\n"
3582
3583 "%BP_main = OpFunction %void None %fun\n"
3584 "%BP_label = OpLabel\n"
3585
3586 "%BP_gl_Invoc = OpLoad %i32 %BP_gl_InvocationID\n"
3587
3588 "%BP_in_col_loc = OpAccessChain %ip_v4f32 %BP_in_color %BP_gl_Invoc\n"
3589 "%BP_out_col_loc = OpAccessChain %op_v4f32 %BP_out_color %BP_gl_Invoc\n"
3590 "%BP_in_col_val = OpLoad %v4f32 %BP_in_col_loc\n"
3591 "%BP_clr_transformed = OpFunctionCall %v4f32 %test_code %BP_in_col_val\n"
3592 "OpStore %BP_out_col_loc %BP_clr_transformed\n"
3593
3594 "%BP_in_pos_loc = OpAccessChain %ip_v4f32 %BP_gl_in %BP_gl_Invoc %c_i32_0\n"
3595 "%BP_out_pos_loc = OpAccessChain %op_v4f32 %BP_gl_out %BP_gl_Invoc %c_i32_0\n"
3596 "%BP_in_pos_val = OpLoad %v4f32 %BP_in_pos_loc\n"
3597 "OpStore %BP_out_pos_loc %BP_in_pos_val\n"
3598
3599 "%BP_cmp = OpIEqual %bool %BP_gl_Invoc %c_i32_0\n"
3600 "OpSelectionMerge %BP_merge_label None\n"
3601 "OpBranchConditional %BP_cmp %BP_if_label %BP_merge_label\n"
3602 "%BP_if_label = OpLabel\n"
3603 "%BP_gl_TessLevelOuterPos_0 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_0\n"
3604 "%BP_gl_TessLevelOuterPos_1 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_1\n"
3605 "%BP_gl_TessLevelOuterPos_2 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_2\n"
3606 "%BP_gl_TessLevelInnerPos_0 = OpAccessChain %op_f32 %BP_gl_TessLevelInner %c_i32_0\n"
3607 "OpStore %BP_gl_TessLevelOuterPos_0 %c_f32_1\n"
3608 "OpStore %BP_gl_TessLevelOuterPos_1 %c_f32_1\n"
3609 "OpStore %BP_gl_TessLevelOuterPos_2 %c_f32_1\n"
3610 "OpStore %BP_gl_TessLevelInnerPos_0 %c_f32_1\n"
3611 "OpBranch %BP_merge_label\n"
3612 "%BP_merge_label = OpLabel\n"
3613 "OpReturn\n"
3614 "OpFunctionEnd\n"
3615 "${testfun}\n";
3616 return tcu::StringTemplate(tessControlShaderBoilerplate).specialize(fragments);
3617 }
3618
3619 // Creates tess-evaluation-shader assembly by specializing a boilerplate
3620 // StringTemplate on fragments, which must (at least) map "testfun" to an
3621 // OpFunction definition for %test_code that takes and returns a %v4f32.
3622 // Boilerplate IDs are prefixed with "BP_" to avoid collisions with fragments.
3623 //
3624 // It roughly corresponds to the following glsl.
3625 //
3626 // #version 450
3627 //
3628 // layout(triangles, equal_spacing, ccw) in;
3629 // layout(location = 1) in vec4 in_color[];
3630 // layout(location = 1) out vec4 out_color;
3631 //
3632 // #define interpolate(val)
3633 // vec4(gl_TessCoord.x) * val[0] + vec4(gl_TessCoord.y) * val[1] +
3634 // vec4(gl_TessCoord.z) * val[2]
3635 //
3636 // void main() {
3637 // gl_Position = vec4(gl_TessCoord.x) * gl_in[0].gl_Position +
3638 // vec4(gl_TessCoord.y) * gl_in[1].gl_Position +
3639 // vec4(gl_TessCoord.z) * gl_in[2].gl_Position;
3640 // out_color = testfun(interpolate(in_color));
3641 // }
makeTessEvalShaderAssembly(const map<string,string> & fragments)3642 string makeTessEvalShaderAssembly(const map<string, string>& fragments)
3643 {
3644 static const char tessEvalBoilerplate[] =
3645 "OpCapability Tessellation\n"
3646 "OpCapability ClipDistance\n"
3647 "OpCapability CullDistance\n"
3648 "OpMemoryModel Logical GLSL450\n"
3649 "OpEntryPoint TessellationEvaluation %BP_main \"main\" %BP_stream %BP_gl_TessCoord %BP_gl_in %BP_out_color %BP_in_color\n"
3650 "OpExecutionMode %BP_main Triangles\n"
3651 "OpExecutionMode %BP_main SpacingEqual\n"
3652 "OpExecutionMode %BP_main VertexOrderCcw\n"
3653 "${debug:opt}\n"
3654 "OpName %BP_main \"main\"\n"
3655 "OpName %test_code \"testfun(vf4;\"\n"
3656 "OpName %BP_gl_PerVertexOut \"gl_PerVertex\"\n"
3657 "OpMemberName %BP_gl_PerVertexOut 0 \"gl_Position\"\n"
3658 "OpMemberName %BP_gl_PerVertexOut 1 \"gl_PointSize\"\n"
3659 "OpMemberName %BP_gl_PerVertexOut 2 \"gl_ClipDistance\"\n"
3660 "OpMemberName %BP_gl_PerVertexOut 3 \"gl_CullDistance\"\n"
3661 "OpName %BP_stream \"\"\n"
3662 "OpName %BP_gl_TessCoord \"gl_TessCoord\"\n"
3663 "OpName %BP_gl_PerVertexIn \"gl_PerVertex\"\n"
3664 "OpMemberName %BP_gl_PerVertexIn 0 \"gl_Position\"\n"
3665 "OpMemberName %BP_gl_PerVertexIn 1 \"gl_PointSize\"\n"
3666 "OpMemberName %BP_gl_PerVertexIn 2 \"gl_ClipDistance\"\n"
3667 "OpMemberName %BP_gl_PerVertexIn 3 \"gl_CullDistance\"\n"
3668 "OpName %BP_gl_in \"gl_in\"\n"
3669 "OpName %BP_out_color \"out_color\"\n"
3670 "OpName %BP_in_color \"in_color\"\n"
3671 "OpMemberDecorate %BP_gl_PerVertexOut 0 BuiltIn Position\n"
3672 "OpMemberDecorate %BP_gl_PerVertexOut 1 BuiltIn PointSize\n"
3673 "OpMemberDecorate %BP_gl_PerVertexOut 2 BuiltIn ClipDistance\n"
3674 "OpMemberDecorate %BP_gl_PerVertexOut 3 BuiltIn CullDistance\n"
3675 "OpDecorate %BP_gl_PerVertexOut Block\n"
3676 "OpDecorate %BP_gl_TessCoord BuiltIn TessCoord\n"
3677 "OpMemberDecorate %BP_gl_PerVertexIn 0 BuiltIn Position\n"
3678 "OpMemberDecorate %BP_gl_PerVertexIn 1 BuiltIn PointSize\n"
3679 "OpMemberDecorate %BP_gl_PerVertexIn 2 BuiltIn ClipDistance\n"
3680 "OpMemberDecorate %BP_gl_PerVertexIn 3 BuiltIn CullDistance\n"
3681 "OpDecorate %BP_gl_PerVertexIn Block\n"
3682 "OpDecorate %BP_out_color Location 1\n"
3683 "OpDecorate %BP_in_color Location 1\n"
3684 "${decoration:opt}\n"
3685 SPIRV_ASSEMBLY_TYPES
3686 SPIRV_ASSEMBLY_CONSTANTS
3687 SPIRV_ASSEMBLY_ARRAYS
3688 "%BP_gl_PerVertexOut = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3689 "%BP_op_gl_PerVertexOut = OpTypePointer Output %BP_gl_PerVertexOut\n"
3690 "%BP_stream = OpVariable %BP_op_gl_PerVertexOut Output\n"
3691 "%BP_gl_TessCoord = OpVariable %ip_v3f32 Input\n"
3692 "%BP_gl_PerVertexIn = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3693 "%BP_a32_gl_PerVertexIn = OpTypeArray %BP_gl_PerVertexIn %c_u32_32\n"
3694 "%BP_ip_a32_gl_PerVertexIn = OpTypePointer Input %BP_a32_gl_PerVertexIn\n"
3695 "%BP_gl_in = OpVariable %BP_ip_a32_gl_PerVertexIn Input\n"
3696 "%BP_out_color = OpVariable %op_v4f32 Output\n"
3697 "%BP_in_color = OpVariable %ip_a32v4f32 Input\n"
3698 "${pre_main:opt}\n"
3699 "%BP_main = OpFunction %void None %fun\n"
3700 "%BP_label = OpLabel\n"
3701 "%BP_gl_TC_0 = OpAccessChain %ip_f32 %BP_gl_TessCoord %c_u32_0\n"
3702 "%BP_gl_TC_1 = OpAccessChain %ip_f32 %BP_gl_TessCoord %c_u32_1\n"
3703 "%BP_gl_TC_2 = OpAccessChain %ip_f32 %BP_gl_TessCoord %c_u32_2\n"
3704 "%BP_gl_in_gl_Pos_0 = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_0 %c_i32_0\n"
3705 "%BP_gl_in_gl_Pos_1 = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_1 %c_i32_0\n"
3706 "%BP_gl_in_gl_Pos_2 = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_2 %c_i32_0\n"
3707
3708 "%BP_gl_OPos = OpAccessChain %op_v4f32 %BP_stream %c_i32_0\n"
3709 "%BP_in_color_0 = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_0\n"
3710 "%BP_in_color_1 = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_1\n"
3711 "%BP_in_color_2 = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_2\n"
3712
3713 "%BP_TC_W_0 = OpLoad %f32 %BP_gl_TC_0\n"
3714 "%BP_TC_W_1 = OpLoad %f32 %BP_gl_TC_1\n"
3715 "%BP_TC_W_2 = OpLoad %f32 %BP_gl_TC_2\n"
3716 "%BP_v4f32_TC_0 = OpCompositeConstruct %v4f32 %BP_TC_W_0 %BP_TC_W_0 %BP_TC_W_0 %BP_TC_W_0\n"
3717 "%BP_v4f32_TC_1 = OpCompositeConstruct %v4f32 %BP_TC_W_1 %BP_TC_W_1 %BP_TC_W_1 %BP_TC_W_1\n"
3718 "%BP_v4f32_TC_2 = OpCompositeConstruct %v4f32 %BP_TC_W_2 %BP_TC_W_2 %BP_TC_W_2 %BP_TC_W_2\n"
3719
3720 "%BP_gl_IP_0 = OpLoad %v4f32 %BP_gl_in_gl_Pos_0\n"
3721 "%BP_gl_IP_1 = OpLoad %v4f32 %BP_gl_in_gl_Pos_1\n"
3722 "%BP_gl_IP_2 = OpLoad %v4f32 %BP_gl_in_gl_Pos_2\n"
3723
3724 "%BP_IP_W_0 = OpFMul %v4f32 %BP_v4f32_TC_0 %BP_gl_IP_0\n"
3725 "%BP_IP_W_1 = OpFMul %v4f32 %BP_v4f32_TC_1 %BP_gl_IP_1\n"
3726 "%BP_IP_W_2 = OpFMul %v4f32 %BP_v4f32_TC_2 %BP_gl_IP_2\n"
3727
3728 "%BP_pos_sum_0 = OpFAdd %v4f32 %BP_IP_W_0 %BP_IP_W_1\n"
3729 "%BP_pos_sum_1 = OpFAdd %v4f32 %BP_pos_sum_0 %BP_IP_W_2\n"
3730
3731 "OpStore %BP_gl_OPos %BP_pos_sum_1\n"
3732
3733 "%BP_IC_0 = OpLoad %v4f32 %BP_in_color_0\n"
3734 "%BP_IC_1 = OpLoad %v4f32 %BP_in_color_1\n"
3735 "%BP_IC_2 = OpLoad %v4f32 %BP_in_color_2\n"
3736
3737 "%BP_IC_W_0 = OpFMul %v4f32 %BP_v4f32_TC_0 %BP_IC_0\n"
3738 "%BP_IC_W_1 = OpFMul %v4f32 %BP_v4f32_TC_1 %BP_IC_1\n"
3739 "%BP_IC_W_2 = OpFMul %v4f32 %BP_v4f32_TC_2 %BP_IC_2\n"
3740
3741 "%BP_col_sum_0 = OpFAdd %v4f32 %BP_IC_W_0 %BP_IC_W_1\n"
3742 "%BP_col_sum_1 = OpFAdd %v4f32 %BP_col_sum_0 %BP_IC_W_2\n"
3743
3744 "%BP_clr_transformed = OpFunctionCall %v4f32 %test_code %BP_col_sum_1\n"
3745
3746 "OpStore %BP_out_color %BP_clr_transformed\n"
3747 "OpReturn\n"
3748 "OpFunctionEnd\n"
3749 "${testfun}\n";
3750 return tcu::StringTemplate(tessEvalBoilerplate).specialize(fragments);
3751 }
3752
3753 // Creates geometry-shader assembly by specializing a boilerplate StringTemplate
3754 // on fragments, which must (at least) map "testfun" to an OpFunction definition
3755 // for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed
3756 // with "BP_" to avoid collisions with fragments.
3757 //
3758 // Derived from this GLSL:
3759 //
3760 // #version 450
3761 // layout(triangles) in;
3762 // layout(triangle_strip, max_vertices = 3) out;
3763 //
3764 // layout(location = 1) in vec4 in_color[];
3765 // layout(location = 1) out vec4 out_color;
3766 //
3767 // void main() {
3768 // gl_Position = gl_in[0].gl_Position;
3769 // out_color = test_fun(in_color[0]);
3770 // EmitVertex();
3771 // gl_Position = gl_in[1].gl_Position;
3772 // out_color = test_fun(in_color[1]);
3773 // EmitVertex();
3774 // gl_Position = gl_in[2].gl_Position;
3775 // out_color = test_fun(in_color[2]);
3776 // EmitVertex();
3777 // EndPrimitive();
3778 // }
makeGeometryShaderAssembly(const map<string,string> & fragments)3779 string makeGeometryShaderAssembly(const map<string, string>& fragments)
3780 {
3781 static const char geometryShaderBoilerplate[] =
3782 "OpCapability Geometry\n"
3783 "OpCapability ClipDistance\n"
3784 "OpCapability CullDistance\n"
3785 "OpMemoryModel Logical GLSL450\n"
3786 "OpEntryPoint Geometry %BP_main \"main\" %BP_out_gl_position %BP_gl_in %BP_out_color %BP_in_color\n"
3787 "OpExecutionMode %BP_main Triangles\n"
3788 "OpExecutionMode %BP_main OutputTriangleStrip\n"
3789 "OpExecutionMode %BP_main OutputVertices 3\n"
3790 "${debug:opt}\n"
3791 "OpName %BP_main \"main\"\n"
3792 "OpName %BP_per_vertex_in \"gl_PerVertex\"\n"
3793 "OpMemberName %BP_per_vertex_in 0 \"gl_Position\"\n"
3794 "OpMemberName %BP_per_vertex_in 1 \"gl_PointSize\"\n"
3795 "OpMemberName %BP_per_vertex_in 2 \"gl_ClipDistance\"\n"
3796 "OpMemberName %BP_per_vertex_in 3 \"gl_CullDistance\"\n"
3797 "OpName %BP_gl_in \"gl_in\"\n"
3798 "OpName %BP_out_color \"out_color\"\n"
3799 "OpName %BP_in_color \"in_color\"\n"
3800 "OpName %test_code \"testfun(vf4;\"\n"
3801 "OpDecorate %BP_out_gl_position BuiltIn Position\n"
3802 "OpMemberDecorate %BP_per_vertex_in 0 BuiltIn Position\n"
3803 "OpMemberDecorate %BP_per_vertex_in 1 BuiltIn PointSize\n"
3804 "OpMemberDecorate %BP_per_vertex_in 2 BuiltIn ClipDistance\n"
3805 "OpMemberDecorate %BP_per_vertex_in 3 BuiltIn CullDistance\n"
3806 "OpDecorate %BP_per_vertex_in Block\n"
3807 "OpDecorate %BP_out_color Location 1\n"
3808 "OpDecorate %BP_in_color Location 1\n"
3809 "${decoration:opt}\n"
3810 SPIRV_ASSEMBLY_TYPES
3811 SPIRV_ASSEMBLY_CONSTANTS
3812 SPIRV_ASSEMBLY_ARRAYS
3813 "%BP_per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3814 "%BP_a3_per_vertex_in = OpTypeArray %BP_per_vertex_in %c_u32_3\n"
3815 "%BP_ip_a3_per_vertex_in = OpTypePointer Input %BP_a3_per_vertex_in\n"
3816
3817 "%BP_gl_in = OpVariable %BP_ip_a3_per_vertex_in Input\n"
3818 "%BP_out_color = OpVariable %op_v4f32 Output\n"
3819 "%BP_in_color = OpVariable %ip_a3v4f32 Input\n"
3820 "%BP_out_gl_position = OpVariable %op_v4f32 Output\n"
3821 "${pre_main:opt}\n"
3822
3823 "%BP_main = OpFunction %void None %fun\n"
3824 "%BP_label = OpLabel\n"
3825 "%BP_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_0 %c_i32_0\n"
3826 "%BP_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_1 %c_i32_0\n"
3827 "%BP_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_2 %c_i32_0\n"
3828
3829 "%BP_in_position_0 = OpLoad %v4f32 %BP_gl_in_0_gl_position\n"
3830 "%BP_in_position_1 = OpLoad %v4f32 %BP_gl_in_1_gl_position\n"
3831 "%BP_in_position_2 = OpLoad %v4f32 %BP_gl_in_2_gl_position \n"
3832
3833 "%BP_in_color_0_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_0\n"
3834 "%BP_in_color_1_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_1\n"
3835 "%BP_in_color_2_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_2\n"
3836
3837 "%BP_in_color_0 = OpLoad %v4f32 %BP_in_color_0_ptr\n"
3838 "%BP_in_color_1 = OpLoad %v4f32 %BP_in_color_1_ptr\n"
3839 "%BP_in_color_2 = OpLoad %v4f32 %BP_in_color_2_ptr\n"
3840
3841 "%BP_transformed_in_color_0 = OpFunctionCall %v4f32 %test_code %BP_in_color_0\n"
3842 "%BP_transformed_in_color_1 = OpFunctionCall %v4f32 %test_code %BP_in_color_1\n"
3843 "%BP_transformed_in_color_2 = OpFunctionCall %v4f32 %test_code %BP_in_color_2\n"
3844
3845
3846 "OpStore %BP_out_gl_position %BP_in_position_0\n"
3847 "OpStore %BP_out_color %BP_transformed_in_color_0\n"
3848 "OpEmitVertex\n"
3849
3850 "OpStore %BP_out_gl_position %BP_in_position_1\n"
3851 "OpStore %BP_out_color %BP_transformed_in_color_1\n"
3852 "OpEmitVertex\n"
3853
3854 "OpStore %BP_out_gl_position %BP_in_position_2\n"
3855 "OpStore %BP_out_color %BP_transformed_in_color_2\n"
3856 "OpEmitVertex\n"
3857
3858 "OpEndPrimitive\n"
3859 "OpReturn\n"
3860 "OpFunctionEnd\n"
3861 "${testfun}\n";
3862 return tcu::StringTemplate(geometryShaderBoilerplate).specialize(fragments);
3863 }
3864
3865 // Creates fragment-shader assembly by specializing a boilerplate StringTemplate
3866 // on fragments, which must (at least) map "testfun" to an OpFunction definition
3867 // for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed
3868 // with "BP_" to avoid collisions with fragments.
3869 //
3870 // Derived from this GLSL:
3871 //
3872 // layout(location = 1) in highp vec4 vtxColor;
3873 // layout(location = 0) out highp vec4 fragColor;
3874 // highp vec4 testfun(highp vec4 x) { return x; }
3875 // void main(void) { fragColor = testfun(vtxColor); }
3876 //
3877 // with modifications including passing vtxColor by value and ripping out
3878 // testfun() definition.
makeFragmentShaderAssembly(const map<string,string> & fragments)3879 string makeFragmentShaderAssembly(const map<string, string>& fragments)
3880 {
3881 static const char fragmentShaderBoilerplate[] =
3882 "OpCapability Shader\n"
3883 "OpMemoryModel Logical GLSL450\n"
3884 "OpEntryPoint Fragment %BP_main \"main\" %BP_vtxColor %BP_fragColor\n"
3885 "OpExecutionMode %BP_main OriginUpperLeft\n"
3886 "${debug:opt}\n"
3887 "OpName %BP_main \"main\"\n"
3888 "OpName %BP_fragColor \"fragColor\"\n"
3889 "OpName %BP_vtxColor \"vtxColor\"\n"
3890 "OpName %test_code \"testfun(vf4;\"\n"
3891 "OpDecorate %BP_fragColor Location 0\n"
3892 "OpDecorate %BP_vtxColor Location 1\n"
3893 "${decoration:opt}\n"
3894 SPIRV_ASSEMBLY_TYPES
3895 SPIRV_ASSEMBLY_CONSTANTS
3896 SPIRV_ASSEMBLY_ARRAYS
3897 "%BP_fragColor = OpVariable %op_v4f32 Output\n"
3898 "%BP_vtxColor = OpVariable %ip_v4f32 Input\n"
3899 "${pre_main:opt}\n"
3900 "%BP_main = OpFunction %void None %fun\n"
3901 "%BP_label_main = OpLabel\n"
3902 "%BP_tmp1 = OpLoad %v4f32 %BP_vtxColor\n"
3903 "%BP_tmp2 = OpFunctionCall %v4f32 %test_code %BP_tmp1\n"
3904 "OpStore %BP_fragColor %BP_tmp2\n"
3905 "OpReturn\n"
3906 "OpFunctionEnd\n"
3907 "${testfun}\n";
3908 return tcu::StringTemplate(fragmentShaderBoilerplate).specialize(fragments);
3909 }
3910
3911 // Creates fragments that specialize into a simple pass-through shader (of any kind).
passthruFragments(void)3912 map<string, string> passthruFragments(void)
3913 {
3914 map<string, string> fragments;
3915 fragments["testfun"] =
3916 // A %test_code function that returns its argument unchanged.
3917 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
3918 "%param1 = OpFunctionParameter %v4f32\n"
3919 "%label_testfun = OpLabel\n"
3920 "OpReturnValue %param1\n"
3921 "OpFunctionEnd\n";
3922 return fragments;
3923 }
3924
3925 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3926 // Vertex shader gets custom code from context, the rest are pass-through.
addShaderCodeCustomVertex(vk::SourceCollections & dst,InstanceContext context)3927 void addShaderCodeCustomVertex(vk::SourceCollections& dst, InstanceContext context)
3928 {
3929 map<string, string> passthru = passthruFragments();
3930 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(context.testCodeFragments);
3931 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
3932 }
3933
3934 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3935 // Tessellation control shader gets custom code from context, the rest are
3936 // pass-through.
addShaderCodeCustomTessControl(vk::SourceCollections & dst,InstanceContext context)3937 void addShaderCodeCustomTessControl(vk::SourceCollections& dst, InstanceContext context)
3938 {
3939 map<string, string> passthru = passthruFragments();
3940 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
3941 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(context.testCodeFragments);
3942 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(passthru);
3943 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
3944 }
3945
3946 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3947 // Tessellation evaluation shader gets custom code from context, the rest are
3948 // pass-through.
addShaderCodeCustomTessEval(vk::SourceCollections & dst,InstanceContext context)3949 void addShaderCodeCustomTessEval(vk::SourceCollections& dst, InstanceContext context)
3950 {
3951 map<string, string> passthru = passthruFragments();
3952 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
3953 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(passthru);
3954 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(context.testCodeFragments);
3955 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
3956 }
3957
3958 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3959 // Geometry shader gets custom code from context, the rest are pass-through.
addShaderCodeCustomGeometry(vk::SourceCollections & dst,InstanceContext context)3960 void addShaderCodeCustomGeometry(vk::SourceCollections& dst, InstanceContext context)
3961 {
3962 map<string, string> passthru = passthruFragments();
3963 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
3964 dst.spirvAsmSources.add("geom") << makeGeometryShaderAssembly(context.testCodeFragments);
3965 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
3966 }
3967
3968 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3969 // Fragment shader gets custom code from context, the rest are pass-through.
addShaderCodeCustomFragment(vk::SourceCollections & dst,InstanceContext context)3970 void addShaderCodeCustomFragment(vk::SourceCollections& dst, InstanceContext context)
3971 {
3972 map<string, string> passthru = passthruFragments();
3973 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
3974 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(context.testCodeFragments);
3975 }
3976
createCombinedModule(vk::SourceCollections & dst,InstanceContext)3977 void createCombinedModule(vk::SourceCollections& dst, InstanceContext)
3978 {
3979 // \todo [2015-12-07 awoloszyn] Make tessellation / geometry conditional
3980 // \todo [2015-12-07 awoloszyn] Remove OpName and OpMemberName at some point
3981 dst.spirvAsmSources.add("module") <<
3982 "OpCapability Shader\n"
3983 "OpCapability ClipDistance\n"
3984 "OpCapability CullDistance\n"
3985 "OpCapability Geometry\n"
3986 "OpCapability Tessellation\n"
3987 "OpMemoryModel Logical GLSL450\n"
3988
3989 "OpEntryPoint Vertex %vert_main \"main\" %vert_Position %vert_vtxColor %vert_color %vert_vtxPosition %vert_vertex_id %vert_instance_id\n"
3990 "OpEntryPoint Geometry %geom_main \"main\" %geom_out_gl_position %geom_gl_in %geom_out_color %geom_in_color\n"
3991 "OpEntryPoint TessellationControl %tessc_main \"main\" %tessc_out_color %tessc_gl_InvocationID %tessc_in_color %tessc_out_position %tessc_in_position %tessc_gl_TessLevelOuter %tessc_gl_TessLevelInner\n"
3992 "OpEntryPoint TessellationEvaluation %tesse_main \"main\" %tesse_stream %tesse_gl_tessCoord %tesse_in_position %tesse_out_color %tesse_in_color \n"
3993 "OpEntryPoint Fragment %frag_main \"main\" %frag_vtxColor %frag_fragColor\n"
3994
3995 "OpExecutionMode %geom_main Triangles\n"
3996 "OpExecutionMode %geom_main OutputTriangleStrip\n"
3997 "OpExecutionMode %geom_main OutputVertices 3\n"
3998
3999 "OpExecutionMode %tessc_main OutputVertices 3\n"
4000
4001 "OpExecutionMode %tesse_main Triangles\n"
4002
4003 "OpExecutionMode %frag_main OriginUpperLeft\n"
4004
4005 "OpName %vert_main \"main\"\n"
4006 "OpName %vert_vtxPosition \"vtxPosition\"\n"
4007 "OpName %vert_Position \"position\"\n"
4008 "OpName %vert_vtxColor \"vtxColor\"\n"
4009 "OpName %vert_color \"color\"\n"
4010 "OpName %vert_vertex_id \"gl_VertexIndex\"\n"
4011 "OpName %vert_instance_id \"gl_InstanceIndex\"\n"
4012 "OpName %geom_main \"main\"\n"
4013 "OpName %geom_per_vertex_in \"gl_PerVertex\"\n"
4014 "OpMemberName %geom_per_vertex_in 0 \"gl_Position\"\n"
4015 "OpMemberName %geom_per_vertex_in 1 \"gl_PointSize\"\n"
4016 "OpMemberName %geom_per_vertex_in 2 \"gl_ClipDistance\"\n"
4017 "OpMemberName %geom_per_vertex_in 3 \"gl_CullDistance\"\n"
4018 "OpName %geom_gl_in \"gl_in\"\n"
4019 "OpName %geom_out_color \"out_color\"\n"
4020 "OpName %geom_in_color \"in_color\"\n"
4021 "OpName %tessc_main \"main\"\n"
4022 "OpName %tessc_out_color \"out_color\"\n"
4023 "OpName %tessc_gl_InvocationID \"gl_InvocationID\"\n"
4024 "OpName %tessc_in_color \"in_color\"\n"
4025 "OpName %tessc_out_position \"out_position\"\n"
4026 "OpName %tessc_in_position \"in_position\"\n"
4027 "OpName %tessc_gl_TessLevelOuter \"gl_TessLevelOuter\"\n"
4028 "OpName %tessc_gl_TessLevelInner \"gl_TessLevelInner\"\n"
4029 "OpName %tesse_main \"main\"\n"
4030 "OpName %tesse_per_vertex_out \"gl_PerVertex\"\n"
4031 "OpMemberName %tesse_per_vertex_out 0 \"gl_Position\"\n"
4032 "OpMemberName %tesse_per_vertex_out 1 \"gl_PointSize\"\n"
4033 "OpMemberName %tesse_per_vertex_out 2 \"gl_ClipDistance\"\n"
4034 "OpMemberName %tesse_per_vertex_out 3 \"gl_CullDistance\"\n"
4035 "OpName %tesse_stream \"\"\n"
4036 "OpName %tesse_gl_tessCoord \"gl_TessCoord\"\n"
4037 "OpName %tesse_in_position \"in_position\"\n"
4038 "OpName %tesse_out_color \"out_color\"\n"
4039 "OpName %tesse_in_color \"in_color\"\n"
4040 "OpName %frag_main \"main\"\n"
4041 "OpName %frag_fragColor \"fragColor\"\n"
4042 "OpName %frag_vtxColor \"vtxColor\"\n"
4043
4044 "; Vertex decorations\n"
4045 "OpDecorate %vert_vtxPosition Location 2\n"
4046 "OpDecorate %vert_Position Location 0\n"
4047 "OpDecorate %vert_vtxColor Location 1\n"
4048 "OpDecorate %vert_color Location 1\n"
4049 "OpDecorate %vert_vertex_id BuiltIn VertexIndex\n"
4050 "OpDecorate %vert_instance_id BuiltIn InstanceIndex\n"
4051
4052 "; Geometry decorations\n"
4053 "OpDecorate %geom_out_gl_position BuiltIn Position\n"
4054 "OpMemberDecorate %geom_per_vertex_in 0 BuiltIn Position\n"
4055 "OpMemberDecorate %geom_per_vertex_in 1 BuiltIn PointSize\n"
4056 "OpMemberDecorate %geom_per_vertex_in 2 BuiltIn ClipDistance\n"
4057 "OpMemberDecorate %geom_per_vertex_in 3 BuiltIn CullDistance\n"
4058 "OpDecorate %geom_per_vertex_in Block\n"
4059 "OpDecorate %geom_out_color Location 1\n"
4060 "OpDecorate %geom_in_color Location 1\n"
4061
4062 "; Tessellation Control decorations\n"
4063 "OpDecorate %tessc_out_color Location 1\n"
4064 "OpDecorate %tessc_gl_InvocationID BuiltIn InvocationId\n"
4065 "OpDecorate %tessc_in_color Location 1\n"
4066 "OpDecorate %tessc_out_position Location 2\n"
4067 "OpDecorate %tessc_in_position Location 2\n"
4068 "OpDecorate %tessc_gl_TessLevelOuter Patch\n"
4069 "OpDecorate %tessc_gl_TessLevelOuter BuiltIn TessLevelOuter\n"
4070 "OpDecorate %tessc_gl_TessLevelInner Patch\n"
4071 "OpDecorate %tessc_gl_TessLevelInner BuiltIn TessLevelInner\n"
4072
4073 "; Tessellation Evaluation decorations\n"
4074 "OpMemberDecorate %tesse_per_vertex_out 0 BuiltIn Position\n"
4075 "OpMemberDecorate %tesse_per_vertex_out 1 BuiltIn PointSize\n"
4076 "OpMemberDecorate %tesse_per_vertex_out 2 BuiltIn ClipDistance\n"
4077 "OpMemberDecorate %tesse_per_vertex_out 3 BuiltIn CullDistance\n"
4078 "OpDecorate %tesse_per_vertex_out Block\n"
4079 "OpDecorate %tesse_gl_tessCoord BuiltIn TessCoord\n"
4080 "OpDecorate %tesse_in_position Location 2\n"
4081 "OpDecorate %tesse_out_color Location 1\n"
4082 "OpDecorate %tesse_in_color Location 1\n"
4083
4084 "; Fragment decorations\n"
4085 "OpDecorate %frag_fragColor Location 0\n"
4086 "OpDecorate %frag_vtxColor Location 1\n"
4087
4088 SPIRV_ASSEMBLY_TYPES
4089 SPIRV_ASSEMBLY_CONSTANTS
4090 SPIRV_ASSEMBLY_ARRAYS
4091
4092 "; Vertex Variables\n"
4093 "%vert_vtxPosition = OpVariable %op_v4f32 Output\n"
4094 "%vert_Position = OpVariable %ip_v4f32 Input\n"
4095 "%vert_vtxColor = OpVariable %op_v4f32 Output\n"
4096 "%vert_color = OpVariable %ip_v4f32 Input\n"
4097 "%vert_vertex_id = OpVariable %ip_i32 Input\n"
4098 "%vert_instance_id = OpVariable %ip_i32 Input\n"
4099
4100 "; Geometry Variables\n"
4101 "%geom_per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4102 "%geom_a3_per_vertex_in = OpTypeArray %geom_per_vertex_in %c_u32_3\n"
4103 "%geom_ip_a3_per_vertex_in = OpTypePointer Input %geom_a3_per_vertex_in\n"
4104 "%geom_gl_in = OpVariable %geom_ip_a3_per_vertex_in Input\n"
4105 "%geom_out_color = OpVariable %op_v4f32 Output\n"
4106 "%geom_in_color = OpVariable %ip_a3v4f32 Input\n"
4107 "%geom_out_gl_position = OpVariable %op_v4f32 Output\n"
4108
4109 "; Tessellation Control Variables\n"
4110 "%tessc_out_color = OpVariable %op_a3v4f32 Output\n"
4111 "%tessc_gl_InvocationID = OpVariable %ip_i32 Input\n"
4112 "%tessc_in_color = OpVariable %ip_a32v4f32 Input\n"
4113 "%tessc_out_position = OpVariable %op_a3v4f32 Output\n"
4114 "%tessc_in_position = OpVariable %ip_a32v4f32 Input\n"
4115 "%tessc_gl_TessLevelOuter = OpVariable %op_a4f32 Output\n"
4116 "%tessc_gl_TessLevelInner = OpVariable %op_a2f32 Output\n"
4117
4118 "; Tessellation Evaluation Decorations\n"
4119 "%tesse_per_vertex_out = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4120 "%tesse_op_per_vertex_out = OpTypePointer Output %tesse_per_vertex_out\n"
4121 "%tesse_stream = OpVariable %tesse_op_per_vertex_out Output\n"
4122 "%tesse_gl_tessCoord = OpVariable %ip_v3f32 Input\n"
4123 "%tesse_in_position = OpVariable %ip_a32v4f32 Input\n"
4124 "%tesse_out_color = OpVariable %op_v4f32 Output\n"
4125 "%tesse_in_color = OpVariable %ip_a32v4f32 Input\n"
4126
4127 "; Fragment Variables\n"
4128 "%frag_fragColor = OpVariable %op_v4f32 Output\n"
4129 "%frag_vtxColor = OpVariable %ip_v4f32 Input\n"
4130
4131 "; Vertex Entry\n"
4132 "%vert_main = OpFunction %void None %fun\n"
4133 "%vert_label = OpLabel\n"
4134 "%vert_tmp_position = OpLoad %v4f32 %vert_Position\n"
4135 "OpStore %vert_vtxPosition %vert_tmp_position\n"
4136 "%vert_tmp_color = OpLoad %v4f32 %vert_color\n"
4137 "OpStore %vert_vtxColor %vert_tmp_color\n"
4138 "OpReturn\n"
4139 "OpFunctionEnd\n"
4140
4141 "; Geometry Entry\n"
4142 "%geom_main = OpFunction %void None %fun\n"
4143 "%geom_label = OpLabel\n"
4144 "%geom_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %geom_gl_in %c_i32_0 %c_i32_0\n"
4145 "%geom_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %geom_gl_in %c_i32_1 %c_i32_0\n"
4146 "%geom_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %geom_gl_in %c_i32_2 %c_i32_0\n"
4147 "%geom_in_position_0 = OpLoad %v4f32 %geom_gl_in_0_gl_position\n"
4148 "%geom_in_position_1 = OpLoad %v4f32 %geom_gl_in_1_gl_position\n"
4149 "%geom_in_position_2 = OpLoad %v4f32 %geom_gl_in_2_gl_position \n"
4150 "%geom_in_color_0_ptr = OpAccessChain %ip_v4f32 %geom_in_color %c_i32_0\n"
4151 "%geom_in_color_1_ptr = OpAccessChain %ip_v4f32 %geom_in_color %c_i32_1\n"
4152 "%geom_in_color_2_ptr = OpAccessChain %ip_v4f32 %geom_in_color %c_i32_2\n"
4153 "%geom_in_color_0 = OpLoad %v4f32 %geom_in_color_0_ptr\n"
4154 "%geom_in_color_1 = OpLoad %v4f32 %geom_in_color_1_ptr\n"
4155 "%geom_in_color_2 = OpLoad %v4f32 %geom_in_color_2_ptr\n"
4156 "OpStore %geom_out_gl_position %geom_in_position_0\n"
4157 "OpStore %geom_out_color %geom_in_color_0\n"
4158 "OpEmitVertex\n"
4159 "OpStore %geom_out_gl_position %geom_in_position_1\n"
4160 "OpStore %geom_out_color %geom_in_color_1\n"
4161 "OpEmitVertex\n"
4162 "OpStore %geom_out_gl_position %geom_in_position_2\n"
4163 "OpStore %geom_out_color %geom_in_color_2\n"
4164 "OpEmitVertex\n"
4165 "OpEndPrimitive\n"
4166 "OpReturn\n"
4167 "OpFunctionEnd\n"
4168
4169 "; Tessellation Control Entry\n"
4170 "%tessc_main = OpFunction %void None %fun\n"
4171 "%tessc_label = OpLabel\n"
4172 "%tessc_invocation_id = OpLoad %i32 %tessc_gl_InvocationID\n"
4173 "%tessc_in_color_ptr = OpAccessChain %ip_v4f32 %tessc_in_color %tessc_invocation_id\n"
4174 "%tessc_in_position_ptr = OpAccessChain %ip_v4f32 %tessc_in_position %tessc_invocation_id\n"
4175 "%tessc_in_color_val = OpLoad %v4f32 %tessc_in_color_ptr\n"
4176 "%tessc_in_position_val = OpLoad %v4f32 %tessc_in_position_ptr\n"
4177 "%tessc_out_color_ptr = OpAccessChain %op_v4f32 %tessc_out_color %tessc_invocation_id\n"
4178 "%tessc_out_position_ptr = OpAccessChain %op_v4f32 %tessc_out_position %tessc_invocation_id\n"
4179 "OpStore %tessc_out_color_ptr %tessc_in_color_val\n"
4180 "OpStore %tessc_out_position_ptr %tessc_in_position_val\n"
4181 "%tessc_is_first_invocation = OpIEqual %bool %tessc_invocation_id %c_i32_0\n"
4182 "OpSelectionMerge %tessc_merge_label None\n"
4183 "OpBranchConditional %tessc_is_first_invocation %tessc_first_invocation %tessc_merge_label\n"
4184 "%tessc_first_invocation = OpLabel\n"
4185 "%tessc_tess_outer_0 = OpAccessChain %op_f32 %tessc_gl_TessLevelOuter %c_i32_0\n"
4186 "%tessc_tess_outer_1 = OpAccessChain %op_f32 %tessc_gl_TessLevelOuter %c_i32_1\n"
4187 "%tessc_tess_outer_2 = OpAccessChain %op_f32 %tessc_gl_TessLevelOuter %c_i32_2\n"
4188 "%tessc_tess_inner = OpAccessChain %op_f32 %tessc_gl_TessLevelInner %c_i32_0\n"
4189 "OpStore %tessc_tess_outer_0 %c_f32_1\n"
4190 "OpStore %tessc_tess_outer_1 %c_f32_1\n"
4191 "OpStore %tessc_tess_outer_2 %c_f32_1\n"
4192 "OpStore %tessc_tess_inner %c_f32_1\n"
4193 "OpBranch %tessc_merge_label\n"
4194 "%tessc_merge_label = OpLabel\n"
4195 "OpReturn\n"
4196 "OpFunctionEnd\n"
4197
4198 "; Tessellation Evaluation Entry\n"
4199 "%tesse_main = OpFunction %void None %fun\n"
4200 "%tesse_label = OpLabel\n"
4201 "%tesse_tc_0_ptr = OpAccessChain %ip_f32 %tesse_gl_tessCoord %c_u32_0\n"
4202 "%tesse_tc_1_ptr = OpAccessChain %ip_f32 %tesse_gl_tessCoord %c_u32_1\n"
4203 "%tesse_tc_2_ptr = OpAccessChain %ip_f32 %tesse_gl_tessCoord %c_u32_2\n"
4204 "%tesse_tc_0 = OpLoad %f32 %tesse_tc_0_ptr\n"
4205 "%tesse_tc_1 = OpLoad %f32 %tesse_tc_1_ptr\n"
4206 "%tesse_tc_2 = OpLoad %f32 %tesse_tc_2_ptr\n"
4207 "%tesse_in_pos_0_ptr = OpAccessChain %ip_v4f32 %tesse_in_position %c_i32_0\n"
4208 "%tesse_in_pos_1_ptr = OpAccessChain %ip_v4f32 %tesse_in_position %c_i32_1\n"
4209 "%tesse_in_pos_2_ptr = OpAccessChain %ip_v4f32 %tesse_in_position %c_i32_2\n"
4210 "%tesse_in_pos_0 = OpLoad %v4f32 %tesse_in_pos_0_ptr\n"
4211 "%tesse_in_pos_1 = OpLoad %v4f32 %tesse_in_pos_1_ptr\n"
4212 "%tesse_in_pos_2 = OpLoad %v4f32 %tesse_in_pos_2_ptr\n"
4213 "%tesse_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_0 %tesse_in_pos_0\n"
4214 "%tesse_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_1 %tesse_in_pos_1\n"
4215 "%tesse_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_2 %tesse_in_pos_2\n"
4216 "%tesse_out_pos_ptr = OpAccessChain %op_v4f32 %tesse_stream %c_i32_0\n"
4217 "%tesse_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %tesse_in_pos_0_weighted %tesse_in_pos_1_weighted\n"
4218 "%tesse_computed_out = OpFAdd %v4f32 %tesse_in_pos_0_plus_pos_1 %tesse_in_pos_2_weighted\n"
4219 "OpStore %tesse_out_pos_ptr %tesse_computed_out\n"
4220 "%tesse_in_clr_0_ptr = OpAccessChain %ip_v4f32 %tesse_in_color %c_i32_0\n"
4221 "%tesse_in_clr_1_ptr = OpAccessChain %ip_v4f32 %tesse_in_color %c_i32_1\n"
4222 "%tesse_in_clr_2_ptr = OpAccessChain %ip_v4f32 %tesse_in_color %c_i32_2\n"
4223 "%tesse_in_clr_0 = OpLoad %v4f32 %tesse_in_clr_0_ptr\n"
4224 "%tesse_in_clr_1 = OpLoad %v4f32 %tesse_in_clr_1_ptr\n"
4225 "%tesse_in_clr_2 = OpLoad %v4f32 %tesse_in_clr_2_ptr\n"
4226 "%tesse_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_0 %tesse_in_clr_0\n"
4227 "%tesse_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_1 %tesse_in_clr_1\n"
4228 "%tesse_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_2 %tesse_in_clr_2\n"
4229 "%tesse_in_clr_0_plus_col_1 = OpFAdd %v4f32 %tesse_in_clr_0_weighted %tesse_in_clr_1_weighted\n"
4230 "%tesse_computed_clr = OpFAdd %v4f32 %tesse_in_clr_0_plus_col_1 %tesse_in_clr_2_weighted\n"
4231 "OpStore %tesse_out_color %tesse_computed_clr\n"
4232 "OpReturn\n"
4233 "OpFunctionEnd\n"
4234
4235 "; Fragment Entry\n"
4236 "%frag_main = OpFunction %void None %fun\n"
4237 "%frag_label_main = OpLabel\n"
4238 "%frag_tmp1 = OpLoad %v4f32 %frag_vtxColor\n"
4239 "OpStore %frag_fragColor %frag_tmp1\n"
4240 "OpReturn\n"
4241 "OpFunctionEnd\n";
4242 }
4243
4244 // This has two shaders of each stage. The first
4245 // is a passthrough, the second inverts the color.
createMultipleEntries(vk::SourceCollections & dst,InstanceContext)4246 void createMultipleEntries(vk::SourceCollections& dst, InstanceContext)
4247 {
4248 dst.spirvAsmSources.add("vert") <<
4249 // This module contains 2 vertex shaders. One that is a passthrough
4250 // and a second that inverts the color of the output (1.0 - color).
4251 "OpCapability Shader\n"
4252 "OpMemoryModel Logical GLSL450\n"
4253 "OpEntryPoint Vertex %main \"vert1\" %Position %vtxColor %color %vtxPosition %vertex_id %instance_id\n"
4254 "OpEntryPoint Vertex %main2 \"vert2\" %Position %vtxColor %color %vtxPosition %vertex_id %instance_id\n"
4255
4256 "OpName %main \"vert1\"\n"
4257 "OpName %main2 \"vert2\"\n"
4258 "OpName %vtxPosition \"vtxPosition\"\n"
4259 "OpName %Position \"position\"\n"
4260 "OpName %vtxColor \"vtxColor\"\n"
4261 "OpName %color \"color\"\n"
4262 "OpName %vertex_id \"gl_VertexIndex\"\n"
4263 "OpName %instance_id \"gl_InstanceIndex\"\n"
4264
4265 "OpDecorate %vtxPosition Location 2\n"
4266 "OpDecorate %Position Location 0\n"
4267 "OpDecorate %vtxColor Location 1\n"
4268 "OpDecorate %color Location 1\n"
4269 "OpDecorate %vertex_id BuiltIn VertexIndex\n"
4270 "OpDecorate %instance_id BuiltIn InstanceIndex\n"
4271 SPIRV_ASSEMBLY_TYPES
4272 SPIRV_ASSEMBLY_CONSTANTS
4273 SPIRV_ASSEMBLY_ARRAYS
4274 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4275 "%vtxPosition = OpVariable %op_v4f32 Output\n"
4276 "%Position = OpVariable %ip_v4f32 Input\n"
4277 "%vtxColor = OpVariable %op_v4f32 Output\n"
4278 "%color = OpVariable %ip_v4f32 Input\n"
4279 "%vertex_id = OpVariable %ip_i32 Input\n"
4280 "%instance_id = OpVariable %ip_i32 Input\n"
4281
4282 "%main = OpFunction %void None %fun\n"
4283 "%label = OpLabel\n"
4284 "%tmp_position = OpLoad %v4f32 %Position\n"
4285 "OpStore %vtxPosition %tmp_position\n"
4286 "%tmp_color = OpLoad %v4f32 %color\n"
4287 "OpStore %vtxColor %tmp_color\n"
4288 "OpReturn\n"
4289 "OpFunctionEnd\n"
4290
4291 "%main2 = OpFunction %void None %fun\n"
4292 "%label2 = OpLabel\n"
4293 "%tmp_position2 = OpLoad %v4f32 %Position\n"
4294 "OpStore %vtxPosition %tmp_position2\n"
4295 "%tmp_color2 = OpLoad %v4f32 %color\n"
4296 "%tmp_color3 = OpFSub %v4f32 %cval %tmp_color2\n"
4297 "%tmp_color4 = OpVectorInsertDynamic %v4f32 %tmp_color3 %c_f32_1 %c_i32_3\n"
4298 "OpStore %vtxColor %tmp_color4\n"
4299 "OpReturn\n"
4300 "OpFunctionEnd\n";
4301
4302 dst.spirvAsmSources.add("frag") <<
4303 // This is a single module that contains 2 fragment shaders.
4304 // One that passes color through and the other that inverts the output
4305 // color (1.0 - color).
4306 "OpCapability Shader\n"
4307 "OpMemoryModel Logical GLSL450\n"
4308 "OpEntryPoint Fragment %main \"frag1\" %vtxColor %fragColor\n"
4309 "OpEntryPoint Fragment %main2 \"frag2\" %vtxColor %fragColor\n"
4310 "OpExecutionMode %main OriginUpperLeft\n"
4311 "OpExecutionMode %main2 OriginUpperLeft\n"
4312
4313 "OpName %main \"frag1\"\n"
4314 "OpName %main2 \"frag2\"\n"
4315 "OpName %fragColor \"fragColor\"\n"
4316 "OpName %vtxColor \"vtxColor\"\n"
4317 "OpDecorate %fragColor Location 0\n"
4318 "OpDecorate %vtxColor Location 1\n"
4319 SPIRV_ASSEMBLY_TYPES
4320 SPIRV_ASSEMBLY_CONSTANTS
4321 SPIRV_ASSEMBLY_ARRAYS
4322 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4323 "%fragColor = OpVariable %op_v4f32 Output\n"
4324 "%vtxColor = OpVariable %ip_v4f32 Input\n"
4325
4326 "%main = OpFunction %void None %fun\n"
4327 "%label_main = OpLabel\n"
4328 "%tmp1 = OpLoad %v4f32 %vtxColor\n"
4329 "OpStore %fragColor %tmp1\n"
4330 "OpReturn\n"
4331 "OpFunctionEnd\n"
4332
4333 "%main2 = OpFunction %void None %fun\n"
4334 "%label_main2 = OpLabel\n"
4335 "%tmp2 = OpLoad %v4f32 %vtxColor\n"
4336 "%tmp3 = OpFSub %v4f32 %cval %tmp2\n"
4337 "%tmp4 = OpVectorInsertDynamic %v4f32 %tmp3 %c_f32_1 %c_i32_3\n"
4338 "OpStore %fragColor %tmp4\n"
4339 "OpReturn\n"
4340 "OpFunctionEnd\n";
4341
4342 dst.spirvAsmSources.add("geom") <<
4343 "OpCapability Geometry\n"
4344 "OpCapability ClipDistance\n"
4345 "OpCapability CullDistance\n"
4346 "OpMemoryModel Logical GLSL450\n"
4347 "OpEntryPoint Geometry %geom1_main \"geom1\" %out_gl_position %gl_in %out_color %in_color\n"
4348 "OpEntryPoint Geometry %geom2_main \"geom2\" %out_gl_position %gl_in %out_color %in_color\n"
4349 "OpExecutionMode %geom1_main Triangles\n"
4350 "OpExecutionMode %geom2_main Triangles\n"
4351 "OpExecutionMode %geom1_main OutputTriangleStrip\n"
4352 "OpExecutionMode %geom2_main OutputTriangleStrip\n"
4353 "OpExecutionMode %geom1_main OutputVertices 3\n"
4354 "OpExecutionMode %geom2_main OutputVertices 3\n"
4355 "OpName %geom1_main \"geom1\"\n"
4356 "OpName %geom2_main \"geom2\"\n"
4357 "OpName %per_vertex_in \"gl_PerVertex\"\n"
4358 "OpMemberName %per_vertex_in 0 \"gl_Position\"\n"
4359 "OpMemberName %per_vertex_in 1 \"gl_PointSize\"\n"
4360 "OpMemberName %per_vertex_in 2 \"gl_ClipDistance\"\n"
4361 "OpMemberName %per_vertex_in 3 \"gl_CullDistance\"\n"
4362 "OpName %gl_in \"gl_in\"\n"
4363 "OpName %out_color \"out_color\"\n"
4364 "OpName %in_color \"in_color\"\n"
4365 "OpDecorate %out_gl_position BuiltIn Position\n"
4366 "OpMemberDecorate %per_vertex_in 0 BuiltIn Position\n"
4367 "OpMemberDecorate %per_vertex_in 1 BuiltIn PointSize\n"
4368 "OpMemberDecorate %per_vertex_in 2 BuiltIn ClipDistance\n"
4369 "OpMemberDecorate %per_vertex_in 3 BuiltIn CullDistance\n"
4370 "OpDecorate %per_vertex_in Block\n"
4371 "OpDecorate %out_color Location 1\n"
4372 "OpDecorate %in_color Location 1\n"
4373 SPIRV_ASSEMBLY_TYPES
4374 SPIRV_ASSEMBLY_CONSTANTS
4375 SPIRV_ASSEMBLY_ARRAYS
4376 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4377 "%per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4378 "%a3_per_vertex_in = OpTypeArray %per_vertex_in %c_u32_3\n"
4379 "%ip_a3_per_vertex_in = OpTypePointer Input %a3_per_vertex_in\n"
4380 "%gl_in = OpVariable %ip_a3_per_vertex_in Input\n"
4381 "%out_color = OpVariable %op_v4f32 Output\n"
4382 "%in_color = OpVariable %ip_a3v4f32 Input\n"
4383 "%out_gl_position = OpVariable %op_v4f32 Output\n"
4384
4385 "%geom1_main = OpFunction %void None %fun\n"
4386 "%geom1_label = OpLabel\n"
4387 "%geom1_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_0 %c_i32_0\n"
4388 "%geom1_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_1 %c_i32_0\n"
4389 "%geom1_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_2 %c_i32_0\n"
4390 "%geom1_in_position_0 = OpLoad %v4f32 %geom1_gl_in_0_gl_position\n"
4391 "%geom1_in_position_1 = OpLoad %v4f32 %geom1_gl_in_1_gl_position\n"
4392 "%geom1_in_position_2 = OpLoad %v4f32 %geom1_gl_in_2_gl_position \n"
4393 "%geom1_in_color_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4394 "%geom1_in_color_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4395 "%geom1_in_color_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4396 "%geom1_in_color_0 = OpLoad %v4f32 %geom1_in_color_0_ptr\n"
4397 "%geom1_in_color_1 = OpLoad %v4f32 %geom1_in_color_1_ptr\n"
4398 "%geom1_in_color_2 = OpLoad %v4f32 %geom1_in_color_2_ptr\n"
4399 "OpStore %out_gl_position %geom1_in_position_0\n"
4400 "OpStore %out_color %geom1_in_color_0\n"
4401 "OpEmitVertex\n"
4402 "OpStore %out_gl_position %geom1_in_position_1\n"
4403 "OpStore %out_color %geom1_in_color_1\n"
4404 "OpEmitVertex\n"
4405 "OpStore %out_gl_position %geom1_in_position_2\n"
4406 "OpStore %out_color %geom1_in_color_2\n"
4407 "OpEmitVertex\n"
4408 "OpEndPrimitive\n"
4409 "OpReturn\n"
4410 "OpFunctionEnd\n"
4411
4412 "%geom2_main = OpFunction %void None %fun\n"
4413 "%geom2_label = OpLabel\n"
4414 "%geom2_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_0 %c_i32_0\n"
4415 "%geom2_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_1 %c_i32_0\n"
4416 "%geom2_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_2 %c_i32_0\n"
4417 "%geom2_in_position_0 = OpLoad %v4f32 %geom2_gl_in_0_gl_position\n"
4418 "%geom2_in_position_1 = OpLoad %v4f32 %geom2_gl_in_1_gl_position\n"
4419 "%geom2_in_position_2 = OpLoad %v4f32 %geom2_gl_in_2_gl_position \n"
4420 "%geom2_in_color_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4421 "%geom2_in_color_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4422 "%geom2_in_color_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4423 "%geom2_in_color_0 = OpLoad %v4f32 %geom2_in_color_0_ptr\n"
4424 "%geom2_in_color_1 = OpLoad %v4f32 %geom2_in_color_1_ptr\n"
4425 "%geom2_in_color_2 = OpLoad %v4f32 %geom2_in_color_2_ptr\n"
4426 "%geom2_transformed_in_color_0 = OpFSub %v4f32 %cval %geom2_in_color_0\n"
4427 "%geom2_transformed_in_color_1 = OpFSub %v4f32 %cval %geom2_in_color_1\n"
4428 "%geom2_transformed_in_color_2 = OpFSub %v4f32 %cval %geom2_in_color_2\n"
4429 "%geom2_transformed_in_color_0_a = OpVectorInsertDynamic %v4f32 %geom2_transformed_in_color_0 %c_f32_1 %c_i32_3\n"
4430 "%geom2_transformed_in_color_1_a = OpVectorInsertDynamic %v4f32 %geom2_transformed_in_color_1 %c_f32_1 %c_i32_3\n"
4431 "%geom2_transformed_in_color_2_a = OpVectorInsertDynamic %v4f32 %geom2_transformed_in_color_2 %c_f32_1 %c_i32_3\n"
4432 "OpStore %out_gl_position %geom2_in_position_0\n"
4433 "OpStore %out_color %geom2_transformed_in_color_0_a\n"
4434 "OpEmitVertex\n"
4435 "OpStore %out_gl_position %geom2_in_position_1\n"
4436 "OpStore %out_color %geom2_transformed_in_color_1_a\n"
4437 "OpEmitVertex\n"
4438 "OpStore %out_gl_position %geom2_in_position_2\n"
4439 "OpStore %out_color %geom2_transformed_in_color_2_a\n"
4440 "OpEmitVertex\n"
4441 "OpEndPrimitive\n"
4442 "OpReturn\n"
4443 "OpFunctionEnd\n";
4444
4445 dst.spirvAsmSources.add("tessc") <<
4446 "OpCapability Tessellation\n"
4447 "OpMemoryModel Logical GLSL450\n"
4448 "OpEntryPoint TessellationControl %tessc1_main \"tessc1\" %out_color %gl_InvocationID %in_color %out_position %in_position %gl_TessLevelOuter %gl_TessLevelInner\n"
4449 "OpEntryPoint TessellationControl %tessc2_main \"tessc2\" %out_color %gl_InvocationID %in_color %out_position %in_position %gl_TessLevelOuter %gl_TessLevelInner\n"
4450 "OpExecutionMode %tessc1_main OutputVertices 3\n"
4451 "OpExecutionMode %tessc2_main OutputVertices 3\n"
4452 "OpName %tessc1_main \"tessc1\"\n"
4453 "OpName %tessc2_main \"tessc2\"\n"
4454 "OpName %out_color \"out_color\"\n"
4455 "OpName %gl_InvocationID \"gl_InvocationID\"\n"
4456 "OpName %in_color \"in_color\"\n"
4457 "OpName %out_position \"out_position\"\n"
4458 "OpName %in_position \"in_position\"\n"
4459 "OpName %gl_TessLevelOuter \"gl_TessLevelOuter\"\n"
4460 "OpName %gl_TessLevelInner \"gl_TessLevelInner\"\n"
4461 "OpDecorate %out_color Location 1\n"
4462 "OpDecorate %gl_InvocationID BuiltIn InvocationId\n"
4463 "OpDecorate %in_color Location 1\n"
4464 "OpDecorate %out_position Location 2\n"
4465 "OpDecorate %in_position Location 2\n"
4466 "OpDecorate %gl_TessLevelOuter Patch\n"
4467 "OpDecorate %gl_TessLevelOuter BuiltIn TessLevelOuter\n"
4468 "OpDecorate %gl_TessLevelInner Patch\n"
4469 "OpDecorate %gl_TessLevelInner BuiltIn TessLevelInner\n"
4470 SPIRV_ASSEMBLY_TYPES
4471 SPIRV_ASSEMBLY_CONSTANTS
4472 SPIRV_ASSEMBLY_ARRAYS
4473 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4474 "%out_color = OpVariable %op_a3v4f32 Output\n"
4475 "%gl_InvocationID = OpVariable %ip_i32 Input\n"
4476 "%in_color = OpVariable %ip_a32v4f32 Input\n"
4477 "%out_position = OpVariable %op_a3v4f32 Output\n"
4478 "%in_position = OpVariable %ip_a32v4f32 Input\n"
4479 "%gl_TessLevelOuter = OpVariable %op_a4f32 Output\n"
4480 "%gl_TessLevelInner = OpVariable %op_a2f32 Output\n"
4481
4482 "%tessc1_main = OpFunction %void None %fun\n"
4483 "%tessc1_label = OpLabel\n"
4484 "%tessc1_invocation_id = OpLoad %i32 %gl_InvocationID\n"
4485 "%tessc1_in_color_ptr = OpAccessChain %ip_v4f32 %in_color %tessc1_invocation_id\n"
4486 "%tessc1_in_position_ptr = OpAccessChain %ip_v4f32 %in_position %tessc1_invocation_id\n"
4487 "%tessc1_in_color_val = OpLoad %v4f32 %tessc1_in_color_ptr\n"
4488 "%tessc1_in_position_val = OpLoad %v4f32 %tessc1_in_position_ptr\n"
4489 "%tessc1_out_color_ptr = OpAccessChain %op_v4f32 %out_color %tessc1_invocation_id\n"
4490 "%tessc1_out_position_ptr = OpAccessChain %op_v4f32 %out_position %tessc1_invocation_id\n"
4491 "OpStore %tessc1_out_color_ptr %tessc1_in_color_val\n"
4492 "OpStore %tessc1_out_position_ptr %tessc1_in_position_val\n"
4493 "%tessc1_is_first_invocation = OpIEqual %bool %tessc1_invocation_id %c_i32_0\n"
4494 "OpSelectionMerge %tessc1_merge_label None\n"
4495 "OpBranchConditional %tessc1_is_first_invocation %tessc1_first_invocation %tessc1_merge_label\n"
4496 "%tessc1_first_invocation = OpLabel\n"
4497 "%tessc1_tess_outer_0 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_0\n"
4498 "%tessc1_tess_outer_1 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_1\n"
4499 "%tessc1_tess_outer_2 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_2\n"
4500 "%tessc1_tess_inner = OpAccessChain %op_f32 %gl_TessLevelInner %c_i32_0\n"
4501 "OpStore %tessc1_tess_outer_0 %c_f32_1\n"
4502 "OpStore %tessc1_tess_outer_1 %c_f32_1\n"
4503 "OpStore %tessc1_tess_outer_2 %c_f32_1\n"
4504 "OpStore %tessc1_tess_inner %c_f32_1\n"
4505 "OpBranch %tessc1_merge_label\n"
4506 "%tessc1_merge_label = OpLabel\n"
4507 "OpReturn\n"
4508 "OpFunctionEnd\n"
4509
4510 "%tessc2_main = OpFunction %void None %fun\n"
4511 "%tessc2_label = OpLabel\n"
4512 "%tessc2_invocation_id = OpLoad %i32 %gl_InvocationID\n"
4513 "%tessc2_in_color_ptr = OpAccessChain %ip_v4f32 %in_color %tessc2_invocation_id\n"
4514 "%tessc2_in_position_ptr = OpAccessChain %ip_v4f32 %in_position %tessc2_invocation_id\n"
4515 "%tessc2_in_color_val = OpLoad %v4f32 %tessc2_in_color_ptr\n"
4516 "%tessc2_in_position_val = OpLoad %v4f32 %tessc2_in_position_ptr\n"
4517 "%tessc2_out_color_ptr = OpAccessChain %op_v4f32 %out_color %tessc2_invocation_id\n"
4518 "%tessc2_out_position_ptr = OpAccessChain %op_v4f32 %out_position %tessc2_invocation_id\n"
4519 "%tessc2_transformed_color = OpFSub %v4f32 %cval %tessc2_in_color_val\n"
4520 "%tessc2_transformed_color_a = OpVectorInsertDynamic %v4f32 %tessc2_transformed_color %c_f32_1 %c_i32_3\n"
4521 "OpStore %tessc2_out_color_ptr %tessc2_transformed_color_a\n"
4522 "OpStore %tessc2_out_position_ptr %tessc2_in_position_val\n"
4523 "%tessc2_is_first_invocation = OpIEqual %bool %tessc2_invocation_id %c_i32_0\n"
4524 "OpSelectionMerge %tessc2_merge_label None\n"
4525 "OpBranchConditional %tessc2_is_first_invocation %tessc2_first_invocation %tessc2_merge_label\n"
4526 "%tessc2_first_invocation = OpLabel\n"
4527 "%tessc2_tess_outer_0 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_0\n"
4528 "%tessc2_tess_outer_1 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_1\n"
4529 "%tessc2_tess_outer_2 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_2\n"
4530 "%tessc2_tess_inner = OpAccessChain %op_f32 %gl_TessLevelInner %c_i32_0\n"
4531 "OpStore %tessc2_tess_outer_0 %c_f32_1\n"
4532 "OpStore %tessc2_tess_outer_1 %c_f32_1\n"
4533 "OpStore %tessc2_tess_outer_2 %c_f32_1\n"
4534 "OpStore %tessc2_tess_inner %c_f32_1\n"
4535 "OpBranch %tessc2_merge_label\n"
4536 "%tessc2_merge_label = OpLabel\n"
4537 "OpReturn\n"
4538 "OpFunctionEnd\n";
4539
4540 dst.spirvAsmSources.add("tesse") <<
4541 "OpCapability Tessellation\n"
4542 "OpCapability ClipDistance\n"
4543 "OpCapability CullDistance\n"
4544 "OpMemoryModel Logical GLSL450\n"
4545 "OpEntryPoint TessellationEvaluation %tesse1_main \"tesse1\" %stream %gl_tessCoord %in_position %out_color %in_color \n"
4546 "OpEntryPoint TessellationEvaluation %tesse2_main \"tesse2\" %stream %gl_tessCoord %in_position %out_color %in_color \n"
4547 "OpExecutionMode %tesse1_main Triangles\n"
4548 "OpExecutionMode %tesse2_main Triangles\n"
4549 "OpName %tesse1_main \"tesse1\"\n"
4550 "OpName %tesse2_main \"tesse2\"\n"
4551 "OpName %per_vertex_out \"gl_PerVertex\"\n"
4552 "OpMemberName %per_vertex_out 0 \"gl_Position\"\n"
4553 "OpMemberName %per_vertex_out 1 \"gl_PointSize\"\n"
4554 "OpMemberName %per_vertex_out 2 \"gl_ClipDistance\"\n"
4555 "OpMemberName %per_vertex_out 3 \"gl_CullDistance\"\n"
4556 "OpName %stream \"\"\n"
4557 "OpName %gl_tessCoord \"gl_TessCoord\"\n"
4558 "OpName %in_position \"in_position\"\n"
4559 "OpName %out_color \"out_color\"\n"
4560 "OpName %in_color \"in_color\"\n"
4561 "OpMemberDecorate %per_vertex_out 0 BuiltIn Position\n"
4562 "OpMemberDecorate %per_vertex_out 1 BuiltIn PointSize\n"
4563 "OpMemberDecorate %per_vertex_out 2 BuiltIn ClipDistance\n"
4564 "OpMemberDecorate %per_vertex_out 3 BuiltIn CullDistance\n"
4565 "OpDecorate %per_vertex_out Block\n"
4566 "OpDecorate %gl_tessCoord BuiltIn TessCoord\n"
4567 "OpDecorate %in_position Location 2\n"
4568 "OpDecorate %out_color Location 1\n"
4569 "OpDecorate %in_color Location 1\n"
4570 SPIRV_ASSEMBLY_TYPES
4571 SPIRV_ASSEMBLY_CONSTANTS
4572 SPIRV_ASSEMBLY_ARRAYS
4573 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4574 "%per_vertex_out = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4575 "%op_per_vertex_out = OpTypePointer Output %per_vertex_out\n"
4576 "%stream = OpVariable %op_per_vertex_out Output\n"
4577 "%gl_tessCoord = OpVariable %ip_v3f32 Input\n"
4578 "%in_position = OpVariable %ip_a32v4f32 Input\n"
4579 "%out_color = OpVariable %op_v4f32 Output\n"
4580 "%in_color = OpVariable %ip_a32v4f32 Input\n"
4581
4582 "%tesse1_main = OpFunction %void None %fun\n"
4583 "%tesse1_label = OpLabel\n"
4584 "%tesse1_tc_0_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_0\n"
4585 "%tesse1_tc_1_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_1\n"
4586 "%tesse1_tc_2_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_2\n"
4587 "%tesse1_tc_0 = OpLoad %f32 %tesse1_tc_0_ptr\n"
4588 "%tesse1_tc_1 = OpLoad %f32 %tesse1_tc_1_ptr\n"
4589 "%tesse1_tc_2 = OpLoad %f32 %tesse1_tc_2_ptr\n"
4590 "%tesse1_in_pos_0_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_0\n"
4591 "%tesse1_in_pos_1_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_1\n"
4592 "%tesse1_in_pos_2_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_2\n"
4593 "%tesse1_in_pos_0 = OpLoad %v4f32 %tesse1_in_pos_0_ptr\n"
4594 "%tesse1_in_pos_1 = OpLoad %v4f32 %tesse1_in_pos_1_ptr\n"
4595 "%tesse1_in_pos_2 = OpLoad %v4f32 %tesse1_in_pos_2_ptr\n"
4596 "%tesse1_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_0 %tesse1_in_pos_0\n"
4597 "%tesse1_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_1 %tesse1_in_pos_1\n"
4598 "%tesse1_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_2 %tesse1_in_pos_2\n"
4599 "%tesse1_out_pos_ptr = OpAccessChain %op_v4f32 %stream %c_i32_0\n"
4600 "%tesse1_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %tesse1_in_pos_0_weighted %tesse1_in_pos_1_weighted\n"
4601 "%tesse1_computed_out = OpFAdd %v4f32 %tesse1_in_pos_0_plus_pos_1 %tesse1_in_pos_2_weighted\n"
4602 "OpStore %tesse1_out_pos_ptr %tesse1_computed_out\n"
4603 "%tesse1_in_clr_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4604 "%tesse1_in_clr_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4605 "%tesse1_in_clr_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4606 "%tesse1_in_clr_0 = OpLoad %v4f32 %tesse1_in_clr_0_ptr\n"
4607 "%tesse1_in_clr_1 = OpLoad %v4f32 %tesse1_in_clr_1_ptr\n"
4608 "%tesse1_in_clr_2 = OpLoad %v4f32 %tesse1_in_clr_2_ptr\n"
4609 "%tesse1_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_0 %tesse1_in_clr_0\n"
4610 "%tesse1_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_1 %tesse1_in_clr_1\n"
4611 "%tesse1_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_2 %tesse1_in_clr_2\n"
4612 "%tesse1_in_clr_0_plus_col_1 = OpFAdd %v4f32 %tesse1_in_clr_0_weighted %tesse1_in_clr_1_weighted\n"
4613 "%tesse1_computed_clr = OpFAdd %v4f32 %tesse1_in_clr_0_plus_col_1 %tesse1_in_clr_2_weighted\n"
4614 "OpStore %out_color %tesse1_computed_clr\n"
4615 "OpReturn\n"
4616 "OpFunctionEnd\n"
4617
4618 "%tesse2_main = OpFunction %void None %fun\n"
4619 "%tesse2_label = OpLabel\n"
4620 "%tesse2_tc_0_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_0\n"
4621 "%tesse2_tc_1_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_1\n"
4622 "%tesse2_tc_2_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_2\n"
4623 "%tesse2_tc_0 = OpLoad %f32 %tesse2_tc_0_ptr\n"
4624 "%tesse2_tc_1 = OpLoad %f32 %tesse2_tc_1_ptr\n"
4625 "%tesse2_tc_2 = OpLoad %f32 %tesse2_tc_2_ptr\n"
4626 "%tesse2_in_pos_0_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_0\n"
4627 "%tesse2_in_pos_1_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_1\n"
4628 "%tesse2_in_pos_2_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_2\n"
4629 "%tesse2_in_pos_0 = OpLoad %v4f32 %tesse2_in_pos_0_ptr\n"
4630 "%tesse2_in_pos_1 = OpLoad %v4f32 %tesse2_in_pos_1_ptr\n"
4631 "%tesse2_in_pos_2 = OpLoad %v4f32 %tesse2_in_pos_2_ptr\n"
4632 "%tesse2_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_0 %tesse2_in_pos_0\n"
4633 "%tesse2_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_1 %tesse2_in_pos_1\n"
4634 "%tesse2_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_2 %tesse2_in_pos_2\n"
4635 "%tesse2_out_pos_ptr = OpAccessChain %op_v4f32 %stream %c_i32_0\n"
4636 "%tesse2_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %tesse2_in_pos_0_weighted %tesse2_in_pos_1_weighted\n"
4637 "%tesse2_computed_out = OpFAdd %v4f32 %tesse2_in_pos_0_plus_pos_1 %tesse2_in_pos_2_weighted\n"
4638 "OpStore %tesse2_out_pos_ptr %tesse2_computed_out\n"
4639 "%tesse2_in_clr_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4640 "%tesse2_in_clr_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4641 "%tesse2_in_clr_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4642 "%tesse2_in_clr_0 = OpLoad %v4f32 %tesse2_in_clr_0_ptr\n"
4643 "%tesse2_in_clr_1 = OpLoad %v4f32 %tesse2_in_clr_1_ptr\n"
4644 "%tesse2_in_clr_2 = OpLoad %v4f32 %tesse2_in_clr_2_ptr\n"
4645 "%tesse2_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_0 %tesse2_in_clr_0\n"
4646 "%tesse2_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_1 %tesse2_in_clr_1\n"
4647 "%tesse2_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_2 %tesse2_in_clr_2\n"
4648 "%tesse2_in_clr_0_plus_col_1 = OpFAdd %v4f32 %tesse2_in_clr_0_weighted %tesse2_in_clr_1_weighted\n"
4649 "%tesse2_computed_clr = OpFAdd %v4f32 %tesse2_in_clr_0_plus_col_1 %tesse2_in_clr_2_weighted\n"
4650 "%tesse2_clr_transformed = OpFSub %v4f32 %cval %tesse2_computed_clr\n"
4651 "%tesse2_clr_transformed_a = OpVectorInsertDynamic %v4f32 %tesse2_clr_transformed %c_f32_1 %c_i32_3\n"
4652 "OpStore %out_color %tesse2_clr_transformed_a\n"
4653 "OpReturn\n"
4654 "OpFunctionEnd\n";
4655 }
4656
4657 // Sets up and runs a Vulkan pipeline, then spot-checks the resulting image.
4658 // Feeds the pipeline a set of colored triangles, which then must occur in the
4659 // rendered image. The surface is cleared before executing the pipeline, so
4660 // whatever the shaders draw can be directly spot-checked.
runAndVerifyDefaultPipeline(Context & context,InstanceContext instance)4661 TestStatus runAndVerifyDefaultPipeline (Context& context, InstanceContext instance)
4662 {
4663 const VkDevice vkDevice = context.getDevice();
4664 const DeviceInterface& vk = context.getDeviceInterface();
4665 const VkQueue queue = context.getUniversalQueue();
4666 const deUint32 queueFamilyIndex = context.getUniversalQueueFamilyIndex();
4667 const tcu::UVec2 renderSize (256, 256);
4668 vector<ModuleHandleSp> modules;
4669 map<VkShaderStageFlagBits, VkShaderModule> moduleByStage;
4670 const int testSpecificSeed = 31354125;
4671 const int seed = context.getTestContext().getCommandLine().getBaseSeed() ^ testSpecificSeed;
4672 bool supportsGeometry = false;
4673 bool supportsTessellation = false;
4674 bool hasTessellation = false;
4675
4676 const VkPhysicalDeviceFeatures& features = context.getDeviceFeatures();
4677 supportsGeometry = features.geometryShader == VK_TRUE;
4678 supportsTessellation = features.tessellationShader == VK_TRUE;
4679 hasTessellation = (instance.requiredStages & VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT) ||
4680 (instance.requiredStages & VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT);
4681
4682 if (hasTessellation && !supportsTessellation)
4683 {
4684 throw tcu::NotSupportedError(std::string("Tessellation not supported"));
4685 }
4686
4687 if ((instance.requiredStages & VK_SHADER_STAGE_GEOMETRY_BIT) &&
4688 !supportsGeometry)
4689 {
4690 throw tcu::NotSupportedError(std::string("Geometry not supported"));
4691 }
4692
4693 de::Random(seed).shuffle(instance.inputColors, instance.inputColors+4);
4694 de::Random(seed).shuffle(instance.outputColors, instance.outputColors+4);
4695 const Vec4 vertexData[] =
4696 {
4697 // Upper left corner:
4698 Vec4(-1.0f, -1.0f, 0.0f, 1.0f), instance.inputColors[0].toVec(),
4699 Vec4(-0.5f, -1.0f, 0.0f, 1.0f), instance.inputColors[0].toVec(),
4700 Vec4(-1.0f, -0.5f, 0.0f, 1.0f), instance.inputColors[0].toVec(),
4701
4702 // Upper right corner:
4703 Vec4(+0.5f, -1.0f, 0.0f, 1.0f), instance.inputColors[1].toVec(),
4704 Vec4(+1.0f, -1.0f, 0.0f, 1.0f), instance.inputColors[1].toVec(),
4705 Vec4(+1.0f, -0.5f, 0.0f, 1.0f), instance.inputColors[1].toVec(),
4706
4707 // Lower left corner:
4708 Vec4(-1.0f, +0.5f, 0.0f, 1.0f), instance.inputColors[2].toVec(),
4709 Vec4(-0.5f, +1.0f, 0.0f, 1.0f), instance.inputColors[2].toVec(),
4710 Vec4(-1.0f, +1.0f, 0.0f, 1.0f), instance.inputColors[2].toVec(),
4711
4712 // Lower right corner:
4713 Vec4(+1.0f, +0.5f, 0.0f, 1.0f), instance.inputColors[3].toVec(),
4714 Vec4(+1.0f, +1.0f, 0.0f, 1.0f), instance.inputColors[3].toVec(),
4715 Vec4(+0.5f, +1.0f, 0.0f, 1.0f), instance.inputColors[3].toVec()
4716 };
4717 const size_t singleVertexDataSize = 2 * sizeof(Vec4);
4718 const size_t vertexCount = sizeof(vertexData) / singleVertexDataSize;
4719
4720 const VkBufferCreateInfo vertexBufferParams =
4721 {
4722 VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType;
4723 DE_NULL, // const void* pNext;
4724 0u, // VkBufferCreateFlags flags;
4725 (VkDeviceSize)sizeof(vertexData), // VkDeviceSize size;
4726 VK_BUFFER_USAGE_VERTEX_BUFFER_BIT, // VkBufferUsageFlags usage;
4727 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
4728 1u, // deUint32 queueFamilyCount;
4729 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices;
4730 };
4731 const Unique<VkBuffer> vertexBuffer (createBuffer(vk, vkDevice, &vertexBufferParams));
4732 const UniquePtr<Allocation> vertexBufferMemory (context.getDefaultAllocator().allocate(getBufferMemoryRequirements(vk, vkDevice, *vertexBuffer), MemoryRequirement::HostVisible));
4733
4734 VK_CHECK(vk.bindBufferMemory(vkDevice, *vertexBuffer, vertexBufferMemory->getMemory(), vertexBufferMemory->getOffset()));
4735
4736 const VkDeviceSize imageSizeBytes = (VkDeviceSize)(sizeof(deUint32)*renderSize.x()*renderSize.y());
4737 const VkBufferCreateInfo readImageBufferParams =
4738 {
4739 VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType;
4740 DE_NULL, // const void* pNext;
4741 0u, // VkBufferCreateFlags flags;
4742 imageSizeBytes, // VkDeviceSize size;
4743 VK_BUFFER_USAGE_TRANSFER_DST_BIT, // VkBufferUsageFlags usage;
4744 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
4745 1u, // deUint32 queueFamilyCount;
4746 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices;
4747 };
4748 const Unique<VkBuffer> readImageBuffer (createBuffer(vk, vkDevice, &readImageBufferParams));
4749 const UniquePtr<Allocation> readImageBufferMemory (context.getDefaultAllocator().allocate(getBufferMemoryRequirements(vk, vkDevice, *readImageBuffer), MemoryRequirement::HostVisible));
4750
4751 VK_CHECK(vk.bindBufferMemory(vkDevice, *readImageBuffer, readImageBufferMemory->getMemory(), readImageBufferMemory->getOffset()));
4752
4753 const VkImageCreateInfo imageParams =
4754 {
4755 VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, // VkStructureType sType;
4756 DE_NULL, // const void* pNext;
4757 0u, // VkImageCreateFlags flags;
4758 VK_IMAGE_TYPE_2D, // VkImageType imageType;
4759 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format;
4760 { renderSize.x(), renderSize.y(), 1 }, // VkExtent3D extent;
4761 1u, // deUint32 mipLevels;
4762 1u, // deUint32 arraySize;
4763 VK_SAMPLE_COUNT_1_BIT, // deUint32 samples;
4764 VK_IMAGE_TILING_OPTIMAL, // VkImageTiling tiling;
4765 VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT|VK_IMAGE_USAGE_TRANSFER_SRC_BIT, // VkImageUsageFlags usage;
4766 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
4767 1u, // deUint32 queueFamilyCount;
4768 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices;
4769 VK_IMAGE_LAYOUT_UNDEFINED, // VkImageLayout initialLayout;
4770 };
4771
4772 const Unique<VkImage> image (createImage(vk, vkDevice, &imageParams));
4773 const UniquePtr<Allocation> imageMemory (context.getDefaultAllocator().allocate(getImageMemoryRequirements(vk, vkDevice, *image), MemoryRequirement::Any));
4774
4775 VK_CHECK(vk.bindImageMemory(vkDevice, *image, imageMemory->getMemory(), imageMemory->getOffset()));
4776
4777 const VkAttachmentDescription colorAttDesc =
4778 {
4779 0u, // VkAttachmentDescriptionFlags flags;
4780 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format;
4781 VK_SAMPLE_COUNT_1_BIT, // deUint32 samples;
4782 VK_ATTACHMENT_LOAD_OP_CLEAR, // VkAttachmentLoadOp loadOp;
4783 VK_ATTACHMENT_STORE_OP_STORE, // VkAttachmentStoreOp storeOp;
4784 VK_ATTACHMENT_LOAD_OP_DONT_CARE, // VkAttachmentLoadOp stencilLoadOp;
4785 VK_ATTACHMENT_STORE_OP_DONT_CARE, // VkAttachmentStoreOp stencilStoreOp;
4786 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout initialLayout;
4787 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout finalLayout;
4788 };
4789 const VkAttachmentReference colorAttRef =
4790 {
4791 0u, // deUint32 attachment;
4792 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout layout;
4793 };
4794 const VkSubpassDescription subpassDesc =
4795 {
4796 0u, // VkSubpassDescriptionFlags flags;
4797 VK_PIPELINE_BIND_POINT_GRAPHICS, // VkPipelineBindPoint pipelineBindPoint;
4798 0u, // deUint32 inputCount;
4799 DE_NULL, // const VkAttachmentReference* pInputAttachments;
4800 1u, // deUint32 colorCount;
4801 &colorAttRef, // const VkAttachmentReference* pColorAttachments;
4802 DE_NULL, // const VkAttachmentReference* pResolveAttachments;
4803 DE_NULL, // const VkAttachmentReference* pDepthStencilAttachment;
4804 0u, // deUint32 preserveCount;
4805 DE_NULL, // const VkAttachmentReference* pPreserveAttachments;
4806
4807 };
4808 const VkRenderPassCreateInfo renderPassParams =
4809 {
4810 VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO, // VkStructureType sType;
4811 DE_NULL, // const void* pNext;
4812 (VkRenderPassCreateFlags)0,
4813 1u, // deUint32 attachmentCount;
4814 &colorAttDesc, // const VkAttachmentDescription* pAttachments;
4815 1u, // deUint32 subpassCount;
4816 &subpassDesc, // const VkSubpassDescription* pSubpasses;
4817 0u, // deUint32 dependencyCount;
4818 DE_NULL, // const VkSubpassDependency* pDependencies;
4819 };
4820 const Unique<VkRenderPass> renderPass (createRenderPass(vk, vkDevice, &renderPassParams));
4821
4822 const VkImageViewCreateInfo colorAttViewParams =
4823 {
4824 VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, // VkStructureType sType;
4825 DE_NULL, // const void* pNext;
4826 0u, // VkImageViewCreateFlags flags;
4827 *image, // VkImage image;
4828 VK_IMAGE_VIEW_TYPE_2D, // VkImageViewType viewType;
4829 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format;
4830 {
4831 VK_COMPONENT_SWIZZLE_R,
4832 VK_COMPONENT_SWIZZLE_G,
4833 VK_COMPONENT_SWIZZLE_B,
4834 VK_COMPONENT_SWIZZLE_A
4835 }, // VkChannelMapping channels;
4836 {
4837 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspectFlags aspectMask;
4838 0u, // deUint32 baseMipLevel;
4839 1u, // deUint32 mipLevels;
4840 0u, // deUint32 baseArrayLayer;
4841 1u, // deUint32 arraySize;
4842 }, // VkImageSubresourceRange subresourceRange;
4843 };
4844 const Unique<VkImageView> colorAttView (createImageView(vk, vkDevice, &colorAttViewParams));
4845
4846
4847 // Pipeline layout
4848 const VkPipelineLayoutCreateInfo pipelineLayoutParams =
4849 {
4850 VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, // VkStructureType sType;
4851 DE_NULL, // const void* pNext;
4852 (VkPipelineLayoutCreateFlags)0,
4853 0u, // deUint32 descriptorSetCount;
4854 DE_NULL, // const VkDescriptorSetLayout* pSetLayouts;
4855 0u, // deUint32 pushConstantRangeCount;
4856 DE_NULL, // const VkPushConstantRange* pPushConstantRanges;
4857 };
4858 const Unique<VkPipelineLayout> pipelineLayout (createPipelineLayout(vk, vkDevice, &pipelineLayoutParams));
4859
4860 // Pipeline
4861 vector<VkPipelineShaderStageCreateInfo> shaderStageParams;
4862 // We need these vectors to make sure that information about specialization constants for each stage can outlive createGraphicsPipeline().
4863 vector<vector<VkSpecializationMapEntry> > specConstantEntries;
4864 vector<VkSpecializationInfo> specializationInfos;
4865 createPipelineShaderStages(vk, vkDevice, instance, context, modules, shaderStageParams);
4866
4867 // And we don't want the reallocation of these vectors to invalidate pointers pointing to their contents.
4868 specConstantEntries.reserve(shaderStageParams.size());
4869 specializationInfos.reserve(shaderStageParams.size());
4870
4871 // Patch the specialization info field in PipelineShaderStageCreateInfos.
4872 for (vector<VkPipelineShaderStageCreateInfo>::iterator stageInfo = shaderStageParams.begin(); stageInfo != shaderStageParams.end(); ++stageInfo)
4873 {
4874 const StageToSpecConstantMap::const_iterator stageIt = instance.specConstants.find(stageInfo->stage);
4875
4876 if (stageIt != instance.specConstants.end())
4877 {
4878 const size_t numSpecConstants = stageIt->second.size();
4879 vector<VkSpecializationMapEntry> entries;
4880 VkSpecializationInfo specInfo;
4881
4882 entries.resize(numSpecConstants);
4883
4884 // Only support 32-bit integers as spec constants now. And their constant IDs are numbered sequentially starting from 0.
4885 for (size_t ndx = 0; ndx < numSpecConstants; ++ndx)
4886 {
4887 entries[ndx].constantID = (deUint32)ndx;
4888 entries[ndx].offset = deUint32(ndx * sizeof(deInt32));
4889 entries[ndx].size = sizeof(deInt32);
4890 }
4891
4892 specConstantEntries.push_back(entries);
4893
4894 specInfo.mapEntryCount = (deUint32)numSpecConstants;
4895 specInfo.pMapEntries = specConstantEntries.back().data();
4896 specInfo.dataSize = numSpecConstants * sizeof(deInt32);
4897 specInfo.pData = stageIt->second.data();
4898 specializationInfos.push_back(specInfo);
4899
4900 stageInfo->pSpecializationInfo = &specializationInfos.back();
4901 }
4902 }
4903 const VkPipelineDepthStencilStateCreateInfo depthStencilParams =
4904 {
4905 VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO, // VkStructureType sType;
4906 DE_NULL, // const void* pNext;
4907 (VkPipelineDepthStencilStateCreateFlags)0,
4908 DE_FALSE, // deUint32 depthTestEnable;
4909 DE_FALSE, // deUint32 depthWriteEnable;
4910 VK_COMPARE_OP_ALWAYS, // VkCompareOp depthCompareOp;
4911 DE_FALSE, // deUint32 depthBoundsTestEnable;
4912 DE_FALSE, // deUint32 stencilTestEnable;
4913 {
4914 VK_STENCIL_OP_KEEP, // VkStencilOp stencilFailOp;
4915 VK_STENCIL_OP_KEEP, // VkStencilOp stencilPassOp;
4916 VK_STENCIL_OP_KEEP, // VkStencilOp stencilDepthFailOp;
4917 VK_COMPARE_OP_ALWAYS, // VkCompareOp stencilCompareOp;
4918 0u, // deUint32 stencilCompareMask;
4919 0u, // deUint32 stencilWriteMask;
4920 0u, // deUint32 stencilReference;
4921 }, // VkStencilOpState front;
4922 {
4923 VK_STENCIL_OP_KEEP, // VkStencilOp stencilFailOp;
4924 VK_STENCIL_OP_KEEP, // VkStencilOp stencilPassOp;
4925 VK_STENCIL_OP_KEEP, // VkStencilOp stencilDepthFailOp;
4926 VK_COMPARE_OP_ALWAYS, // VkCompareOp stencilCompareOp;
4927 0u, // deUint32 stencilCompareMask;
4928 0u, // deUint32 stencilWriteMask;
4929 0u, // deUint32 stencilReference;
4930 }, // VkStencilOpState back;
4931 -1.0f, // float minDepthBounds;
4932 +1.0f, // float maxDepthBounds;
4933 };
4934 const VkViewport viewport0 =
4935 {
4936 0.0f, // float originX;
4937 0.0f, // float originY;
4938 (float)renderSize.x(), // float width;
4939 (float)renderSize.y(), // float height;
4940 0.0f, // float minDepth;
4941 1.0f, // float maxDepth;
4942 };
4943 const VkRect2D scissor0 =
4944 {
4945 {
4946 0u, // deInt32 x;
4947 0u, // deInt32 y;
4948 }, // VkOffset2D offset;
4949 {
4950 renderSize.x(), // deInt32 width;
4951 renderSize.y(), // deInt32 height;
4952 }, // VkExtent2D extent;
4953 };
4954 const VkPipelineViewportStateCreateInfo viewportParams =
4955 {
4956 VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO, // VkStructureType sType;
4957 DE_NULL, // const void* pNext;
4958 (VkPipelineViewportStateCreateFlags)0,
4959 1u, // deUint32 viewportCount;
4960 &viewport0,
4961 1u,
4962 &scissor0
4963 };
4964 const VkSampleMask sampleMask = ~0u;
4965 const VkPipelineMultisampleStateCreateInfo multisampleParams =
4966 {
4967 VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO, // VkStructureType sType;
4968 DE_NULL, // const void* pNext;
4969 (VkPipelineMultisampleStateCreateFlags)0,
4970 VK_SAMPLE_COUNT_1_BIT, // VkSampleCountFlagBits rasterSamples;
4971 DE_FALSE, // deUint32 sampleShadingEnable;
4972 0.0f, // float minSampleShading;
4973 &sampleMask, // const VkSampleMask* pSampleMask;
4974 DE_FALSE, // VkBool32 alphaToCoverageEnable;
4975 DE_FALSE, // VkBool32 alphaToOneEnable;
4976 };
4977 const VkPipelineRasterizationStateCreateInfo rasterParams =
4978 {
4979 VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO, // VkStructureType sType;
4980 DE_NULL, // const void* pNext;
4981 (VkPipelineRasterizationStateCreateFlags)0,
4982 DE_TRUE, // deUint32 depthClipEnable;
4983 DE_FALSE, // deUint32 rasterizerDiscardEnable;
4984 VK_POLYGON_MODE_FILL, // VkFillMode fillMode;
4985 VK_CULL_MODE_NONE, // VkCullMode cullMode;
4986 VK_FRONT_FACE_COUNTER_CLOCKWISE, // VkFrontFace frontFace;
4987 VK_FALSE, // VkBool32 depthBiasEnable;
4988 0.0f, // float depthBias;
4989 0.0f, // float depthBiasClamp;
4990 0.0f, // float slopeScaledDepthBias;
4991 1.0f, // float lineWidth;
4992 };
4993 const VkPrimitiveTopology topology = hasTessellation? VK_PRIMITIVE_TOPOLOGY_PATCH_LIST: VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
4994 const VkPipelineInputAssemblyStateCreateInfo inputAssemblyParams =
4995 {
4996 VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO, // VkStructureType sType;
4997 DE_NULL, // const void* pNext;
4998 (VkPipelineInputAssemblyStateCreateFlags)0,
4999 topology, // VkPrimitiveTopology topology;
5000 DE_FALSE, // deUint32 primitiveRestartEnable;
5001 };
5002 const VkVertexInputBindingDescription vertexBinding0 =
5003 {
5004 0u, // deUint32 binding;
5005 deUint32(singleVertexDataSize), // deUint32 strideInBytes;
5006 VK_VERTEX_INPUT_RATE_VERTEX // VkVertexInputStepRate stepRate;
5007 };
5008 const VkVertexInputAttributeDescription vertexAttrib0[2] =
5009 {
5010 {
5011 0u, // deUint32 location;
5012 0u, // deUint32 binding;
5013 VK_FORMAT_R32G32B32A32_SFLOAT, // VkFormat format;
5014 0u // deUint32 offsetInBytes;
5015 },
5016 {
5017 1u, // deUint32 location;
5018 0u, // deUint32 binding;
5019 VK_FORMAT_R32G32B32A32_SFLOAT, // VkFormat format;
5020 sizeof(Vec4), // deUint32 offsetInBytes;
5021 }
5022 };
5023
5024 const VkPipelineVertexInputStateCreateInfo vertexInputStateParams =
5025 {
5026 VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO, // VkStructureType sType;
5027 DE_NULL, // const void* pNext;
5028 (VkPipelineVertexInputStateCreateFlags)0,
5029 1u, // deUint32 bindingCount;
5030 &vertexBinding0, // const VkVertexInputBindingDescription* pVertexBindingDescriptions;
5031 2u, // deUint32 attributeCount;
5032 vertexAttrib0, // const VkVertexInputAttributeDescription* pVertexAttributeDescriptions;
5033 };
5034 const VkPipelineColorBlendAttachmentState attBlendParams =
5035 {
5036 DE_FALSE, // deUint32 blendEnable;
5037 VK_BLEND_FACTOR_ONE, // VkBlend srcBlendColor;
5038 VK_BLEND_FACTOR_ZERO, // VkBlend destBlendColor;
5039 VK_BLEND_OP_ADD, // VkBlendOp blendOpColor;
5040 VK_BLEND_FACTOR_ONE, // VkBlend srcBlendAlpha;
5041 VK_BLEND_FACTOR_ZERO, // VkBlend destBlendAlpha;
5042 VK_BLEND_OP_ADD, // VkBlendOp blendOpAlpha;
5043 (VK_COLOR_COMPONENT_R_BIT|
5044 VK_COLOR_COMPONENT_G_BIT|
5045 VK_COLOR_COMPONENT_B_BIT|
5046 VK_COLOR_COMPONENT_A_BIT), // VkChannelFlags channelWriteMask;
5047 };
5048 const VkPipelineColorBlendStateCreateInfo blendParams =
5049 {
5050 VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO, // VkStructureType sType;
5051 DE_NULL, // const void* pNext;
5052 (VkPipelineColorBlendStateCreateFlags)0,
5053 DE_FALSE, // VkBool32 logicOpEnable;
5054 VK_LOGIC_OP_COPY, // VkLogicOp logicOp;
5055 1u, // deUint32 attachmentCount;
5056 &attBlendParams, // const VkPipelineColorBlendAttachmentState* pAttachments;
5057 { 0.0f, 0.0f, 0.0f, 0.0f }, // float blendConst[4];
5058 };
5059 const VkPipelineTessellationStateCreateInfo tessellationState =
5060 {
5061 VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO,
5062 DE_NULL,
5063 (VkPipelineTessellationStateCreateFlags)0,
5064 3u
5065 };
5066
5067 const VkPipelineTessellationStateCreateInfo* tessellationInfo = hasTessellation ? &tessellationState: DE_NULL;
5068 const VkGraphicsPipelineCreateInfo pipelineParams =
5069 {
5070 VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, // VkStructureType sType;
5071 DE_NULL, // const void* pNext;
5072 0u, // VkPipelineCreateFlags flags;
5073 (deUint32)shaderStageParams.size(), // deUint32 stageCount;
5074 &shaderStageParams[0], // const VkPipelineShaderStageCreateInfo* pStages;
5075 &vertexInputStateParams, // const VkPipelineVertexInputStateCreateInfo* pVertexInputState;
5076 &inputAssemblyParams, // const VkPipelineInputAssemblyStateCreateInfo* pInputAssemblyState;
5077 tessellationInfo, // const VkPipelineTessellationStateCreateInfo* pTessellationState;
5078 &viewportParams, // const VkPipelineViewportStateCreateInfo* pViewportState;
5079 &rasterParams, // const VkPipelineRasterStateCreateInfo* pRasterState;
5080 &multisampleParams, // const VkPipelineMultisampleStateCreateInfo* pMultisampleState;
5081 &depthStencilParams, // const VkPipelineDepthStencilStateCreateInfo* pDepthStencilState;
5082 &blendParams, // const VkPipelineColorBlendStateCreateInfo* pColorBlendState;
5083 (const VkPipelineDynamicStateCreateInfo*)DE_NULL, // const VkPipelineDynamicStateCreateInfo* pDynamicState;
5084 *pipelineLayout, // VkPipelineLayout layout;
5085 *renderPass, // VkRenderPass renderPass;
5086 0u, // deUint32 subpass;
5087 DE_NULL, // VkPipeline basePipelineHandle;
5088 0u, // deInt32 basePipelineIndex;
5089 };
5090
5091 const Unique<VkPipeline> pipeline (createGraphicsPipeline(vk, vkDevice, DE_NULL, &pipelineParams));
5092
5093 // Framebuffer
5094 const VkFramebufferCreateInfo framebufferParams =
5095 {
5096 VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO, // VkStructureType sType;
5097 DE_NULL, // const void* pNext;
5098 (VkFramebufferCreateFlags)0,
5099 *renderPass, // VkRenderPass renderPass;
5100 1u, // deUint32 attachmentCount;
5101 &*colorAttView, // const VkImageView* pAttachments;
5102 (deUint32)renderSize.x(), // deUint32 width;
5103 (deUint32)renderSize.y(), // deUint32 height;
5104 1u, // deUint32 layers;
5105 };
5106 const Unique<VkFramebuffer> framebuffer (createFramebuffer(vk, vkDevice, &framebufferParams));
5107
5108 const VkCommandPoolCreateInfo cmdPoolParams =
5109 {
5110 VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO, // VkStructureType sType;
5111 DE_NULL, // const void* pNext;
5112 VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, // VkCmdPoolCreateFlags flags;
5113 queueFamilyIndex, // deUint32 queueFamilyIndex;
5114 };
5115 const Unique<VkCommandPool> cmdPool (createCommandPool(vk, vkDevice, &cmdPoolParams));
5116
5117 // Command buffer
5118 const VkCommandBufferAllocateInfo cmdBufParams =
5119 {
5120 VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO, // VkStructureType sType;
5121 DE_NULL, // const void* pNext;
5122 *cmdPool, // VkCmdPool pool;
5123 VK_COMMAND_BUFFER_LEVEL_PRIMARY, // VkCmdBufferLevel level;
5124 1u, // deUint32 count;
5125 };
5126 const Unique<VkCommandBuffer> cmdBuf (allocateCommandBuffer(vk, vkDevice, &cmdBufParams));
5127
5128 const VkCommandBufferBeginInfo cmdBufBeginParams =
5129 {
5130 VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, // VkStructureType sType;
5131 DE_NULL, // const void* pNext;
5132 (VkCommandBufferUsageFlags)0,
5133 (const VkCommandBufferInheritanceInfo*)DE_NULL,
5134 };
5135
5136 // Record commands
5137 VK_CHECK(vk.beginCommandBuffer(*cmdBuf, &cmdBufBeginParams));
5138
5139 {
5140 const VkMemoryBarrier vertFlushBarrier =
5141 {
5142 VK_STRUCTURE_TYPE_MEMORY_BARRIER, // VkStructureType sType;
5143 DE_NULL, // const void* pNext;
5144 VK_ACCESS_HOST_WRITE_BIT, // VkMemoryOutputFlags outputMask;
5145 VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT, // VkMemoryInputFlags inputMask;
5146 };
5147 const VkImageMemoryBarrier colorAttBarrier =
5148 {
5149 VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType;
5150 DE_NULL, // const void* pNext;
5151 0u, // VkMemoryOutputFlags outputMask;
5152 VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, // VkMemoryInputFlags inputMask;
5153 VK_IMAGE_LAYOUT_UNDEFINED, // VkImageLayout oldLayout;
5154 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout newLayout;
5155 queueFamilyIndex, // deUint32 srcQueueFamilyIndex;
5156 queueFamilyIndex, // deUint32 destQueueFamilyIndex;
5157 *image, // VkImage image;
5158 {
5159 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspect aspect;
5160 0u, // deUint32 baseMipLevel;
5161 1u, // deUint32 mipLevels;
5162 0u, // deUint32 baseArraySlice;
5163 1u, // deUint32 arraySize;
5164 } // VkImageSubresourceRange subresourceRange;
5165 };
5166 vk.cmdPipelineBarrier(*cmdBuf, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, (VkDependencyFlags)0, 1, &vertFlushBarrier, 0, (const VkBufferMemoryBarrier*)DE_NULL, 1, &colorAttBarrier);
5167 }
5168
5169 {
5170 const VkClearValue clearValue = makeClearValueColorF32(0.125f, 0.25f, 0.75f, 1.0f);
5171 const VkRenderPassBeginInfo passBeginParams =
5172 {
5173 VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO, // VkStructureType sType;
5174 DE_NULL, // const void* pNext;
5175 *renderPass, // VkRenderPass renderPass;
5176 *framebuffer, // VkFramebuffer framebuffer;
5177 { { 0, 0 }, { renderSize.x(), renderSize.y() } }, // VkRect2D renderArea;
5178 1u, // deUint32 clearValueCount;
5179 &clearValue, // const VkClearValue* pClearValues;
5180 };
5181 vk.cmdBeginRenderPass(*cmdBuf, &passBeginParams, VK_SUBPASS_CONTENTS_INLINE);
5182 }
5183
5184 vk.cmdBindPipeline(*cmdBuf, VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline);
5185 {
5186 const VkDeviceSize bindingOffset = 0;
5187 vk.cmdBindVertexBuffers(*cmdBuf, 0u, 1u, &vertexBuffer.get(), &bindingOffset);
5188 }
5189 vk.cmdDraw(*cmdBuf, deUint32(vertexCount), 1u /*run pipeline once*/, 0u /*first vertex*/, 0u /*first instanceIndex*/);
5190 vk.cmdEndRenderPass(*cmdBuf);
5191
5192 {
5193 const VkImageMemoryBarrier renderFinishBarrier =
5194 {
5195 VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType;
5196 DE_NULL, // const void* pNext;
5197 VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, // VkMemoryOutputFlags outputMask;
5198 VK_ACCESS_TRANSFER_READ_BIT, // VkMemoryInputFlags inputMask;
5199 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout oldLayout;
5200 VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, // VkImageLayout newLayout;
5201 queueFamilyIndex, // deUint32 srcQueueFamilyIndex;
5202 queueFamilyIndex, // deUint32 destQueueFamilyIndex;
5203 *image, // VkImage image;
5204 {
5205 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspectFlags aspectMask;
5206 0u, // deUint32 baseMipLevel;
5207 1u, // deUint32 mipLevels;
5208 0u, // deUint32 baseArraySlice;
5209 1u, // deUint32 arraySize;
5210 } // VkImageSubresourceRange subresourceRange;
5211 };
5212 vk.cmdPipelineBarrier(*cmdBuf, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 0, (const VkBufferMemoryBarrier*)DE_NULL, 1, &renderFinishBarrier);
5213 }
5214
5215 {
5216 const VkBufferImageCopy copyParams =
5217 {
5218 (VkDeviceSize)0u, // VkDeviceSize bufferOffset;
5219 (deUint32)renderSize.x(), // deUint32 bufferRowLength;
5220 (deUint32)renderSize.y(), // deUint32 bufferImageHeight;
5221 {
5222 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspect aspect;
5223 0u, // deUint32 mipLevel;
5224 0u, // deUint32 arrayLayer;
5225 1u, // deUint32 arraySize;
5226 }, // VkImageSubresourceCopy imageSubresource;
5227 { 0u, 0u, 0u }, // VkOffset3D imageOffset;
5228 { renderSize.x(), renderSize.y(), 1u } // VkExtent3D imageExtent;
5229 };
5230 vk.cmdCopyImageToBuffer(*cmdBuf, *image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, *readImageBuffer, 1u, ©Params);
5231 }
5232
5233 {
5234 const VkBufferMemoryBarrier copyFinishBarrier =
5235 {
5236 VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, // VkStructureType sType;
5237 DE_NULL, // const void* pNext;
5238 VK_ACCESS_TRANSFER_WRITE_BIT, // VkMemoryOutputFlags outputMask;
5239 VK_ACCESS_HOST_READ_BIT, // VkMemoryInputFlags inputMask;
5240 queueFamilyIndex, // deUint32 srcQueueFamilyIndex;
5241 queueFamilyIndex, // deUint32 destQueueFamilyIndex;
5242 *readImageBuffer, // VkBuffer buffer;
5243 0u, // VkDeviceSize offset;
5244 imageSizeBytes // VkDeviceSize size;
5245 };
5246 vk.cmdPipelineBarrier(*cmdBuf, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, (VkDependencyFlags)0, 0, (const VkMemoryBarrier*)DE_NULL, 1, ©FinishBarrier, 0, (const VkImageMemoryBarrier*)DE_NULL);
5247 }
5248
5249 VK_CHECK(vk.endCommandBuffer(*cmdBuf));
5250
5251 // Upload vertex data
5252 {
5253 const VkMappedMemoryRange range =
5254 {
5255 VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, // VkStructureType sType;
5256 DE_NULL, // const void* pNext;
5257 vertexBufferMemory->getMemory(), // VkDeviceMemory mem;
5258 0, // VkDeviceSize offset;
5259 (VkDeviceSize)sizeof(vertexData), // VkDeviceSize size;
5260 };
5261 void* vertexBufPtr = vertexBufferMemory->getHostPtr();
5262
5263 deMemcpy(vertexBufPtr, &vertexData[0], sizeof(vertexData));
5264 VK_CHECK(vk.flushMappedMemoryRanges(vkDevice, 1u, &range));
5265 }
5266
5267 // Submit & wait for completion
5268 {
5269 const VkFenceCreateInfo fenceParams =
5270 {
5271 VK_STRUCTURE_TYPE_FENCE_CREATE_INFO, // VkStructureType sType;
5272 DE_NULL, // const void* pNext;
5273 0u, // VkFenceCreateFlags flags;
5274 };
5275 const Unique<VkFence> fence (createFence(vk, vkDevice, &fenceParams));
5276 const VkSubmitInfo submitInfo =
5277 {
5278 VK_STRUCTURE_TYPE_SUBMIT_INFO,
5279 DE_NULL,
5280 0u,
5281 (const VkSemaphore*)DE_NULL,
5282 (const VkPipelineStageFlags*)DE_NULL,
5283 1u,
5284 &cmdBuf.get(),
5285 0u,
5286 (const VkSemaphore*)DE_NULL,
5287 };
5288
5289 VK_CHECK(vk.queueSubmit(queue, 1u, &submitInfo, *fence));
5290 VK_CHECK(vk.waitForFences(vkDevice, 1u, &fence.get(), DE_TRUE, ~0ull));
5291 }
5292
5293 const void* imagePtr = readImageBufferMemory->getHostPtr();
5294 const tcu::ConstPixelBufferAccess pixelBuffer(tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNORM_INT8),
5295 renderSize.x(), renderSize.y(), 1, imagePtr);
5296 // Log image
5297 {
5298 const VkMappedMemoryRange range =
5299 {
5300 VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, // VkStructureType sType;
5301 DE_NULL, // const void* pNext;
5302 readImageBufferMemory->getMemory(), // VkDeviceMemory mem;
5303 0, // VkDeviceSize offset;
5304 imageSizeBytes, // VkDeviceSize size;
5305 };
5306
5307 VK_CHECK(vk.invalidateMappedMemoryRanges(vkDevice, 1u, &range));
5308 context.getTestContext().getLog() << TestLog::Image("Result", "Result", pixelBuffer);
5309 }
5310
5311 const RGBA threshold(1, 1, 1, 1);
5312 const RGBA upperLeft(pixelBuffer.getPixel(1, 1));
5313 if (!tcu::compareThreshold(upperLeft, instance.outputColors[0], threshold))
5314 return TestStatus::fail("Upper left corner mismatch");
5315
5316 const RGBA upperRight(pixelBuffer.getPixel(pixelBuffer.getWidth() - 1, 1));
5317 if (!tcu::compareThreshold(upperRight, instance.outputColors[1], threshold))
5318 return TestStatus::fail("Upper right corner mismatch");
5319
5320 const RGBA lowerLeft(pixelBuffer.getPixel(1, pixelBuffer.getHeight() - 1));
5321 if (!tcu::compareThreshold(lowerLeft, instance.outputColors[2], threshold))
5322 return TestStatus::fail("Lower left corner mismatch");
5323
5324 const RGBA lowerRight(pixelBuffer.getPixel(pixelBuffer.getWidth() - 1, pixelBuffer.getHeight() - 1));
5325 if (!tcu::compareThreshold(lowerRight, instance.outputColors[3], threshold))
5326 return TestStatus::fail("Lower right corner mismatch");
5327
5328 return TestStatus::pass("Rendered output matches input");
5329 }
5330
createTestsForAllStages(const std::string & name,const RGBA (& inputColors)[4],const RGBA (& outputColors)[4],const map<string,string> & testCodeFragments,const vector<deInt32> & specConstants,tcu::TestCaseGroup * tests)5331 void createTestsForAllStages (const std::string& name, const RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments, const vector<deInt32>& specConstants, tcu::TestCaseGroup* tests)
5332 {
5333 const ShaderElement vertFragPipelineStages[] =
5334 {
5335 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
5336 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
5337 };
5338
5339 const ShaderElement tessPipelineStages[] =
5340 {
5341 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
5342 ShaderElement("tessc", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
5343 ShaderElement("tesse", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
5344 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
5345 };
5346
5347 const ShaderElement geomPipelineStages[] =
5348 {
5349 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
5350 ShaderElement("geom", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
5351 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
5352 };
5353
5354 StageToSpecConstantMap specConstantMap;
5355
5356 specConstantMap[VK_SHADER_STAGE_VERTEX_BIT] = specConstants;
5357 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_vert", "", addShaderCodeCustomVertex, runAndVerifyDefaultPipeline,
5358 createInstanceContext(vertFragPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5359
5360 specConstantMap.clear();
5361 specConstantMap[VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT] = specConstants;
5362 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_tessc", "", addShaderCodeCustomTessControl, runAndVerifyDefaultPipeline,
5363 createInstanceContext(tessPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5364
5365 specConstantMap.clear();
5366 specConstantMap[VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT] = specConstants;
5367 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_tesse", "", addShaderCodeCustomTessEval, runAndVerifyDefaultPipeline,
5368 createInstanceContext(tessPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5369
5370 specConstantMap.clear();
5371 specConstantMap[VK_SHADER_STAGE_GEOMETRY_BIT] = specConstants;
5372 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_geom", "", addShaderCodeCustomGeometry, runAndVerifyDefaultPipeline,
5373 createInstanceContext(geomPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5374
5375 specConstantMap.clear();
5376 specConstantMap[VK_SHADER_STAGE_FRAGMENT_BIT] = specConstants;
5377 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_frag", "", addShaderCodeCustomFragment, runAndVerifyDefaultPipeline,
5378 createInstanceContext(vertFragPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5379 }
5380
createTestsForAllStages(const std::string & name,const RGBA (& inputColors)[4],const RGBA (& outputColors)[4],const map<string,string> & testCodeFragments,tcu::TestCaseGroup * tests)5381 inline void createTestsForAllStages (const std::string& name, const RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments, tcu::TestCaseGroup* tests)
5382 {
5383 vector<deInt32> noSpecConstants;
5384 createTestsForAllStages(name, inputColors, outputColors, testCodeFragments, noSpecConstants, tests);
5385 }
5386
5387 } // anonymous
5388
createOpSourceTests(tcu::TestContext & testCtx)5389 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
5390 {
5391 struct NameCodePair { string name, code; };
5392 RGBA defaultColors[4];
5393 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
5394 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
5395 map<string, string> fragments = passthruFragments();
5396 const NameCodePair tests[] =
5397 {
5398 {"unknown", "OpSource Unknown 321"},
5399 {"essl", "OpSource ESSL 310"},
5400 {"glsl", "OpSource GLSL 450"},
5401 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
5402 {"opencl_c", "OpSource OpenCL_C 120"},
5403 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
5404 {"file", opsourceGLSLWithFile},
5405 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
5406 // Longest possible source string: SPIR-V limits instructions to 65535
5407 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
5408 // contain 65530 UTF8 characters (one word each) plus one last word
5409 // containing 3 ASCII characters and \0.
5410 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
5411 };
5412
5413 getDefaultColors(defaultColors);
5414 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5415 {
5416 fragments["debug"] = tests[testNdx].code;
5417 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5418 }
5419
5420 return opSourceTests.release();
5421 }
5422
createOpSourceContinuedTests(tcu::TestContext & testCtx)5423 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
5424 {
5425 struct NameCodePair { string name, code; };
5426 RGBA defaultColors[4];
5427 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
5428 map<string, string> fragments = passthruFragments();
5429 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
5430 const NameCodePair tests[] =
5431 {
5432 {"empty", opsource + "OpSourceContinued \"\""},
5433 {"short", opsource + "OpSourceContinued \"abcde\""},
5434 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
5435 // Longest possible source string: SPIR-V limits instructions to 65535
5436 // words, of which the first one is OpSourceContinued/length; the rest
5437 // will contain 65533 UTF8 characters (one word each) plus one last word
5438 // containing 3 ASCII characters and \0.
5439 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
5440 };
5441
5442 getDefaultColors(defaultColors);
5443 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5444 {
5445 fragments["debug"] = tests[testNdx].code;
5446 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5447 }
5448
5449 return opSourceTests.release();
5450 }
5451
createOpNoLineTests(tcu::TestContext & testCtx)5452 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
5453 {
5454 RGBA defaultColors[4];
5455 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
5456 map<string, string> fragments;
5457 getDefaultColors(defaultColors);
5458 fragments["debug"] =
5459 "%name = OpString \"name\"\n";
5460
5461 fragments["pre_main"] =
5462 "OpNoLine\n"
5463 "OpNoLine\n"
5464 "OpLine %name 1 1\n"
5465 "OpNoLine\n"
5466 "OpLine %name 1 1\n"
5467 "OpLine %name 1 1\n"
5468 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5469 "OpNoLine\n"
5470 "OpLine %name 1 1\n"
5471 "OpNoLine\n"
5472 "OpLine %name 1 1\n"
5473 "OpLine %name 1 1\n"
5474 "%second_param1 = OpFunctionParameter %v4f32\n"
5475 "OpNoLine\n"
5476 "OpNoLine\n"
5477 "%label_secondfunction = OpLabel\n"
5478 "OpNoLine\n"
5479 "OpReturnValue %second_param1\n"
5480 "OpFunctionEnd\n"
5481 "OpNoLine\n"
5482 "OpNoLine\n";
5483
5484 fragments["testfun"] =
5485 // A %test_code function that returns its argument unchanged.
5486 "OpNoLine\n"
5487 "OpNoLine\n"
5488 "OpLine %name 1 1\n"
5489 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5490 "OpNoLine\n"
5491 "%param1 = OpFunctionParameter %v4f32\n"
5492 "OpNoLine\n"
5493 "OpNoLine\n"
5494 "%label_testfun = OpLabel\n"
5495 "OpNoLine\n"
5496 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5497 "OpReturnValue %val1\n"
5498 "OpFunctionEnd\n"
5499 "OpLine %name 1 1\n"
5500 "OpNoLine\n";
5501
5502 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
5503
5504 return opLineTests.release();
5505 }
5506
5507
createOpLineTests(tcu::TestContext & testCtx)5508 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
5509 {
5510 RGBA defaultColors[4];
5511 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
5512 map<string, string> fragments;
5513 std::vector<std::pair<std::string, std::string> > problemStrings;
5514
5515 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
5516 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
5517 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
5518 getDefaultColors(defaultColors);
5519
5520 fragments["debug"] =
5521 "%other_name = OpString \"other_name\"\n";
5522
5523 fragments["pre_main"] =
5524 "OpLine %file_name 32 0\n"
5525 "OpLine %file_name 32 32\n"
5526 "OpLine %file_name 32 40\n"
5527 "OpLine %other_name 32 40\n"
5528 "OpLine %other_name 0 100\n"
5529 "OpLine %other_name 0 4294967295\n"
5530 "OpLine %other_name 4294967295 0\n"
5531 "OpLine %other_name 32 40\n"
5532 "OpLine %file_name 0 0\n"
5533 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5534 "OpLine %file_name 1 0\n"
5535 "%second_param1 = OpFunctionParameter %v4f32\n"
5536 "OpLine %file_name 1 3\n"
5537 "OpLine %file_name 1 2\n"
5538 "%label_secondfunction = OpLabel\n"
5539 "OpLine %file_name 0 2\n"
5540 "OpReturnValue %second_param1\n"
5541 "OpFunctionEnd\n"
5542 "OpLine %file_name 0 2\n"
5543 "OpLine %file_name 0 2\n";
5544
5545 fragments["testfun"] =
5546 // A %test_code function that returns its argument unchanged.
5547 "OpLine %file_name 1 0\n"
5548 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5549 "OpLine %file_name 16 330\n"
5550 "%param1 = OpFunctionParameter %v4f32\n"
5551 "OpLine %file_name 14 442\n"
5552 "%label_testfun = OpLabel\n"
5553 "OpLine %file_name 11 1024\n"
5554 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5555 "OpLine %file_name 2 97\n"
5556 "OpReturnValue %val1\n"
5557 "OpFunctionEnd\n"
5558 "OpLine %file_name 5 32\n";
5559
5560 for (size_t i = 0; i < problemStrings.size(); ++i)
5561 {
5562 map<string, string> testFragments = fragments;
5563 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
5564 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
5565 }
5566
5567 return opLineTests.release();
5568 }
5569
createOpConstantNullTests(tcu::TestContext & testCtx)5570 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
5571 {
5572 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
5573 RGBA colors[4];
5574
5575
5576 const char functionStart[] =
5577 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5578 "%param1 = OpFunctionParameter %v4f32\n"
5579 "%lbl = OpLabel\n";
5580
5581 const char functionEnd[] =
5582 "OpReturnValue %transformed_param\n"
5583 "OpFunctionEnd\n";
5584
5585 struct NameConstantsCode
5586 {
5587 string name;
5588 string constants;
5589 string code;
5590 };
5591
5592 NameConstantsCode tests[] =
5593 {
5594 {
5595 "vec4",
5596 "%cnull = OpConstantNull %v4f32\n",
5597 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
5598 },
5599 {
5600 "float",
5601 "%cnull = OpConstantNull %f32\n",
5602 "%vp = OpVariable %fp_v4f32 Function\n"
5603 "%v = OpLoad %v4f32 %vp\n"
5604 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
5605 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
5606 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
5607 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
5608 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
5609 },
5610 {
5611 "bool",
5612 "%cnull = OpConstantNull %bool\n",
5613 "%v = OpVariable %fp_v4f32 Function\n"
5614 " OpStore %v %param1\n"
5615 " OpSelectionMerge %false_label None\n"
5616 " OpBranchConditional %cnull %true_label %false_label\n"
5617 "%true_label = OpLabel\n"
5618 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
5619 " OpBranch %false_label\n"
5620 "%false_label = OpLabel\n"
5621 "%transformed_param = OpLoad %v4f32 %v\n"
5622 },
5623 {
5624 "i32",
5625 "%cnull = OpConstantNull %i32\n",
5626 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
5627 "%b = OpIEqual %bool %cnull %c_i32_0\n"
5628 " OpSelectionMerge %false_label None\n"
5629 " OpBranchConditional %b %true_label %false_label\n"
5630 "%true_label = OpLabel\n"
5631 " OpStore %v %param1\n"
5632 " OpBranch %false_label\n"
5633 "%false_label = OpLabel\n"
5634 "%transformed_param = OpLoad %v4f32 %v\n"
5635 },
5636 {
5637 "struct",
5638 "%stype = OpTypeStruct %f32 %v4f32\n"
5639 "%fp_stype = OpTypePointer Function %stype\n"
5640 "%cnull = OpConstantNull %stype\n",
5641 "%v = OpVariable %fp_stype Function %cnull\n"
5642 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
5643 "%f_val = OpLoad %v4f32 %f\n"
5644 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
5645 },
5646 {
5647 "array",
5648 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
5649 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
5650 "%cnull = OpConstantNull %a4_v4f32\n",
5651 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
5652 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5653 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
5654 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
5655 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
5656 "%f_val = OpLoad %v4f32 %f\n"
5657 "%f1_val = OpLoad %v4f32 %f1\n"
5658 "%f2_val = OpLoad %v4f32 %f2\n"
5659 "%f3_val = OpLoad %v4f32 %f3\n"
5660 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
5661 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
5662 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
5663 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
5664 },
5665 {
5666 "matrix",
5667 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5668 "%cnull = OpConstantNull %mat4x4_f32\n",
5669 // Our null matrix * any vector should result in a zero vector.
5670 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
5671 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
5672 }
5673 };
5674
5675 getHalfColorsFullAlpha(colors);
5676
5677 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5678 {
5679 map<string, string> fragments;
5680 fragments["pre_main"] = tests[testNdx].constants;
5681 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5682 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
5683 }
5684 return opConstantNullTests.release();
5685 }
createOpConstantCompositeTests(tcu::TestContext & testCtx)5686 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
5687 {
5688 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
5689 RGBA inputColors[4];
5690 RGBA outputColors[4];
5691
5692
5693 const char functionStart[] =
5694 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5695 "%param1 = OpFunctionParameter %v4f32\n"
5696 "%lbl = OpLabel\n";
5697
5698 const char functionEnd[] =
5699 "OpReturnValue %transformed_param\n"
5700 "OpFunctionEnd\n";
5701
5702 struct NameConstantsCode
5703 {
5704 string name;
5705 string constants;
5706 string code;
5707 };
5708
5709 NameConstantsCode tests[] =
5710 {
5711 {
5712 "vec4",
5713
5714 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
5715 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
5716 },
5717 {
5718 "struct",
5719
5720 "%stype = OpTypeStruct %v4f32 %f32\n"
5721 "%fp_stype = OpTypePointer Function %stype\n"
5722 "%f32_n_1 = OpConstant %f32 -1.0\n"
5723 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5724 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
5725 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
5726
5727 "%v = OpVariable %fp_stype Function %cval\n"
5728 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5729 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
5730 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
5731 "%f32_val = OpLoad %f32 %f32_ptr\n"
5732 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
5733 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
5734 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
5735 },
5736 {
5737 // [1|0|0|0.5] [x] = x + 0.5
5738 // [0|1|0|0.5] [y] = y + 0.5
5739 // [0|0|1|0.5] [z] = z + 0.5
5740 // [0|0|0|1 ] [1] = 1
5741 "matrix",
5742
5743 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5744 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
5745 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
5746 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
5747 "%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"
5748 "%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",
5749
5750 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
5751 },
5752 {
5753 "array",
5754
5755 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5756 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5757 "%f32_n_1 = OpConstant %f32 -1.0\n"
5758 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5759 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
5760
5761 "%v = OpVariable %fp_a4f32 Function %carr\n"
5762 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
5763 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
5764 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5765 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
5766 "%f_val = OpLoad %f32 %f\n"
5767 "%f1_val = OpLoad %f32 %f1\n"
5768 "%f2_val = OpLoad %f32 %f2\n"
5769 "%f3_val = OpLoad %f32 %f3\n"
5770 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
5771 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
5772 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
5773 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
5774 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5775 },
5776 {
5777 //
5778 // [
5779 // {
5780 // 0.0,
5781 // [ 1.0, 1.0, 1.0, 1.0]
5782 // },
5783 // {
5784 // 1.0,
5785 // [ 0.0, 0.5, 0.0, 0.0]
5786 // }, // ^^^
5787 // {
5788 // 0.0,
5789 // [ 1.0, 1.0, 1.0, 1.0]
5790 // }
5791 // ]
5792 "array_of_struct_of_array",
5793
5794 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5795 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5796 "%stype = OpTypeStruct %f32 %a4f32\n"
5797 "%a3stype = OpTypeArray %stype %c_u32_3\n"
5798 "%fp_a3stype = OpTypePointer Function %a3stype\n"
5799 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
5800 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5801 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
5802 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
5803 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
5804
5805 "%v = OpVariable %fp_a3stype Function %carr\n"
5806 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
5807 "%f_l = OpLoad %f32 %f\n"
5808 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f_l\n"
5809 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5810 }
5811 };
5812
5813 getHalfColorsFullAlpha(inputColors);
5814 outputColors[0] = RGBA(255, 255, 255, 255);
5815 outputColors[1] = RGBA(255, 127, 127, 255);
5816 outputColors[2] = RGBA(127, 255, 127, 255);
5817 outputColors[3] = RGBA(127, 127, 255, 255);
5818
5819 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5820 {
5821 map<string, string> fragments;
5822 fragments["pre_main"] = tests[testNdx].constants;
5823 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5824 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
5825 }
5826 return opConstantCompositeTests.release();
5827 }
5828
createSelectionBlockOrderTests(tcu::TestContext & testCtx)5829 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
5830 {
5831 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
5832 RGBA inputColors[4];
5833 RGBA outputColors[4];
5834 map<string, string> fragments;
5835
5836 // vec4 test_code(vec4 param) {
5837 // vec4 result = param;
5838 // for (int i = 0; i < 4; ++i) {
5839 // if (i == 0) result[i] = 0.;
5840 // else result[i] = 1. - result[i];
5841 // }
5842 // return result;
5843 // }
5844 const char function[] =
5845 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5846 "%param1 = OpFunctionParameter %v4f32\n"
5847 "%lbl = OpLabel\n"
5848 "%iptr = OpVariable %fp_i32 Function\n"
5849 "%result = OpVariable %fp_v4f32 Function\n"
5850 " OpStore %iptr %c_i32_0\n"
5851 " OpStore %result %param1\n"
5852 " OpBranch %loop\n"
5853
5854 // Loop entry block.
5855 "%loop = OpLabel\n"
5856 "%ival = OpLoad %i32 %iptr\n"
5857 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5858 " OpLoopMerge %exit %loop None\n"
5859 " OpBranchConditional %lt_4 %if_entry %exit\n"
5860
5861 // Merge block for loop.
5862 "%exit = OpLabel\n"
5863 "%ret = OpLoad %v4f32 %result\n"
5864 " OpReturnValue %ret\n"
5865
5866 // If-statement entry block.
5867 "%if_entry = OpLabel\n"
5868 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5869 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
5870 " OpSelectionMerge %if_exit None\n"
5871 " OpBranchConditional %eq_0 %if_true %if_false\n"
5872
5873 // False branch for if-statement.
5874 "%if_false = OpLabel\n"
5875 "%val = OpLoad %f32 %loc\n"
5876 "%sub = OpFSub %f32 %c_f32_1 %val\n"
5877 " OpStore %loc %sub\n"
5878 " OpBranch %if_exit\n"
5879
5880 // Merge block for if-statement.
5881 "%if_exit = OpLabel\n"
5882 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5883 " OpStore %iptr %ival_next\n"
5884 " OpBranch %loop\n"
5885
5886 // True branch for if-statement.
5887 "%if_true = OpLabel\n"
5888 " OpStore %loc %c_f32_0\n"
5889 " OpBranch %if_exit\n"
5890
5891 " OpFunctionEnd\n";
5892
5893 fragments["testfun"] = function;
5894
5895 inputColors[0] = RGBA(127, 127, 127, 0);
5896 inputColors[1] = RGBA(127, 0, 0, 0);
5897 inputColors[2] = RGBA(0, 127, 0, 0);
5898 inputColors[3] = RGBA(0, 0, 127, 0);
5899
5900 outputColors[0] = RGBA(0, 128, 128, 255);
5901 outputColors[1] = RGBA(0, 255, 255, 255);
5902 outputColors[2] = RGBA(0, 128, 255, 255);
5903 outputColors[3] = RGBA(0, 255, 128, 255);
5904
5905 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5906
5907 return group.release();
5908 }
5909
createSwitchBlockOrderTests(tcu::TestContext & testCtx)5910 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
5911 {
5912 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
5913 RGBA inputColors[4];
5914 RGBA outputColors[4];
5915 map<string, string> fragments;
5916
5917 const char typesAndConstants[] =
5918 "%c_f32_p2 = OpConstant %f32 0.2\n"
5919 "%c_f32_p4 = OpConstant %f32 0.4\n"
5920 "%c_f32_p6 = OpConstant %f32 0.6\n"
5921 "%c_f32_p8 = OpConstant %f32 0.8\n";
5922
5923 // vec4 test_code(vec4 param) {
5924 // vec4 result = param;
5925 // for (int i = 0; i < 4; ++i) {
5926 // switch (i) {
5927 // case 0: result[i] += .2; break;
5928 // case 1: result[i] += .6; break;
5929 // case 2: result[i] += .4; break;
5930 // case 3: result[i] += .8; break;
5931 // default: break; // unreachable
5932 // }
5933 // }
5934 // return result;
5935 // }
5936 const char function[] =
5937 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5938 "%param1 = OpFunctionParameter %v4f32\n"
5939 "%lbl = OpLabel\n"
5940 "%iptr = OpVariable %fp_i32 Function\n"
5941 "%result = OpVariable %fp_v4f32 Function\n"
5942 " OpStore %iptr %c_i32_0\n"
5943 " OpStore %result %param1\n"
5944 " OpBranch %loop\n"
5945
5946 // Loop entry block.
5947 "%loop = OpLabel\n"
5948 "%ival = OpLoad %i32 %iptr\n"
5949 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5950 " OpLoopMerge %exit %loop None\n"
5951 " OpBranchConditional %lt_4 %switch_entry %exit\n"
5952
5953 // Merge block for loop.
5954 "%exit = OpLabel\n"
5955 "%ret = OpLoad %v4f32 %result\n"
5956 " OpReturnValue %ret\n"
5957
5958 // Switch-statement entry block.
5959 "%switch_entry = OpLabel\n"
5960 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5961 "%val = OpLoad %f32 %loc\n"
5962 " OpSelectionMerge %switch_exit None\n"
5963 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5964
5965 "%case2 = OpLabel\n"
5966 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
5967 " OpStore %loc %addp4\n"
5968 " OpBranch %switch_exit\n"
5969
5970 "%switch_default = OpLabel\n"
5971 " OpUnreachable\n"
5972
5973 "%case3 = OpLabel\n"
5974 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
5975 " OpStore %loc %addp8\n"
5976 " OpBranch %switch_exit\n"
5977
5978 "%case0 = OpLabel\n"
5979 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
5980 " OpStore %loc %addp2\n"
5981 " OpBranch %switch_exit\n"
5982
5983 // Merge block for switch-statement.
5984 "%switch_exit = OpLabel\n"
5985 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5986 " OpStore %iptr %ival_next\n"
5987 " OpBranch %loop\n"
5988
5989 "%case1 = OpLabel\n"
5990 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
5991 " OpStore %loc %addp6\n"
5992 " OpBranch %switch_exit\n"
5993
5994 " OpFunctionEnd\n";
5995
5996 fragments["pre_main"] = typesAndConstants;
5997 fragments["testfun"] = function;
5998
5999 inputColors[0] = RGBA(127, 27, 127, 51);
6000 inputColors[1] = RGBA(127, 0, 0, 51);
6001 inputColors[2] = RGBA(0, 27, 0, 51);
6002 inputColors[3] = RGBA(0, 0, 127, 51);
6003
6004 outputColors[0] = RGBA(178, 180, 229, 255);
6005 outputColors[1] = RGBA(178, 153, 102, 255);
6006 outputColors[2] = RGBA(51, 180, 102, 255);
6007 outputColors[3] = RGBA(51, 153, 229, 255);
6008
6009 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
6010
6011 return group.release();
6012 }
6013
createDecorationGroupTests(tcu::TestContext & testCtx)6014 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
6015 {
6016 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
6017 RGBA inputColors[4];
6018 RGBA outputColors[4];
6019 map<string, string> fragments;
6020
6021 const char decorations[] =
6022 "OpDecorate %array_group ArrayStride 4\n"
6023 "OpDecorate %struct_member_group Offset 0\n"
6024 "%array_group = OpDecorationGroup\n"
6025 "%struct_member_group = OpDecorationGroup\n"
6026
6027 "OpDecorate %group1 RelaxedPrecision\n"
6028 "OpDecorate %group3 RelaxedPrecision\n"
6029 "OpDecorate %group3 Invariant\n"
6030 "OpDecorate %group3 Restrict\n"
6031 "%group0 = OpDecorationGroup\n"
6032 "%group1 = OpDecorationGroup\n"
6033 "%group3 = OpDecorationGroup\n";
6034
6035 const char typesAndConstants[] =
6036 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
6037 "%struct1 = OpTypeStruct %a3f32\n"
6038 "%struct2 = OpTypeStruct %a3f32\n"
6039 "%fp_struct1 = OpTypePointer Function %struct1\n"
6040 "%fp_struct2 = OpTypePointer Function %struct2\n"
6041 "%c_f32_2 = OpConstant %f32 2.\n"
6042 "%c_f32_n2 = OpConstant %f32 -2.\n"
6043
6044 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
6045 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
6046 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
6047 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
6048
6049 const char function[] =
6050 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6051 "%param = OpFunctionParameter %v4f32\n"
6052 "%entry = OpLabel\n"
6053 "%result = OpVariable %fp_v4f32 Function\n"
6054 "%v_struct1 = OpVariable %fp_struct1 Function\n"
6055 "%v_struct2 = OpVariable %fp_struct2 Function\n"
6056 " OpStore %result %param\n"
6057 " OpStore %v_struct1 %c_struct1\n"
6058 " OpStore %v_struct2 %c_struct2\n"
6059 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_2\n"
6060 "%val1 = OpLoad %f32 %ptr1\n"
6061 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
6062 "%val2 = OpLoad %f32 %ptr2\n"
6063 "%addvalues = OpFAdd %f32 %val1 %val2\n"
6064 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
6065 "%val = OpLoad %f32 %ptr\n"
6066 "%addresult = OpFAdd %f32 %addvalues %val\n"
6067 " OpStore %ptr %addresult\n"
6068 "%ret = OpLoad %v4f32 %result\n"
6069 " OpReturnValue %ret\n"
6070 " OpFunctionEnd\n";
6071
6072 struct CaseNameDecoration
6073 {
6074 string name;
6075 string decoration;
6076 };
6077
6078 CaseNameDecoration tests[] =
6079 {
6080 {
6081 "same_decoration_group_on_multiple_types",
6082 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
6083 },
6084 {
6085 "empty_decoration_group",
6086 "OpGroupDecorate %group0 %a3f32\n"
6087 "OpGroupDecorate %group0 %result\n"
6088 },
6089 {
6090 "one_element_decoration_group",
6091 "OpGroupDecorate %array_group %a3f32\n"
6092 },
6093 {
6094 "multiple_elements_decoration_group",
6095 "OpGroupDecorate %group3 %v_struct1\n"
6096 },
6097 {
6098 "multiple_decoration_groups_on_same_variable",
6099 "OpGroupDecorate %group0 %v_struct2\n"
6100 "OpGroupDecorate %group1 %v_struct2\n"
6101 "OpGroupDecorate %group3 %v_struct2\n"
6102 },
6103 {
6104 "same_decoration_group_multiple_times",
6105 "OpGroupDecorate %group1 %addvalues\n"
6106 "OpGroupDecorate %group1 %addvalues\n"
6107 "OpGroupDecorate %group1 %addvalues\n"
6108 },
6109
6110 };
6111
6112 getHalfColorsFullAlpha(inputColors);
6113 getHalfColorsFullAlpha(outputColors);
6114
6115 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
6116 {
6117 fragments["decoration"] = decorations + tests[idx].decoration;
6118 fragments["pre_main"] = typesAndConstants;
6119 fragments["testfun"] = function;
6120
6121 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
6122 }
6123
6124 return group.release();
6125 }
6126
6127 struct SpecConstantTwoIntGraphicsCase
6128 {
6129 const char* caseName;
6130 const char* scDefinition0;
6131 const char* scDefinition1;
6132 const char* scResultType;
6133 const char* scOperation;
6134 deInt32 scActualValue0;
6135 deInt32 scActualValue1;
6136 const char* resultOperation;
6137 RGBA expectedColors[4];
6138
SpecConstantTwoIntGraphicsCasevkt::SpirVAssembly::SpecConstantTwoIntGraphicsCase6139 SpecConstantTwoIntGraphicsCase (const char* name,
6140 const char* definition0,
6141 const char* definition1,
6142 const char* resultType,
6143 const char* operation,
6144 deInt32 value0,
6145 deInt32 value1,
6146 const char* resultOp,
6147 const RGBA (&output)[4])
6148 : caseName (name)
6149 , scDefinition0 (definition0)
6150 , scDefinition1 (definition1)
6151 , scResultType (resultType)
6152 , scOperation (operation)
6153 , scActualValue0 (value0)
6154 , scActualValue1 (value1)
6155 , resultOperation (resultOp)
6156 {
6157 expectedColors[0] = output[0];
6158 expectedColors[1] = output[1];
6159 expectedColors[2] = output[2];
6160 expectedColors[3] = output[3];
6161 }
6162 };
6163
createSpecConstantTests(tcu::TestContext & testCtx)6164 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
6165 {
6166 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
6167 vector<SpecConstantTwoIntGraphicsCase> cases;
6168 RGBA inputColors[4];
6169 RGBA outputColors0[4];
6170 RGBA outputColors1[4];
6171 RGBA outputColors2[4];
6172
6173 const char decorations1[] =
6174 "OpDecorate %sc_0 SpecId 0\n"
6175 "OpDecorate %sc_1 SpecId 1\n";
6176
6177 const char typesAndConstants1[] =
6178 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
6179 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
6180 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
6181
6182 const char function1[] =
6183 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6184 "%param = OpFunctionParameter %v4f32\n"
6185 "%label = OpLabel\n"
6186 "%result = OpVariable %fp_v4f32 Function\n"
6187 " OpStore %result %param\n"
6188 "%gen = ${GEN_RESULT}\n"
6189 "%index = OpIAdd %i32 %gen %c_i32_1\n"
6190 "%loc = OpAccessChain %fp_f32 %result %index\n"
6191 "%val = OpLoad %f32 %loc\n"
6192 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6193 " OpStore %loc %add\n"
6194 "%ret = OpLoad %v4f32 %result\n"
6195 " OpReturnValue %ret\n"
6196 " OpFunctionEnd\n";
6197
6198 inputColors[0] = RGBA(127, 127, 127, 255);
6199 inputColors[1] = RGBA(127, 0, 0, 255);
6200 inputColors[2] = RGBA(0, 127, 0, 255);
6201 inputColors[3] = RGBA(0, 0, 127, 255);
6202
6203 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
6204 outputColors0[0] = RGBA(255, 127, 127, 255);
6205 outputColors0[1] = RGBA(255, 0, 0, 255);
6206 outputColors0[2] = RGBA(128, 127, 0, 255);
6207 outputColors0[3] = RGBA(128, 0, 127, 255);
6208
6209 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
6210 outputColors1[0] = RGBA(127, 255, 127, 255);
6211 outputColors1[1] = RGBA(127, 128, 0, 255);
6212 outputColors1[2] = RGBA(0, 255, 0, 255);
6213 outputColors1[3] = RGBA(0, 128, 127, 255);
6214
6215 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
6216 outputColors2[0] = RGBA(127, 127, 255, 255);
6217 outputColors2[1] = RGBA(127, 0, 128, 255);
6218 outputColors2[2] = RGBA(0, 127, 128, 255);
6219 outputColors2[3] = RGBA(0, 0, 255, 255);
6220
6221 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
6222 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
6223 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
6224
6225 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
6226 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
6227 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
6228 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
6229 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
6230 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6231 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6232 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
6233 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
6234 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
6235 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
6236 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
6237 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
6238 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
6239 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
6240 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
6241 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6242 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
6243 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
6244 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
6245 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6246 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
6247 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
6248 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6249 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6250 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6251 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6252 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
6253 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
6254 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
6255 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
6256 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
6257 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
6258
6259 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
6260 {
6261 map<string, string> specializations;
6262 map<string, string> fragments;
6263 vector<deInt32> specConstants;
6264
6265 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
6266 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
6267 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
6268 specializations["SC_OP"] = cases[caseNdx].scOperation;
6269 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
6270
6271 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
6272 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
6273 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
6274
6275 specConstants.push_back(cases[caseNdx].scActualValue0);
6276 specConstants.push_back(cases[caseNdx].scActualValue1);
6277
6278 createTestsForAllStages(cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants, group.get());
6279 }
6280
6281 const char decorations2[] =
6282 "OpDecorate %sc_0 SpecId 0\n"
6283 "OpDecorate %sc_1 SpecId 1\n"
6284 "OpDecorate %sc_2 SpecId 2\n";
6285
6286 const char typesAndConstants2[] =
6287 "%v3i32 = OpTypeVector %i32 3\n"
6288
6289 "%sc_0 = OpSpecConstant %i32 0\n"
6290 "%sc_1 = OpSpecConstant %i32 0\n"
6291 "%sc_2 = OpSpecConstant %i32 0\n"
6292
6293 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
6294 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
6295 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
6296 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
6297 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %sc_vec3_1 1 0 4\n" // (0, sc_0, sc_1)
6298 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2 5 1 2\n" // (sc_2, sc_0, sc_1)
6299 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
6300 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
6301 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
6302 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
6303 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
6304
6305 const char function2[] =
6306 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6307 "%param = OpFunctionParameter %v4f32\n"
6308 "%label = OpLabel\n"
6309 "%result = OpVariable %fp_v4f32 Function\n"
6310 " OpStore %result %param\n"
6311 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
6312 "%val = OpLoad %f32 %loc\n"
6313 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6314 " OpStore %loc %add\n"
6315 "%ret = OpLoad %v4f32 %result\n"
6316 " OpReturnValue %ret\n"
6317 " OpFunctionEnd\n";
6318
6319 map<string, string> fragments;
6320 vector<deInt32> specConstants;
6321
6322 fragments["decoration"] = decorations2;
6323 fragments["pre_main"] = typesAndConstants2;
6324 fragments["testfun"] = function2;
6325
6326 specConstants.push_back(56789);
6327 specConstants.push_back(-2);
6328 specConstants.push_back(56788);
6329
6330 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
6331
6332 return group.release();
6333 }
6334
createOpPhiTests(tcu::TestContext & testCtx)6335 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
6336 {
6337 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
6338 RGBA inputColors[4];
6339 RGBA outputColors1[4];
6340 RGBA outputColors2[4];
6341 RGBA outputColors3[4];
6342 map<string, string> fragments1;
6343 map<string, string> fragments2;
6344 map<string, string> fragments3;
6345
6346 const char typesAndConstants1[] =
6347 "%c_f32_p2 = OpConstant %f32 0.2\n"
6348 "%c_f32_p4 = OpConstant %f32 0.4\n"
6349 "%c_f32_p5 = OpConstant %f32 0.5\n"
6350 "%c_f32_p8 = OpConstant %f32 0.8\n";
6351
6352 // vec4 test_code(vec4 param) {
6353 // vec4 result = param;
6354 // for (int i = 0; i < 4; ++i) {
6355 // float operand;
6356 // switch (i) {
6357 // case 0: operand = .2; break;
6358 // case 1: operand = .5; break;
6359 // case 2: operand = .4; break;
6360 // case 3: operand = .0; break;
6361 // default: break; // unreachable
6362 // }
6363 // result[i] += operand;
6364 // }
6365 // return result;
6366 // }
6367 const char function1[] =
6368 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6369 "%param1 = OpFunctionParameter %v4f32\n"
6370 "%lbl = OpLabel\n"
6371 "%iptr = OpVariable %fp_i32 Function\n"
6372 "%result = OpVariable %fp_v4f32 Function\n"
6373 " OpStore %iptr %c_i32_0\n"
6374 " OpStore %result %param1\n"
6375 " OpBranch %loop\n"
6376
6377 "%loop = OpLabel\n"
6378 "%ival = OpLoad %i32 %iptr\n"
6379 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6380 " OpLoopMerge %exit %loop None\n"
6381 " OpBranchConditional %lt_4 %entry %exit\n"
6382
6383 "%entry = OpLabel\n"
6384 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6385 "%val = OpLoad %f32 %loc\n"
6386 " OpSelectionMerge %phi None\n"
6387 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6388
6389 "%case0 = OpLabel\n"
6390 " OpBranch %phi\n"
6391 "%case1 = OpLabel\n"
6392 " OpBranch %phi\n"
6393 "%case2 = OpLabel\n"
6394 " OpBranch %phi\n"
6395 "%case3 = OpLabel\n"
6396 " OpBranch %phi\n"
6397
6398 "%default = OpLabel\n"
6399 " OpUnreachable\n"
6400
6401 "%phi = OpLabel\n"
6402 "%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
6403 "%add = OpFAdd %f32 %val %operand\n"
6404 " OpStore %loc %add\n"
6405 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6406 " OpStore %iptr %ival_next\n"
6407 " OpBranch %loop\n"
6408
6409 "%exit = OpLabel\n"
6410 "%ret = OpLoad %v4f32 %result\n"
6411 " OpReturnValue %ret\n"
6412
6413 " OpFunctionEnd\n";
6414
6415 fragments1["pre_main"] = typesAndConstants1;
6416 fragments1["testfun"] = function1;
6417
6418 getHalfColorsFullAlpha(inputColors);
6419
6420 outputColors1[0] = RGBA(178, 255, 229, 255);
6421 outputColors1[1] = RGBA(178, 127, 102, 255);
6422 outputColors1[2] = RGBA(51, 255, 102, 255);
6423 outputColors1[3] = RGBA(51, 127, 229, 255);
6424
6425 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
6426
6427 const char typesAndConstants2[] =
6428 "%c_f32_p2 = OpConstant %f32 0.2\n";
6429
6430 // Add .4 to the second element of the given parameter.
6431 const char function2[] =
6432 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6433 "%param = OpFunctionParameter %v4f32\n"
6434 "%entry = OpLabel\n"
6435 "%result = OpVariable %fp_v4f32 Function\n"
6436 " OpStore %result %param\n"
6437 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6438 "%val = OpLoad %f32 %loc\n"
6439 " OpBranch %phi\n"
6440
6441 "%phi = OpLabel\n"
6442 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
6443 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
6444 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
6445 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
6446 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
6447 " OpLoopMerge %exit %phi None\n"
6448 " OpBranchConditional %still_loop %phi %exit\n"
6449
6450 "%exit = OpLabel\n"
6451 " OpStore %loc %accum\n"
6452 "%ret = OpLoad %v4f32 %result\n"
6453 " OpReturnValue %ret\n"
6454
6455 " OpFunctionEnd\n";
6456
6457 fragments2["pre_main"] = typesAndConstants2;
6458 fragments2["testfun"] = function2;
6459
6460 outputColors2[0] = RGBA(127, 229, 127, 255);
6461 outputColors2[1] = RGBA(127, 102, 0, 255);
6462 outputColors2[2] = RGBA(0, 229, 0, 255);
6463 outputColors2[3] = RGBA(0, 102, 127, 255);
6464
6465 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
6466
6467 const char typesAndConstants3[] =
6468 "%true = OpConstantTrue %bool\n"
6469 "%false = OpConstantFalse %bool\n"
6470 "%c_f32_p2 = OpConstant %f32 0.2\n";
6471
6472 // Swap the second and the third element of the given parameter.
6473 const char function3[] =
6474 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6475 "%param = OpFunctionParameter %v4f32\n"
6476 "%entry = OpLabel\n"
6477 "%result = OpVariable %fp_v4f32 Function\n"
6478 " OpStore %result %param\n"
6479 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6480 "%a_init = OpLoad %f32 %a_loc\n"
6481 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
6482 "%b_init = OpLoad %f32 %b_loc\n"
6483 " OpBranch %phi\n"
6484
6485 "%phi = OpLabel\n"
6486 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
6487 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
6488 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
6489 " OpLoopMerge %exit %phi None\n"
6490 " OpBranchConditional %still_loop %phi %exit\n"
6491
6492 "%exit = OpLabel\n"
6493 " OpStore %a_loc %a_next\n"
6494 " OpStore %b_loc %b_next\n"
6495 "%ret = OpLoad %v4f32 %result\n"
6496 " OpReturnValue %ret\n"
6497
6498 " OpFunctionEnd\n";
6499
6500 fragments3["pre_main"] = typesAndConstants3;
6501 fragments3["testfun"] = function3;
6502
6503 outputColors3[0] = RGBA(127, 127, 127, 255);
6504 outputColors3[1] = RGBA(127, 0, 0, 255);
6505 outputColors3[2] = RGBA(0, 0, 127, 255);
6506 outputColors3[3] = RGBA(0, 127, 0, 255);
6507
6508 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
6509
6510 return group.release();
6511 }
6512
createNoContractionTests(tcu::TestContext & testCtx)6513 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
6514 {
6515 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
6516 RGBA inputColors[4];
6517 RGBA outputColors[4];
6518
6519 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
6520 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
6521 // only have 23-bit fraction.) So it will be rounded to 1. Or 0x1.fffffc. Then the final result is 0 or -0x1p-24.
6522 // On the contrary, the result will be 2^-46, which is a normalized number perfectly representable as 32-bit float.
6523 const char constantsAndTypes[] =
6524 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
6525 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
6526 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
6527 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
6528 "%c_f32_n1pn24 = OpConstant %f32 -0x1p-24\n"
6529 ;
6530
6531 const char function[] =
6532 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6533 "%param = OpFunctionParameter %v4f32\n"
6534 "%label = OpLabel\n"
6535 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
6536 "%var2 = OpVariable %fp_f32 Function\n"
6537 "%red = OpCompositeExtract %f32 %param 0\n"
6538 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
6539 " OpStore %var2 %plus_red\n"
6540 "%val1 = OpLoad %f32 %var1\n"
6541 "%val2 = OpLoad %f32 %var2\n"
6542 "%mul = OpFMul %f32 %val1 %val2\n"
6543 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
6544 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
6545 "%isn1n24 = OpFOrdEqual %bool %add %c_f32_n1pn24\n"
6546 "%success = OpLogicalOr %bool %is0 %isn1n24\n"
6547 "%v4success = OpCompositeConstruct %v4bool %success %success %success %success\n"
6548 "%ret = OpSelect %v4f32 %v4success %c_vec4_0 %c_vec4_1\n"
6549 " OpReturnValue %ret\n"
6550 " OpFunctionEnd\n";
6551
6552 struct CaseNameDecoration
6553 {
6554 string name;
6555 string decoration;
6556 };
6557
6558
6559 CaseNameDecoration tests[] = {
6560 {"multiplication", "OpDecorate %mul NoContraction"},
6561 {"addition", "OpDecorate %add NoContraction"},
6562 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
6563 };
6564
6565 getHalfColorsFullAlpha(inputColors);
6566
6567 for (deUint8 idx = 0; idx < 4; ++idx)
6568 {
6569 inputColors[idx].setRed(0);
6570 outputColors[idx] = RGBA(0, 0, 0, 255);
6571 }
6572
6573 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
6574 {
6575 map<string, string> fragments;
6576
6577 fragments["decoration"] = tests[testNdx].decoration;
6578 fragments["pre_main"] = constantsAndTypes;
6579 fragments["testfun"] = function;
6580
6581 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
6582 }
6583
6584 return group.release();
6585 }
6586
createMemoryAccessTests(tcu::TestContext & testCtx)6587 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
6588 {
6589 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
6590 RGBA colors[4];
6591
6592 const char constantsAndTypes[] =
6593 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
6594 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
6595 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
6596 "%fp_stype = OpTypePointer Function %stype\n";
6597
6598 const char function[] =
6599 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6600 "%param1 = OpFunctionParameter %v4f32\n"
6601 "%lbl = OpLabel\n"
6602 "%v1 = OpVariable %fp_v4f32 Function\n"
6603 "%v2 = OpVariable %fp_a2f32 Function\n"
6604 "%v3 = OpVariable %fp_f32 Function\n"
6605 "%v = OpVariable %fp_stype Function\n"
6606 "%vv = OpVariable %fp_stype Function\n"
6607 "%vvv = OpVariable %fp_f32 Function\n"
6608
6609 " OpStore %v1 %c_v4f32_1_1_1_1\n"
6610 " OpStore %v2 %c_a2f32_1\n"
6611 " OpStore %v3 %c_f32_1\n"
6612
6613 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
6614 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
6615 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
6616 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
6617 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
6618 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
6619
6620 " OpStore %p_v4f32 %v1_v ${access_type}\n"
6621 " OpStore %p_a2f32 %v2_v ${access_type}\n"
6622 " OpStore %p_f32 %v3_v ${access_type}\n"
6623
6624 " OpCopyMemory %vv %v ${access_type}\n"
6625 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
6626
6627 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
6628 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
6629 "%v_f32_3 = OpLoad %f32 %vvv\n"
6630
6631 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
6632 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
6633 " OpReturnValue %ret2\n"
6634 " OpFunctionEnd\n";
6635
6636 struct NameMemoryAccess
6637 {
6638 string name;
6639 string accessType;
6640 };
6641
6642
6643 NameMemoryAccess tests[] =
6644 {
6645 { "none", "" },
6646 { "volatile", "Volatile" },
6647 { "aligned", "Aligned 1" },
6648 { "volatile_aligned", "Volatile|Aligned 1" },
6649 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
6650 { "volatile_nontemporal", "Volatile|Nontemporal" },
6651 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
6652 };
6653
6654 getHalfColorsFullAlpha(colors);
6655
6656 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
6657 {
6658 map<string, string> fragments;
6659 map<string, string> memoryAccess;
6660 memoryAccess["access_type"] = tests[testNdx].accessType;
6661
6662 fragments["pre_main"] = constantsAndTypes;
6663 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
6664 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
6665 }
6666 return memoryAccessTests.release();
6667 }
createOpUndefTests(tcu::TestContext & testCtx)6668 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
6669 {
6670 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
6671 RGBA defaultColors[4];
6672 map<string, string> fragments;
6673 getDefaultColors(defaultColors);
6674
6675 // First, simple cases that don't do anything with the OpUndef result.
6676 fragments["testfun"] =
6677 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6678 "%param1 = OpFunctionParameter %v4f32\n"
6679 "%label_testfun = OpLabel\n"
6680 "%undef = OpUndef %type\n"
6681 "OpReturnValue %param1\n"
6682 "OpFunctionEnd\n"
6683 ;
6684 struct NameCodePair { string name, code; };
6685 const NameCodePair tests[] =
6686 {
6687 {"bool", "%type = OpTypeBool"},
6688 {"vec2uint32", "%type = OpTypeVector %u32 2"},
6689 {"image", "%type = OpTypeImage %f32 2D 0 0 0 1 Unknown"},
6690 {"sampler", "%type = OpTypeSampler"},
6691 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 1 Unknown\n" "%type = OpTypeSampledImage %img"},
6692 {"pointer", "%type = OpTypePointer Function %i32"},
6693 {"runtimearray", "%type = OpTypeRuntimeArray %f32"},
6694 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100"},
6695 {"struct", "%type = OpTypeStruct %f32 %i32 %u32"}};
6696 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
6697 {
6698 fragments["pre_main"] = tests[testNdx].code;
6699 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
6700 }
6701 fragments.clear();
6702
6703 fragments["testfun"] =
6704 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6705 "%param1 = OpFunctionParameter %v4f32\n"
6706 "%label_testfun = OpLabel\n"
6707 "%undef = OpUndef %f32\n"
6708 "%zero = OpFMul %f32 %undef %c_f32_0\n"
6709 "%is_nan = OpIsNan %bool %zero\n" //OpUndef may result in NaN which may turn %zero into Nan.
6710 "%actually_zero = OpSelect %f32 %is_nan %c_f32_0 %zero\n"
6711 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6712 "%b = OpFAdd %f32 %a %actually_zero\n"
6713 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
6714 "OpReturnValue %ret\n"
6715 "OpFunctionEnd\n"
6716 ;
6717 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6718
6719 fragments["testfun"] =
6720 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6721 "%param1 = OpFunctionParameter %v4f32\n"
6722 "%label_testfun = OpLabel\n"
6723 "%undef = OpUndef %i32\n"
6724 "%zero = OpIMul %i32 %undef %c_i32_0\n"
6725 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6726 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6727 "OpReturnValue %ret\n"
6728 "OpFunctionEnd\n"
6729 ;
6730 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6731
6732 fragments["testfun"] =
6733 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6734 "%param1 = OpFunctionParameter %v4f32\n"
6735 "%label_testfun = OpLabel\n"
6736 "%undef = OpUndef %u32\n"
6737 "%zero = OpIMul %u32 %undef %c_i32_0\n"
6738 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6739 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6740 "OpReturnValue %ret\n"
6741 "OpFunctionEnd\n"
6742 ;
6743 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6744
6745 fragments["testfun"] =
6746 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6747 "%param1 = OpFunctionParameter %v4f32\n"
6748 "%label_testfun = OpLabel\n"
6749 "%undef = OpUndef %v4f32\n"
6750 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
6751 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
6752 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
6753 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
6754 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
6755 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6756 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6757 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6758 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6759 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6760 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
6761 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
6762 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
6763 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6764 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6765 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6766 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6767 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6768 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6769 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6770 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6771 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6772 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6773 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6774 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6775 "OpReturnValue %ret\n"
6776 "OpFunctionEnd\n"
6777 ;
6778 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6779
6780 fragments["pre_main"] =
6781 "%v2f32 = OpTypeVector %f32 2\n"
6782 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
6783 fragments["testfun"] =
6784 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6785 "%param1 = OpFunctionParameter %v4f32\n"
6786 "%label_testfun = OpLabel\n"
6787 "%undef = OpUndef %m2x2f32\n"
6788 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
6789 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
6790 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
6791 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
6792 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
6793 "%is_nan_0 = OpIsNan %bool %zero_0\n"
6794 "%is_nan_1 = OpIsNan %bool %zero_1\n"
6795 "%is_nan_2 = OpIsNan %bool %zero_2\n"
6796 "%is_nan_3 = OpIsNan %bool %zero_3\n"
6797 "%actually_zero_0 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_0\n"
6798 "%actually_zero_1 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_1\n"
6799 "%actually_zero_2 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_2\n"
6800 "%actually_zero_3 = OpSelect %f32 %is_nan_0 %c_f32_0 %zero_3\n"
6801 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6802 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6803 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6804 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6805 "%sum_0 = OpFAdd %f32 %param1_0 %actually_zero_0\n"
6806 "%sum_1 = OpFAdd %f32 %param1_1 %actually_zero_1\n"
6807 "%sum_2 = OpFAdd %f32 %param1_2 %actually_zero_2\n"
6808 "%sum_3 = OpFAdd %f32 %param1_3 %actually_zero_3\n"
6809 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6810 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6811 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6812 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6813 "OpReturnValue %ret\n"
6814 "OpFunctionEnd\n"
6815 ;
6816 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
6817
6818 return opUndefTests.release();
6819 }
6820
createOpQuantizeSingleOptionTests(tcu::TestCaseGroup * testCtx)6821 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
6822 {
6823 const RGBA inputColors[4] =
6824 {
6825 RGBA(0, 0, 0, 255),
6826 RGBA(0, 0, 255, 255),
6827 RGBA(0, 255, 0, 255),
6828 RGBA(0, 255, 255, 255)
6829 };
6830
6831 const RGBA expectedColors[4] =
6832 {
6833 RGBA(255, 0, 0, 255),
6834 RGBA(255, 0, 0, 255),
6835 RGBA(255, 0, 0, 255),
6836 RGBA(255, 0, 0, 255)
6837 };
6838
6839 const struct SingleFP16Possibility
6840 {
6841 const char* name;
6842 const char* constant; // Value to assign to %test_constant.
6843 float valueAsFloat;
6844 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
6845 } tests[] =
6846 {
6847 {
6848 "negative",
6849 "-0x1.3p1\n",
6850 -constructNormalizedFloat(1, 0x300000),
6851 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6852 }, // -19
6853 {
6854 "positive",
6855 "0x1.0p7\n",
6856 constructNormalizedFloat(7, 0x000000),
6857 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6858 }, // +128
6859 // SPIR-V requires that OpQuantizeToF16 flushes
6860 // any numbers that would end up denormalized in F16 to zero.
6861 {
6862 "denorm",
6863 "0x0.0006p-126\n",
6864 std::ldexp(1.5f, -140),
6865 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6866 }, // denorm
6867 {
6868 "negative_denorm",
6869 "-0x0.0006p-126\n",
6870 -std::ldexp(1.5f, -140),
6871 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6872 }, // -denorm
6873 {
6874 "too_small",
6875 "0x1.0p-16\n",
6876 std::ldexp(1.0f, -16),
6877 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6878 }, // too small positive
6879 {
6880 "negative_too_small",
6881 "-0x1.0p-32\n",
6882 -std::ldexp(1.0f, -32),
6883 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6884 }, // too small negative
6885 {
6886 "negative_inf",
6887 "-0x1.0p128\n",
6888 -std::ldexp(1.0f, 128),
6889
6890 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6891 "%inf = OpIsInf %bool %c\n"
6892 "%cond = OpLogicalAnd %bool %gz %inf\n"
6893 }, // -inf to -inf
6894 {
6895 "inf",
6896 "0x1.0p128\n",
6897 std::ldexp(1.0f, 128),
6898
6899 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6900 "%inf = OpIsInf %bool %c\n"
6901 "%cond = OpLogicalAnd %bool %gz %inf\n"
6902 }, // +inf to +inf
6903 {
6904 "round_to_negative_inf",
6905 "-0x1.0p32\n",
6906 -std::ldexp(1.0f, 32),
6907
6908 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6909 "%inf = OpIsInf %bool %c\n"
6910 "%cond = OpLogicalAnd %bool %gz %inf\n"
6911 }, // round to -inf
6912 {
6913 "round_to_inf",
6914 "0x1.0p16\n",
6915 std::ldexp(1.0f, 16),
6916
6917 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6918 "%inf = OpIsInf %bool %c\n"
6919 "%cond = OpLogicalAnd %bool %gz %inf\n"
6920 }, // round to +inf
6921 {
6922 "nan",
6923 "0x1.1p128\n",
6924 std::numeric_limits<float>::quiet_NaN(),
6925
6926 // Test for any NaN value, as NaNs are not preserved
6927 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6928 "%cond = OpIsNan %bool %direct_quant\n"
6929 }, // nan
6930 {
6931 "negative_nan",
6932 "-0x1.0001p128\n",
6933 std::numeric_limits<float>::quiet_NaN(),
6934
6935 // Test for any NaN value, as NaNs are not preserved
6936 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6937 "%cond = OpIsNan %bool %direct_quant\n"
6938 } // -nan
6939 };
6940 const char* constants =
6941 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
6942
6943 StringTemplate function (
6944 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6945 "%param1 = OpFunctionParameter %v4f32\n"
6946 "%label_testfun = OpLabel\n"
6947 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6948 "%b = OpFAdd %f32 %test_constant %a\n"
6949 "%c = OpQuantizeToF16 %f32 %b\n"
6950 "${condition}\n"
6951 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6952 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
6953 " OpReturnValue %retval\n"
6954 "OpFunctionEnd\n"
6955 );
6956
6957 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
6958 const char* specConstants =
6959 "%test_constant = OpSpecConstant %f32 0.\n"
6960 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
6961
6962 StringTemplate specConstantFunction(
6963 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6964 "%param1 = OpFunctionParameter %v4f32\n"
6965 "%label_testfun = OpLabel\n"
6966 "${condition}\n"
6967 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
6968 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1\n"
6969 " OpReturnValue %retval\n"
6970 "OpFunctionEnd\n"
6971 );
6972
6973 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6974 {
6975 map<string, string> codeSpecialization;
6976 map<string, string> fragments;
6977 codeSpecialization["condition"] = tests[idx].condition;
6978 fragments["testfun"] = function.specialize(codeSpecialization);
6979 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
6980 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6981 }
6982
6983 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6984 {
6985 map<string, string> codeSpecialization;
6986 map<string, string> fragments;
6987 vector<deInt32> passConstants;
6988 deInt32 specConstant;
6989
6990 codeSpecialization["condition"] = tests[idx].condition;
6991 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
6992 fragments["decoration"] = specDecorations;
6993 fragments["pre_main"] = specConstants;
6994
6995 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
6996 passConstants.push_back(specConstant);
6997
6998 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6999 }
7000 }
7001
createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup * testCtx)7002 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
7003 {
7004 RGBA inputColors[4] = {
7005 RGBA(0, 0, 0, 255),
7006 RGBA(0, 0, 255, 255),
7007 RGBA(0, 255, 0, 255),
7008 RGBA(0, 255, 255, 255)
7009 };
7010
7011 RGBA expectedColors[4] =
7012 {
7013 RGBA(255, 0, 0, 255),
7014 RGBA(255, 0, 0, 255),
7015 RGBA(255, 0, 0, 255),
7016 RGBA(255, 0, 0, 255)
7017 };
7018
7019 struct DualFP16Possibility
7020 {
7021 const char* name;
7022 const char* input;
7023 float inputAsFloat;
7024 const char* possibleOutput1;
7025 const char* possibleOutput2;
7026 } tests[] = {
7027 {
7028 "positive_round_up_or_round_down",
7029 "0x1.3003p8",
7030 constructNormalizedFloat(8, 0x300300),
7031 "0x1.304p8",
7032 "0x1.3p8"
7033 },
7034 {
7035 "negative_round_up_or_round_down",
7036 "-0x1.6008p-7",
7037 -constructNormalizedFloat(-7, 0x600800),
7038 "-0x1.6p-7",
7039 "-0x1.604p-7"
7040 },
7041 {
7042 "carry_bit",
7043 "0x1.01ep2",
7044 constructNormalizedFloat(2, 0x01e000),
7045 "0x1.01cp2",
7046 "0x1.02p2"
7047 },
7048 {
7049 "carry_to_exponent",
7050 "0x1.ffep1",
7051 constructNormalizedFloat(1, 0xffe000),
7052 "0x1.ffcp1",
7053 "0x1.0p2"
7054 },
7055 };
7056 StringTemplate constants (
7057 "%input_const = OpConstant %f32 ${input}\n"
7058 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7059 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7060 );
7061
7062 StringTemplate specConstants (
7063 "%input_const = OpSpecConstant %f32 0.\n"
7064 "%possible_solution1 = OpConstant %f32 ${output1}\n"
7065 "%possible_solution2 = OpConstant %f32 ${output2}\n"
7066 );
7067
7068 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
7069
7070 const char* function =
7071 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7072 "%param1 = OpFunctionParameter %v4f32\n"
7073 "%label_testfun = OpLabel\n"
7074 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7075 // For the purposes of this test we assume that 0.f will always get
7076 // faithfully passed through the pipeline stages.
7077 "%b = OpFAdd %f32 %input_const %a\n"
7078 "%c = OpQuantizeToF16 %f32 %b\n"
7079 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
7080 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
7081 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
7082 "%v4cond = OpCompositeConstruct %v4bool %cond %cond %cond %cond\n"
7083 "%retval = OpSelect %v4f32 %v4cond %c_v4f32_1_0_0_1 %param1"
7084 " OpReturnValue %retval\n"
7085 "OpFunctionEnd\n";
7086
7087 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7088 map<string, string> fragments;
7089 map<string, string> constantSpecialization;
7090
7091 constantSpecialization["input"] = tests[idx].input;
7092 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7093 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7094 fragments["testfun"] = function;
7095 fragments["pre_main"] = constants.specialize(constantSpecialization);
7096 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7097 }
7098
7099 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7100 map<string, string> fragments;
7101 map<string, string> constantSpecialization;
7102 vector<deInt32> passConstants;
7103 deInt32 specConstant;
7104
7105 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7106 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7107 fragments["testfun"] = function;
7108 fragments["decoration"] = specDecorations;
7109 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
7110
7111 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
7112 passConstants.push_back(specConstant);
7113
7114 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7115 }
7116 }
7117
createOpQuantizeTests(tcu::TestContext & testCtx)7118 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
7119 {
7120 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
7121 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
7122 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
7123 return opQuantizeTests.release();
7124 }
7125
7126 struct ShaderPermutation
7127 {
7128 deUint8 vertexPermutation;
7129 deUint8 geometryPermutation;
7130 deUint8 tesscPermutation;
7131 deUint8 tessePermutation;
7132 deUint8 fragmentPermutation;
7133 };
7134
getShaderPermutation(deUint8 inputValue)7135 ShaderPermutation getShaderPermutation(deUint8 inputValue)
7136 {
7137 ShaderPermutation permutation =
7138 {
7139 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
7140 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
7141 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
7142 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
7143 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
7144 };
7145 return permutation;
7146 }
7147
createModuleTests(tcu::TestContext & testCtx)7148 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
7149 {
7150 RGBA defaultColors[4];
7151 RGBA invertedColors[4];
7152 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
7153
7154 const ShaderElement combinedPipeline[] =
7155 {
7156 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
7157 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
7158 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7159 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7160 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
7161 };
7162
7163 getDefaultColors(defaultColors);
7164 getInvertedDefaultColors(invertedColors);
7165 addFunctionCaseWithPrograms<InstanceContext>(moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline, createInstanceContext(combinedPipeline, map<string, string>()));
7166
7167 const char* numbers[] =
7168 {
7169 "1", "2"
7170 };
7171
7172 for (deInt8 idx = 0; idx < 32; ++idx)
7173 {
7174 ShaderPermutation permutation = getShaderPermutation(idx);
7175 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
7176 const ShaderElement pipeline[] =
7177 {
7178 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
7179 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
7180 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7181 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7182 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
7183 };
7184
7185 // If there are an even number of swaps, then it should be no-op.
7186 // If there are an odd number, the color should be flipped.
7187 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
7188 {
7189 addFunctionCaseWithPrograms<InstanceContext>(moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline, createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
7190 }
7191 else
7192 {
7193 addFunctionCaseWithPrograms<InstanceContext>(moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline, createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
7194 }
7195 }
7196 return moduleTests.release();
7197 }
7198
createLoopTests(tcu::TestContext & testCtx)7199 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
7200 {
7201 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
7202 RGBA defaultColors[4];
7203 getDefaultColors(defaultColors);
7204 map<string, string> fragments;
7205 fragments["pre_main"] =
7206 "%c_f32_5 = OpConstant %f32 5.\n";
7207
7208 // A loop with a single block. The Continue Target is the loop block
7209 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
7210 // -- the "continue construct" forms the entire loop.
7211 fragments["testfun"] =
7212 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7213 "%param1 = OpFunctionParameter %v4f32\n"
7214
7215 "%entry = OpLabel\n"
7216 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7217 "OpBranch %loop\n"
7218
7219 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7220 "%loop = OpLabel\n"
7221 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7222 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
7223 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7224 "%val = OpFAdd %f32 %val1 %delta\n"
7225 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
7226 "%count__ = OpISub %i32 %count %c_i32_1\n"
7227 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7228 "OpLoopMerge %exit %loop None\n"
7229 "OpBranchConditional %again %loop %exit\n"
7230
7231 "%exit = OpLabel\n"
7232 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7233 "OpReturnValue %result\n"
7234
7235 "OpFunctionEnd\n"
7236 ;
7237 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
7238
7239 // Body comprised of multiple basic blocks.
7240 const StringTemplate multiBlock(
7241 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7242 "%param1 = OpFunctionParameter %v4f32\n"
7243
7244 "%entry = OpLabel\n"
7245 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7246 "OpBranch %loop\n"
7247
7248 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7249 "%loop = OpLabel\n"
7250 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
7251 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
7252 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
7253 // There are several possibilities for the Continue Target below. Each
7254 // will be specialized into a separate test case.
7255 "OpLoopMerge %exit ${continue_target} None\n"
7256 "OpBranch %if\n"
7257
7258 "%if = OpLabel\n"
7259 ";delta_next = (delta > 0) ? -1 : 1;\n"
7260 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
7261 "OpSelectionMerge %gather DontFlatten\n"
7262 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
7263
7264 "%odd = OpLabel\n"
7265 "OpBranch %gather\n"
7266
7267 "%even = OpLabel\n"
7268 "OpBranch %gather\n"
7269
7270 "%gather = OpLabel\n"
7271 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
7272 "%val = OpFAdd %f32 %val1 %delta\n"
7273 "%count__ = OpISub %i32 %count %c_i32_1\n"
7274 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7275 "OpBranchConditional %again %loop %exit\n"
7276
7277 "%exit = OpLabel\n"
7278 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7279 "OpReturnValue %result\n"
7280
7281 "OpFunctionEnd\n");
7282
7283 map<string, string> continue_target;
7284
7285 // The Continue Target is the loop block itself.
7286 continue_target["continue_target"] = "%loop";
7287 fragments["testfun"] = multiBlock.specialize(continue_target);
7288 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
7289
7290 // The Continue Target is at the end of the loop.
7291 continue_target["continue_target"] = "%gather";
7292 fragments["testfun"] = multiBlock.specialize(continue_target);
7293 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
7294
7295 // A loop with continue statement.
7296 fragments["testfun"] =
7297 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7298 "%param1 = OpFunctionParameter %v4f32\n"
7299
7300 "%entry = OpLabel\n"
7301 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7302 "OpBranch %loop\n"
7303
7304 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
7305 "%loop = OpLabel\n"
7306 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7307 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
7308 "OpLoopMerge %exit %continue None\n"
7309 "OpBranch %if\n"
7310
7311 "%if = OpLabel\n"
7312 ";skip if %count==2\n"
7313 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
7314 "OpSelectionMerge %continue DontFlatten\n"
7315 "OpBranchConditional %eq2 %continue %body\n"
7316
7317 "%body = OpLabel\n"
7318 "%fcount = OpConvertSToF %f32 %count\n"
7319 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7320 "OpBranch %continue\n"
7321
7322 "%continue = OpLabel\n"
7323 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
7324 "%count__ = OpISub %i32 %count %c_i32_1\n"
7325 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7326 "OpBranchConditional %again %loop %exit\n"
7327
7328 "%exit = OpLabel\n"
7329 "%same = OpFSub %f32 %val %c_f32_8\n"
7330 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7331 "OpReturnValue %result\n"
7332 "OpFunctionEnd\n";
7333 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
7334
7335 // A loop with break.
7336 fragments["testfun"] =
7337 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7338 "%param1 = OpFunctionParameter %v4f32\n"
7339
7340 "%entry = OpLabel\n"
7341 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7342 "%dot = OpDot %f32 %param1 %param1\n"
7343 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7344 "%zero = OpConvertFToU %u32 %div\n"
7345 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7346 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7347 "OpBranch %loop\n"
7348
7349 ";adds 4 and 3 to %val0 (exits early)\n"
7350 "%loop = OpLabel\n"
7351 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7352 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7353 "OpLoopMerge %exit %continue None\n"
7354 "OpBranch %if\n"
7355
7356 "%if = OpLabel\n"
7357 ";end loop if %count==%two\n"
7358 "%above2 = OpSGreaterThan %bool %count %two\n"
7359 "OpSelectionMerge %continue DontFlatten\n"
7360 "OpBranchConditional %above2 %body %exit\n"
7361
7362 "%body = OpLabel\n"
7363 "%fcount = OpConvertSToF %f32 %count\n"
7364 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7365 "OpBranch %continue\n"
7366
7367 "%continue = OpLabel\n"
7368 "%count__ = OpISub %i32 %count %c_i32_1\n"
7369 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7370 "OpBranchConditional %again %loop %exit\n"
7371
7372 "%exit = OpLabel\n"
7373 "%val_post = OpPhi %f32 %val2 %continue %val1 %if\n"
7374 "%same = OpFSub %f32 %val_post %c_f32_7\n"
7375 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7376 "OpReturnValue %result\n"
7377 "OpFunctionEnd\n";
7378 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
7379
7380 // A loop with return.
7381 fragments["testfun"] =
7382 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7383 "%param1 = OpFunctionParameter %v4f32\n"
7384
7385 "%entry = OpLabel\n"
7386 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7387 "%dot = OpDot %f32 %param1 %param1\n"
7388 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7389 "%zero = OpConvertFToU %u32 %div\n"
7390 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7391 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7392 "OpBranch %loop\n"
7393
7394 ";returns early without modifying %param1\n"
7395 "%loop = OpLabel\n"
7396 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7397 "%val1 = OpPhi %f32 %val0 %entry %val2 %continue\n"
7398 "OpLoopMerge %exit %continue None\n"
7399 "OpBranch %if\n"
7400
7401 "%if = OpLabel\n"
7402 ";return if %count==%two\n"
7403 "%above2 = OpSGreaterThan %bool %count %two\n"
7404 "OpSelectionMerge %continue DontFlatten\n"
7405 "OpBranchConditional %above2 %body %early_exit\n"
7406
7407 "%early_exit = OpLabel\n"
7408 "OpReturnValue %param1\n"
7409
7410 "%body = OpLabel\n"
7411 "%fcount = OpConvertSToF %f32 %count\n"
7412 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7413 "OpBranch %continue\n"
7414
7415 "%continue = OpLabel\n"
7416 "%count__ = OpISub %i32 %count %c_i32_1\n"
7417 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7418 "OpBranchConditional %again %loop %exit\n"
7419
7420 "%exit = OpLabel\n"
7421 ";should never get here, so return an incorrect result\n"
7422 "%result = OpVectorInsertDynamic %v4f32 %param1 %val2 %c_i32_0\n"
7423 "OpReturnValue %result\n"
7424 "OpFunctionEnd\n";
7425 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
7426
7427 return testGroup.release();
7428 }
7429
7430 // Adds a new test to group using custom fragments for the tessellation-control
7431 // stage and passthrough fragments for all other stages. Uses default colors
7432 // for input and expected output.
addTessCtrlTest(tcu::TestCaseGroup * group,const char * name,const map<string,string> & fragments)7433 void addTessCtrlTest(tcu::TestCaseGroup* group, const char* name, const map<string, string>& fragments)
7434 {
7435 RGBA defaultColors[4];
7436 getDefaultColors(defaultColors);
7437 const ShaderElement pipelineStages[] =
7438 {
7439 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
7440 ShaderElement("tessc", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7441 ShaderElement("tesse", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7442 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
7443 };
7444
7445 addFunctionCaseWithPrograms<InstanceContext>(group, name, "", addShaderCodeCustomTessControl,
7446 runAndVerifyDefaultPipeline, createInstanceContext(
7447 pipelineStages, defaultColors, defaultColors, fragments, StageToSpecConstantMap()));
7448 }
7449
7450 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
createBarrierTests(tcu::TestContext & testCtx)7451 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
7452 {
7453 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
7454 map<string, string> fragments;
7455
7456 // A barrier inside a function body.
7457 fragments["pre_main"] =
7458 "%Workgroup = OpConstant %i32 2\n"
7459 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
7460 fragments["testfun"] =
7461 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7462 "%param1 = OpFunctionParameter %v4f32\n"
7463 "%label_testfun = OpLabel\n"
7464 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7465 "OpReturnValue %param1\n"
7466 "OpFunctionEnd\n";
7467 addTessCtrlTest(testGroup.get(), "in_function", fragments);
7468
7469 // Common setup code for the following tests.
7470 fragments["pre_main"] =
7471 "%Workgroup = OpConstant %i32 2\n"
7472 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7473 "%c_f32_5 = OpConstant %f32 5.\n";
7474 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
7475 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7476 "%param1 = OpFunctionParameter %v4f32\n"
7477 "%entry = OpLabel\n"
7478 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7479 "%dot = OpDot %f32 %param1 %param1\n"
7480 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7481 "%zero = OpConvertFToU %u32 %div\n";
7482
7483 // Barriers inside OpSwitch branches.
7484 fragments["testfun"] =
7485 setupPercentZero +
7486 "OpSelectionMerge %switch_exit None\n"
7487 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
7488
7489 "%case1 = OpLabel\n"
7490 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7491 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7492 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7493 "OpBranch %switch_exit\n"
7494
7495 "%switch_default = OpLabel\n"
7496 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7497 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7498 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7499 "OpBranch %switch_exit\n"
7500
7501 "%case0 = OpLabel\n"
7502 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7503 "OpBranch %switch_exit\n"
7504
7505 "%switch_exit = OpLabel\n"
7506 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
7507 "OpReturnValue %ret\n"
7508 "OpFunctionEnd\n";
7509 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
7510
7511 // Barriers inside if-then-else.
7512 fragments["testfun"] =
7513 setupPercentZero +
7514 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
7515 "OpSelectionMerge %exit DontFlatten\n"
7516 "OpBranchConditional %eq0 %then %else\n"
7517
7518 "%else = OpLabel\n"
7519 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7520 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7521 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7522 "OpBranch %exit\n"
7523
7524 "%then = OpLabel\n"
7525 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7526 "OpBranch %exit\n"
7527
7528 "%exit = OpLabel\n"
7529 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
7530 "OpReturnValue %ret\n"
7531 "OpFunctionEnd\n";
7532 addTessCtrlTest(testGroup.get(), "in_if", fragments);
7533
7534 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
7535 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
7536 fragments["testfun"] =
7537 setupPercentZero +
7538 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
7539 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
7540 "OpSelectionMerge %exit DontFlatten\n"
7541 "OpBranchConditional %thread0 %then %else\n"
7542
7543 "%else = OpLabel\n"
7544 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7545 "OpBranch %exit\n"
7546
7547 "%then = OpLabel\n"
7548 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
7549 "OpBranch %exit\n"
7550
7551 "%exit = OpLabel\n"
7552 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
7553 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7554 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
7555 "OpReturnValue %ret\n"
7556 "OpFunctionEnd\n";
7557 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
7558
7559 // A barrier inside a loop.
7560 fragments["pre_main"] =
7561 "%Workgroup = OpConstant %i32 2\n"
7562 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7563 "%c_f32_10 = OpConstant %f32 10.\n";
7564 fragments["testfun"] =
7565 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7566 "%param1 = OpFunctionParameter %v4f32\n"
7567 "%entry = OpLabel\n"
7568 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7569 "OpBranch %loop\n"
7570
7571 ";adds 4, 3, 2, and 1 to %val0\n"
7572 "%loop = OpLabel\n"
7573 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7574 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7575 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7576 "%fcount = OpConvertSToF %f32 %count\n"
7577 "%val = OpFAdd %f32 %val1 %fcount\n"
7578 "%count__ = OpISub %i32 %count %c_i32_1\n"
7579 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7580 "OpLoopMerge %exit %loop None\n"
7581 "OpBranchConditional %again %loop %exit\n"
7582
7583 "%exit = OpLabel\n"
7584 "%same = OpFSub %f32 %val %c_f32_10\n"
7585 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7586 "OpReturnValue %ret\n"
7587 "OpFunctionEnd\n";
7588 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
7589
7590 return testGroup.release();
7591 }
7592
7593 // Test for the OpFRem instruction.
createFRemTests(tcu::TestContext & testCtx)7594 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
7595 {
7596 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
7597 map<string, string> fragments;
7598 RGBA inputColors[4];
7599 RGBA outputColors[4];
7600
7601 fragments["pre_main"] =
7602 "%c_f32_3 = OpConstant %f32 3.0\n"
7603 "%c_f32_n3 = OpConstant %f32 -3.0\n"
7604 "%c_f32_4 = OpConstant %f32 4.0\n"
7605 "%c_f32_p75 = OpConstant %f32 0.75\n"
7606 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
7607 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
7608 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
7609
7610 // The test does the following.
7611 // vec4 result = (param1 * 8.0) - 4.0;
7612 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
7613 fragments["testfun"] =
7614 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7615 "%param1 = OpFunctionParameter %v4f32\n"
7616 "%label_testfun = OpLabel\n"
7617 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
7618 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
7619 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
7620 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
7621 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
7622 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
7623 "OpReturnValue %xy_0_1\n"
7624 "OpFunctionEnd\n";
7625
7626
7627 inputColors[0] = RGBA(16, 16, 0, 255);
7628 inputColors[1] = RGBA(232, 232, 0, 255);
7629 inputColors[2] = RGBA(232, 16, 0, 255);
7630 inputColors[3] = RGBA(16, 232, 0, 255);
7631
7632 outputColors[0] = RGBA(64, 64, 0, 255);
7633 outputColors[1] = RGBA(255, 255, 0, 255);
7634 outputColors[2] = RGBA(255, 64, 0, 255);
7635 outputColors[3] = RGBA(64, 255, 0, 255);
7636
7637 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
7638 return testGroup.release();
7639 }
7640
createInstructionTests(tcu::TestContext & testCtx)7641 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
7642 {
7643 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
7644 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
7645 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
7646
7647 computeTests->addChild(createOpNopGroup(testCtx));
7648 computeTests->addChild(createOpLineGroup(testCtx));
7649 computeTests->addChild(createOpNoLineGroup(testCtx));
7650 computeTests->addChild(createOpConstantNullGroup(testCtx));
7651 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
7652 computeTests->addChild(createOpConstantUsageGroup(testCtx));
7653 computeTests->addChild(createSpecConstantGroup(testCtx));
7654 computeTests->addChild(createOpSourceGroup(testCtx));
7655 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
7656 computeTests->addChild(createDecorationGroupGroup(testCtx));
7657 computeTests->addChild(createOpPhiGroup(testCtx));
7658 computeTests->addChild(createLoopControlGroup(testCtx));
7659 computeTests->addChild(createFunctionControlGroup(testCtx));
7660 computeTests->addChild(createSelectionControlGroup(testCtx));
7661 computeTests->addChild(createBlockOrderGroup(testCtx));
7662 computeTests->addChild(createMultipleShaderGroup(testCtx));
7663 computeTests->addChild(createMemoryAccessGroup(testCtx));
7664 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
7665 computeTests->addChild(createOpCopyObjectGroup(testCtx));
7666 computeTests->addChild(createNoContractionGroup(testCtx));
7667 computeTests->addChild(createOpUndefGroup(testCtx));
7668 computeTests->addChild(createOpUnreachableGroup(testCtx));
7669 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
7670 computeTests ->addChild(createOpFRemGroup(testCtx));
7671
7672 RGBA defaultColors[4];
7673 getDefaultColors(defaultColors);
7674
7675 de::MovePtr<tcu::TestCaseGroup> opnopTests (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
7676 map<string, string> opNopFragments;
7677 opNopFragments["testfun"] =
7678 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7679 "%param1 = OpFunctionParameter %v4f32\n"
7680 "%label_testfun = OpLabel\n"
7681 "OpNop\n"
7682 "OpNop\n"
7683 "OpNop\n"
7684 "OpNop\n"
7685 "OpNop\n"
7686 "OpNop\n"
7687 "OpNop\n"
7688 "OpNop\n"
7689 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7690 "%b = OpFAdd %f32 %a %a\n"
7691 "OpNop\n"
7692 "%c = OpFSub %f32 %b %a\n"
7693 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
7694 "OpNop\n"
7695 "OpNop\n"
7696 "OpReturnValue %ret\n"
7697 "OpFunctionEnd\n"
7698 ;
7699 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, opnopTests.get());
7700
7701
7702 graphicsTests->addChild(opnopTests.release());
7703 graphicsTests->addChild(createOpSourceTests(testCtx));
7704 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
7705 graphicsTests->addChild(createOpLineTests(testCtx));
7706 graphicsTests->addChild(createOpNoLineTests(testCtx));
7707 graphicsTests->addChild(createOpConstantNullTests(testCtx));
7708 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
7709 graphicsTests->addChild(createMemoryAccessTests(testCtx));
7710 graphicsTests->addChild(createOpUndefTests(testCtx));
7711 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
7712 graphicsTests->addChild(createModuleTests(testCtx));
7713 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
7714 graphicsTests->addChild(createOpPhiTests(testCtx));
7715 graphicsTests->addChild(createNoContractionTests(testCtx));
7716 graphicsTests->addChild(createOpQuantizeTests(testCtx));
7717 graphicsTests->addChild(createLoopTests(testCtx));
7718 graphicsTests->addChild(createSpecConstantTests(testCtx));
7719 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
7720 graphicsTests->addChild(createBarrierTests(testCtx));
7721 graphicsTests->addChild(createDecorationGroupTests(testCtx));
7722 graphicsTests->addChild(createFRemTests(testCtx));
7723
7724 instructionTests->addChild(computeTests.release());
7725 instructionTests->addChild(graphicsTests.release());
7726
7727 return instructionTests.release();
7728 }
7729
7730 } // SpirVAssembly
7731 } // vkt
7732