1 /*-------------------------------------------------------------------------
2 * Vulkan Conformance Tests
3 * ------------------------
5 * Copyright (c) 2015 Google Inc.
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and/or associated documentation files (the
9 * "Materials"), to deal in the Materials without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sublicense, and/or sell copies of the Materials, and to
12 * permit persons to whom the Materials are furnished to do so, subject to
13 * the following conditions:
15 * The above copyright notice(s) and this permission notice shall be
16 * included in all copies or substantial portions of the Materials.
18 * The Materials are Confidential Information as defined by the
19 * Khronos Membership Agreement until designated non-confidential by
20 * Khronos, at which point this condition clause shall be removed.
22 * THE MATERIALS ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
23 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
24 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
25 * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
26 * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
27 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
28 * MATERIALS OR THE USE OR OTHER DEALINGS IN THE MATERIALS.
32 * \brief SPIR-V Assembly Tests for Instructions (special opcode/operand)
33 *//*--------------------------------------------------------------------*/
35 #include "vktSpvAsmInstructionTests.hpp"
37 #include "tcuCommandLine.hpp"
38 #include "tcuFormatUtil.hpp"
39 #include "tcuRGBA.hpp"
40 #include "tcuStringTemplate.hpp"
41 #include "tcuTestLog.hpp"
42 #include "tcuVectorUtil.hpp"
45 #include "vkDeviceUtil.hpp"
46 #include "vkMemUtil.hpp"
47 #include "vkPlatform.hpp"
48 #include "vkPrograms.hpp"
49 #include "vkQueryUtil.hpp"
51 #include "vkRefUtil.hpp"
52 #include "vkStrUtil.hpp"
53 #include "vkTypeUtil.hpp"
55 #include "deRandom.hpp"
56 #include "deStringUtil.hpp"
57 #include "deUniquePtr.hpp"
58 #include "tcuStringTemplate.hpp"
61 #include "vktSpvAsmComputeShaderCase.hpp"
62 #include "vktSpvAsmComputeShaderTestUtil.hpp"
63 #include "vktTestCaseUtil.hpp"
73 namespace SpirVAssembly
87 using tcu::TestStatus;
90 using tcu::StringTemplate;
93 typedef Unique<VkShaderModule> ModuleHandleUp;
94 typedef de::SharedPtr<ModuleHandleUp> ModuleHandleSp;
96 template<typename T> T randomScalar (de::Random& rnd, T minValue, T maxValue);
97 template<> inline float randomScalar (de::Random& rnd, float minValue, float maxValue) { return rnd.getFloat(minValue, maxValue); }
98 template<> inline deInt32 randomScalar (de::Random& rnd, deInt32 minValue, deInt32 maxValue) { return rnd.getInt(minValue, maxValue); }
101 static void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0)
103 T* const typedPtr = (T*)dst;
104 for (int ndx = 0; ndx < numValues; ndx++)
105 typedPtr[offset + ndx] = randomScalar<T>(rnd, minValue, maxValue);
113 CaseParameter (const char* case_, const string& param_) : name(case_), param(param_) {}
116 // Assembly code used for testing OpNop, OpConstant{Null|Composite}, Op[No]Line, OpSource[Continued], OpSourceExtension, OpUndef is based on GLSL source code:
120 // layout(std140, set = 0, binding = 0) readonly buffer Input {
123 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
127 // layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
130 // uint x = gl_GlobalInvocationID.x;
131 // output_data.elements[x] = -input_data.elements[x];
134 static const char* const s_ShaderPreamble =
135 "OpCapability Shader\n"
136 "OpMemoryModel Logical GLSL450\n"
137 "OpEntryPoint GLCompute %main \"main\" %id\n"
138 "OpExecutionMode %main LocalSize 1 1 1\n";
140 static const char* const s_CommonTypes =
141 "%bool = OpTypeBool\n"
142 "%void = OpTypeVoid\n"
143 "%voidf = OpTypeFunction %void\n"
144 "%u32 = OpTypeInt 32 0\n"
145 "%i32 = OpTypeInt 32 1\n"
146 "%f32 = OpTypeFloat 32\n"
147 "%uvec3 = OpTypeVector %u32 3\n"
148 "%fvec3 = OpTypeVector %f32 3\n"
149 "%uvec3ptr = OpTypePointer Input %uvec3\n"
150 "%f32ptr = OpTypePointer Uniform %f32\n"
151 "%f32arr = OpTypeRuntimeArray %f32\n";
153 // Declares two uniform variables (indata, outdata) of type "struct { float[] }". Depends on type "f32arr" (for "float[]").
154 static const char* const s_InputOutputBuffer =
155 "%inbuf = OpTypeStruct %f32arr\n"
156 "%inbufptr = OpTypePointer Uniform %inbuf\n"
157 "%indata = OpVariable %inbufptr Uniform\n"
158 "%outbuf = OpTypeStruct %f32arr\n"
159 "%outbufptr = OpTypePointer Uniform %outbuf\n"
160 "%outdata = OpVariable %outbufptr Uniform\n";
162 // Declares buffer type and layout for uniform variables indata and outdata. Both of them are SSBO bounded to descriptor set 0.
163 // indata is at binding point 0, while outdata is at 1.
164 static const char* const s_InputOutputBufferTraits =
165 "OpDecorate %inbuf BufferBlock\n"
166 "OpDecorate %indata DescriptorSet 0\n"
167 "OpDecorate %indata Binding 0\n"
168 "OpDecorate %outbuf BufferBlock\n"
169 "OpDecorate %outdata DescriptorSet 0\n"
170 "OpDecorate %outdata Binding 1\n"
171 "OpDecorate %f32arr ArrayStride 4\n"
172 "OpMemberDecorate %inbuf 0 Offset 0\n"
173 "OpMemberDecorate %outbuf 0 Offset 0\n";
175 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
177 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
178 ComputeShaderSpec spec;
179 de::Random rnd (deStringHash(group->getName()));
180 const int numElements = 100;
181 vector<float> positiveFloats (numElements, 0);
182 vector<float> negativeFloats (numElements, 0);
184 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
186 for (size_t ndx = 0; ndx < numElements; ++ndx)
187 negativeFloats[ndx] = -positiveFloats[ndx];
190 string(s_ShaderPreamble) +
192 "OpSource GLSL 430\n"
193 "OpName %main \"main\"\n"
194 "OpName %id \"gl_GlobalInvocationID\"\n"
196 "OpDecorate %id BuiltIn GlobalInvocationId\n"
198 + string(s_InputOutputBufferTraits) + string(s_CommonTypes)
200 + string(s_InputOutputBuffer) +
202 "%id = OpVariable %uvec3ptr Input\n"
203 "%zero = OpConstant %i32 0\n"
205 "%main = OpFunction %void None %voidf\n"
207 "%idval = OpLoad %uvec3 %id\n"
208 "%x = OpCompositeExtract %u32 %idval 0\n"
210 " OpNop\n" // Inside a function body
212 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
213 "%inval = OpLoad %f32 %inloc\n"
214 "%neg = OpFNegate %f32 %inval\n"
215 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
216 " OpStore %outloc %neg\n"
219 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
220 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
221 spec.numWorkGroups = IVec3(numElements, 1, 1);
223 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
225 return group.release();
228 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
230 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
231 ComputeShaderSpec spec;
232 de::Random rnd (deStringHash(group->getName()));
233 const int numElements = 100;
234 vector<float> positiveFloats (numElements, 0);
235 vector<float> negativeFloats (numElements, 0);
237 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
239 for (size_t ndx = 0; ndx < numElements; ++ndx)
240 negativeFloats[ndx] = -positiveFloats[ndx];
243 string(s_ShaderPreamble) +
245 "%fname1 = OpString \"negateInputs.comp\"\n"
246 "%fname2 = OpString \"negateInputs\"\n"
248 "OpSource GLSL 430\n"
249 "OpName %main \"main\"\n"
250 "OpName %id \"gl_GlobalInvocationID\"\n"
252 "OpDecorate %id BuiltIn GlobalInvocationId\n"
254 + string(s_InputOutputBufferTraits) +
256 "OpLine %fname1 0 0\n" // At the earliest possible position
258 + string(s_CommonTypes) + string(s_InputOutputBuffer) +
260 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
261 "OpLine %fname2 1 0\n" // Different filenames
262 "OpLine %fname1 1000 100000\n"
264 "%id = OpVariable %uvec3ptr Input\n"
265 "%zero = OpConstant %i32 0\n"
267 "OpLine %fname1 1 1\n" // Before a function
269 "%main = OpFunction %void None %voidf\n"
272 "OpLine %fname1 1 1\n" // In a function
274 "%idval = OpLoad %uvec3 %id\n"
275 "%x = OpCompositeExtract %u32 %idval 0\n"
276 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
277 "%inval = OpLoad %f32 %inloc\n"
278 "%neg = OpFNegate %f32 %inval\n"
279 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
280 " OpStore %outloc %neg\n"
283 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
284 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
285 spec.numWorkGroups = IVec3(numElements, 1, 1);
287 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
289 return group.release();
292 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
294 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
295 ComputeShaderSpec spec;
296 de::Random rnd (deStringHash(group->getName()));
297 const int numElements = 100;
298 vector<float> positiveFloats (numElements, 0);
299 vector<float> negativeFloats (numElements, 0);
301 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
303 for (size_t ndx = 0; ndx < numElements; ++ndx)
304 negativeFloats[ndx] = -positiveFloats[ndx];
307 string(s_ShaderPreamble) +
309 "%fname = OpString \"negateInputs.comp\"\n"
311 "OpSource GLSL 430\n"
312 "OpName %main \"main\"\n"
313 "OpName %id \"gl_GlobalInvocationID\"\n"
315 "OpDecorate %id BuiltIn GlobalInvocationId\n"
317 + string(s_InputOutputBufferTraits) +
319 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
321 + string(s_CommonTypes) + string(s_InputOutputBuffer) +
323 "OpLine %fname 0 1\n"
324 "OpNoLine\n" // Immediately following a preceding OpLine
326 "OpLine %fname 1000 1\n"
328 "%id = OpVariable %uvec3ptr Input\n"
329 "%zero = OpConstant %i32 0\n"
331 "OpNoLine\n" // Contents after the previous OpLine
333 "%main = OpFunction %void None %voidf\n"
335 "%idval = OpLoad %uvec3 %id\n"
336 "%x = OpCompositeExtract %u32 %idval 0\n"
338 "OpNoLine\n" // Multiple OpNoLine
342 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
343 "%inval = OpLoad %f32 %inloc\n"
344 "%neg = OpFNegate %f32 %inval\n"
345 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
346 " OpStore %outloc %neg\n"
349 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
350 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
351 spec.numWorkGroups = IVec3(numElements, 1, 1);
353 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
355 return group.release();
358 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
360 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
361 vector<CaseParameter> cases;
362 const int numElements = 100;
363 vector<float> inputFloats1 (numElements, 0);
364 vector<float> inputFloats2 (numElements, 0);
365 vector<float> outputFloats (numElements, 0);
366 const StringTemplate shaderTemplate (
367 string(s_ShaderPreamble) +
369 "OpName %main \"main\"\n"
370 "OpName %id \"gl_GlobalInvocationID\"\n"
372 "OpDecorate %id BuiltIn GlobalInvocationId\n"
376 "OpDecorate %inbuf1 BufferBlock\n"
377 "OpDecorate %indata1 DescriptorSet 0\n"
378 "OpDecorate %indata1 Binding 0\n"
379 "OpDecorate %inbuf2 BufferBlock\n"
380 "OpDecorate %indata2 DescriptorSet 0\n"
381 "OpDecorate %indata2 Binding 1\n"
382 "OpDecorate %outbuf BufferBlock\n"
383 "OpDecorate %outdata DescriptorSet 0\n"
384 "OpDecorate %outdata Binding 2\n"
385 "OpDecorate %f32arr ArrayStride 4\n"
386 "OpMemberDecorate %inbuf1 0 Offset 0\n"
387 "OpMemberDecorate %inbuf2 0 Offset 0\n"
388 "OpMemberDecorate %outbuf 0 Offset 0\n"
390 + string(s_CommonTypes) +
392 "%inbuf1 = OpTypeStruct %f32arr\n"
393 "%inbufptr1 = OpTypePointer Uniform %inbuf1\n"
394 "%indata1 = OpVariable %inbufptr1 Uniform\n"
395 "%inbuf2 = OpTypeStruct %f32arr\n"
396 "%inbufptr2 = OpTypePointer Uniform %inbuf2\n"
397 "%indata2 = OpVariable %inbufptr2 Uniform\n"
398 "%outbuf = OpTypeStruct %f32arr\n"
399 "%outbufptr = OpTypePointer Uniform %outbuf\n"
400 "%outdata = OpVariable %outbufptr Uniform\n"
402 "%id = OpVariable %uvec3ptr Input\n"
403 "%zero = OpConstant %i32 0\n"
404 "%c_f_m1 = OpConstant %f32 -1.\n"
406 "%main = OpFunction %void None %voidf\n"
408 "%idval = OpLoad %uvec3 %id\n"
409 "%x = OpCompositeExtract %u32 %idval 0\n"
410 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
411 "%inval1 = OpLoad %f32 %inloc1\n"
412 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
413 "%inval2 = OpLoad %f32 %inloc2\n"
414 "%mul = OpFMul %f32 %inval1 %inval2\n"
415 "%add = OpFAdd %f32 %mul %c_f_m1\n"
416 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
417 " OpStore %outloc %add\n"
421 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
422 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
423 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
425 for (size_t ndx = 0; ndx < numElements; ++ndx)
427 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
428 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
429 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
430 // conducted separately and the result is rounded to 1. So the final result will be 0.f.
431 // If the operation is combined into a precise fused multiply-add, then the result would be
432 // 2^-46 (0xa8800000).
433 outputFloats[ndx] = 0.f;
436 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
438 map<string, string> specializations;
439 ComputeShaderSpec spec;
441 specializations["DECORATION"] = cases[caseNdx].param;
442 spec.assembly = shaderTemplate.specialize(specializations);
443 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
444 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
445 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
446 spec.numWorkGroups = IVec3(numElements, 1, 1);
448 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
450 return group.release();
453 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
455 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
456 ComputeShaderSpec spec;
457 de::Random rnd (deStringHash(group->getName()));
458 const int numElements = 200;
459 vector<float> inputFloats1 (numElements, 0);
460 vector<float> inputFloats2 (numElements, 0);
461 vector<float> outputFloats (numElements, 0);
463 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
464 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
466 for (size_t ndx = 0; ndx < numElements; ++ndx)
468 // Guard against divisors near zero.
469 if (std::fabs(inputFloats2[ndx]) < 1e-3)
470 inputFloats2[ndx] = 8.f;
472 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
473 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
477 string(s_ShaderPreamble) +
479 "OpName %main \"main\"\n"
480 "OpName %id \"gl_GlobalInvocationID\"\n"
482 "OpDecorate %id BuiltIn GlobalInvocationId\n"
484 "OpDecorate %inbuf1 BufferBlock\n"
485 "OpDecorate %indata1 DescriptorSet 0\n"
486 "OpDecorate %indata1 Binding 0\n"
487 "OpDecorate %inbuf2 BufferBlock\n"
488 "OpDecorate %indata2 DescriptorSet 0\n"
489 "OpDecorate %indata2 Binding 1\n"
490 "OpDecorate %outbuf BufferBlock\n"
491 "OpDecorate %outdata DescriptorSet 0\n"
492 "OpDecorate %outdata Binding 2\n"
493 "OpDecorate %f32arr ArrayStride 4\n"
494 "OpMemberDecorate %inbuf1 0 Offset 0\n"
495 "OpMemberDecorate %inbuf2 0 Offset 0\n"
496 "OpMemberDecorate %outbuf 0 Offset 0\n"
498 + string(s_CommonTypes) +
500 "%inbuf1 = OpTypeStruct %f32arr\n"
501 "%inbufptr1 = OpTypePointer Uniform %inbuf1\n"
502 "%indata1 = OpVariable %inbufptr1 Uniform\n"
503 "%inbuf2 = OpTypeStruct %f32arr\n"
504 "%inbufptr2 = OpTypePointer Uniform %inbuf2\n"
505 "%indata2 = OpVariable %inbufptr2 Uniform\n"
506 "%outbuf = OpTypeStruct %f32arr\n"
507 "%outbufptr = OpTypePointer Uniform %outbuf\n"
508 "%outdata = OpVariable %outbufptr Uniform\n"
510 "%id = OpVariable %uvec3ptr Input\n"
511 "%zero = OpConstant %i32 0\n"
513 "%main = OpFunction %void None %voidf\n"
515 "%idval = OpLoad %uvec3 %id\n"
516 "%x = OpCompositeExtract %u32 %idval 0\n"
517 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
518 "%inval1 = OpLoad %f32 %inloc1\n"
519 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
520 "%inval2 = OpLoad %f32 %inloc2\n"
521 "%rem = OpFRem %f32 %inval1 %inval2\n"
522 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
523 " OpStore %outloc %rem\n"
527 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
528 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
529 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
530 spec.numWorkGroups = IVec3(numElements, 1, 1);
532 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
534 return group.release();
537 // Copy contents in the input buffer to the output buffer.
538 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
540 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
541 de::Random rnd (deStringHash(group->getName()));
542 const int numElements = 100;
544 // 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.
545 ComputeShaderSpec spec1;
546 vector<Vec4> inputFloats1 (numElements);
547 vector<Vec4> outputFloats1 (numElements);
549 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
551 for (size_t ndx = 0; ndx < numElements; ++ndx)
552 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
555 string(s_ShaderPreamble) +
557 "OpName %main \"main\"\n"
558 "OpName %id \"gl_GlobalInvocationID\"\n"
560 "OpDecorate %id BuiltIn GlobalInvocationId\n"
561 "OpDecorate %vec4arr ArrayStride 16\n"
563 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
565 "%vec4 = OpTypeVector %f32 4\n"
566 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
567 "%vec4ptr_f = OpTypePointer Function %vec4\n"
568 "%vec4arr = OpTypeRuntimeArray %vec4\n"
569 "%inbuf = OpTypeStruct %vec4arr\n"
570 "%inbufptr = OpTypePointer Uniform %inbuf\n"
571 "%indata = OpVariable %inbufptr Uniform\n"
572 "%outbuf = OpTypeStruct %vec4arr\n"
573 "%outbufptr = OpTypePointer Uniform %outbuf\n"
574 "%outdata = OpVariable %outbufptr Uniform\n"
576 "%id = OpVariable %uvec3ptr Input\n"
577 "%zero = OpConstant %i32 0\n"
578 "%c_f_0 = OpConstant %f32 0.\n"
579 "%c_f_0_5 = OpConstant %f32 0.5\n"
580 "%c_f_1_5 = OpConstant %f32 1.5\n"
581 "%c_f_2_5 = OpConstant %f32 2.5\n"
582 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
584 "%main = OpFunction %void None %voidf\n"
586 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
587 "%idval = OpLoad %uvec3 %id\n"
588 "%x = OpCompositeExtract %u32 %idval 0\n"
589 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
590 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
591 " OpCopyMemory %v_vec4 %inloc\n"
592 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
593 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
594 " OpStore %outloc %add\n"
598 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
599 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
600 spec1.numWorkGroups = IVec3(numElements, 1, 1);
602 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
604 // The following case copies a float[100] variable from the input buffer to the output buffer.
605 ComputeShaderSpec spec2;
606 vector<float> inputFloats2 (numElements);
607 vector<float> outputFloats2 (numElements);
609 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
611 for (size_t ndx = 0; ndx < numElements; ++ndx)
612 outputFloats2[ndx] = inputFloats2[ndx];
615 string(s_ShaderPreamble) +
617 "OpName %main \"main\"\n"
618 "OpName %id \"gl_GlobalInvocationID\"\n"
620 "OpDecorate %id BuiltIn GlobalInvocationId\n"
621 "OpDecorate %f32arr100 ArrayStride 4\n"
623 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
625 "%hundred = OpConstant %u32 100\n"
626 "%f32arr100 = OpTypeArray %f32 %hundred\n"
627 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
628 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
629 "%inbuf = OpTypeStruct %f32arr100\n"
630 "%inbufptr = OpTypePointer Uniform %inbuf\n"
631 "%indata = OpVariable %inbufptr Uniform\n"
632 "%outbuf = OpTypeStruct %f32arr100\n"
633 "%outbufptr = OpTypePointer Uniform %outbuf\n"
634 "%outdata = OpVariable %outbufptr Uniform\n"
636 "%id = OpVariable %uvec3ptr Input\n"
637 "%zero = OpConstant %i32 0\n"
639 "%main = OpFunction %void None %voidf\n"
641 "%var = OpVariable %f32arr100ptr_f Function\n"
642 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
643 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
644 " OpCopyMemory %var %inarr\n"
645 " OpCopyMemory %outarr %var\n"
649 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
650 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
651 spec2.numWorkGroups = IVec3(1, 1, 1);
653 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
655 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
656 ComputeShaderSpec spec3;
657 vector<float> inputFloats3 (16);
658 vector<float> outputFloats3 (16);
660 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
662 for (size_t ndx = 0; ndx < 16; ++ndx)
663 outputFloats3[ndx] = inputFloats3[ndx];
666 string(s_ShaderPreamble) +
668 "OpName %main \"main\"\n"
669 "OpName %id \"gl_GlobalInvocationID\"\n"
671 "OpDecorate %id BuiltIn GlobalInvocationId\n"
672 "OpMemberDecorate %inbuf 0 Offset 0\n"
673 "OpMemberDecorate %inbuf 1 Offset 16\n"
674 "OpMemberDecorate %inbuf 2 Offset 32\n"
675 "OpMemberDecorate %inbuf 3 Offset 48\n"
676 "OpMemberDecorate %outbuf 0 Offset 0\n"
677 "OpMemberDecorate %outbuf 1 Offset 16\n"
678 "OpMemberDecorate %outbuf 2 Offset 32\n"
679 "OpMemberDecorate %outbuf 3 Offset 48\n"
681 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
683 "%vec4 = OpTypeVector %f32 4\n"
684 "%inbuf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
685 "%inbufptr = OpTypePointer Uniform %inbuf\n"
686 "%indata = OpVariable %inbufptr Uniform\n"
687 "%outbuf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
688 "%outbufptr = OpTypePointer Uniform %outbuf\n"
689 "%outdata = OpVariable %outbufptr Uniform\n"
690 "%vec4stptr = OpTypePointer Function %inbuf\n"
692 "%id = OpVariable %uvec3ptr Input\n"
693 "%zero = OpConstant %i32 0\n"
695 "%main = OpFunction %void None %voidf\n"
697 "%var = OpVariable %vec4stptr Function\n"
698 " OpCopyMemory %var %indata\n"
699 " OpCopyMemory %outdata %var\n"
703 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
704 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
705 spec3.numWorkGroups = IVec3(1, 1, 1);
707 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
709 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
710 ComputeShaderSpec spec4;
711 vector<float> inputFloats4 (numElements);
712 vector<float> outputFloats4 (numElements);
714 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
716 for (size_t ndx = 0; ndx < numElements; ++ndx)
717 outputFloats4[ndx] = -inputFloats4[ndx];
720 string(s_ShaderPreamble) +
722 "OpName %main \"main\"\n"
723 "OpName %id \"gl_GlobalInvocationID\"\n"
725 "OpDecorate %id BuiltIn GlobalInvocationId\n"
727 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
729 "%f32ptr_f = OpTypePointer Function %f32\n"
730 "%id = OpVariable %uvec3ptr Input\n"
731 "%zero = OpConstant %i32 0\n"
733 "%main = OpFunction %void None %voidf\n"
735 "%var = OpVariable %f32ptr_f Function\n"
736 "%idval = OpLoad %uvec3 %id\n"
737 "%x = OpCompositeExtract %u32 %idval 0\n"
738 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
739 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
740 " OpCopyMemory %var %inloc\n"
741 "%val = OpLoad %f32 %var\n"
742 "%neg = OpFNegate %f32 %val\n"
743 " OpStore %outloc %neg\n"
747 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
748 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
749 spec4.numWorkGroups = IVec3(numElements, 1, 1);
751 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
753 return group.release();
756 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
758 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
759 ComputeShaderSpec spec;
760 de::Random rnd (deStringHash(group->getName()));
761 const int numElements = 100;
762 vector<float> inputFloats (numElements, 0);
763 vector<float> outputFloats (numElements, 0);
765 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
767 for (size_t ndx = 0; ndx < numElements; ++ndx)
768 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
771 string(s_ShaderPreamble) +
773 "OpName %main \"main\"\n"
774 "OpName %id \"gl_GlobalInvocationID\"\n"
776 "OpDecorate %id BuiltIn GlobalInvocationId\n"
778 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
780 "%fmat = OpTypeMatrix %fvec3 3\n"
781 "%three = OpConstant %u32 3\n"
782 "%farr = OpTypeArray %f32 %three\n"
783 "%fst = OpTypeStruct %f32 %f32\n"
785 + string(s_InputOutputBuffer) +
787 "%id = OpVariable %uvec3ptr Input\n"
788 "%zero = OpConstant %i32 0\n"
789 "%c_f = OpConstant %f32 1.5\n"
790 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
791 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
792 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
793 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
795 "%main = OpFunction %void None %voidf\n"
797 "%c_f_copy = OpCopyObject %f32 %c_f\n"
798 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
799 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
800 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
801 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
802 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
803 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
804 "%farr_elem = OpCompositeExtract %f32 %c_fmat_copy 2\n"
805 "%fst_elem = OpCompositeExtract %f32 %c_fmat_copy 1\n"
806 // Add up. 1.5 * 5 = 7.5.
807 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
808 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
809 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
810 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
812 "%idval = OpLoad %uvec3 %id\n"
813 "%x = OpCompositeExtract %u32 %idval 0\n"
814 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
815 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
816 "%inval = OpLoad %f32 %inloc\n"
817 "%add = OpFAdd %f32 %add4 %inval\n"
818 " OpStore %outloc %add\n"
821 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
822 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
823 spec.numWorkGroups = IVec3(numElements, 1, 1);
825 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
827 return group.release();
829 // Assembly code used for testing OpUnreachable is based on GLSL source code:
833 // layout(std140, set = 0, binding = 0) readonly buffer Input {
836 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
840 // void not_called_func() {
841 // // place OpUnreachable here
844 // uint modulo4(uint val) {
845 // switch (val % uint(4)) {
850 // default: return 100; // place OpUnreachable here
856 // // place OpUnreachable here
860 // uint x = gl_GlobalInvocationID.x;
861 // if (const5() > modulo4(1000)) {
862 // output_data.elements[x] = -input_data.elements[x];
864 // // place OpUnreachable here
865 // output_data.elements[x] = input_data.elements[x];
869 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
871 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
872 ComputeShaderSpec spec;
873 de::Random rnd (deStringHash(group->getName()));
874 const int numElements = 100;
875 vector<float> positiveFloats (numElements, 0);
876 vector<float> negativeFloats (numElements, 0);
878 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
880 for (size_t ndx = 0; ndx < numElements; ++ndx)
881 negativeFloats[ndx] = -positiveFloats[ndx];
884 string(s_ShaderPreamble) +
886 "OpSource GLSL 430\n"
887 "OpName %main \"main\"\n"
888 "OpName %func_not_called_func \"not_called_func(\"\n"
889 "OpName %func_modulo4 \"modulo4(u1;\"\n"
890 "OpName %func_const5 \"const5(\"\n"
891 "OpName %id \"gl_GlobalInvocationID\"\n"
893 "OpDecorate %id BuiltIn GlobalInvocationId\n"
895 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
897 "%u32ptr = OpTypePointer Function %u32\n"
898 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
899 "%unitf = OpTypeFunction %u32\n"
901 "%id = OpVariable %uvec3ptr Input\n"
902 "%zero = OpConstant %u32 0\n"
903 "%one = OpConstant %u32 1\n"
904 "%two = OpConstant %u32 2\n"
905 "%three = OpConstant %u32 3\n"
906 "%four = OpConstant %u32 4\n"
907 "%five = OpConstant %u32 5\n"
908 "%hundred = OpConstant %u32 100\n"
909 "%thousand = OpConstant %u32 1000\n"
911 + string(s_InputOutputBuffer) +
914 "%main = OpFunction %void None %voidf\n"
915 "%main_entry = OpLabel\n"
916 "%idval = OpLoad %uvec3 %id\n"
917 "%x = OpCompositeExtract %u32 %idval 0\n"
918 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
919 "%inval = OpLoad %f32 %inloc\n"
920 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
921 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
922 "%v_thousand = OpVariable %u32ptr Function %thousand\n"
923 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %v_thousand\n"
924 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
925 " OpSelectionMerge %if_end None\n"
926 " OpBranchConditional %cmp_gt %if_true %if_false\n"
927 "%if_true = OpLabel\n"
928 "%negate = OpFNegate %f32 %inval\n"
929 " OpStore %outloc %negate\n"
930 " OpBranch %if_end\n"
931 "%if_false = OpLabel\n"
932 " OpUnreachable\n" // Unreachable else branch for if statement
933 "%if_end = OpLabel\n"
937 // not_called_function()
938 "%func_not_called_func = OpFunction %void None %voidf\n"
939 "%not_called_func_entry = OpLabel\n"
940 " OpUnreachable\n" // Unreachable entry block in not called static function
944 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
945 "%valptr = OpFunctionParameter %u32ptr\n"
946 "%modulo4_entry = OpLabel\n"
947 "%val = OpLoad %u32 %valptr\n"
948 "%modulo = OpUMod %u32 %val %four\n"
949 " OpSelectionMerge %switch_merge None\n"
950 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
952 " OpReturnValue %three\n"
954 " OpReturnValue %two\n"
956 " OpReturnValue %one\n"
958 " OpReturnValue %zero\n"
959 "%default = OpLabel\n"
960 " OpUnreachable\n" // Unreachable default case for switch statement
961 "%switch_merge = OpLabel\n"
962 " OpUnreachable\n" // Unreachable merge block for switch statement
966 "%func_const5 = OpFunction %u32 None %unitf\n"
967 "%const5_entry = OpLabel\n"
968 " OpReturnValue %five\n"
969 "%unreachable = OpLabel\n"
970 " OpUnreachable\n" // Unreachable block in function
972 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
973 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
974 spec.numWorkGroups = IVec3(numElements, 1, 1);
976 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
978 return group.release();
981 // Assembly code used for testing decoration group is based on GLSL source code:
985 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
988 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
991 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
994 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
997 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
1000 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
1001 // float elements[];
1005 // uint x = gl_GlobalInvocationID.x;
1006 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
1008 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
1010 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
1011 ComputeShaderSpec spec;
1012 de::Random rnd (deStringHash(group->getName()));
1013 const int numElements = 100;
1014 vector<float> inputFloats0 (numElements, 0);
1015 vector<float> inputFloats1 (numElements, 0);
1016 vector<float> inputFloats2 (numElements, 0);
1017 vector<float> inputFloats3 (numElements, 0);
1018 vector<float> inputFloats4 (numElements, 0);
1019 vector<float> outputFloats (numElements, 0);
1021 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
1022 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
1023 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
1024 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
1025 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
1027 for (size_t ndx = 0; ndx < numElements; ++ndx)
1028 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
1031 string(s_ShaderPreamble) +
1033 "OpSource GLSL 430\n"
1034 "OpName %main \"main\"\n"
1035 "OpName %id \"gl_GlobalInvocationID\"\n"
1037 // Not using group decoration on variable.
1038 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1039 // Not using group decoration on type.
1040 "OpDecorate %f32arr ArrayStride 4\n"
1042 "OpDecorate %groups BufferBlock\n"
1043 "OpDecorate %groupm Offset 0\n"
1044 "%groups = OpDecorationGroup\n"
1045 "%groupm = OpDecorationGroup\n"
1047 // Group decoration on multiple structs.
1048 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
1049 // Group decoration on multiple struct members.
1050 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
1052 "OpDecorate %group1 DescriptorSet 0\n"
1053 "OpDecorate %group3 DescriptorSet 0\n"
1054 "OpDecorate %group3 NonWritable\n"
1055 "OpDecorate %group3 Restrict\n"
1056 "%group0 = OpDecorationGroup\n"
1057 "%group1 = OpDecorationGroup\n"
1058 "%group3 = OpDecorationGroup\n"
1060 // Applying the same decoration group multiple times.
1061 "OpGroupDecorate %group1 %outdata\n"
1062 "OpGroupDecorate %group1 %outdata\n"
1063 "OpGroupDecorate %group1 %outdata\n"
1064 "OpDecorate %outdata DescriptorSet 0\n"
1065 "OpDecorate %outdata Binding 5\n"
1066 // Applying decoration group containing nothing.
1067 "OpGroupDecorate %group0 %indata0\n"
1068 "OpDecorate %indata0 DescriptorSet 0\n"
1069 "OpDecorate %indata0 Binding 0\n"
1070 // Applying decoration group containing one decoration.
1071 "OpGroupDecorate %group1 %indata1\n"
1072 "OpDecorate %indata1 Binding 1\n"
1073 // Applying decoration group containing multiple decorations.
1074 "OpGroupDecorate %group3 %indata2 %indata3\n"
1075 "OpDecorate %indata2 Binding 2\n"
1076 "OpDecorate %indata3 Binding 3\n"
1077 // Applying multiple decoration groups (with overlapping).
1078 "OpGroupDecorate %group0 %indata4\n"
1079 "OpGroupDecorate %group1 %indata4\n"
1080 "OpGroupDecorate %group3 %indata4\n"
1081 "OpDecorate %indata4 Binding 4\n"
1083 + string(s_CommonTypes) +
1085 "%id = OpVariable %uvec3ptr Input\n"
1086 "%zero = OpConstant %i32 0\n"
1088 "%outbuf = OpTypeStruct %f32arr\n"
1089 "%outbufptr = OpTypePointer Uniform %outbuf\n"
1090 "%outdata = OpVariable %outbufptr Uniform\n"
1091 "%inbuf0 = OpTypeStruct %f32arr\n"
1092 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
1093 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
1094 "%inbuf1 = OpTypeStruct %f32arr\n"
1095 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
1096 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
1097 "%inbuf2 = OpTypeStruct %f32arr\n"
1098 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
1099 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
1100 "%inbuf3 = OpTypeStruct %f32arr\n"
1101 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
1102 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
1103 "%inbuf4 = OpTypeStruct %f32arr\n"
1104 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
1105 "%indata4 = OpVariable %inbufptr Uniform\n"
1107 "%main = OpFunction %void None %voidf\n"
1108 "%label = OpLabel\n"
1109 "%idval = OpLoad %uvec3 %id\n"
1110 "%x = OpCompositeExtract %u32 %idval 0\n"
1111 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
1112 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1113 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1114 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1115 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
1116 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1117 "%inval0 = OpLoad %f32 %inloc0\n"
1118 "%inval1 = OpLoad %f32 %inloc1\n"
1119 "%inval2 = OpLoad %f32 %inloc2\n"
1120 "%inval3 = OpLoad %f32 %inloc3\n"
1121 "%inval4 = OpLoad %f32 %inloc4\n"
1122 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
1123 "%add1 = OpFAdd %f32 %add0 %inval2\n"
1124 "%add2 = OpFAdd %f32 %add1 %inval3\n"
1125 "%add = OpFAdd %f32 %add2 %inval4\n"
1126 " OpStore %outloc %add\n"
1129 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
1130 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1131 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1132 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1133 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
1134 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1135 spec.numWorkGroups = IVec3(numElements, 1, 1);
1137 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
1139 return group.release();
1142 struct SpecConstantTwoIntCase
1144 const char* caseName;
1145 const char* scDefinition0;
1146 const char* scDefinition1;
1147 const char* scResultType;
1148 const char* scOperation;
1149 deInt32 scActualValue0;
1150 deInt32 scActualValue1;
1151 const char* resultOperation;
1152 vector<deInt32> expectedOutput;
1154 SpecConstantTwoIntCase (const char* name,
1155 const char* definition0,
1156 const char* definition1,
1157 const char* resultType,
1158 const char* operation,
1161 const char* resultOp,
1162 const vector<deInt32>& output)
1164 , scDefinition0 (definition0)
1165 , scDefinition1 (definition1)
1166 , scResultType (resultType)
1167 , scOperation (operation)
1168 , scActualValue0 (value0)
1169 , scActualValue1 (value1)
1170 , resultOperation (resultOp)
1171 , expectedOutput (output) {}
1174 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
1176 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
1177 vector<SpecConstantTwoIntCase> cases;
1178 de::Random rnd (deStringHash(group->getName()));
1179 const int numElements = 100;
1180 vector<deInt32> inputInts (numElements, 0);
1181 vector<deInt32> outputInts1 (numElements, 0);
1182 vector<deInt32> outputInts2 (numElements, 0);
1183 vector<deInt32> outputInts3 (numElements, 0);
1184 vector<deInt32> outputInts4 (numElements, 0);
1185 const StringTemplate shaderTemplate (
1186 string(s_ShaderPreamble) +
1188 "OpName %main \"main\"\n"
1189 "OpName %id \"gl_GlobalInvocationID\"\n"
1191 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1192 "OpDecorate %sc_0 SpecId 0\n"
1193 "OpDecorate %sc_1 SpecId 1\n"
1194 "OpDecorate %i32arr ArrayStride 4\n"
1196 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1198 "%i32ptr = OpTypePointer Uniform %i32\n"
1199 "%i32arr = OpTypeRuntimeArray %i32\n"
1200 "%boolptr = OpTypePointer Uniform %bool\n"
1201 "%boolarr = OpTypeRuntimeArray %bool\n"
1202 "%inbuf = OpTypeStruct %i32arr\n"
1203 "%inbufptr = OpTypePointer Uniform %inbuf\n"
1204 "%indata = OpVariable %inbufptr Uniform\n"
1205 "%outbuf = OpTypeStruct %i32arr\n"
1206 "%outbufptr = OpTypePointer Uniform %outbuf\n"
1207 "%outdata = OpVariable %outbufptr Uniform\n"
1209 "%id = OpVariable %uvec3ptr Input\n"
1210 "%zero = OpConstant %i32 0\n"
1212 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
1213 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
1214 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
1216 "%main = OpFunction %void None %voidf\n"
1217 "%label = OpLabel\n"
1218 "%idval = OpLoad %uvec3 %id\n"
1219 "%x = OpCompositeExtract %u32 %idval 0\n"
1220 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
1221 "%inval = OpLoad %i32 %inloc\n"
1222 "%final = ${GEN_RESULT}\n"
1223 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1224 " OpStore %outloc %final\n"
1226 " OpFunctionEnd\n");
1228 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
1230 for (size_t ndx = 0; ndx < numElements; ++ndx)
1232 outputInts1[ndx] = inputInts[ndx] + 42;
1233 outputInts2[ndx] = inputInts[ndx];
1234 outputInts3[ndx] = inputInts[ndx] - 11200;
1235 outputInts4[ndx] = inputInts[ndx] + 1;
1238 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
1239 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
1240 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
1242 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
1243 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
1244 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
1245 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
1246 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
1247 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
1248 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
1249 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
1250 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
1251 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
1252 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
1253 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
1254 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
1255 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
1256 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
1257 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
1258 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
1259 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
1260 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
1261 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
1262 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
1263 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
1264 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
1265 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
1266 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
1267 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
1268 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
1269 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
1270 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
1271 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
1272 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
1273 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
1275 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
1277 map<string, string> specializations;
1278 ComputeShaderSpec spec;
1280 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
1281 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
1282 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
1283 specializations["SC_OP"] = cases[caseNdx].scOperation;
1284 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
1286 spec.assembly = shaderTemplate.specialize(specializations);
1287 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
1288 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
1289 spec.numWorkGroups = IVec3(numElements, 1, 1);
1290 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
1291 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
1293 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec));
1296 ComputeShaderSpec spec;
1299 string(s_ShaderPreamble) +
1301 "OpName %main \"main\"\n"
1302 "OpName %id \"gl_GlobalInvocationID\"\n"
1304 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1305 "OpDecorate %sc_0 SpecId 0\n"
1306 "OpDecorate %sc_1 SpecId 1\n"
1307 "OpDecorate %sc_2 SpecId 2\n"
1308 "OpDecorate %i32arr ArrayStride 4\n"
1310 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1312 "%ivec3 = OpTypeVector %i32 3\n"
1313 "%i32ptr = OpTypePointer Uniform %i32\n"
1314 "%i32arr = OpTypeRuntimeArray %i32\n"
1315 "%boolptr = OpTypePointer Uniform %bool\n"
1316 "%boolarr = OpTypeRuntimeArray %bool\n"
1317 "%inbuf = OpTypeStruct %i32arr\n"
1318 "%inbufptr = OpTypePointer Uniform %inbuf\n"
1319 "%indata = OpVariable %inbufptr Uniform\n"
1320 "%outbuf = OpTypeStruct %i32arr\n"
1321 "%outbufptr = OpTypePointer Uniform %outbuf\n"
1322 "%outdata = OpVariable %outbufptr Uniform\n"
1324 "%id = OpVariable %uvec3ptr Input\n"
1325 "%zero = OpConstant %i32 0\n"
1326 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
1328 "%sc_0 = OpSpecConstant %i32 0\n"
1329 "%sc_1 = OpSpecConstant %i32 0\n"
1330 "%sc_2 = OpSpecConstant %i32 0\n"
1331 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
1332 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
1333 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
1334 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %sc_vec3_1 1 0 4\n" // (0, sc_0, sc_1)
1335 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2 5 1 2\n" // (sc_2, sc_0, sc_1)
1336 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
1337 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
1338 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
1339 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
1340 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
1342 "%main = OpFunction %void None %voidf\n"
1343 "%label = OpLabel\n"
1344 "%idval = OpLoad %uvec3 %id\n"
1345 "%x = OpCompositeExtract %u32 %idval 0\n"
1346 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
1347 "%inval = OpLoad %i32 %inloc\n"
1348 "%final = OpIAdd %i32 %inval %sc_final\n"
1349 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1350 " OpStore %outloc %final\n"
1353 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
1354 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
1355 spec.numWorkGroups = IVec3(numElements, 1, 1);
1356 spec.specConstants.push_back(123);
1357 spec.specConstants.push_back(56);
1358 spec.specConstants.push_back(-77);
1360 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
1362 return group.release();
1365 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
1367 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
1368 ComputeShaderSpec spec1;
1369 ComputeShaderSpec spec2;
1370 ComputeShaderSpec spec3;
1371 de::Random rnd (deStringHash(group->getName()));
1372 const int numElements = 100;
1373 vector<float> inputFloats (numElements, 0);
1374 vector<float> outputFloats1 (numElements, 0);
1375 vector<float> outputFloats2 (numElements, 0);
1376 vector<float> outputFloats3 (numElements, 0);
1378 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
1380 for (size_t ndx = 0; ndx < numElements; ++ndx)
1384 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
1385 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
1386 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
1389 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
1390 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
1394 string(s_ShaderPreamble) +
1396 "OpSource GLSL 430\n"
1397 "OpName %main \"main\"\n"
1398 "OpName %id \"gl_GlobalInvocationID\"\n"
1400 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1402 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1404 "%id = OpVariable %uvec3ptr Input\n"
1405 "%zero = OpConstant %i32 0\n"
1406 "%three = OpConstant %u32 3\n"
1407 "%constf5p5 = OpConstant %f32 5.5\n"
1408 "%constf20p5 = OpConstant %f32 20.5\n"
1409 "%constf1p75 = OpConstant %f32 1.75\n"
1410 "%constf8p5 = OpConstant %f32 8.5\n"
1411 "%constf6p5 = OpConstant %f32 6.5\n"
1413 "%main = OpFunction %void None %voidf\n"
1414 "%entry = OpLabel\n"
1415 "%idval = OpLoad %uvec3 %id\n"
1416 "%x = OpCompositeExtract %u32 %idval 0\n"
1417 "%selector = OpUMod %u32 %x %three\n"
1418 " OpSelectionMerge %phi None\n"
1419 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
1421 // Case 1 before OpPhi.
1422 "%case1 = OpLabel\n"
1425 "%default = OpLabel\n"
1429 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
1430 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1431 "%inval = OpLoad %f32 %inloc\n"
1432 "%add = OpFAdd %f32 %inval %operand\n"
1433 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1434 " OpStore %outloc %add\n"
1437 // Case 0 after OpPhi.
1438 "%case0 = OpLabel\n"
1442 // Case 2 after OpPhi.
1443 "%case2 = OpLabel\n"
1447 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1448 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
1449 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1451 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
1454 string(s_ShaderPreamble) +
1456 "OpName %main \"main\"\n"
1457 "OpName %id \"gl_GlobalInvocationID\"\n"
1459 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1461 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1463 "%id = OpVariable %uvec3ptr Input\n"
1464 "%zero = OpConstant %i32 0\n"
1465 "%one = OpConstant %i32 1\n"
1466 "%three = OpConstant %i32 3\n"
1467 "%constf6p5 = OpConstant %f32 6.5\n"
1469 "%main = OpFunction %void None %voidf\n"
1470 "%entry = OpLabel\n"
1471 "%idval = OpLoad %uvec3 %id\n"
1472 "%x = OpCompositeExtract %u32 %idval 0\n"
1473 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1474 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1475 "%inval = OpLoad %f32 %inloc\n"
1479 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
1480 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
1481 "%step_next = OpIAdd %i32 %step %one\n"
1482 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
1483 "%still_loop = OpSLessThan %bool %step %three\n"
1484 " OpLoopMerge %exit %phi None\n"
1485 " OpBranchConditional %still_loop %phi %exit\n"
1488 " OpStore %outloc %accum\n"
1491 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1492 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1493 spec2.numWorkGroups = IVec3(numElements, 1, 1);
1495 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
1498 string(s_ShaderPreamble) +
1500 "OpName %main \"main\"\n"
1501 "OpName %id \"gl_GlobalInvocationID\"\n"
1503 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1505 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1507 "%f32ptr_f = OpTypePointer Function %f32\n"
1508 "%id = OpVariable %uvec3ptr Input\n"
1509 "%true = OpConstantTrue %bool\n"
1510 "%false = OpConstantFalse %bool\n"
1511 "%zero = OpConstant %i32 0\n"
1512 "%constf8p5 = OpConstant %f32 8.5\n"
1514 "%main = OpFunction %void None %voidf\n"
1515 "%entry = OpLabel\n"
1516 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
1517 "%idval = OpLoad %uvec3 %id\n"
1518 "%x = OpCompositeExtract %u32 %idval 0\n"
1519 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1520 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1521 "%a_init = OpLoad %f32 %inloc\n"
1522 "%b_init = OpLoad %f32 %b\n"
1526 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
1527 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
1528 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
1529 " OpLoopMerge %exit %phi None\n"
1530 " OpBranchConditional %still_loop %phi %exit\n"
1533 "%sub = OpFSub %f32 %a_next %b_next\n"
1534 " OpStore %outloc %sub\n"
1537 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1538 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
1539 spec3.numWorkGroups = IVec3(numElements, 1, 1);
1541 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
1543 return group.release();
1546 // Assembly code used for testing block order is based on GLSL source code:
1550 // layout(std140, set = 0, binding = 0) readonly buffer Input {
1551 // float elements[];
1553 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
1554 // float elements[];
1558 // uint x = gl_GlobalInvocationID.x;
1559 // output_data.elements[x] = input_data.elements[x];
1560 // if (x > uint(50)) {
1561 // switch (x % uint(3)) {
1562 // case 0: output_data.elements[x] += 1.5f; break;
1563 // case 1: output_data.elements[x] += 42.f; break;
1564 // case 2: output_data.elements[x] -= 27.f; break;
1568 // output_data.elements[x] = -input_data.elements[x];
1571 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
1573 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
1574 ComputeShaderSpec spec;
1575 de::Random rnd (deStringHash(group->getName()));
1576 const int numElements = 100;
1577 vector<float> inputFloats (numElements, 0);
1578 vector<float> outputFloats (numElements, 0);
1580 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
1582 for (size_t ndx = 0; ndx <= 50; ++ndx)
1583 outputFloats[ndx] = -inputFloats[ndx];
1585 for (size_t ndx = 51; ndx < numElements; ++ndx)
1589 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
1590 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
1591 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
1597 string(s_ShaderPreamble) +
1599 "OpSource GLSL 430\n"
1600 "OpName %main \"main\"\n"
1601 "OpName %id \"gl_GlobalInvocationID\"\n"
1603 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1605 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1607 "%u32ptr = OpTypePointer Function %u32\n"
1608 "%u32ptr_input = OpTypePointer Input %u32\n"
1610 + string(s_InputOutputBuffer) +
1612 "%id = OpVariable %uvec3ptr Input\n"
1613 "%zero = OpConstant %i32 0\n"
1614 "%const3 = OpConstant %u32 3\n"
1615 "%const50 = OpConstant %u32 50\n"
1616 "%constf1p5 = OpConstant %f32 1.5\n"
1617 "%constf27 = OpConstant %f32 27.0\n"
1618 "%constf42 = OpConstant %f32 42.0\n"
1620 "%main = OpFunction %void None %voidf\n"
1623 "%entry = OpLabel\n"
1625 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
1626 "%xvar = OpVariable %u32ptr Function\n"
1627 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
1628 "%x = OpLoad %u32 %xptr\n"
1629 " OpStore %xvar %x\n"
1631 "%cmp = OpUGreaterThan %bool %x %const50\n"
1632 " OpSelectionMerge %if_merge None\n"
1633 " OpBranchConditional %cmp %if_true %if_false\n"
1635 // Merge block for switch-statement: placed at the beginning.
1636 "%switch_merge = OpLabel\n"
1637 " OpBranch %if_merge\n"
1639 // Case 1 for switch-statement.
1640 "%case1 = OpLabel\n"
1641 "%x_1 = OpLoad %u32 %xvar\n"
1642 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
1643 "%inval_1 = OpLoad %f32 %inloc_1\n"
1644 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
1645 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
1646 " OpStore %outloc_1 %addf42\n"
1647 " OpBranch %switch_merge\n"
1649 // False branch for if-statement: placed in the middle of switch cases and before true branch.
1650 "%if_false = OpLabel\n"
1651 "%x_f = OpLoad %u32 %xvar\n"
1652 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
1653 "%inval_f = OpLoad %f32 %inloc_f\n"
1654 "%negate = OpFNegate %f32 %inval_f\n"
1655 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
1656 " OpStore %outloc_f %negate\n"
1657 " OpBranch %if_merge\n"
1659 // Merge block for if-statement: placed in the middle of true and false branch.
1660 "%if_merge = OpLabel\n"
1663 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
1664 "%if_true = OpLabel\n"
1665 "%xval_t = OpLoad %u32 %xvar\n"
1666 "%mod = OpUMod %u32 %xval_t %const3\n"
1667 " OpSelectionMerge %switch_merge None\n"
1668 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
1670 // Case 2 for switch-statement.
1671 "%case2 = OpLabel\n"
1672 "%x_2 = OpLoad %u32 %xvar\n"
1673 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
1674 "%inval_2 = OpLoad %f32 %inloc_2\n"
1675 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
1676 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
1677 " OpStore %outloc_2 %subf27\n"
1678 " OpBranch %switch_merge\n"
1680 // Default case for switch-statement: placed in the middle of normal cases.
1681 "%default = OpLabel\n"
1682 " OpBranch %switch_merge\n"
1684 // Case 0 for switch-statement: out of order.
1685 "%case0 = OpLabel\n"
1686 "%x_0 = OpLoad %u32 %xvar\n"
1687 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
1688 "%inval_0 = OpLoad %f32 %inloc_0\n"
1689 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
1690 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
1691 " OpStore %outloc_0 %addf1p5\n"
1692 " OpBranch %switch_merge\n"
1695 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1696 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1697 spec.numWorkGroups = IVec3(numElements, 1, 1);
1699 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
1701 return group.release();
1704 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
1706 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
1707 ComputeShaderSpec spec1;
1708 ComputeShaderSpec spec2;
1709 de::Random rnd (deStringHash(group->getName()));
1710 const int numElements = 100;
1711 vector<float> inputFloats (numElements, 0);
1712 vector<float> outputFloats1 (numElements, 0);
1713 vector<float> outputFloats2 (numElements, 0);
1714 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
1716 for (size_t ndx = 0; ndx < numElements; ++ndx)
1718 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
1719 outputFloats2[ndx] = -inputFloats[ndx];
1722 const string assembly(
1723 "OpCapability Shader\n"
1724 "OpMemoryModel Logical GLSL450\n"
1725 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
1726 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
1727 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
1728 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexIndex %instanceIndex\n"
1729 "OpExecutionMode %comp_main1 LocalSize 1 1 1\n"
1730 "OpExecutionMode %comp_main2 LocalSize 1 1 1\n"
1732 "OpName %comp_main1 \"entrypoint1\"\n"
1733 "OpName %comp_main2 \"entrypoint2\"\n"
1734 "OpName %vert_main \"entrypoint2\"\n"
1735 "OpName %id \"gl_GlobalInvocationID\"\n"
1736 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
1737 "OpName %vertexIndex \"gl_VertexIndex\"\n"
1738 "OpName %instanceIndex \"gl_InstanceIndex\"\n"
1739 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
1740 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
1741 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
1743 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1744 "OpDecorate %vertexIndex BuiltIn VertexIndex\n"
1745 "OpDecorate %instanceIndex BuiltIn InstanceIndex\n"
1746 "OpDecorate %vert_builtin_st Block\n"
1747 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
1748 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
1749 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
1751 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1753 "%zero = OpConstant %i32 0\n"
1754 "%one = OpConstant %u32 1\n"
1755 "%c_f32_1 = OpConstant %f32 1\n"
1757 "%i32ptr = OpTypePointer Input %i32\n"
1758 "%vec4 = OpTypeVector %f32 4\n"
1759 "%vec4ptr = OpTypePointer Output %vec4\n"
1760 "%f32arr1 = OpTypeArray %f32 %one\n"
1761 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
1762 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
1763 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
1765 "%id = OpVariable %uvec3ptr Input\n"
1766 "%vertexIndex = OpVariable %i32ptr Input\n"
1767 "%instanceIndex = OpVariable %i32ptr Input\n"
1768 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
1770 // gl_Position = vec4(1.);
1771 "%vert_main = OpFunction %void None %voidf\n"
1772 "%vert_entry = OpLabel\n"
1773 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
1774 " OpStore %position %c_vec4_1\n"
1779 "%comp_main1 = OpFunction %void None %voidf\n"
1780 "%comp1_entry = OpLabel\n"
1781 "%idval1 = OpLoad %uvec3 %id\n"
1782 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
1783 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
1784 "%inval1 = OpLoad %f32 %inloc1\n"
1785 "%add = OpFAdd %f32 %inval1 %inval1\n"
1786 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
1787 " OpStore %outloc1 %add\n"
1792 "%comp_main2 = OpFunction %void None %voidf\n"
1793 "%comp2_entry = OpLabel\n"
1794 "%idval2 = OpLoad %uvec3 %id\n"
1795 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
1796 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
1797 "%inval2 = OpLoad %f32 %inloc2\n"
1798 "%neg = OpFNegate %f32 %inval2\n"
1799 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
1800 " OpStore %outloc2 %neg\n"
1802 " OpFunctionEnd\n");
1804 spec1.assembly = assembly;
1805 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1806 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
1807 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1808 spec1.entryPoint = "entrypoint1";
1810 spec2.assembly = assembly;
1811 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1812 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1813 spec2.numWorkGroups = IVec3(numElements, 1, 1);
1814 spec2.entryPoint = "entrypoint2";
1816 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
1817 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
1819 return group.release();
1822 inline std::string makeLongUTF8String (size_t num4ByteChars)
1824 // An example of a longest valid UTF-8 character. Be explicit about the
1825 // character type because Microsoft compilers can otherwise interpret the
1826 // character string as being over wide (16-bit) characters. Ideally, we
1827 // would just use a C++11 UTF-8 string literal, but we want to support older
1828 // Microsoft compilers.
1829 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
1830 std::string longString;
1831 longString.reserve(num4ByteChars * 4);
1832 for (size_t count = 0; count < num4ByteChars; count++)
1834 longString += earthAfrica;
1839 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
1841 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
1842 vector<CaseParameter> cases;
1843 de::Random rnd (deStringHash(group->getName()));
1844 const int numElements = 100;
1845 vector<float> positiveFloats (numElements, 0);
1846 vector<float> negativeFloats (numElements, 0);
1847 const StringTemplate shaderTemplate (
1848 "OpCapability Shader\n"
1849 "OpMemoryModel Logical GLSL450\n"
1851 "OpEntryPoint GLCompute %main \"main\" %id\n"
1852 "OpExecutionMode %main LocalSize 1 1 1\n"
1856 "OpName %main \"main\"\n"
1857 "OpName %id \"gl_GlobalInvocationID\"\n"
1859 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1861 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1863 "%id = OpVariable %uvec3ptr Input\n"
1864 "%zero = OpConstant %i32 0\n"
1866 "%main = OpFunction %void None %voidf\n"
1867 "%label = OpLabel\n"
1868 "%idval = OpLoad %uvec3 %id\n"
1869 "%x = OpCompositeExtract %u32 %idval 0\n"
1870 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1871 "%inval = OpLoad %f32 %inloc\n"
1872 "%neg = OpFNegate %f32 %inval\n"
1873 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1874 " OpStore %outloc %neg\n"
1876 " OpFunctionEnd\n");
1878 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
1879 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
1880 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
1881 "OpSource GLSL 430 %fname"));
1882 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
1883 "OpSource GLSL 430 %fname"));
1884 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
1885 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
1886 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
1887 "OpSource GLSL 430 %fname \"\""));
1888 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
1889 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
1890 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
1891 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
1892 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
1893 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
1894 "OpSourceContinued \"id main() {}\""));
1895 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
1896 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
1897 "OpSourceContinued \"\""));
1898 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
1899 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
1900 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
1901 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
1902 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
1903 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
1904 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
1905 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
1906 "OpSourceContinued \"void\"\n"
1907 "OpSourceContinued \"main()\"\n"
1908 "OpSourceContinued \"{}\""));
1909 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
1910 "OpSource GLSL 430 %fname \"\"\n"
1911 "OpSourceContinued \"#version 430\nvoid main() {}\""));
1913 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
1915 for (size_t ndx = 0; ndx < numElements; ++ndx)
1916 negativeFloats[ndx] = -positiveFloats[ndx];
1918 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
1920 map<string, string> specializations;
1921 ComputeShaderSpec spec;
1923 specializations["SOURCE"] = cases[caseNdx].param;
1924 spec.assembly = shaderTemplate.specialize(specializations);
1925 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
1926 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
1927 spec.numWorkGroups = IVec3(numElements, 1, 1);
1929 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
1932 return group.release();
1935 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
1937 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
1938 vector<CaseParameter> cases;
1939 de::Random rnd (deStringHash(group->getName()));
1940 const int numElements = 100;
1941 vector<float> inputFloats (numElements, 0);
1942 vector<float> outputFloats (numElements, 0);
1943 const StringTemplate shaderTemplate (
1944 string(s_ShaderPreamble) +
1946 "OpSourceExtension \"${EXTENSION}\"\n"
1948 "OpName %main \"main\"\n"
1949 "OpName %id \"gl_GlobalInvocationID\"\n"
1951 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1953 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1955 "%id = OpVariable %uvec3ptr Input\n"
1956 "%zero = OpConstant %i32 0\n"
1958 "%main = OpFunction %void None %voidf\n"
1959 "%label = OpLabel\n"
1960 "%idval = OpLoad %uvec3 %id\n"
1961 "%x = OpCompositeExtract %u32 %idval 0\n"
1962 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1963 "%inval = OpLoad %f32 %inloc\n"
1964 "%neg = OpFNegate %f32 %inval\n"
1965 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1966 " OpStore %outloc %neg\n"
1968 " OpFunctionEnd\n");
1970 cases.push_back(CaseParameter("empty_extension", ""));
1971 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
1972 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
1973 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
1974 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
1976 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
1978 for (size_t ndx = 0; ndx < numElements; ++ndx)
1979 outputFloats[ndx] = -inputFloats[ndx];
1981 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
1983 map<string, string> specializations;
1984 ComputeShaderSpec spec;
1986 specializations["EXTENSION"] = cases[caseNdx].param;
1987 spec.assembly = shaderTemplate.specialize(specializations);
1988 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1989 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1990 spec.numWorkGroups = IVec3(numElements, 1, 1);
1992 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
1995 return group.release();
1998 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
1999 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
2001 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
2002 vector<CaseParameter> cases;
2003 de::Random rnd (deStringHash(group->getName()));
2004 const int numElements = 100;
2005 vector<float> positiveFloats (numElements, 0);
2006 vector<float> negativeFloats (numElements, 0);
2007 const StringTemplate shaderTemplate (
2008 string(s_ShaderPreamble) +
2010 "OpSource GLSL 430\n"
2011 "OpName %main \"main\"\n"
2012 "OpName %id \"gl_GlobalInvocationID\"\n"
2014 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2016 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2019 "%null = OpConstantNull %type\n"
2021 "%id = OpVariable %uvec3ptr Input\n"
2022 "%zero = OpConstant %i32 0\n"
2024 "%main = OpFunction %void None %voidf\n"
2025 "%label = OpLabel\n"
2026 "%idval = OpLoad %uvec3 %id\n"
2027 "%x = OpCompositeExtract %u32 %idval 0\n"
2028 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2029 "%inval = OpLoad %f32 %inloc\n"
2030 "%neg = OpFNegate %f32 %inval\n"
2031 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2032 " OpStore %outloc %neg\n"
2034 " OpFunctionEnd\n");
2036 cases.push_back(CaseParameter("bool", "%type = OpTypeBool"));
2037 cases.push_back(CaseParameter("sint32", "%type = OpTypeInt 32 1"));
2038 cases.push_back(CaseParameter("uint32", "%type = OpTypeInt 32 0"));
2039 cases.push_back(CaseParameter("float32", "%type = OpTypeFloat 32"));
2040 cases.push_back(CaseParameter("vec4float32", "%type = OpTypeVector %f32 4"));
2041 cases.push_back(CaseParameter("vec3bool", "%type = OpTypeVector %bool 3"));
2042 cases.push_back(CaseParameter("vec2uint32", "%type = OpTypeVector %u32 2"));
2043 cases.push_back(CaseParameter("matrix", "%type = OpTypeMatrix %fvec3 3"));
2044 cases.push_back(CaseParameter("array", "%100 = OpConstant %u32 100\n"
2045 "%type = OpTypeArray %i32 %100"));
2046 cases.push_back(CaseParameter("struct", "%type = OpTypeStruct %f32 %i32 %u32"));
2047 cases.push_back(CaseParameter("pointer", "%type = OpTypePointer Function %i32"));
2049 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2051 for (size_t ndx = 0; ndx < numElements; ++ndx)
2052 negativeFloats[ndx] = -positiveFloats[ndx];
2054 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2056 map<string, string> specializations;
2057 ComputeShaderSpec spec;
2059 specializations["TYPE"] = cases[caseNdx].param;
2060 spec.assembly = shaderTemplate.specialize(specializations);
2061 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2062 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2063 spec.numWorkGroups = IVec3(numElements, 1, 1);
2065 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2068 return group.release();
2071 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
2072 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
2074 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
2075 vector<CaseParameter> cases;
2076 de::Random rnd (deStringHash(group->getName()));
2077 const int numElements = 100;
2078 vector<float> positiveFloats (numElements, 0);
2079 vector<float> negativeFloats (numElements, 0);
2080 const StringTemplate shaderTemplate (
2081 string(s_ShaderPreamble) +
2083 "OpSource GLSL 430\n"
2084 "OpName %main \"main\"\n"
2085 "OpName %id \"gl_GlobalInvocationID\"\n"
2087 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2089 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2091 "%id = OpVariable %uvec3ptr Input\n"
2092 "%zero = OpConstant %i32 0\n"
2096 "%main = OpFunction %void None %voidf\n"
2097 "%label = OpLabel\n"
2098 "%idval = OpLoad %uvec3 %id\n"
2099 "%x = OpCompositeExtract %u32 %idval 0\n"
2100 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2101 "%inval = OpLoad %f32 %inloc\n"
2102 "%neg = OpFNegate %f32 %inval\n"
2103 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2104 " OpStore %outloc %neg\n"
2106 " OpFunctionEnd\n");
2108 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
2109 "%const = OpConstantComposite %uvec3 %five %zero %five"));
2110 cases.push_back(CaseParameter("matrix", "%m3uvec3 = OpTypeMatrix %fvec3 3\n"
2111 "%ten = OpConstant %u32 10\n"
2112 "%vec = OpConstantComposite %uvec3 %ten %zero %ten\n"
2113 "%mat = OpConstantComposite %m3uvec3 %vec %vec %vec"));
2114 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %fvec3 2\n"
2115 "%struct = OpTypeStruct %u32 %f32 %uvec3 %m2vec3\n"
2116 "%one = OpConstant %u32 1\n"
2117 "%point5 = OpConstant %f32 0.5\n"
2118 "%vec = OpConstantComposite %uvec3 %one %one %zero\n"
2119 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
2120 "%const = OpConstantComposite %struct %one %point5 %vec %mat"));
2121 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
2122 "%st2 = OpTypeStruct %i32 %i32\n"
2123 "%struct = OpTypeStruct %st1 %st2\n"
2124 "%point5 = OpConstant %f32 0.5\n"
2125 "%one = OpConstant %u32 1\n"
2126 "%ten = OpConstant %i32 10\n"
2127 "%st1val = OpConstantComposite %st1 %one %point5\n"
2128 "%st2val = OpConstantComposite %st2 %ten %ten\n"
2129 "%const = OpConstantComposite %struct %st1val %st2val"));
2131 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2133 for (size_t ndx = 0; ndx < numElements; ++ndx)
2134 negativeFloats[ndx] = -positiveFloats[ndx];
2136 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2138 map<string, string> specializations;
2139 ComputeShaderSpec spec;
2141 specializations["CONSTANT"] = cases[caseNdx].param;
2142 spec.assembly = shaderTemplate.specialize(specializations);
2143 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2144 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2145 spec.numWorkGroups = IVec3(numElements, 1, 1);
2147 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2150 return group.release();
2153 // Creates a floating point number with the given exponent, and significand
2154 // bits set. It can only create normalized numbers. Only the least significant
2155 // 24 bits of the significand will be examined. The final bit of the
2156 // significand will also be ignored. This allows alignment to be written
2157 // similarly to C99 hex-floats.
2158 // For example if you wanted to write 0x1.7f34p-12 you would call
2159 // constructNormalizedFloat(-12, 0x7f3400)
2160 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
2164 for (deInt32 idx = 0; idx < 23; ++idx)
2166 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -(idx + 1));
2170 return std::ldexp(f, exponent);
2173 // Compare instruction for the OpQuantizeF16 compute exact case.
2174 // Returns true if the output is what is expected from the test case.
2175 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs)
2177 if (outputAllocs.size() != 1)
2180 // We really just need this for size because we cannot compare Nans.
2181 const BufferSp& expectedOutput = expectedOutputs[0];
2182 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
2184 if (expectedOutput->getNumBytes() != 4*sizeof(float)) {
2188 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
2189 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
2193 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
2194 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
2198 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
2199 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
2203 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
2204 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
2211 // Checks that every output from a test-case is a float NaN.
2212 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs)
2214 if (outputAllocs.size() != 1)
2217 // We really just need this for size because we cannot compare Nans.
2218 const BufferSp& expectedOutput = expectedOutputs[0];
2219 const float* output_as_float = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
2221 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
2223 if (!isnan(output_as_float[idx]))
2232 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
2233 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
2235 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
2237 const std::string shader (
2238 string(s_ShaderPreamble) +
2240 "OpSource GLSL 430\n"
2241 "OpName %main \"main\"\n"
2242 "OpName %id \"gl_GlobalInvocationID\"\n"
2244 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2246 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2248 "%id = OpVariable %uvec3ptr Input\n"
2249 "%zero = OpConstant %i32 0\n"
2251 "%main = OpFunction %void None %voidf\n"
2252 "%label = OpLabel\n"
2253 "%idval = OpLoad %uvec3 %id\n"
2254 "%x = OpCompositeExtract %u32 %idval 0\n"
2255 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2256 "%inval = OpLoad %f32 %inloc\n"
2257 "%quant = OpQuantizeToF16 %f32 %inval\n"
2258 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2259 " OpStore %outloc %quant\n"
2261 " OpFunctionEnd\n");
2264 ComputeShaderSpec spec;
2265 const deUint32 numElements = 100;
2266 vector<float> infinities;
2267 vector<float> results;
2269 infinities.reserve(numElements);
2270 results.reserve(numElements);
2272 for (size_t idx = 0; idx < numElements; ++idx)
2277 infinities.push_back(std::numeric_limits<float>::infinity());
2278 results.push_back(std::numeric_limits<float>::infinity());
2281 infinities.push_back(-std::numeric_limits<float>::infinity());
2282 results.push_back(-std::numeric_limits<float>::infinity());
2285 infinities.push_back(std::ldexp(1.0f, 16));
2286 results.push_back(std::numeric_limits<float>::infinity());
2289 infinities.push_back(std::ldexp(-1.0f, 32));
2290 results.push_back(-std::numeric_limits<float>::infinity());
2295 spec.assembly = shader;
2296 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
2297 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
2298 spec.numWorkGroups = IVec3(numElements, 1, 1);
2300 group->addChild(new SpvAsmComputeShaderCase(
2301 testCtx, "infinities", "Check that infinities propagated and created", spec));
2305 ComputeShaderSpec spec;
2307 const deUint32 numElements = 100;
2309 nans.reserve(numElements);
2311 for (size_t idx = 0; idx < numElements; ++idx)
2315 nans.push_back(std::numeric_limits<float>::quiet_NaN());
2319 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
2323 spec.assembly = shader;
2324 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
2325 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
2326 spec.numWorkGroups = IVec3(numElements, 1, 1);
2327 spec.verifyIO = &compareNan;
2329 group->addChild(new SpvAsmComputeShaderCase(
2330 testCtx, "propagated_nans", "Check that nans are propagated", spec));
2334 ComputeShaderSpec spec;
2335 vector<float> small;
2336 vector<float> zeros;
2337 const deUint32 numElements = 100;
2339 small.reserve(numElements);
2340 zeros.reserve(numElements);
2342 for (size_t idx = 0; idx < numElements; ++idx)
2347 small.push_back(0.f);
2348 zeros.push_back(0.f);
2351 small.push_back(-0.f);
2352 zeros.push_back(-0.f);
2355 small.push_back(std::ldexp(1.0f, -16));
2356 zeros.push_back(0.f);
2359 small.push_back(std::ldexp(-1.0f, -32));
2360 zeros.push_back(-0.f);
2363 small.push_back(std::ldexp(1.0f, -127));
2364 zeros.push_back(0.f);
2367 small.push_back(-std::ldexp(1.0f, -128));
2368 zeros.push_back(-0.f);
2373 spec.assembly = shader;
2374 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
2375 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
2376 spec.numWorkGroups = IVec3(numElements, 1, 1);
2378 group->addChild(new SpvAsmComputeShaderCase(
2379 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
2383 ComputeShaderSpec spec;
2384 vector<float> exact;
2385 const deUint32 numElements = 200;
2387 exact.reserve(numElements);
2389 for (size_t idx = 0; idx < numElements; ++idx)
2390 exact.push_back(static_cast<float>(idx - 100));
2392 spec.assembly = shader;
2393 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
2394 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
2395 spec.numWorkGroups = IVec3(numElements, 1, 1);
2397 group->addChild(new SpvAsmComputeShaderCase(
2398 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
2402 ComputeShaderSpec spec;
2403 vector<float> inputs;
2404 const deUint32 numElements = 4;
2406 inputs.push_back(constructNormalizedFloat(8, 0x300300));
2407 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
2408 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
2409 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
2411 spec.assembly = shader;
2412 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
2413 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2414 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
2415 spec.numWorkGroups = IVec3(numElements, 1, 1);
2417 group->addChild(new SpvAsmComputeShaderCase(
2418 testCtx, "rounded", "Check that are rounded when needed", spec));
2421 return group.release();
2424 // Performs a bitwise copy of source to the destination type Dest.
2425 template <typename Dest, typename Src>
2426 Dest bitwiseCast(Src source)
2429 DE_STATIC_ASSERT(sizeof(source) == sizeof(dest));
2430 deMemcpy(&dest, &source, sizeof(dest));
2434 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
2436 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
2438 const std::string shader (
2439 string(s_ShaderPreamble) +
2441 "OpName %main \"main\"\n"
2442 "OpName %id \"gl_GlobalInvocationID\"\n"
2444 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2446 "OpDecorate %sc_0 SpecId 0\n"
2447 "OpDecorate %sc_1 SpecId 1\n"
2448 "OpDecorate %sc_2 SpecId 2\n"
2449 "OpDecorate %sc_3 SpecId 3\n"
2450 "OpDecorate %sc_4 SpecId 4\n"
2451 "OpDecorate %sc_5 SpecId 5\n"
2453 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2455 "%id = OpVariable %uvec3ptr Input\n"
2456 "%zero = OpConstant %i32 0\n"
2457 "%c_u32_6 = OpConstant %u32 6\n"
2459 "%sc_0 = OpSpecConstant %f32 0.\n"
2460 "%sc_1 = OpSpecConstant %f32 0.\n"
2461 "%sc_2 = OpSpecConstant %f32 0.\n"
2462 "%sc_3 = OpSpecConstant %f32 0.\n"
2463 "%sc_4 = OpSpecConstant %f32 0.\n"
2464 "%sc_5 = OpSpecConstant %f32 0.\n"
2466 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
2467 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
2468 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
2469 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
2470 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
2471 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
2473 "%main = OpFunction %void None %voidf\n"
2474 "%label = OpLabel\n"
2475 "%idval = OpLoad %uvec3 %id\n"
2476 "%x = OpCompositeExtract %u32 %idval 0\n"
2477 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2478 "%selector = OpUMod %u32 %x %c_u32_6\n"
2479 " OpSelectionMerge %exit None\n"
2480 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
2482 "%case0 = OpLabel\n"
2483 " OpStore %outloc %sc_0_quant\n"
2486 "%case1 = OpLabel\n"
2487 " OpStore %outloc %sc_1_quant\n"
2490 "%case2 = OpLabel\n"
2491 " OpStore %outloc %sc_2_quant\n"
2494 "%case3 = OpLabel\n"
2495 " OpStore %outloc %sc_3_quant\n"
2498 "%case4 = OpLabel\n"
2499 " OpStore %outloc %sc_4_quant\n"
2502 "%case5 = OpLabel\n"
2503 " OpStore %outloc %sc_5_quant\n"
2509 " OpFunctionEnd\n");
2512 ComputeShaderSpec spec;
2513 const deUint8 numCases = 4;
2514 vector<float> inputs (numCases, 0.f);
2515 vector<float> outputs;
2517 spec.numWorkGroups = IVec3(numCases, 1, 1);
2519 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
2520 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
2521 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
2522 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
2524 outputs.push_back(std::numeric_limits<float>::infinity());
2525 outputs.push_back(-std::numeric_limits<float>::infinity());
2526 outputs.push_back(std::numeric_limits<float>::infinity());
2527 outputs.push_back(-std::numeric_limits<float>::infinity());
2529 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2530 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2532 group->addChild(new SpvAsmComputeShaderCase(
2533 testCtx, "infinities", "Check that infinities propagated and created", spec));
2537 ComputeShaderSpec spec;
2538 const deUint8 numCases = 2;
2539 vector<float> inputs (numCases, 0.f);
2540 vector<float> outputs;
2542 spec.numWorkGroups = IVec3(numCases, 1, 1);
2543 spec.verifyIO = &compareNan;
2545 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
2546 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
2548 for (deUint8 idx = 0; idx < numCases; ++idx)
2549 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
2551 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2552 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2554 group->addChild(new SpvAsmComputeShaderCase(
2555 testCtx, "propagated_nans", "Check that nans are propagated", spec));
2559 ComputeShaderSpec spec;
2560 const deUint8 numCases = 6;
2561 vector<float> inputs (numCases, 0.f);
2562 vector<float> outputs;
2564 spec.numWorkGroups = IVec3(numCases, 1, 1);
2566 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
2567 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
2568 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
2569 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
2570 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
2571 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
2573 outputs.push_back(0.f);
2574 outputs.push_back(-0.f);
2575 outputs.push_back(0.f);
2576 outputs.push_back(-0.f);
2577 outputs.push_back(0.f);
2578 outputs.push_back(-0.f);
2580 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2581 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2583 group->addChild(new SpvAsmComputeShaderCase(
2584 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
2588 ComputeShaderSpec spec;
2589 const deUint8 numCases = 6;
2590 vector<float> inputs (numCases, 0.f);
2591 vector<float> outputs;
2593 spec.numWorkGroups = IVec3(numCases, 1, 1);
2595 for (deUint8 idx = 0; idx < 6; ++idx)
2597 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
2598 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
2599 outputs.push_back(f);
2602 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2603 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2605 group->addChild(new SpvAsmComputeShaderCase(
2606 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
2610 ComputeShaderSpec spec;
2611 const deUint8 numCases = 4;
2612 vector<float> inputs (numCases, 0.f);
2613 vector<float> outputs;
2615 spec.numWorkGroups = IVec3(numCases, 1, 1);
2616 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
2618 outputs.push_back(constructNormalizedFloat(8, 0x300300));
2619 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
2620 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
2621 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
2623 for (deUint8 idx = 0; idx < numCases; ++idx)
2624 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
2626 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2627 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2629 group->addChild(new SpvAsmComputeShaderCase(
2630 testCtx, "rounded", "Check that are rounded when needed", spec));
2633 return group.release();
2636 // Checks that constant null/composite values can be used in computation.
2637 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
2639 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
2640 ComputeShaderSpec spec;
2641 de::Random rnd (deStringHash(group->getName()));
2642 const int numElements = 100;
2643 vector<float> positiveFloats (numElements, 0);
2644 vector<float> negativeFloats (numElements, 0);
2646 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2648 for (size_t ndx = 0; ndx < numElements; ++ndx)
2649 negativeFloats[ndx] = -positiveFloats[ndx];
2652 "OpCapability Shader\n"
2653 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
2654 "OpMemoryModel Logical GLSL450\n"
2655 "OpEntryPoint GLCompute %main \"main\" %id\n"
2656 "OpExecutionMode %main LocalSize 1 1 1\n"
2658 "OpSource GLSL 430\n"
2659 "OpName %main \"main\"\n"
2660 "OpName %id \"gl_GlobalInvocationID\"\n"
2662 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2664 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
2666 "%fmat = OpTypeMatrix %fvec3 3\n"
2667 "%ten = OpConstant %u32 10\n"
2668 "%f32arr10 = OpTypeArray %f32 %ten\n"
2669 "%fst = OpTypeStruct %f32 %f32\n"
2671 + string(s_InputOutputBuffer) +
2673 "%id = OpVariable %uvec3ptr Input\n"
2674 "%zero = OpConstant %i32 0\n"
2676 // Create a bunch of null values
2677 "%unull = OpConstantNull %u32\n"
2678 "%fnull = OpConstantNull %f32\n"
2679 "%vnull = OpConstantNull %fvec3\n"
2680 "%mnull = OpConstantNull %fmat\n"
2681 "%anull = OpConstantNull %f32arr10\n"
2682 "%snull = OpConstantComposite %fst %fnull %fnull\n"
2684 "%main = OpFunction %void None %voidf\n"
2685 "%label = OpLabel\n"
2686 "%idval = OpLoad %uvec3 %id\n"
2687 "%x = OpCompositeExtract %u32 %idval 0\n"
2688 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2689 "%inval = OpLoad %f32 %inloc\n"
2690 "%neg = OpFNegate %f32 %inval\n"
2692 // Get the abs() of (a certain element of) those null values
2693 "%unull_cov = OpConvertUToF %f32 %unull\n"
2694 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
2695 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
2696 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
2697 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
2698 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
2699 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
2700 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
2701 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
2702 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
2703 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
2706 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
2707 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
2708 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
2709 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
2710 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
2711 "%final = OpFAdd %f32 %add5 %snull_abs\n"
2713 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2714 " OpStore %outloc %final\n" // write to output
2717 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2718 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2719 spec.numWorkGroups = IVec3(numElements, 1, 1);
2721 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
2723 return group.release();
2726 // Assembly code used for testing loop control is based on GLSL source code:
2729 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2730 // float elements[];
2732 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2733 // float elements[];
2737 // uint x = gl_GlobalInvocationID.x;
2738 // output_data.elements[x] = input_data.elements[x];
2739 // for (uint i = 0; i < 4; ++i)
2740 // output_data.elements[x] += 1.f;
2742 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
2744 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
2745 vector<CaseParameter> cases;
2746 de::Random rnd (deStringHash(group->getName()));
2747 const int numElements = 100;
2748 vector<float> inputFloats (numElements, 0);
2749 vector<float> outputFloats (numElements, 0);
2750 const StringTemplate shaderTemplate (
2751 string(s_ShaderPreamble) +
2753 "OpSource GLSL 430\n"
2754 "OpName %main \"main\"\n"
2755 "OpName %id \"gl_GlobalInvocationID\"\n"
2757 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2759 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2761 "%u32ptr = OpTypePointer Function %u32\n"
2763 "%id = OpVariable %uvec3ptr Input\n"
2764 "%zero = OpConstant %i32 0\n"
2765 "%uzero = OpConstant %u32 0\n"
2766 "%one = OpConstant %i32 1\n"
2767 "%constf1 = OpConstant %f32 1.0\n"
2768 "%four = OpConstant %u32 4\n"
2770 "%main = OpFunction %void None %voidf\n"
2771 "%entry = OpLabel\n"
2772 "%i = OpVariable %u32ptr Function\n"
2773 " OpStore %i %uzero\n"
2775 "%idval = OpLoad %uvec3 %id\n"
2776 "%x = OpCompositeExtract %u32 %idval 0\n"
2777 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2778 "%inval = OpLoad %f32 %inloc\n"
2779 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2780 " OpStore %outloc %inval\n"
2781 " OpBranch %loop_entry\n"
2783 "%loop_entry = OpLabel\n"
2784 "%i_val = OpLoad %u32 %i\n"
2785 "%cmp_lt = OpULessThan %bool %i_val %four\n"
2786 " OpLoopMerge %loop_merge %loop_entry ${CONTROL}\n"
2787 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
2788 "%loop_body = OpLabel\n"
2789 "%outval = OpLoad %f32 %outloc\n"
2790 "%addf1 = OpFAdd %f32 %outval %constf1\n"
2791 " OpStore %outloc %addf1\n"
2792 "%new_i = OpIAdd %u32 %i_val %one\n"
2793 " OpStore %i %new_i\n"
2794 " OpBranch %loop_entry\n"
2795 "%loop_merge = OpLabel\n"
2797 " OpFunctionEnd\n");
2799 cases.push_back(CaseParameter("none", "None"));
2800 cases.push_back(CaseParameter("unroll", "Unroll"));
2801 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
2802 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
2804 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2806 for (size_t ndx = 0; ndx < numElements; ++ndx)
2807 outputFloats[ndx] = inputFloats[ndx] + 4.f;
2809 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2811 map<string, string> specializations;
2812 ComputeShaderSpec spec;
2814 specializations["CONTROL"] = cases[caseNdx].param;
2815 spec.assembly = shaderTemplate.specialize(specializations);
2816 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2817 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2818 spec.numWorkGroups = IVec3(numElements, 1, 1);
2820 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2823 return group.release();
2826 // Assembly code used for testing selection control is based on GLSL source code:
2829 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2830 // float elements[];
2832 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2833 // float elements[];
2837 // uint x = gl_GlobalInvocationID.x;
2838 // float val = input_data.elements[x];
2840 // output_data.elements[x] = val + 1.f;
2842 // output_data.elements[x] = val - 1.f;
2844 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
2846 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
2847 vector<CaseParameter> cases;
2848 de::Random rnd (deStringHash(group->getName()));
2849 const int numElements = 100;
2850 vector<float> inputFloats (numElements, 0);
2851 vector<float> outputFloats (numElements, 0);
2852 const StringTemplate shaderTemplate (
2853 string(s_ShaderPreamble) +
2855 "OpSource GLSL 430\n"
2856 "OpName %main \"main\"\n"
2857 "OpName %id \"gl_GlobalInvocationID\"\n"
2859 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2861 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2863 "%id = OpVariable %uvec3ptr Input\n"
2864 "%zero = OpConstant %i32 0\n"
2865 "%constf1 = OpConstant %f32 1.0\n"
2866 "%constf10 = OpConstant %f32 10.0\n"
2868 "%main = OpFunction %void None %voidf\n"
2869 "%entry = OpLabel\n"
2870 "%idval = OpLoad %uvec3 %id\n"
2871 "%x = OpCompositeExtract %u32 %idval 0\n"
2872 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2873 "%inval = OpLoad %f32 %inloc\n"
2874 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2875 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
2877 " OpSelectionMerge %if_end ${CONTROL}\n"
2878 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2879 "%if_true = OpLabel\n"
2880 "%addf1 = OpFAdd %f32 %inval %constf1\n"
2881 " OpStore %outloc %addf1\n"
2882 " OpBranch %if_end\n"
2883 "%if_false = OpLabel\n"
2884 "%subf1 = OpFSub %f32 %inval %constf1\n"
2885 " OpStore %outloc %subf1\n"
2886 " OpBranch %if_end\n"
2887 "%if_end = OpLabel\n"
2889 " OpFunctionEnd\n");
2891 cases.push_back(CaseParameter("none", "None"));
2892 cases.push_back(CaseParameter("flatten", "Flatten"));
2893 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
2894 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
2896 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2898 for (size_t ndx = 0; ndx < numElements; ++ndx)
2899 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
2901 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2903 map<string, string> specializations;
2904 ComputeShaderSpec spec;
2906 specializations["CONTROL"] = cases[caseNdx].param;
2907 spec.assembly = shaderTemplate.specialize(specializations);
2908 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2909 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2910 spec.numWorkGroups = IVec3(numElements, 1, 1);
2912 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2915 return group.release();
2918 // Assembly code used for testing function control is based on GLSL source code:
2922 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2923 // float elements[];
2925 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2926 // float elements[];
2929 // float const10() { return 10.f; }
2932 // uint x = gl_GlobalInvocationID.x;
2933 // output_data.elements[x] = input_data.elements[x] + const10();
2935 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
2937 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
2938 vector<CaseParameter> cases;
2939 de::Random rnd (deStringHash(group->getName()));
2940 const int numElements = 100;
2941 vector<float> inputFloats (numElements, 0);
2942 vector<float> outputFloats (numElements, 0);
2943 const StringTemplate shaderTemplate (
2944 string(s_ShaderPreamble) +
2946 "OpSource GLSL 430\n"
2947 "OpName %main \"main\"\n"
2948 "OpName %func_const10 \"const10(\"\n"
2949 "OpName %id \"gl_GlobalInvocationID\"\n"
2951 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2953 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2955 "%f32f = OpTypeFunction %f32\n"
2956 "%id = OpVariable %uvec3ptr Input\n"
2957 "%zero = OpConstant %i32 0\n"
2958 "%constf10 = OpConstant %f32 10.0\n"
2960 "%main = OpFunction %void None %voidf\n"
2961 "%entry = OpLabel\n"
2962 "%idval = OpLoad %uvec3 %id\n"
2963 "%x = OpCompositeExtract %u32 %idval 0\n"
2964 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2965 "%inval = OpLoad %f32 %inloc\n"
2966 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
2967 "%fadd = OpFAdd %f32 %inval %ret_10\n"
2968 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2969 " OpStore %outloc %fadd\n"
2973 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
2974 "%label = OpLabel\n"
2975 " OpReturnValue %constf10\n"
2976 " OpFunctionEnd\n");
2978 cases.push_back(CaseParameter("none", "None"));
2979 cases.push_back(CaseParameter("inline", "Inline"));
2980 cases.push_back(CaseParameter("dont_inline", "DontInline"));
2981 cases.push_back(CaseParameter("pure", "Pure"));
2982 cases.push_back(CaseParameter("const", "Const"));
2983 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
2984 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
2985 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
2986 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
2988 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2990 for (size_t ndx = 0; ndx < numElements; ++ndx)
2991 outputFloats[ndx] = inputFloats[ndx] + 10.f;
2993 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2995 map<string, string> specializations;
2996 ComputeShaderSpec spec;
2998 specializations["CONTROL"] = cases[caseNdx].param;
2999 spec.assembly = shaderTemplate.specialize(specializations);
3000 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3001 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3002 spec.numWorkGroups = IVec3(numElements, 1, 1);
3004 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3007 return group.release();
3010 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
3012 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
3013 vector<CaseParameter> cases;
3014 de::Random rnd (deStringHash(group->getName()));
3015 const int numElements = 100;
3016 vector<float> inputFloats (numElements, 0);
3017 vector<float> outputFloats (numElements, 0);
3018 const StringTemplate shaderTemplate (
3019 string(s_ShaderPreamble) +
3021 "OpSource GLSL 430\n"
3022 "OpName %main \"main\"\n"
3023 "OpName %id \"gl_GlobalInvocationID\"\n"
3025 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3027 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
3029 "%f32ptr_f = OpTypePointer Function %f32\n"
3031 "%id = OpVariable %uvec3ptr Input\n"
3032 "%zero = OpConstant %i32 0\n"
3033 "%four = OpConstant %i32 4\n"
3035 "%main = OpFunction %void None %voidf\n"
3036 "%label = OpLabel\n"
3037 "%copy = OpVariable %f32ptr_f Function\n"
3038 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
3039 "%x = OpCompositeExtract %u32 %idval 0\n"
3040 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3041 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3042 " OpCopyMemory %copy %inloc ${ACCESS}\n"
3043 "%val1 = OpLoad %f32 %copy\n"
3044 "%val2 = OpLoad %f32 %inloc\n"
3045 "%add = OpFAdd %f32 %val1 %val2\n"
3046 " OpStore %outloc %add ${ACCESS}\n"
3048 " OpFunctionEnd\n");
3050 cases.push_back(CaseParameter("null", ""));
3051 cases.push_back(CaseParameter("none", "None"));
3052 cases.push_back(CaseParameter("volatile", "Volatile"));
3053 cases.push_back(CaseParameter("aligned", "Aligned 4"));
3054 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
3055 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
3056 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
3058 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3060 for (size_t ndx = 0; ndx < numElements; ++ndx)
3061 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
3063 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3065 map<string, string> specializations;
3066 ComputeShaderSpec spec;
3068 specializations["ACCESS"] = cases[caseNdx].param;
3069 spec.assembly = shaderTemplate.specialize(specializations);
3070 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3071 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3072 spec.numWorkGroups = IVec3(numElements, 1, 1);
3074 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3077 return group.release();
3080 // Checks that we can get undefined values for various types, without exercising a computation with it.
3081 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
3083 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
3084 vector<CaseParameter> cases;
3085 de::Random rnd (deStringHash(group->getName()));
3086 const int numElements = 100;
3087 vector<float> positiveFloats (numElements, 0);
3088 vector<float> negativeFloats (numElements, 0);
3089 const StringTemplate shaderTemplate (
3090 string(s_ShaderPreamble) +
3092 "OpSource GLSL 430\n"
3093 "OpName %main \"main\"\n"
3094 "OpName %id \"gl_GlobalInvocationID\"\n"
3096 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3098 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
3102 "%id = OpVariable %uvec3ptr Input\n"
3103 "%zero = OpConstant %i32 0\n"
3105 "%main = OpFunction %void None %voidf\n"
3106 "%label = OpLabel\n"
3108 "%undef = OpUndef %type\n"
3110 "%idval = OpLoad %uvec3 %id\n"
3111 "%x = OpCompositeExtract %u32 %idval 0\n"
3113 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3114 "%inval = OpLoad %f32 %inloc\n"
3115 "%neg = OpFNegate %f32 %inval\n"
3116 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3117 " OpStore %outloc %neg\n"
3119 " OpFunctionEnd\n");
3121 cases.push_back(CaseParameter("bool", "%type = OpTypeBool"));
3122 cases.push_back(CaseParameter("sint32", "%type = OpTypeInt 32 1"));
3123 cases.push_back(CaseParameter("uint32", "%type = OpTypeInt 32 0"));
3124 cases.push_back(CaseParameter("float32", "%type = OpTypeFloat 32"));
3125 cases.push_back(CaseParameter("vec4float32", "%type = OpTypeVector %f32 4"));
3126 cases.push_back(CaseParameter("vec2uint32", "%type = OpTypeVector %u32 2"));
3127 cases.push_back(CaseParameter("matrix", "%type = OpTypeMatrix %fvec3 3"));
3128 cases.push_back(CaseParameter("image", "%type = OpTypeImage %f32 2D 0 0 0 0 Unknown"));
3129 cases.push_back(CaseParameter("sampler", "%type = OpTypeSampler"));
3130 cases.push_back(CaseParameter("sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 0 Unknown\n"
3131 "%type = OpTypeSampledImage %img"));
3132 cases.push_back(CaseParameter("array", "%100 = OpConstant %u32 100\n"
3133 "%type = OpTypeArray %i32 %100"));
3134 cases.push_back(CaseParameter("runtimearray", "%type = OpTypeRuntimeArray %f32"));
3135 cases.push_back(CaseParameter("struct", "%type = OpTypeStruct %f32 %i32 %u32"));
3136 cases.push_back(CaseParameter("pointer", "%type = OpTypePointer Function %i32"));
3138 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3140 for (size_t ndx = 0; ndx < numElements; ++ndx)
3141 negativeFloats[ndx] = -positiveFloats[ndx];
3143 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3145 map<string, string> specializations;
3146 ComputeShaderSpec spec;
3148 specializations["TYPE"] = cases[caseNdx].param;
3149 spec.assembly = shaderTemplate.specialize(specializations);
3150 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3151 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3152 spec.numWorkGroups = IVec3(numElements, 1, 1);
3154 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3157 return group.release();
3159 typedef std::pair<std::string, VkShaderStageFlagBits> EntryToStage;
3160 typedef map<string, vector<EntryToStage> > ModuleMap;
3161 typedef map<VkShaderStageFlagBits, vector<deInt32> > StageToSpecConstantMap;
3163 // Context for a specific test instantiation. For example, an instantiation
3164 // may test colors yellow/magenta/cyan/mauve in a tesselation shader
3165 // with an entry point named 'main_to_the_main'
3166 struct InstanceContext
3168 // Map of modules to what entry_points we care to use from those modules.
3169 ModuleMap moduleMap;
3170 RGBA inputColors[4];
3171 RGBA outputColors[4];
3172 // Concrete SPIR-V code to test via boilerplate specialization.
3173 map<string, string> testCodeFragments;
3174 StageToSpecConstantMap specConstants;
3175 bool hasTessellation;
3176 VkShaderStageFlagBits requiredStages;
3178 InstanceContext (const RGBA (&inputs)[4], const RGBA (&outputs)[4], const map<string, string>& testCodeFragments_, const StageToSpecConstantMap& specConstants_)
3179 : testCodeFragments (testCodeFragments_)
3180 , specConstants (specConstants_)
3181 , hasTessellation (false)
3182 , requiredStages (static_cast<VkShaderStageFlagBits>(0))
3184 inputColors[0] = inputs[0];
3185 inputColors[1] = inputs[1];
3186 inputColors[2] = inputs[2];
3187 inputColors[3] = inputs[3];
3189 outputColors[0] = outputs[0];
3190 outputColors[1] = outputs[1];
3191 outputColors[2] = outputs[2];
3192 outputColors[3] = outputs[3];
3195 InstanceContext (const InstanceContext& other)
3196 : moduleMap (other.moduleMap)
3197 , testCodeFragments (other.testCodeFragments)
3198 , specConstants (other.specConstants)
3199 , hasTessellation (other.hasTessellation)
3200 , requiredStages (other.requiredStages)
3202 inputColors[0] = other.inputColors[0];
3203 inputColors[1] = other.inputColors[1];
3204 inputColors[2] = other.inputColors[2];
3205 inputColors[3] = other.inputColors[3];
3207 outputColors[0] = other.outputColors[0];
3208 outputColors[1] = other.outputColors[1];
3209 outputColors[2] = other.outputColors[2];
3210 outputColors[3] = other.outputColors[3];
3214 // A description of a shader to be used for a single stage of the graphics pipeline.
3215 struct ShaderElement
3217 // The module that contains this shader entrypoint.
3220 // The name of the entrypoint.
3223 // Which shader stage this entry point represents.
3224 VkShaderStageFlagBits stage;
3226 ShaderElement (const string& moduleName_, const string& entryPoint_, VkShaderStageFlagBits shaderStage_)
3227 : moduleName(moduleName_)
3228 , entryName(entryPoint_)
3229 , stage(shaderStage_)
3234 void getDefaultColors (RGBA (&colors)[4])
3236 colors[0] = RGBA::white();
3237 colors[1] = RGBA::red();
3238 colors[2] = RGBA::green();
3239 colors[3] = RGBA::blue();
3242 void getHalfColorsFullAlpha (RGBA (&colors)[4])
3244 colors[0] = RGBA(127, 127, 127, 255);
3245 colors[1] = RGBA(127, 0, 0, 255);
3246 colors[2] = RGBA(0, 127, 0, 255);
3247 colors[3] = RGBA(0, 0, 127, 255);
3250 void getInvertedDefaultColors (RGBA (&colors)[4])
3252 colors[0] = RGBA(0, 0, 0, 255);
3253 colors[1] = RGBA(0, 255, 255, 255);
3254 colors[2] = RGBA(255, 0, 255, 255);
3255 colors[3] = RGBA(255, 255, 0, 255);
3258 // Turns a statically sized array of ShaderElements into an instance-context
3259 // by setting up the mapping of modules to their contained shaders and stages.
3260 // The inputs and expected outputs are given by inputColors and outputColors
3262 InstanceContext createInstanceContext (const ShaderElement (&elements)[N], const RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments, const StageToSpecConstantMap& specConstants)
3264 InstanceContext ctx (inputColors, outputColors, testCodeFragments, specConstants);
3265 for (size_t i = 0; i < N; ++i)
3267 ctx.moduleMap[elements[i].moduleName].push_back(std::make_pair(elements[i].entryName, elements[i].stage));
3268 ctx.requiredStages = static_cast<VkShaderStageFlagBits>(ctx.requiredStages | elements[i].stage);
3274 inline InstanceContext createInstanceContext (const ShaderElement (&elements)[N], RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments)
3276 return createInstanceContext(elements, inputColors, outputColors, testCodeFragments, StageToSpecConstantMap());
3279 // The same as createInstanceContext above, but with default colors.
3281 InstanceContext createInstanceContext (const ShaderElement (&elements)[N], const map<string, string>& testCodeFragments)
3283 RGBA defaultColors[4];
3284 getDefaultColors(defaultColors);
3285 return createInstanceContext(elements, defaultColors, defaultColors, testCodeFragments);
3288 // For the current InstanceContext, constructs the required modules and shader stage create infos.
3289 void createPipelineShaderStages (const DeviceInterface& vk, const VkDevice vkDevice, InstanceContext& instance, Context& context, vector<ModuleHandleSp>& modules, vector<VkPipelineShaderStageCreateInfo>& createInfos)
3291 for (ModuleMap::const_iterator moduleNdx = instance.moduleMap.begin(); moduleNdx != instance.moduleMap.end(); ++moduleNdx)
3293 const ModuleHandleSp mod(new Unique<VkShaderModule>(createShaderModule(vk, vkDevice, context.getBinaryCollection().get(moduleNdx->first), 0)));
3294 modules.push_back(ModuleHandleSp(mod));
3295 for (vector<EntryToStage>::const_iterator shaderNdx = moduleNdx->second.begin(); shaderNdx != moduleNdx->second.end(); ++shaderNdx)
3297 const EntryToStage& stage = *shaderNdx;
3298 const VkPipelineShaderStageCreateInfo shaderParam =
3300 VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, // VkStructureType sType;
3301 DE_NULL, // const void* pNext;
3302 (VkPipelineShaderStageCreateFlags)0,
3303 stage.second, // VkShaderStageFlagBits stage;
3304 **modules.back(), // VkShaderModule module;
3305 stage.first.c_str(), // const char* pName;
3306 (const VkSpecializationInfo*)DE_NULL,
3308 createInfos.push_back(shaderParam);
3313 #define SPIRV_ASSEMBLY_TYPES \
3314 "%void = OpTypeVoid\n" \
3315 "%bool = OpTypeBool\n" \
3317 "%i32 = OpTypeInt 32 1\n" \
3318 "%u32 = OpTypeInt 32 0\n" \
3320 "%f32 = OpTypeFloat 32\n" \
3321 "%v3f32 = OpTypeVector %f32 3\n" \
3322 "%v4f32 = OpTypeVector %f32 4\n" \
3324 "%v4f32_function = OpTypeFunction %v4f32 %v4f32\n" \
3325 "%fun = OpTypeFunction %void\n" \
3327 "%ip_f32 = OpTypePointer Input %f32\n" \
3328 "%ip_i32 = OpTypePointer Input %i32\n" \
3329 "%ip_v3f32 = OpTypePointer Input %v3f32\n" \
3330 "%ip_v4f32 = OpTypePointer Input %v4f32\n" \
3332 "%op_f32 = OpTypePointer Output %f32\n" \
3333 "%op_v4f32 = OpTypePointer Output %v4f32\n" \
3335 "%fp_f32 = OpTypePointer Function %f32\n" \
3336 "%fp_i32 = OpTypePointer Function %i32\n" \
3337 "%fp_v4f32 = OpTypePointer Function %v4f32\n"
3339 #define SPIRV_ASSEMBLY_CONSTANTS \
3340 "%c_f32_1 = OpConstant %f32 1.0\n" \
3341 "%c_f32_0 = OpConstant %f32 0.0\n" \
3342 "%c_f32_0_5 = OpConstant %f32 0.5\n" \
3343 "%c_f32_n1 = OpConstant %f32 -1.\n" \
3344 "%c_f32_7 = OpConstant %f32 7.0\n" \
3345 "%c_f32_8 = OpConstant %f32 8.0\n" \
3346 "%c_i32_0 = OpConstant %i32 0\n" \
3347 "%c_i32_1 = OpConstant %i32 1\n" \
3348 "%c_i32_2 = OpConstant %i32 2\n" \
3349 "%c_i32_3 = OpConstant %i32 3\n" \
3350 "%c_i32_4 = OpConstant %i32 4\n" \
3351 "%c_u32_0 = OpConstant %u32 0\n" \
3352 "%c_u32_1 = OpConstant %u32 1\n" \
3353 "%c_u32_2 = OpConstant %u32 2\n" \
3354 "%c_u32_3 = OpConstant %u32 3\n" \
3355 "%c_u32_32 = OpConstant %u32 32\n" \
3356 "%c_u32_4 = OpConstant %u32 4\n" \
3357 "%c_u32_31_bits = OpConstant %u32 0x7FFFFFFF\n" \
3358 "%c_v4f32_1_1_1_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n" \
3359 "%c_v4f32_1_0_0_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_1\n" \
3360 "%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"
3362 #define SPIRV_ASSEMBLY_ARRAYS \
3363 "%a1f32 = OpTypeArray %f32 %c_u32_1\n" \
3364 "%a2f32 = OpTypeArray %f32 %c_u32_2\n" \
3365 "%a3v4f32 = OpTypeArray %v4f32 %c_u32_3\n" \
3366 "%a4f32 = OpTypeArray %f32 %c_u32_4\n" \
3367 "%a32v4f32 = OpTypeArray %v4f32 %c_u32_32\n" \
3368 "%ip_a3v4f32 = OpTypePointer Input %a3v4f32\n" \
3369 "%ip_a32v4f32 = OpTypePointer Input %a32v4f32\n" \
3370 "%op_a2f32 = OpTypePointer Output %a2f32\n" \
3371 "%op_a3v4f32 = OpTypePointer Output %a3v4f32\n" \
3372 "%op_a4f32 = OpTypePointer Output %a4f32\n"
3374 // Creates vertex-shader assembly by specializing a boilerplate StringTemplate
3375 // on fragments, which must (at least) map "testfun" to an OpFunction definition
3376 // for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed
3377 // with "BP_" to avoid collisions with fragments.
3379 // It corresponds roughly to this GLSL:
3381 // layout(location = 0) in vec4 position;
3382 // layout(location = 1) in vec4 color;
3383 // layout(location = 1) out highp vec4 vtxColor;
3384 // void main (void) { gl_Position = position; vtxColor = test_func(color); }
3385 string makeVertexShaderAssembly(const map<string, string>& fragments)
3387 // \todo [2015-11-23 awoloszyn] Remove OpName once these have stabalized
3388 static const char vertexShaderBoilerplate[] =
3389 "OpCapability Shader\n"
3390 "OpMemoryModel Logical GLSL450\n"
3391 "OpEntryPoint Vertex %main \"main\" %BP_stream %BP_position %BP_vtx_color %BP_color %BP_gl_VertexIndex %BP_gl_InstanceIndex\n"
3393 "OpName %main \"main\"\n"
3394 "OpName %BP_gl_PerVertex \"gl_PerVertex\"\n"
3395 "OpMemberName %BP_gl_PerVertex 0 \"gl_Position\"\n"
3396 "OpMemberName %BP_gl_PerVertex 1 \"gl_PointSize\"\n"
3397 "OpMemberName %BP_gl_PerVertex 2 \"gl_ClipDistance\"\n"
3398 "OpMemberName %BP_gl_PerVertex 3 \"gl_CullDistance\"\n"
3399 "OpName %test_code \"testfun(vf4;\"\n"
3400 "OpName %BP_stream \"\"\n"
3401 "OpName %BP_position \"position\"\n"
3402 "OpName %BP_vtx_color \"vtxColor\"\n"
3403 "OpName %BP_color \"color\"\n"
3404 "OpName %BP_gl_VertexIndex \"gl_VertexIndex\"\n"
3405 "OpName %BP_gl_InstanceIndex \"gl_InstanceIndex\"\n"
3406 "OpMemberDecorate %BP_gl_PerVertex 0 BuiltIn Position\n"
3407 "OpMemberDecorate %BP_gl_PerVertex 1 BuiltIn PointSize\n"
3408 "OpMemberDecorate %BP_gl_PerVertex 2 BuiltIn ClipDistance\n"
3409 "OpMemberDecorate %BP_gl_PerVertex 3 BuiltIn CullDistance\n"
3410 "OpDecorate %BP_gl_PerVertex Block\n"
3411 "OpDecorate %BP_position Location 0\n"
3412 "OpDecorate %BP_vtx_color Location 1\n"
3413 "OpDecorate %BP_color Location 1\n"
3414 "OpDecorate %BP_gl_VertexIndex BuiltIn VertexIndex\n"
3415 "OpDecorate %BP_gl_InstanceIndex BuiltIn InstanceIndex\n"
3416 "${decoration:opt}\n"
3417 SPIRV_ASSEMBLY_TYPES
3418 SPIRV_ASSEMBLY_CONSTANTS
3419 SPIRV_ASSEMBLY_ARRAYS
3420 "%BP_gl_PerVertex = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3421 "%BP_op_gl_PerVertex = OpTypePointer Output %BP_gl_PerVertex\n"
3422 "%BP_stream = OpVariable %BP_op_gl_PerVertex Output\n"
3423 "%BP_position = OpVariable %ip_v4f32 Input\n"
3424 "%BP_vtx_color = OpVariable %op_v4f32 Output\n"
3425 "%BP_color = OpVariable %ip_v4f32 Input\n"
3426 "%BP_gl_VertexIndex = OpVariable %ip_i32 Input\n"
3427 "%BP_gl_InstanceIndex = OpVariable %ip_i32 Input\n"
3429 "%main = OpFunction %void None %fun\n"
3430 "%BP_label = OpLabel\n"
3431 "%BP_pos = OpLoad %v4f32 %BP_position\n"
3432 "%BP_gl_pos = OpAccessChain %op_v4f32 %BP_stream %c_i32_0\n"
3433 "OpStore %BP_gl_pos %BP_pos\n"
3434 "%BP_col = OpLoad %v4f32 %BP_color\n"
3435 "%BP_col_transformed = OpFunctionCall %v4f32 %test_code %BP_col\n"
3436 "OpStore %BP_vtx_color %BP_col_transformed\n"
3440 return tcu::StringTemplate(vertexShaderBoilerplate).specialize(fragments);
3443 // Creates tess-control-shader assembly by specializing a boilerplate
3444 // StringTemplate on fragments, which must (at least) map "testfun" to an
3445 // OpFunction definition for %test_code that takes and returns a %v4f32.
3446 // Boilerplate IDs are prefixed with "BP_" to avoid collisions with fragments.
3448 // It roughly corresponds to the following GLSL.
3451 // layout(vertices = 3) out;
3452 // layout(location = 1) in vec4 in_color[];
3453 // layout(location = 1) out vec4 out_color[];
3456 // out_color[gl_InvocationID] = testfun(in_color[gl_InvocationID]);
3457 // gl_out[gl_InvocationID].gl_Position = gl_in[gl_InvocationID].gl_Position;
3458 // if (gl_InvocationID == 0) {
3459 // gl_TessLevelOuter[0] = 1.0;
3460 // gl_TessLevelOuter[1] = 1.0;
3461 // gl_TessLevelOuter[2] = 1.0;
3462 // gl_TessLevelInner[0] = 1.0;
3465 string makeTessControlShaderAssembly (const map<string, string>& fragments)
3467 static const char tessControlShaderBoilerplate[] =
3468 "OpCapability Tessellation\n"
3469 "OpMemoryModel Logical GLSL450\n"
3470 "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"
3471 "OpExecutionMode %BP_main OutputVertices 3\n"
3473 "OpName %BP_main \"main\"\n"
3474 "OpName %test_code \"testfun(vf4;\"\n"
3475 "OpName %BP_out_color \"out_color\"\n"
3476 "OpName %BP_gl_InvocationID \"gl_InvocationID\"\n"
3477 "OpName %BP_in_color \"in_color\"\n"
3478 "OpName %BP_gl_PerVertex \"gl_PerVertex\"\n"
3479 "OpMemberName %BP_gl_PerVertex 0 \"gl_Position\"\n"
3480 "OpMemberName %BP_gl_PerVertex 1 \"gl_PointSize\"\n"
3481 "OpMemberName %BP_gl_PerVertex 2 \"gl_ClipDistance\"\n"
3482 "OpMemberName %BP_gl_PerVertex 3 \"gl_CullDistance\"\n"
3483 "OpName %BP_gl_out \"gl_out\"\n"
3484 "OpName %BP_gl_PVOut \"gl_PerVertex\"\n"
3485 "OpMemberName %BP_gl_PVOut 0 \"gl_Position\"\n"
3486 "OpMemberName %BP_gl_PVOut 1 \"gl_PointSize\"\n"
3487 "OpMemberName %BP_gl_PVOut 2 \"gl_ClipDistance\"\n"
3488 "OpMemberName %BP_gl_PVOut 3 \"gl_CullDistance\"\n"
3489 "OpName %BP_gl_in \"gl_in\"\n"
3490 "OpName %BP_gl_TessLevelOuter \"gl_TessLevelOuter\"\n"
3491 "OpName %BP_gl_TessLevelInner \"gl_TessLevelInner\"\n"
3492 "OpDecorate %BP_out_color Location 1\n"
3493 "OpDecorate %BP_gl_InvocationID BuiltIn InvocationId\n"
3494 "OpDecorate %BP_in_color Location 1\n"
3495 "OpMemberDecorate %BP_gl_PerVertex 0 BuiltIn Position\n"
3496 "OpMemberDecorate %BP_gl_PerVertex 1 BuiltIn PointSize\n"
3497 "OpMemberDecorate %BP_gl_PerVertex 2 BuiltIn ClipDistance\n"
3498 "OpMemberDecorate %BP_gl_PerVertex 3 BuiltIn CullDistance\n"
3499 "OpDecorate %BP_gl_PerVertex Block\n"
3500 "OpMemberDecorate %BP_gl_PVOut 0 BuiltIn Position\n"
3501 "OpMemberDecorate %BP_gl_PVOut 1 BuiltIn PointSize\n"
3502 "OpMemberDecorate %BP_gl_PVOut 2 BuiltIn ClipDistance\n"
3503 "OpMemberDecorate %BP_gl_PVOut 3 BuiltIn CullDistance\n"
3504 "OpDecorate %BP_gl_PVOut Block\n"
3505 "OpDecorate %BP_gl_TessLevelOuter Patch\n"
3506 "OpDecorate %BP_gl_TessLevelOuter BuiltIn TessLevelOuter\n"
3507 "OpDecorate %BP_gl_TessLevelInner Patch\n"
3508 "OpDecorate %BP_gl_TessLevelInner BuiltIn TessLevelInner\n"
3509 "${decoration:opt}\n"
3510 SPIRV_ASSEMBLY_TYPES
3511 SPIRV_ASSEMBLY_CONSTANTS
3512 SPIRV_ASSEMBLY_ARRAYS
3513 "%BP_out_color = OpVariable %op_a3v4f32 Output\n"
3514 "%BP_gl_InvocationID = OpVariable %ip_i32 Input\n"
3515 "%BP_in_color = OpVariable %ip_a32v4f32 Input\n"
3516 "%BP_gl_PerVertex = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3517 "%BP_a3_gl_PerVertex = OpTypeArray %BP_gl_PerVertex %c_u32_3\n"
3518 "%BP_op_a3_gl_PerVertex = OpTypePointer Output %BP_a3_gl_PerVertex\n"
3519 "%BP_gl_out = OpVariable %BP_op_a3_gl_PerVertex Output\n"
3520 "%BP_gl_PVOut = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3521 "%BP_a32_gl_PVOut = OpTypeArray %BP_gl_PVOut %c_u32_32\n"
3522 "%BP_ip_a32_gl_PVOut = OpTypePointer Input %BP_a32_gl_PVOut\n"
3523 "%BP_gl_in = OpVariable %BP_ip_a32_gl_PVOut Input\n"
3524 "%BP_gl_TessLevelOuter = OpVariable %op_a4f32 Output\n"
3525 "%BP_gl_TessLevelInner = OpVariable %op_a2f32 Output\n"
3528 "%BP_main = OpFunction %void None %fun\n"
3529 "%BP_label = OpLabel\n"
3531 "%BP_gl_Invoc = OpLoad %i32 %BP_gl_InvocationID\n"
3533 "%BP_in_col_loc = OpAccessChain %ip_v4f32 %BP_in_color %BP_gl_Invoc\n"
3534 "%BP_out_col_loc = OpAccessChain %op_v4f32 %BP_out_color %BP_gl_Invoc\n"
3535 "%BP_in_col_val = OpLoad %v4f32 %BP_in_col_loc\n"
3536 "%BP_clr_transformed = OpFunctionCall %v4f32 %test_code %BP_in_col_val\n"
3537 "OpStore %BP_out_col_loc %BP_clr_transformed\n"
3539 "%BP_in_pos_loc = OpAccessChain %ip_v4f32 %BP_gl_in %BP_gl_Invoc %c_i32_0\n"
3540 "%BP_out_pos_loc = OpAccessChain %op_v4f32 %BP_gl_out %BP_gl_Invoc %c_i32_0\n"
3541 "%BP_in_pos_val = OpLoad %v4f32 %BP_in_pos_loc\n"
3542 "OpStore %BP_out_pos_loc %BP_in_pos_val\n"
3544 "%BP_cmp = OpIEqual %bool %BP_gl_Invoc %c_i32_0\n"
3545 "OpSelectionMerge %BP_merge_label None\n"
3546 "OpBranchConditional %BP_cmp %BP_if_label %BP_merge_label\n"
3547 "%BP_if_label = OpLabel\n"
3548 "%BP_gl_TessLevelOuterPos_0 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_0\n"
3549 "%BP_gl_TessLevelOuterPos_1 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_1\n"
3550 "%BP_gl_TessLevelOuterPos_2 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_2\n"
3551 "%BP_gl_TessLevelInnerPos_0 = OpAccessChain %op_f32 %BP_gl_TessLevelInner %c_i32_0\n"
3552 "OpStore %BP_gl_TessLevelOuterPos_0 %c_f32_1\n"
3553 "OpStore %BP_gl_TessLevelOuterPos_1 %c_f32_1\n"
3554 "OpStore %BP_gl_TessLevelOuterPos_2 %c_f32_1\n"
3555 "OpStore %BP_gl_TessLevelInnerPos_0 %c_f32_1\n"
3556 "OpBranch %BP_merge_label\n"
3557 "%BP_merge_label = OpLabel\n"
3561 return tcu::StringTemplate(tessControlShaderBoilerplate).specialize(fragments);
3564 // Creates tess-evaluation-shader assembly by specializing a boilerplate
3565 // StringTemplate on fragments, which must (at least) map "testfun" to an
3566 // OpFunction definition for %test_code that takes and returns a %v4f32.
3567 // Boilerplate IDs are prefixed with "BP_" to avoid collisions with fragments.
3569 // It roughly corresponds to the following glsl.
3573 // layout(triangles, equal_spacing, ccw) in;
3574 // layout(location = 1) in vec4 in_color[];
3575 // layout(location = 1) out vec4 out_color;
3577 // #define interpolate(val)
3578 // vec4(gl_TessCoord.x) * val[0] + vec4(gl_TessCoord.y) * val[1] +
3579 // vec4(gl_TessCoord.z) * val[2]
3582 // gl_Position = vec4(gl_TessCoord.x) * gl_in[0].gl_Position +
3583 // vec4(gl_TessCoord.y) * gl_in[1].gl_Position +
3584 // vec4(gl_TessCoord.z) * gl_in[2].gl_Position;
3585 // out_color = testfun(interpolate(in_color));
3587 string makeTessEvalShaderAssembly(const map<string, string>& fragments)
3589 static const char tessEvalBoilerplate[] =
3590 "OpCapability Tessellation\n"
3591 "OpMemoryModel Logical GLSL450\n"
3592 "OpEntryPoint TessellationEvaluation %BP_main \"main\" %BP_stream %BP_gl_TessCoord %BP_gl_in %BP_out_color %BP_in_color\n"
3593 "OpExecutionMode %BP_main Triangles\n"
3594 "OpExecutionMode %BP_main SpacingEqual\n"
3595 "OpExecutionMode %BP_main VertexOrderCcw\n"
3597 "OpName %BP_main \"main\"\n"
3598 "OpName %test_code \"testfun(vf4;\"\n"
3599 "OpName %BP_gl_PerVertexOut \"gl_PerVertex\"\n"
3600 "OpMemberName %BP_gl_PerVertexOut 0 \"gl_Position\"\n"
3601 "OpMemberName %BP_gl_PerVertexOut 1 \"gl_PointSize\"\n"
3602 "OpMemberName %BP_gl_PerVertexOut 2 \"gl_ClipDistance\"\n"
3603 "OpMemberName %BP_gl_PerVertexOut 3 \"gl_CullDistance\"\n"
3604 "OpName %BP_stream \"\"\n"
3605 "OpName %BP_gl_TessCoord \"gl_TessCoord\"\n"
3606 "OpName %BP_gl_PerVertexIn \"gl_PerVertex\"\n"
3607 "OpMemberName %BP_gl_PerVertexIn 0 \"gl_Position\"\n"
3608 "OpMemberName %BP_gl_PerVertexIn 1 \"gl_PointSize\"\n"
3609 "OpMemberName %BP_gl_PerVertexIn 2 \"gl_ClipDistance\"\n"
3610 "OpMemberName %BP_gl_PerVertexIn 3 \"gl_CullDistance\"\n"
3611 "OpName %BP_gl_in \"gl_in\"\n"
3612 "OpName %BP_out_color \"out_color\"\n"
3613 "OpName %BP_in_color \"in_color\"\n"
3614 "OpMemberDecorate %BP_gl_PerVertexOut 0 BuiltIn Position\n"
3615 "OpMemberDecorate %BP_gl_PerVertexOut 1 BuiltIn PointSize\n"
3616 "OpMemberDecorate %BP_gl_PerVertexOut 2 BuiltIn ClipDistance\n"
3617 "OpMemberDecorate %BP_gl_PerVertexOut 3 BuiltIn CullDistance\n"
3618 "OpDecorate %BP_gl_PerVertexOut Block\n"
3619 "OpDecorate %BP_gl_TessCoord BuiltIn TessCoord\n"
3620 "OpMemberDecorate %BP_gl_PerVertexIn 0 BuiltIn Position\n"
3621 "OpMemberDecorate %BP_gl_PerVertexIn 1 BuiltIn PointSize\n"
3622 "OpMemberDecorate %BP_gl_PerVertexIn 2 BuiltIn ClipDistance\n"
3623 "OpMemberDecorate %BP_gl_PerVertexIn 3 BuiltIn CullDistance\n"
3624 "OpDecorate %BP_gl_PerVertexIn Block\n"
3625 "OpDecorate %BP_out_color Location 1\n"
3626 "OpDecorate %BP_in_color Location 1\n"
3627 "${decoration:opt}\n"
3628 SPIRV_ASSEMBLY_TYPES
3629 SPIRV_ASSEMBLY_CONSTANTS
3630 SPIRV_ASSEMBLY_ARRAYS
3631 "%BP_gl_PerVertexOut = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3632 "%BP_op_gl_PerVertexOut = OpTypePointer Output %BP_gl_PerVertexOut\n"
3633 "%BP_stream = OpVariable %BP_op_gl_PerVertexOut Output\n"
3634 "%BP_gl_TessCoord = OpVariable %ip_v3f32 Input\n"
3635 "%BP_gl_PerVertexIn = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3636 "%BP_a32_gl_PerVertexIn = OpTypeArray %BP_gl_PerVertexIn %c_u32_32\n"
3637 "%BP_ip_a32_gl_PerVertexIn = OpTypePointer Input %BP_a32_gl_PerVertexIn\n"
3638 "%BP_gl_in = OpVariable %BP_ip_a32_gl_PerVertexIn Input\n"
3639 "%BP_out_color = OpVariable %op_v4f32 Output\n"
3640 "%BP_in_color = OpVariable %ip_a32v4f32 Input\n"
3642 "%BP_main = OpFunction %void None %fun\n"
3643 "%BP_label = OpLabel\n"
3644 "%BP_gl_TC_0 = OpAccessChain %ip_f32 %BP_gl_TessCoord %c_u32_0\n"
3645 "%BP_gl_TC_1 = OpAccessChain %ip_f32 %BP_gl_TessCoord %c_u32_1\n"
3646 "%BP_gl_TC_2 = OpAccessChain %ip_f32 %BP_gl_TessCoord %c_u32_2\n"
3647 "%BP_gl_in_gl_Pos_0 = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_0 %c_i32_0\n"
3648 "%BP_gl_in_gl_Pos_1 = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_1 %c_i32_0\n"
3649 "%BP_gl_in_gl_Pos_2 = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_2 %c_i32_0\n"
3651 "%BP_gl_OPos = OpAccessChain %op_v4f32 %BP_stream %c_i32_0\n"
3652 "%BP_in_color_0 = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_0\n"
3653 "%BP_in_color_1 = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_1\n"
3654 "%BP_in_color_2 = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_2\n"
3656 "%BP_TC_W_0 = OpLoad %f32 %BP_gl_TC_0\n"
3657 "%BP_TC_W_1 = OpLoad %f32 %BP_gl_TC_1\n"
3658 "%BP_TC_W_2 = OpLoad %f32 %BP_gl_TC_2\n"
3659 "%BP_v4f32_TC_0 = OpCompositeConstruct %v4f32 %BP_TC_W_0 %BP_TC_W_0 %BP_TC_W_0 %BP_TC_W_0\n"
3660 "%BP_v4f32_TC_1 = OpCompositeConstruct %v4f32 %BP_TC_W_1 %BP_TC_W_1 %BP_TC_W_1 %BP_TC_W_1\n"
3661 "%BP_v4f32_TC_2 = OpCompositeConstruct %v4f32 %BP_TC_W_2 %BP_TC_W_2 %BP_TC_W_2 %BP_TC_W_2\n"
3663 "%BP_gl_IP_0 = OpLoad %v4f32 %BP_gl_in_gl_Pos_0\n"
3664 "%BP_gl_IP_1 = OpLoad %v4f32 %BP_gl_in_gl_Pos_1\n"
3665 "%BP_gl_IP_2 = OpLoad %v4f32 %BP_gl_in_gl_Pos_2\n"
3667 "%BP_IP_W_0 = OpFMul %v4f32 %BP_v4f32_TC_0 %BP_gl_IP_0\n"
3668 "%BP_IP_W_1 = OpFMul %v4f32 %BP_v4f32_TC_1 %BP_gl_IP_1\n"
3669 "%BP_IP_W_2 = OpFMul %v4f32 %BP_v4f32_TC_2 %BP_gl_IP_2\n"
3671 "%BP_pos_sum_0 = OpFAdd %v4f32 %BP_IP_W_0 %BP_IP_W_1\n"
3672 "%BP_pos_sum_1 = OpFAdd %v4f32 %BP_pos_sum_0 %BP_IP_W_2\n"
3674 "OpStore %BP_gl_OPos %BP_pos_sum_1\n"
3676 "%BP_IC_0 = OpLoad %v4f32 %BP_in_color_0\n"
3677 "%BP_IC_1 = OpLoad %v4f32 %BP_in_color_1\n"
3678 "%BP_IC_2 = OpLoad %v4f32 %BP_in_color_2\n"
3680 "%BP_IC_W_0 = OpFMul %v4f32 %BP_v4f32_TC_0 %BP_IC_0\n"
3681 "%BP_IC_W_1 = OpFMul %v4f32 %BP_v4f32_TC_1 %BP_IC_1\n"
3682 "%BP_IC_W_2 = OpFMul %v4f32 %BP_v4f32_TC_2 %BP_IC_2\n"
3684 "%BP_col_sum_0 = OpFAdd %v4f32 %BP_IC_W_0 %BP_IC_W_1\n"
3685 "%BP_col_sum_1 = OpFAdd %v4f32 %BP_col_sum_0 %BP_IC_W_2\n"
3687 "%BP_clr_transformed = OpFunctionCall %v4f32 %test_code %BP_col_sum_1\n"
3689 "OpStore %BP_out_color %BP_clr_transformed\n"
3693 return tcu::StringTemplate(tessEvalBoilerplate).specialize(fragments);
3696 // Creates geometry-shader assembly by specializing a boilerplate StringTemplate
3697 // on fragments, which must (at least) map "testfun" to an OpFunction definition
3698 // for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed
3699 // with "BP_" to avoid collisions with fragments.
3701 // Derived from this GLSL:
3704 // layout(triangles) in;
3705 // layout(triangle_strip, max_vertices = 3) out;
3707 // layout(location = 1) in vec4 in_color[];
3708 // layout(location = 1) out vec4 out_color;
3711 // gl_Position = gl_in[0].gl_Position;
3712 // out_color = test_fun(in_color[0]);
3714 // gl_Position = gl_in[1].gl_Position;
3715 // out_color = test_fun(in_color[1]);
3717 // gl_Position = gl_in[2].gl_Position;
3718 // out_color = test_fun(in_color[2]);
3722 string makeGeometryShaderAssembly(const map<string, string>& fragments)
3724 static const char geometryShaderBoilerplate[] =
3725 "OpCapability Geometry\n"
3726 "OpMemoryModel Logical GLSL450\n"
3727 "OpEntryPoint Geometry %BP_main \"main\" %BP_out_gl_position %BP_gl_in %BP_out_color %BP_in_color\n"
3728 "OpExecutionMode %BP_main Triangles\n"
3729 "OpExecutionMode %BP_main Invocations 0\n"
3730 "OpExecutionMode %BP_main OutputTriangleStrip\n"
3731 "OpExecutionMode %BP_main OutputVertices 3\n"
3733 "OpName %BP_main \"main\"\n"
3734 "OpName %BP_per_vertex_in \"gl_PerVertex\"\n"
3735 "OpMemberName %BP_per_vertex_in 0 \"gl_Position\"\n"
3736 "OpMemberName %BP_per_vertex_in 1 \"gl_PointSize\"\n"
3737 "OpMemberName %BP_per_vertex_in 2 \"gl_ClipDistance\"\n"
3738 "OpMemberName %BP_per_vertex_in 3 \"gl_CullDistance\"\n"
3739 "OpName %BP_gl_in \"gl_in\"\n"
3740 "OpName %BP_out_color \"out_color\"\n"
3741 "OpName %BP_in_color \"in_color\"\n"
3742 "OpName %test_code \"testfun(vf4;\"\n"
3743 "OpDecorate %BP_out_gl_position BuiltIn Position\n"
3744 "OpMemberDecorate %BP_per_vertex_in 0 BuiltIn Position\n"
3745 "OpMemberDecorate %BP_per_vertex_in 1 BuiltIn PointSize\n"
3746 "OpMemberDecorate %BP_per_vertex_in 2 BuiltIn ClipDistance\n"
3747 "OpMemberDecorate %BP_per_vertex_in 3 BuiltIn CullDistance\n"
3748 "OpDecorate %BP_per_vertex_in Block\n"
3749 "OpDecorate %BP_out_color Location 1\n"
3750 "OpDecorate %BP_out_color Stream 0\n"
3751 "OpDecorate %BP_in_color Location 1\n"
3752 "${decoration:opt}\n"
3753 SPIRV_ASSEMBLY_TYPES
3754 SPIRV_ASSEMBLY_CONSTANTS
3755 SPIRV_ASSEMBLY_ARRAYS
3756 "%BP_per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3757 "%BP_a3_per_vertex_in = OpTypeArray %BP_per_vertex_in %c_u32_3\n"
3758 "%BP_ip_a3_per_vertex_in = OpTypePointer Input %BP_a3_per_vertex_in\n"
3760 "%BP_gl_in = OpVariable %BP_ip_a3_per_vertex_in Input\n"
3761 "%BP_out_color = OpVariable %op_v4f32 Output\n"
3762 "%BP_in_color = OpVariable %ip_a3v4f32 Input\n"
3763 "%BP_out_gl_position = OpVariable %op_v4f32 Output\n"
3766 "%BP_main = OpFunction %void None %fun\n"
3767 "%BP_label = OpLabel\n"
3768 "%BP_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_0 %c_i32_0\n"
3769 "%BP_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_1 %c_i32_0\n"
3770 "%BP_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_2 %c_i32_0\n"
3772 "%BP_in_position_0 = OpLoad %v4f32 %BP_gl_in_0_gl_position\n"
3773 "%BP_in_position_1 = OpLoad %v4f32 %BP_gl_in_1_gl_position\n"
3774 "%BP_in_position_2 = OpLoad %v4f32 %BP_gl_in_2_gl_position \n"
3776 "%BP_in_color_0_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_0\n"
3777 "%BP_in_color_1_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_1\n"
3778 "%BP_in_color_2_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_2\n"
3780 "%BP_in_color_0 = OpLoad %v4f32 %BP_in_color_0_ptr\n"
3781 "%BP_in_color_1 = OpLoad %v4f32 %BP_in_color_1_ptr\n"
3782 "%BP_in_color_2 = OpLoad %v4f32 %BP_in_color_2_ptr\n"
3784 "%BP_transformed_in_color_0 = OpFunctionCall %v4f32 %test_code %BP_in_color_0\n"
3785 "%BP_transformed_in_color_1 = OpFunctionCall %v4f32 %test_code %BP_in_color_1\n"
3786 "%BP_transformed_in_color_2 = OpFunctionCall %v4f32 %test_code %BP_in_color_2\n"
3789 "OpStore %BP_out_gl_position %BP_in_position_0\n"
3790 "OpStore %BP_out_color %BP_transformed_in_color_0\n"
3793 "OpStore %BP_out_gl_position %BP_in_position_1\n"
3794 "OpStore %BP_out_color %BP_transformed_in_color_1\n"
3797 "OpStore %BP_out_gl_position %BP_in_position_2\n"
3798 "OpStore %BP_out_color %BP_transformed_in_color_2\n"
3805 return tcu::StringTemplate(geometryShaderBoilerplate).specialize(fragments);
3808 // Creates fragment-shader assembly by specializing a boilerplate StringTemplate
3809 // on fragments, which must (at least) map "testfun" to an OpFunction definition
3810 // for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed
3811 // with "BP_" to avoid collisions with fragments.
3813 // Derived from this GLSL:
3815 // layout(location = 1) in highp vec4 vtxColor;
3816 // layout(location = 0) out highp vec4 fragColor;
3817 // highp vec4 testfun(highp vec4 x) { return x; }
3818 // void main(void) { fragColor = testfun(vtxColor); }
3820 // with modifications including passing vtxColor by value and ripping out
3821 // testfun() definition.
3822 string makeFragmentShaderAssembly(const map<string, string>& fragments)
3824 static const char fragmentShaderBoilerplate[] =
3825 "OpCapability Shader\n"
3826 "OpMemoryModel Logical GLSL450\n"
3827 "OpEntryPoint Fragment %BP_main \"main\" %BP_vtxColor %BP_fragColor\n"
3828 "OpExecutionMode %BP_main OriginUpperLeft\n"
3830 "OpName %BP_main \"main\"\n"
3831 "OpName %BP_fragColor \"fragColor\"\n"
3832 "OpName %BP_vtxColor \"vtxColor\"\n"
3833 "OpName %test_code \"testfun(vf4;\"\n"
3834 "OpDecorate %BP_fragColor Location 0\n"
3835 "OpDecorate %BP_vtxColor Location 1\n"
3836 "${decoration:opt}\n"
3837 SPIRV_ASSEMBLY_TYPES
3838 SPIRV_ASSEMBLY_CONSTANTS
3839 SPIRV_ASSEMBLY_ARRAYS
3840 "%BP_fragColor = OpVariable %op_v4f32 Output\n"
3841 "%BP_vtxColor = OpVariable %ip_v4f32 Input\n"
3843 "%BP_main = OpFunction %void None %fun\n"
3844 "%BP_label_main = OpLabel\n"
3845 "%BP_tmp1 = OpLoad %v4f32 %BP_vtxColor\n"
3846 "%BP_tmp2 = OpFunctionCall %v4f32 %test_code %BP_tmp1\n"
3847 "OpStore %BP_fragColor %BP_tmp2\n"
3851 return tcu::StringTemplate(fragmentShaderBoilerplate).specialize(fragments);
3854 // Creates fragments that specialize into a simple pass-through shader (of any kind).
3855 map<string, string> passthruFragments(void)
3857 map<string, string> fragments;
3858 fragments["testfun"] =
3859 // A %test_code function that returns its argument unchanged.
3860 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
3861 "%param1 = OpFunctionParameter %v4f32\n"
3862 "%label_testfun = OpLabel\n"
3863 "OpReturnValue %param1\n"
3868 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3869 // Vertex shader gets custom code from context, the rest are pass-through.
3870 void addShaderCodeCustomVertex(vk::SourceCollections& dst, InstanceContext context)
3872 map<string, string> passthru = passthruFragments();
3873 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(context.testCodeFragments);
3874 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
3877 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3878 // Tessellation control shader gets custom code from context, the rest are
3880 void addShaderCodeCustomTessControl(vk::SourceCollections& dst, InstanceContext context)
3882 map<string, string> passthru = passthruFragments();
3883 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
3884 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(context.testCodeFragments);
3885 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(passthru);
3886 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
3889 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3890 // Tessellation evaluation shader gets custom code from context, the rest are
3892 void addShaderCodeCustomTessEval(vk::SourceCollections& dst, InstanceContext context)
3894 map<string, string> passthru = passthruFragments();
3895 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
3896 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(passthru);
3897 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(context.testCodeFragments);
3898 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
3901 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3902 // Geometry shader gets custom code from context, the rest are pass-through.
3903 void addShaderCodeCustomGeometry(vk::SourceCollections& dst, InstanceContext context)
3905 map<string, string> passthru = passthruFragments();
3906 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
3907 dst.spirvAsmSources.add("geom") << makeGeometryShaderAssembly(context.testCodeFragments);
3908 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
3911 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3912 // Fragment shader gets custom code from context, the rest are pass-through.
3913 void addShaderCodeCustomFragment(vk::SourceCollections& dst, InstanceContext context)
3915 map<string, string> passthru = passthruFragments();
3916 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
3917 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(context.testCodeFragments);
3920 void createCombinedModule(vk::SourceCollections& dst, InstanceContext)
3922 // \todo [2015-12-07 awoloszyn] Make tessellation / geometry conditional
3923 // \todo [2015-12-07 awoloszyn] Remove OpName and OpMemberName at some point
3924 dst.spirvAsmSources.add("module") <<
3925 "OpCapability Shader\n"
3926 "OpCapability Geometry\n"
3927 "OpCapability Tessellation\n"
3928 "OpMemoryModel Logical GLSL450\n"
3930 "OpEntryPoint Vertex %vert_main \"main\" %vert_Position %vert_vtxColor %vert_color %vert_vtxPosition %vert_vertex_id %vert_instance_id\n"
3931 "OpEntryPoint Geometry %geom_main \"main\" %geom_out_gl_position %geom_gl_in %geom_out_color %geom_in_color\n"
3932 "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"
3933 "OpEntryPoint TessellationEvaluation %tesse_main \"main\" %tesse_stream %tesse_gl_tessCoord %tesse_in_position %tesse_out_color %tesse_in_color \n"
3934 "OpEntryPoint Fragment %frag_main \"main\" %frag_vtxColor %frag_fragColor\n"
3936 "OpExecutionMode %geom_main Triangles\n"
3937 "OpExecutionMode %geom_main Invocations 0\n"
3938 "OpExecutionMode %geom_main OutputTriangleStrip\n"
3939 "OpExecutionMode %geom_main OutputVertices 3\n"
3941 "OpExecutionMode %tessc_main OutputVertices 3\n"
3943 "OpExecutionMode %tesse_main Triangles\n"
3945 "OpExecutionMode %frag_main OriginUpperLeft\n"
3947 "; Vertex decorations\n"
3948 "OpName %vert_main \"main\"\n"
3949 "OpName %vert_vtxPosition \"vtxPosition\"\n"
3950 "OpName %vert_Position \"position\"\n"
3951 "OpName %vert_vtxColor \"vtxColor\"\n"
3952 "OpName %vert_color \"color\"\n"
3953 "OpName %vert_vertex_id \"gl_VertexIndex\"\n"
3954 "OpName %vert_instance_id \"gl_InstanceIndex\"\n"
3955 "OpDecorate %vert_vtxPosition Location 2\n"
3956 "OpDecorate %vert_Position Location 0\n"
3957 "OpDecorate %vert_vtxColor Location 1\n"
3958 "OpDecorate %vert_color Location 1\n"
3959 "OpDecorate %vert_vertex_id BuiltIn VertexIndex\n"
3960 "OpDecorate %vert_instance_id BuiltIn InstanceIndex\n"
3962 "; Geometry decorations\n"
3963 "OpName %geom_main \"main\"\n"
3964 "OpName %geom_per_vertex_in \"gl_PerVertex\"\n"
3965 "OpMemberName %geom_per_vertex_in 0 \"gl_Position\"\n"
3966 "OpMemberName %geom_per_vertex_in 1 \"gl_PointSize\"\n"
3967 "OpMemberName %geom_per_vertex_in 2 \"gl_ClipDistance\"\n"
3968 "OpMemberName %geom_per_vertex_in 3 \"gl_CullDistance\"\n"
3969 "OpName %geom_gl_in \"gl_in\"\n"
3970 "OpName %geom_out_color \"out_color\"\n"
3971 "OpName %geom_in_color \"in_color\"\n"
3972 "OpDecorate %geom_out_gl_position BuiltIn Position\n"
3973 "OpMemberDecorate %geom_per_vertex_in 0 BuiltIn Position\n"
3974 "OpMemberDecorate %geom_per_vertex_in 1 BuiltIn PointSize\n"
3975 "OpMemberDecorate %geom_per_vertex_in 2 BuiltIn ClipDistance\n"
3976 "OpMemberDecorate %geom_per_vertex_in 3 BuiltIn CullDistance\n"
3977 "OpDecorate %geom_per_vertex_in Block\n"
3978 "OpDecorate %geom_out_color Location 1\n"
3979 "OpDecorate %geom_out_color Stream 0\n"
3980 "OpDecorate %geom_in_color Location 1\n"
3982 "; Tessellation Control decorations\n"
3983 "OpName %tessc_main \"main\"\n"
3984 "OpName %tessc_out_color \"out_color\"\n"
3985 "OpName %tessc_gl_InvocationID \"gl_InvocationID\"\n"
3986 "OpName %tessc_in_color \"in_color\"\n"
3987 "OpName %tessc_out_position \"out_position\"\n"
3988 "OpName %tessc_in_position \"in_position\"\n"
3989 "OpName %tessc_gl_TessLevelOuter \"gl_TessLevelOuter\"\n"
3990 "OpName %tessc_gl_TessLevelInner \"gl_TessLevelInner\"\n"
3991 "OpDecorate %tessc_out_color Location 1\n"
3992 "OpDecorate %tessc_gl_InvocationID BuiltIn InvocationId\n"
3993 "OpDecorate %tessc_in_color Location 1\n"
3994 "OpDecorate %tessc_out_position Location 2\n"
3995 "OpDecorate %tessc_in_position Location 2\n"
3996 "OpDecorate %tessc_gl_TessLevelOuter Patch\n"
3997 "OpDecorate %tessc_gl_TessLevelOuter BuiltIn TessLevelOuter\n"
3998 "OpDecorate %tessc_gl_TessLevelInner Patch\n"
3999 "OpDecorate %tessc_gl_TessLevelInner BuiltIn TessLevelInner\n"
4001 "; Tessellation Evaluation decorations\n"
4002 "OpName %tesse_main \"main\"\n"
4003 "OpName %tesse_per_vertex_out \"gl_PerVertex\"\n"
4004 "OpMemberName %tesse_per_vertex_out 0 \"gl_Position\"\n"
4005 "OpMemberName %tesse_per_vertex_out 1 \"gl_PointSize\"\n"
4006 "OpMemberName %tesse_per_vertex_out 2 \"gl_ClipDistance\"\n"
4007 "OpMemberName %tesse_per_vertex_out 3 \"gl_CullDistance\"\n"
4008 "OpName %tesse_stream \"\"\n"
4009 "OpName %tesse_gl_tessCoord \"gl_TessCoord\"\n"
4010 "OpName %tesse_in_position \"in_position\"\n"
4011 "OpName %tesse_out_color \"out_color\"\n"
4012 "OpName %tesse_in_color \"in_color\"\n"
4013 "OpMemberDecorate %tesse_per_vertex_out 0 BuiltIn Position\n"
4014 "OpMemberDecorate %tesse_per_vertex_out 1 BuiltIn PointSize\n"
4015 "OpMemberDecorate %tesse_per_vertex_out 2 BuiltIn ClipDistance\n"
4016 "OpMemberDecorate %tesse_per_vertex_out 3 BuiltIn CullDistance\n"
4017 "OpDecorate %tesse_per_vertex_out Block\n"
4018 "OpDecorate %tesse_gl_tessCoord BuiltIn TessCoord\n"
4019 "OpDecorate %tesse_in_position Location 2\n"
4020 "OpDecorate %tesse_out_color Location 1\n"
4021 "OpDecorate %tesse_in_color Location 1\n"
4023 "; Fragment decorations\n"
4024 "OpName %frag_main \"main\"\n"
4025 "OpName %frag_fragColor \"fragColor\"\n"
4026 "OpName %frag_vtxColor \"vtxColor\"\n"
4027 "OpDecorate %frag_fragColor Location 0\n"
4028 "OpDecorate %frag_vtxColor Location 1\n"
4030 SPIRV_ASSEMBLY_TYPES
4031 SPIRV_ASSEMBLY_CONSTANTS
4032 SPIRV_ASSEMBLY_ARRAYS
4034 "; Vertex Variables\n"
4035 "%vert_vtxPosition = OpVariable %op_v4f32 Output\n"
4036 "%vert_Position = OpVariable %ip_v4f32 Input\n"
4037 "%vert_vtxColor = OpVariable %op_v4f32 Output\n"
4038 "%vert_color = OpVariable %ip_v4f32 Input\n"
4039 "%vert_vertex_id = OpVariable %ip_i32 Input\n"
4040 "%vert_instance_id = OpVariable %ip_i32 Input\n"
4042 "; Geometry Variables\n"
4043 "%geom_per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4044 "%geom_a3_per_vertex_in = OpTypeArray %geom_per_vertex_in %c_u32_3\n"
4045 "%geom_ip_a3_per_vertex_in = OpTypePointer Input %geom_a3_per_vertex_in\n"
4046 "%geom_gl_in = OpVariable %geom_ip_a3_per_vertex_in Input\n"
4047 "%geom_out_color = OpVariable %op_v4f32 Output\n"
4048 "%geom_in_color = OpVariable %ip_a3v4f32 Input\n"
4049 "%geom_out_gl_position = OpVariable %op_v4f32 Output\n"
4051 "; Tessellation Control Variables\n"
4052 "%tessc_out_color = OpVariable %op_a3v4f32 Output\n"
4053 "%tessc_gl_InvocationID = OpVariable %ip_i32 Input\n"
4054 "%tessc_in_color = OpVariable %ip_a32v4f32 Input\n"
4055 "%tessc_out_position = OpVariable %op_a3v4f32 Output\n"
4056 "%tessc_in_position = OpVariable %ip_a32v4f32 Input\n"
4057 "%tessc_gl_TessLevelOuter = OpVariable %op_a4f32 Output\n"
4058 "%tessc_gl_TessLevelInner = OpVariable %op_a2f32 Output\n"
4060 "; Tessellation Evaluation Decorations\n"
4061 "%tesse_per_vertex_out = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4062 "%tesse_op_per_vertex_out = OpTypePointer Output %tesse_per_vertex_out\n"
4063 "%tesse_stream = OpVariable %tesse_op_per_vertex_out Output\n"
4064 "%tesse_gl_tessCoord = OpVariable %ip_v3f32 Input\n"
4065 "%tesse_in_position = OpVariable %ip_a32v4f32 Input\n"
4066 "%tesse_out_color = OpVariable %op_v4f32 Output\n"
4067 "%tesse_in_color = OpVariable %ip_a32v4f32 Input\n"
4069 "; Fragment Variables\n"
4070 "%frag_fragColor = OpVariable %op_v4f32 Output\n"
4071 "%frag_vtxColor = OpVariable %ip_v4f32 Input\n"
4074 "%vert_main = OpFunction %void None %fun\n"
4075 "%vert_label = OpLabel\n"
4076 "%vert_tmp_position = OpLoad %v4f32 %vert_Position\n"
4077 "OpStore %vert_vtxPosition %vert_tmp_position\n"
4078 "%vert_tmp_color = OpLoad %v4f32 %vert_color\n"
4079 "OpStore %vert_vtxColor %vert_tmp_color\n"
4083 "; Geometry Entry\n"
4084 "%geom_main = OpFunction %void None %fun\n"
4085 "%geom_label = OpLabel\n"
4086 "%geom_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %geom_gl_in %c_i32_0 %c_i32_0\n"
4087 "%geom_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %geom_gl_in %c_i32_1 %c_i32_0\n"
4088 "%geom_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %geom_gl_in %c_i32_2 %c_i32_0\n"
4089 "%geom_in_position_0 = OpLoad %v4f32 %geom_gl_in_0_gl_position\n"
4090 "%geom_in_position_1 = OpLoad %v4f32 %geom_gl_in_1_gl_position\n"
4091 "%geom_in_position_2 = OpLoad %v4f32 %geom_gl_in_2_gl_position \n"
4092 "%geom_in_color_0_ptr = OpAccessChain %ip_v4f32 %geom_in_color %c_i32_0\n"
4093 "%geom_in_color_1_ptr = OpAccessChain %ip_v4f32 %geom_in_color %c_i32_1\n"
4094 "%geom_in_color_2_ptr = OpAccessChain %ip_v4f32 %geom_in_color %c_i32_2\n"
4095 "%geom_in_color_0 = OpLoad %v4f32 %geom_in_color_0_ptr\n"
4096 "%geom_in_color_1 = OpLoad %v4f32 %geom_in_color_1_ptr\n"
4097 "%geom_in_color_2 = OpLoad %v4f32 %geom_in_color_2_ptr\n"
4098 "OpStore %geom_out_gl_position %geom_in_position_0\n"
4099 "OpStore %geom_out_color %geom_in_color_0\n"
4101 "OpStore %geom_out_gl_position %geom_in_position_1\n"
4102 "OpStore %geom_out_color %geom_in_color_1\n"
4104 "OpStore %geom_out_gl_position %geom_in_position_2\n"
4105 "OpStore %geom_out_color %geom_in_color_2\n"
4111 "; Tessellation Control Entry\n"
4112 "%tessc_main = OpFunction %void None %fun\n"
4113 "%tessc_label = OpLabel\n"
4114 "%tessc_invocation_id = OpLoad %i32 %tessc_gl_InvocationID\n"
4115 "%tessc_in_color_ptr = OpAccessChain %ip_v4f32 %tessc_in_color %tessc_invocation_id\n"
4116 "%tessc_in_position_ptr = OpAccessChain %ip_v4f32 %tessc_in_position %tessc_invocation_id\n"
4117 "%tessc_in_color_val = OpLoad %v4f32 %tessc_in_color_ptr\n"
4118 "%tessc_in_position_val = OpLoad %v4f32 %tessc_in_position_ptr\n"
4119 "%tessc_out_color_ptr = OpAccessChain %op_v4f32 %tessc_out_color %tessc_invocation_id\n"
4120 "%tessc_out_position_ptr = OpAccessChain %op_v4f32 %tessc_out_position %tessc_invocation_id\n"
4121 "OpStore %tessc_out_color_ptr %tessc_in_color_val\n"
4122 "OpStore %tessc_out_position_ptr %tessc_in_position_val\n"
4123 "%tessc_is_first_invocation = OpIEqual %bool %tessc_invocation_id %c_i32_0\n"
4124 "OpSelectionMerge %tessc_merge_label None\n"
4125 "OpBranchConditional %tessc_is_first_invocation %tessc_first_invocation %tessc_merge_label\n"
4126 "%tessc_first_invocation = OpLabel\n"
4127 "%tessc_tess_outer_0 = OpAccessChain %op_f32 %tessc_gl_TessLevelOuter %c_i32_0\n"
4128 "%tessc_tess_outer_1 = OpAccessChain %op_f32 %tessc_gl_TessLevelOuter %c_i32_1\n"
4129 "%tessc_tess_outer_2 = OpAccessChain %op_f32 %tessc_gl_TessLevelOuter %c_i32_2\n"
4130 "%tessc_tess_inner = OpAccessChain %op_f32 %tessc_gl_TessLevelInner %c_i32_0\n"
4131 "OpStore %tessc_tess_outer_0 %c_f32_1\n"
4132 "OpStore %tessc_tess_outer_1 %c_f32_1\n"
4133 "OpStore %tessc_tess_outer_2 %c_f32_1\n"
4134 "OpStore %tessc_tess_inner %c_f32_1\n"
4135 "OpBranch %tessc_merge_label\n"
4136 "%tessc_merge_label = OpLabel\n"
4140 "; Tessellation Evaluation Entry\n"
4141 "%tesse_main = OpFunction %void None %fun\n"
4142 "%tesse_label = OpLabel\n"
4143 "%tesse_tc_0_ptr = OpAccessChain %ip_f32 %tesse_gl_tessCoord %c_u32_0\n"
4144 "%tesse_tc_1_ptr = OpAccessChain %ip_f32 %tesse_gl_tessCoord %c_u32_1\n"
4145 "%tesse_tc_2_ptr = OpAccessChain %ip_f32 %tesse_gl_tessCoord %c_u32_2\n"
4146 "%tesse_tc_0 = OpLoad %f32 %tesse_tc_0_ptr\n"
4147 "%tesse_tc_1 = OpLoad %f32 %tesse_tc_1_ptr\n"
4148 "%tesse_tc_2 = OpLoad %f32 %tesse_tc_2_ptr\n"
4149 "%tesse_in_pos_0_ptr = OpAccessChain %ip_v4f32 %tesse_in_position %c_i32_0\n"
4150 "%tesse_in_pos_1_ptr = OpAccessChain %ip_v4f32 %tesse_in_position %c_i32_1\n"
4151 "%tesse_in_pos_2_ptr = OpAccessChain %ip_v4f32 %tesse_in_position %c_i32_2\n"
4152 "%tesse_in_pos_0 = OpLoad %v4f32 %tesse_in_pos_0_ptr\n"
4153 "%tesse_in_pos_1 = OpLoad %v4f32 %tesse_in_pos_1_ptr\n"
4154 "%tesse_in_pos_2 = OpLoad %v4f32 %tesse_in_pos_2_ptr\n"
4155 "%tesse_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_0 %tesse_in_pos_0\n"
4156 "%tesse_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_1 %tesse_in_pos_1\n"
4157 "%tesse_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_2 %tesse_in_pos_2\n"
4158 "%tesse_out_pos_ptr = OpAccessChain %op_v4f32 %tesse_stream %c_i32_0\n"
4159 "%tesse_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %tesse_in_pos_0_weighted %tesse_in_pos_1_weighted\n"
4160 "%tesse_computed_out = OpFAdd %v4f32 %tesse_in_pos_0_plus_pos_1 %tesse_in_pos_2_weighted\n"
4161 "OpStore %tesse_out_pos_ptr %tesse_computed_out\n"
4162 "%tesse_in_clr_0_ptr = OpAccessChain %ip_v4f32 %tesse_in_color %c_i32_0\n"
4163 "%tesse_in_clr_1_ptr = OpAccessChain %ip_v4f32 %tesse_in_color %c_i32_1\n"
4164 "%tesse_in_clr_2_ptr = OpAccessChain %ip_v4f32 %tesse_in_color %c_i32_2\n"
4165 "%tesse_in_clr_0 = OpLoad %v4f32 %tesse_in_clr_0_ptr\n"
4166 "%tesse_in_clr_1 = OpLoad %v4f32 %tesse_in_clr_1_ptr\n"
4167 "%tesse_in_clr_2 = OpLoad %v4f32 %tesse_in_clr_2_ptr\n"
4168 "%tesse_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_0 %tesse_in_clr_0\n"
4169 "%tesse_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_1 %tesse_in_clr_1\n"
4170 "%tesse_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_2 %tesse_in_clr_2\n"
4171 "%tesse_in_clr_0_plus_col_1 = OpFAdd %v4f32 %tesse_in_clr_0_weighted %tesse_in_clr_1_weighted\n"
4172 "%tesse_computed_clr = OpFAdd %v4f32 %tesse_in_clr_0_plus_col_1 %tesse_in_clr_2_weighted\n"
4173 "OpStore %tesse_out_color %tesse_computed_clr\n"
4177 "; Fragment Entry\n"
4178 "%frag_main = OpFunction %void None %fun\n"
4179 "%frag_label_main = OpLabel\n"
4180 "%frag_tmp1 = OpLoad %v4f32 %frag_vtxColor\n"
4181 "OpStore %frag_fragColor %frag_tmp1\n"
4186 // This has two shaders of each stage. The first
4187 // is a passthrough, the second inverts the color.
4188 void createMultipleEntries(vk::SourceCollections& dst, InstanceContext)
4190 dst.spirvAsmSources.add("vert") <<
4191 // This module contains 2 vertex shaders. One that is a passthrough
4192 // and a second that inverts the color of the output (1.0 - color).
4193 "OpCapability Shader\n"
4194 "OpMemoryModel Logical GLSL450\n"
4195 "OpEntryPoint Vertex %main \"vert1\" %Position %vtxColor %color %vtxPosition %vertex_id %instance_id\n"
4196 "OpEntryPoint Vertex %main2 \"vert2\" %Position %vtxColor %color %vtxPosition %vertex_id %instance_id\n"
4198 "OpName %main \"vert1\"\n"
4199 "OpName %main2 \"vert2\"\n"
4200 "OpName %vtxPosition \"vtxPosition\"\n"
4201 "OpName %Position \"position\"\n"
4202 "OpName %vtxColor \"vtxColor\"\n"
4203 "OpName %color \"color\"\n"
4204 "OpName %vertex_id \"gl_VertexIndex\"\n"
4205 "OpName %instance_id \"gl_InstanceIndex\"\n"
4207 "OpDecorate %vtxPosition Location 2\n"
4208 "OpDecorate %Position Location 0\n"
4209 "OpDecorate %vtxColor Location 1\n"
4210 "OpDecorate %color Location 1\n"
4211 "OpDecorate %vertex_id BuiltIn VertexIndex\n"
4212 "OpDecorate %instance_id BuiltIn InstanceIndex\n"
4213 SPIRV_ASSEMBLY_TYPES
4214 SPIRV_ASSEMBLY_CONSTANTS
4215 SPIRV_ASSEMBLY_ARRAYS
4216 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4217 "%vtxPosition = OpVariable %op_v4f32 Output\n"
4218 "%Position = OpVariable %ip_v4f32 Input\n"
4219 "%vtxColor = OpVariable %op_v4f32 Output\n"
4220 "%color = OpVariable %ip_v4f32 Input\n"
4221 "%vertex_id = OpVariable %ip_i32 Input\n"
4222 "%instance_id = OpVariable %ip_i32 Input\n"
4224 "%main = OpFunction %void None %fun\n"
4225 "%label = OpLabel\n"
4226 "%tmp_position = OpLoad %v4f32 %Position\n"
4227 "OpStore %vtxPosition %tmp_position\n"
4228 "%tmp_color = OpLoad %v4f32 %color\n"
4229 "OpStore %vtxColor %tmp_color\n"
4233 "%main2 = OpFunction %void None %fun\n"
4234 "%label2 = OpLabel\n"
4235 "%tmp_position2 = OpLoad %v4f32 %Position\n"
4236 "OpStore %vtxPosition %tmp_position2\n"
4237 "%tmp_color2 = OpLoad %v4f32 %color\n"
4238 "%tmp_color3 = OpFSub %v4f32 %cval %tmp_color2\n"
4239 "OpStore %vtxColor %tmp_color3\n"
4243 dst.spirvAsmSources.add("frag") <<
4244 // This is a single module that contains 2 fragment shaders.
4245 // One that passes color through and the other that inverts the output
4246 // color (1.0 - color).
4247 "OpCapability Shader\n"
4248 "OpMemoryModel Logical GLSL450\n"
4249 "OpEntryPoint Fragment %main \"frag1\" %vtxColor %fragColor\n"
4250 "OpEntryPoint Fragment %main2 \"frag2\" %vtxColor %fragColor\n"
4251 "OpExecutionMode %main OriginUpperLeft\n"
4252 "OpExecutionMode %main2 OriginUpperLeft\n"
4254 "OpName %main \"frag1\"\n"
4255 "OpName %main2 \"frag2\"\n"
4256 "OpName %fragColor \"fragColor\"\n"
4257 "OpName %vtxColor \"vtxColor\"\n"
4258 "OpDecorate %fragColor Location 0\n"
4259 "OpDecorate %vtxColor Location 1\n"
4260 SPIRV_ASSEMBLY_TYPES
4261 SPIRV_ASSEMBLY_CONSTANTS
4262 SPIRV_ASSEMBLY_ARRAYS
4263 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4264 "%fragColor = OpVariable %op_v4f32 Output\n"
4265 "%vtxColor = OpVariable %ip_v4f32 Input\n"
4267 "%main = OpFunction %void None %fun\n"
4268 "%label_main = OpLabel\n"
4269 "%tmp1 = OpLoad %v4f32 %vtxColor\n"
4270 "OpStore %fragColor %tmp1\n"
4274 "%main2 = OpFunction %void None %fun\n"
4275 "%label_main2 = OpLabel\n"
4276 "%tmp2 = OpLoad %v4f32 %vtxColor\n"
4277 "%tmp3 = OpFSub %v4f32 %cval %tmp2\n"
4278 "OpStore %fragColor %tmp3\n"
4282 dst.spirvAsmSources.add("geom") <<
4283 "OpCapability Geometry\n"
4284 "OpMemoryModel Logical GLSL450\n"
4285 "OpEntryPoint Geometry %geom1_main \"geom1\" %out_gl_position %gl_in %out_color %in_color\n"
4286 "OpEntryPoint Geometry %geom2_main \"geom2\" %out_gl_position %gl_in %out_color %in_color\n"
4287 "OpExecutionMode %geom1_main Triangles\n"
4288 "OpExecutionMode %geom2_main Triangles\n"
4289 "OpExecutionMode %geom1_main Invocations 0\n"
4290 "OpExecutionMode %geom2_main Invocations 0\n"
4291 "OpExecutionMode %geom1_main OutputTriangleStrip\n"
4292 "OpExecutionMode %geom2_main OutputTriangleStrip\n"
4293 "OpExecutionMode %geom1_main OutputVertices 3\n"
4294 "OpExecutionMode %geom2_main OutputVertices 3\n"
4295 "OpName %geom1_main \"geom1\"\n"
4296 "OpName %geom2_main \"geom2\"\n"
4297 "OpName %per_vertex_in \"gl_PerVertex\"\n"
4298 "OpMemberName %per_vertex_in 0 \"gl_Position\"\n"
4299 "OpMemberName %per_vertex_in 1 \"gl_PointSize\"\n"
4300 "OpMemberName %per_vertex_in 2 \"gl_ClipDistance\"\n"
4301 "OpMemberName %per_vertex_in 3 \"gl_CullDistance\"\n"
4302 "OpName %gl_in \"gl_in\"\n"
4303 "OpName %out_color \"out_color\"\n"
4304 "OpName %in_color \"in_color\"\n"
4305 "OpDecorate %out_gl_position BuiltIn Position\n"
4306 "OpMemberDecorate %per_vertex_in 0 BuiltIn Position\n"
4307 "OpMemberDecorate %per_vertex_in 1 BuiltIn PointSize\n"
4308 "OpMemberDecorate %per_vertex_in 2 BuiltIn ClipDistance\n"
4309 "OpMemberDecorate %per_vertex_in 3 BuiltIn CullDistance\n"
4310 "OpDecorate %per_vertex_in Block\n"
4311 "OpDecorate %out_color Location 1\n"
4312 "OpDecorate %out_color Stream 0\n"
4313 "OpDecorate %in_color Location 1\n"
4314 SPIRV_ASSEMBLY_TYPES
4315 SPIRV_ASSEMBLY_CONSTANTS
4316 SPIRV_ASSEMBLY_ARRAYS
4317 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4318 "%per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4319 "%a3_per_vertex_in = OpTypeArray %per_vertex_in %c_u32_3\n"
4320 "%ip_a3_per_vertex_in = OpTypePointer Input %a3_per_vertex_in\n"
4321 "%gl_in = OpVariable %ip_a3_per_vertex_in Input\n"
4322 "%out_color = OpVariable %op_v4f32 Output\n"
4323 "%in_color = OpVariable %ip_a3v4f32 Input\n"
4324 "%out_gl_position = OpVariable %op_v4f32 Output\n"
4326 "%geom1_main = OpFunction %void None %fun\n"
4327 "%geom1_label = OpLabel\n"
4328 "%geom1_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_0 %c_i32_0\n"
4329 "%geom1_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_1 %c_i32_0\n"
4330 "%geom1_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_2 %c_i32_0\n"
4331 "%geom1_in_position_0 = OpLoad %v4f32 %geom1_gl_in_0_gl_position\n"
4332 "%geom1_in_position_1 = OpLoad %v4f32 %geom1_gl_in_1_gl_position\n"
4333 "%geom1_in_position_2 = OpLoad %v4f32 %geom1_gl_in_2_gl_position \n"
4334 "%geom1_in_color_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4335 "%geom1_in_color_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4336 "%geom1_in_color_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4337 "%geom1_in_color_0 = OpLoad %v4f32 %geom1_in_color_0_ptr\n"
4338 "%geom1_in_color_1 = OpLoad %v4f32 %geom1_in_color_1_ptr\n"
4339 "%geom1_in_color_2 = OpLoad %v4f32 %geom1_in_color_2_ptr\n"
4340 "OpStore %out_gl_position %geom1_in_position_0\n"
4341 "OpStore %out_color %geom1_in_color_0\n"
4343 "OpStore %out_gl_position %geom1_in_position_1\n"
4344 "OpStore %out_color %geom1_in_color_1\n"
4346 "OpStore %out_gl_position %geom1_in_position_2\n"
4347 "OpStore %out_color %geom1_in_color_2\n"
4353 "%geom2_main = OpFunction %void None %fun\n"
4354 "%geom2_label = OpLabel\n"
4355 "%geom2_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_0 %c_i32_0\n"
4356 "%geom2_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_1 %c_i32_0\n"
4357 "%geom2_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_2 %c_i32_0\n"
4358 "%geom2_in_position_0 = OpLoad %v4f32 %geom2_gl_in_0_gl_position\n"
4359 "%geom2_in_position_1 = OpLoad %v4f32 %geom2_gl_in_1_gl_position\n"
4360 "%geom2_in_position_2 = OpLoad %v4f32 %geom2_gl_in_2_gl_position \n"
4361 "%geom2_in_color_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4362 "%geom2_in_color_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4363 "%geom2_in_color_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4364 "%geom2_in_color_0 = OpLoad %v4f32 %geom2_in_color_0_ptr\n"
4365 "%geom2_in_color_1 = OpLoad %v4f32 %geom2_in_color_1_ptr\n"
4366 "%geom2_in_color_2 = OpLoad %v4f32 %geom2_in_color_2_ptr\n"
4367 "%geom2_transformed_in_color_0 = OpFSub %v4f32 %cval %geom2_in_color_0\n"
4368 "%geom2_transformed_in_color_1 = OpFSub %v4f32 %cval %geom2_in_color_1\n"
4369 "%geom2_transformed_in_color_2 = OpFSub %v4f32 %cval %geom2_in_color_2\n"
4370 "OpStore %out_gl_position %geom2_in_position_0\n"
4371 "OpStore %out_color %geom2_transformed_in_color_0\n"
4373 "OpStore %out_gl_position %geom2_in_position_1\n"
4374 "OpStore %out_color %geom2_transformed_in_color_1\n"
4376 "OpStore %out_gl_position %geom2_in_position_2\n"
4377 "OpStore %out_color %geom2_transformed_in_color_2\n"
4383 dst.spirvAsmSources.add("tessc") <<
4384 "OpCapability Tessellation\n"
4385 "OpMemoryModel Logical GLSL450\n"
4386 "OpEntryPoint TessellationControl %tessc1_main \"tessc1\" %out_color %gl_InvocationID %in_color %out_position %in_position %gl_TessLevelOuter %gl_TessLevelInner\n"
4387 "OpEntryPoint TessellationControl %tessc2_main \"tessc2\" %out_color %gl_InvocationID %in_color %out_position %in_position %gl_TessLevelOuter %gl_TessLevelInner\n"
4388 "OpExecutionMode %tessc1_main OutputVertices 3\n"
4389 "OpExecutionMode %tessc2_main OutputVertices 3\n"
4390 "OpName %tessc1_main \"tessc1\"\n"
4391 "OpName %tessc2_main \"tessc2\"\n"
4392 "OpName %out_color \"out_color\"\n"
4393 "OpName %gl_InvocationID \"gl_InvocationID\"\n"
4394 "OpName %in_color \"in_color\"\n"
4395 "OpName %out_position \"out_position\"\n"
4396 "OpName %in_position \"in_position\"\n"
4397 "OpName %gl_TessLevelOuter \"gl_TessLevelOuter\"\n"
4398 "OpName %gl_TessLevelInner \"gl_TessLevelInner\"\n"
4399 "OpDecorate %out_color Location 1\n"
4400 "OpDecorate %gl_InvocationID BuiltIn InvocationId\n"
4401 "OpDecorate %in_color Location 1\n"
4402 "OpDecorate %out_position Location 2\n"
4403 "OpDecorate %in_position Location 2\n"
4404 "OpDecorate %gl_TessLevelOuter Patch\n"
4405 "OpDecorate %gl_TessLevelOuter BuiltIn TessLevelOuter\n"
4406 "OpDecorate %gl_TessLevelInner Patch\n"
4407 "OpDecorate %gl_TessLevelInner BuiltIn TessLevelInner\n"
4408 SPIRV_ASSEMBLY_TYPES
4409 SPIRV_ASSEMBLY_CONSTANTS
4410 SPIRV_ASSEMBLY_ARRAYS
4411 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4412 "%out_color = OpVariable %op_a3v4f32 Output\n"
4413 "%gl_InvocationID = OpVariable %ip_i32 Input\n"
4414 "%in_color = OpVariable %ip_a32v4f32 Input\n"
4415 "%out_position = OpVariable %op_a3v4f32 Output\n"
4416 "%in_position = OpVariable %ip_a32v4f32 Input\n"
4417 "%gl_TessLevelOuter = OpVariable %op_a4f32 Output\n"
4418 "%gl_TessLevelInner = OpVariable %op_a2f32 Output\n"
4420 "%tessc1_main = OpFunction %void None %fun\n"
4421 "%tessc1_label = OpLabel\n"
4422 "%tessc1_invocation_id = OpLoad %i32 %gl_InvocationID\n"
4423 "%tessc1_in_color_ptr = OpAccessChain %ip_v4f32 %in_color %tessc1_invocation_id\n"
4424 "%tessc1_in_position_ptr = OpAccessChain %ip_v4f32 %in_position %tessc1_invocation_id\n"
4425 "%tessc1_in_color_val = OpLoad %v4f32 %tessc1_in_color_ptr\n"
4426 "%tessc1_in_position_val = OpLoad %v4f32 %tessc1_in_position_ptr\n"
4427 "%tessc1_out_color_ptr = OpAccessChain %op_v4f32 %out_color %tessc1_invocation_id\n"
4428 "%tessc1_out_position_ptr = OpAccessChain %op_v4f32 %out_position %tessc1_invocation_id\n"
4429 "OpStore %tessc1_out_color_ptr %tessc1_in_color_val\n"
4430 "OpStore %tessc1_out_position_ptr %tessc1_in_position_val\n"
4431 "%tessc1_is_first_invocation = OpIEqual %bool %tessc1_invocation_id %c_i32_0\n"
4432 "OpSelectionMerge %tessc1_merge_label None\n"
4433 "OpBranchConditional %tessc1_is_first_invocation %tessc1_first_invocation %tessc1_merge_label\n"
4434 "%tessc1_first_invocation = OpLabel\n"
4435 "%tessc1_tess_outer_0 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_0\n"
4436 "%tessc1_tess_outer_1 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_1\n"
4437 "%tessc1_tess_outer_2 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_2\n"
4438 "%tessc1_tess_inner = OpAccessChain %op_f32 %gl_TessLevelInner %c_i32_0\n"
4439 "OpStore %tessc1_tess_outer_0 %c_f32_1\n"
4440 "OpStore %tessc1_tess_outer_1 %c_f32_1\n"
4441 "OpStore %tessc1_tess_outer_2 %c_f32_1\n"
4442 "OpStore %tessc1_tess_inner %c_f32_1\n"
4443 "OpBranch %tessc1_merge_label\n"
4444 "%tessc1_merge_label = OpLabel\n"
4448 "%tessc2_main = OpFunction %void None %fun\n"
4449 "%tessc2_label = OpLabel\n"
4450 "%tessc2_invocation_id = OpLoad %i32 %gl_InvocationID\n"
4451 "%tessc2_in_color_ptr = OpAccessChain %ip_v4f32 %in_color %tessc2_invocation_id\n"
4452 "%tessc2_in_position_ptr = OpAccessChain %ip_v4f32 %in_position %tessc2_invocation_id\n"
4453 "%tessc2_in_color_val = OpLoad %v4f32 %tessc2_in_color_ptr\n"
4454 "%tessc2_in_position_val = OpLoad %v4f32 %tessc2_in_position_ptr\n"
4455 "%tessc2_out_color_ptr = OpAccessChain %op_v4f32 %out_color %tessc2_invocation_id\n"
4456 "%tessc2_out_position_ptr = OpAccessChain %op_v4f32 %out_position %tessc2_invocation_id\n"
4457 "%tessc2_transformed_color = OpFSub %v4f32 %cval %tessc2_in_color_val\n"
4458 "OpStore %tessc2_out_color_ptr %tessc2_transformed_color\n"
4459 "OpStore %tessc2_out_position_ptr %tessc2_in_position_val\n"
4460 "%tessc2_is_first_invocation = OpIEqual %bool %tessc2_invocation_id %c_i32_0\n"
4461 "OpSelectionMerge %tessc2_merge_label None\n"
4462 "OpBranchConditional %tessc2_is_first_invocation %tessc2_first_invocation %tessc2_merge_label\n"
4463 "%tessc2_first_invocation = OpLabel\n"
4464 "%tessc2_tess_outer_0 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_0\n"
4465 "%tessc2_tess_outer_1 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_1\n"
4466 "%tessc2_tess_outer_2 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_2\n"
4467 "%tessc2_tess_inner = OpAccessChain %op_f32 %gl_TessLevelInner %c_i32_0\n"
4468 "OpStore %tessc2_tess_outer_0 %c_f32_1\n"
4469 "OpStore %tessc2_tess_outer_1 %c_f32_1\n"
4470 "OpStore %tessc2_tess_outer_2 %c_f32_1\n"
4471 "OpStore %tessc2_tess_inner %c_f32_1\n"
4472 "OpBranch %tessc2_merge_label\n"
4473 "%tessc2_merge_label = OpLabel\n"
4477 dst.spirvAsmSources.add("tesse") <<
4478 "OpCapability Tessellation\n"
4479 "OpMemoryModel Logical GLSL450\n"
4480 "OpEntryPoint TessellationEvaluation %tesse1_main \"tesse1\" %stream %gl_tessCoord %in_position %out_color %in_color \n"
4481 "OpEntryPoint TessellationEvaluation %tesse2_main \"tesse2\" %stream %gl_tessCoord %in_position %out_color %in_color \n"
4482 "OpExecutionMode %tesse1_main Triangles\n"
4483 "OpExecutionMode %tesse2_main Triangles\n"
4484 "OpName %tesse1_main \"tesse1\"\n"
4485 "OpName %tesse2_main \"tesse2\"\n"
4486 "OpName %per_vertex_out \"gl_PerVertex\"\n"
4487 "OpMemberName %per_vertex_out 0 \"gl_Position\"\n"
4488 "OpMemberName %per_vertex_out 1 \"gl_PointSize\"\n"
4489 "OpMemberName %per_vertex_out 2 \"gl_ClipDistance\"\n"
4490 "OpMemberName %per_vertex_out 3 \"gl_CullDistance\"\n"
4491 "OpName %stream \"\"\n"
4492 "OpName %gl_tessCoord \"gl_TessCoord\"\n"
4493 "OpName %in_position \"in_position\"\n"
4494 "OpName %out_color \"out_color\"\n"
4495 "OpName %in_color \"in_color\"\n"
4496 "OpMemberDecorate %per_vertex_out 0 BuiltIn Position\n"
4497 "OpMemberDecorate %per_vertex_out 1 BuiltIn PointSize\n"
4498 "OpMemberDecorate %per_vertex_out 2 BuiltIn ClipDistance\n"
4499 "OpMemberDecorate %per_vertex_out 3 BuiltIn CullDistance\n"
4500 "OpDecorate %per_vertex_out Block\n"
4501 "OpDecorate %gl_tessCoord BuiltIn TessCoord\n"
4502 "OpDecorate %in_position Location 2\n"
4503 "OpDecorate %out_color Location 1\n"
4504 "OpDecorate %in_color Location 1\n"
4505 SPIRV_ASSEMBLY_TYPES
4506 SPIRV_ASSEMBLY_CONSTANTS
4507 SPIRV_ASSEMBLY_ARRAYS
4508 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4509 "%per_vertex_out = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4510 "%op_per_vertex_out = OpTypePointer Output %per_vertex_out\n"
4511 "%stream = OpVariable %op_per_vertex_out Output\n"
4512 "%gl_tessCoord = OpVariable %ip_v3f32 Input\n"
4513 "%in_position = OpVariable %ip_a32v4f32 Input\n"
4514 "%out_color = OpVariable %op_v4f32 Output\n"
4515 "%in_color = OpVariable %ip_a32v4f32 Input\n"
4517 "%tesse1_main = OpFunction %void None %fun\n"
4518 "%tesse1_label = OpLabel\n"
4519 "%tesse1_tc_0_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_0\n"
4520 "%tesse1_tc_1_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_1\n"
4521 "%tesse1_tc_2_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_2\n"
4522 "%tesse1_tc_0 = OpLoad %f32 %tesse1_tc_0_ptr\n"
4523 "%tesse1_tc_1 = OpLoad %f32 %tesse1_tc_1_ptr\n"
4524 "%tesse1_tc_2 = OpLoad %f32 %tesse1_tc_2_ptr\n"
4525 "%tesse1_in_pos_0_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_0\n"
4526 "%tesse1_in_pos_1_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_1\n"
4527 "%tesse1_in_pos_2_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_2\n"
4528 "%tesse1_in_pos_0 = OpLoad %v4f32 %tesse1_in_pos_0_ptr\n"
4529 "%tesse1_in_pos_1 = OpLoad %v4f32 %tesse1_in_pos_1_ptr\n"
4530 "%tesse1_in_pos_2 = OpLoad %v4f32 %tesse1_in_pos_2_ptr\n"
4531 "%tesse1_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_0 %tesse1_in_pos_0\n"
4532 "%tesse1_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_1 %tesse1_in_pos_1\n"
4533 "%tesse1_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_2 %tesse1_in_pos_2\n"
4534 "%tesse1_out_pos_ptr = OpAccessChain %op_v4f32 %stream %c_i32_0\n"
4535 "%tesse1_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %tesse1_in_pos_0_weighted %tesse1_in_pos_1_weighted\n"
4536 "%tesse1_computed_out = OpFAdd %v4f32 %tesse1_in_pos_0_plus_pos_1 %tesse1_in_pos_2_weighted\n"
4537 "OpStore %tesse1_out_pos_ptr %tesse1_computed_out\n"
4538 "%tesse1_in_clr_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4539 "%tesse1_in_clr_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4540 "%tesse1_in_clr_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4541 "%tesse1_in_clr_0 = OpLoad %v4f32 %tesse1_in_clr_0_ptr\n"
4542 "%tesse1_in_clr_1 = OpLoad %v4f32 %tesse1_in_clr_1_ptr\n"
4543 "%tesse1_in_clr_2 = OpLoad %v4f32 %tesse1_in_clr_2_ptr\n"
4544 "%tesse1_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_0 %tesse1_in_clr_0\n"
4545 "%tesse1_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_1 %tesse1_in_clr_1\n"
4546 "%tesse1_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_2 %tesse1_in_clr_2\n"
4547 "%tesse1_in_clr_0_plus_col_1 = OpFAdd %v4f32 %tesse1_in_clr_0_weighted %tesse1_in_clr_1_weighted\n"
4548 "%tesse1_computed_clr = OpFAdd %v4f32 %tesse1_in_clr_0_plus_col_1 %tesse1_in_clr_2_weighted\n"
4549 "OpStore %out_color %tesse1_computed_clr\n"
4553 "%tesse2_main = OpFunction %void None %fun\n"
4554 "%tesse2_label = OpLabel\n"
4555 "%tesse2_tc_0_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_0\n"
4556 "%tesse2_tc_1_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_1\n"
4557 "%tesse2_tc_2_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_2\n"
4558 "%tesse2_tc_0 = OpLoad %f32 %tesse2_tc_0_ptr\n"
4559 "%tesse2_tc_1 = OpLoad %f32 %tesse2_tc_1_ptr\n"
4560 "%tesse2_tc_2 = OpLoad %f32 %tesse2_tc_2_ptr\n"
4561 "%tesse2_in_pos_0_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_0\n"
4562 "%tesse2_in_pos_1_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_1\n"
4563 "%tesse2_in_pos_2_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_2\n"
4564 "%tesse2_in_pos_0 = OpLoad %v4f32 %tesse2_in_pos_0_ptr\n"
4565 "%tesse2_in_pos_1 = OpLoad %v4f32 %tesse2_in_pos_1_ptr\n"
4566 "%tesse2_in_pos_2 = OpLoad %v4f32 %tesse2_in_pos_2_ptr\n"
4567 "%tesse2_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_0 %tesse2_in_pos_0\n"
4568 "%tesse2_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_1 %tesse2_in_pos_1\n"
4569 "%tesse2_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_2 %tesse2_in_pos_2\n"
4570 "%tesse2_out_pos_ptr = OpAccessChain %op_v4f32 %stream %c_i32_0\n"
4571 "%tesse2_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %tesse2_in_pos_0_weighted %tesse2_in_pos_1_weighted\n"
4572 "%tesse2_computed_out = OpFAdd %v4f32 %tesse2_in_pos_0_plus_pos_1 %tesse2_in_pos_2_weighted\n"
4573 "OpStore %tesse2_out_pos_ptr %tesse2_computed_out\n"
4574 "%tesse2_in_clr_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4575 "%tesse2_in_clr_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4576 "%tesse2_in_clr_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4577 "%tesse2_in_clr_0 = OpLoad %v4f32 %tesse2_in_clr_0_ptr\n"
4578 "%tesse2_in_clr_1 = OpLoad %v4f32 %tesse2_in_clr_1_ptr\n"
4579 "%tesse2_in_clr_2 = OpLoad %v4f32 %tesse2_in_clr_2_ptr\n"
4580 "%tesse2_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_0 %tesse2_in_clr_0\n"
4581 "%tesse2_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_1 %tesse2_in_clr_1\n"
4582 "%tesse2_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_2 %tesse2_in_clr_2\n"
4583 "%tesse2_in_clr_0_plus_col_1 = OpFAdd %v4f32 %tesse2_in_clr_0_weighted %tesse2_in_clr_1_weighted\n"
4584 "%tesse2_computed_clr = OpFAdd %v4f32 %tesse2_in_clr_0_plus_col_1 %tesse2_in_clr_2_weighted\n"
4585 "%tesse2_clr_transformed = OpFSub %v4f32 %cval %tesse2_computed_clr\n"
4586 "OpStore %out_color %tesse2_clr_transformed\n"
4591 // Sets up and runs a Vulkan pipeline, then spot-checks the resulting image.
4592 // Feeds the pipeline a set of colored triangles, which then must occur in the
4593 // rendered image. The surface is cleared before executing the pipeline, so
4594 // whatever the shaders draw can be directly spot-checked.
4595 TestStatus runAndVerifyDefaultPipeline (Context& context, InstanceContext instance)
4597 const VkDevice vkDevice = context.getDevice();
4598 const DeviceInterface& vk = context.getDeviceInterface();
4599 const VkQueue queue = context.getUniversalQueue();
4600 const deUint32 queueFamilyIndex = context.getUniversalQueueFamilyIndex();
4601 const tcu::UVec2 renderSize (256, 256);
4602 vector<ModuleHandleSp> modules;
4603 map<VkShaderStageFlagBits, VkShaderModule> moduleByStage;
4604 const int testSpecificSeed = 31354125;
4605 const int seed = context.getTestContext().getCommandLine().getBaseSeed() ^ testSpecificSeed;
4606 bool supportsGeometry = false;
4607 bool supportsTessellation = false;
4608 bool hasTessellation = false;
4610 const VkPhysicalDeviceFeatures& features = context.getDeviceFeatures();
4611 supportsGeometry = features.geometryShader == VK_TRUE;
4612 supportsTessellation = features.tessellationShader == VK_TRUE;
4613 hasTessellation = (instance.requiredStages & VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT) ||
4614 (instance.requiredStages & VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT);
4616 if (hasTessellation && !supportsTessellation)
4618 throw tcu::NotSupportedError(std::string("Tessellation not supported"));
4621 if ((instance.requiredStages & VK_SHADER_STAGE_GEOMETRY_BIT) &&
4624 throw tcu::NotSupportedError(std::string("Geometry not supported"));
4627 de::Random(seed).shuffle(instance.inputColors, instance.inputColors+4);
4628 de::Random(seed).shuffle(instance.outputColors, instance.outputColors+4);
4629 const Vec4 vertexData[] =
4631 // Upper left corner:
4632 Vec4(-1.0f, -1.0f, 0.0f, 1.0f), instance.inputColors[0].toVec(),
4633 Vec4(-0.5f, -1.0f, 0.0f, 1.0f), instance.inputColors[0].toVec(),
4634 Vec4(-1.0f, -0.5f, 0.0f, 1.0f), instance.inputColors[0].toVec(),
4636 // Upper right corner:
4637 Vec4(+0.5f, -1.0f, 0.0f, 1.0f), instance.inputColors[1].toVec(),
4638 Vec4(+1.0f, -1.0f, 0.0f, 1.0f), instance.inputColors[1].toVec(),
4639 Vec4(+1.0f, -0.5f, 0.0f, 1.0f), instance.inputColors[1].toVec(),
4641 // Lower left corner:
4642 Vec4(-1.0f, +0.5f, 0.0f, 1.0f), instance.inputColors[2].toVec(),
4643 Vec4(-0.5f, +1.0f, 0.0f, 1.0f), instance.inputColors[2].toVec(),
4644 Vec4(-1.0f, +1.0f, 0.0f, 1.0f), instance.inputColors[2].toVec(),
4646 // Lower right corner:
4647 Vec4(+1.0f, +0.5f, 0.0f, 1.0f), instance.inputColors[3].toVec(),
4648 Vec4(+1.0f, +1.0f, 0.0f, 1.0f), instance.inputColors[3].toVec(),
4649 Vec4(+0.5f, +1.0f, 0.0f, 1.0f), instance.inputColors[3].toVec()
4651 const size_t singleVertexDataSize = 2 * sizeof(Vec4);
4652 const size_t vertexCount = sizeof(vertexData) / singleVertexDataSize;
4654 const VkBufferCreateInfo vertexBufferParams =
4656 VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType;
4657 DE_NULL, // const void* pNext;
4658 0u, // VkBufferCreateFlags flags;
4659 (VkDeviceSize)sizeof(vertexData), // VkDeviceSize size;
4660 VK_BUFFER_USAGE_VERTEX_BUFFER_BIT, // VkBufferUsageFlags usage;
4661 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
4662 1u, // deUint32 queueFamilyCount;
4663 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices;
4665 const Unique<VkBuffer> vertexBuffer (createBuffer(vk, vkDevice, &vertexBufferParams));
4666 const UniquePtr<Allocation> vertexBufferMemory (context.getDefaultAllocator().allocate(getBufferMemoryRequirements(vk, vkDevice, *vertexBuffer), MemoryRequirement::HostVisible));
4668 VK_CHECK(vk.bindBufferMemory(vkDevice, *vertexBuffer, vertexBufferMemory->getMemory(), vertexBufferMemory->getOffset()));
4670 const VkDeviceSize imageSizeBytes = (VkDeviceSize)(sizeof(deUint32)*renderSize.x()*renderSize.y());
4671 const VkBufferCreateInfo readImageBufferParams =
4673 VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType;
4674 DE_NULL, // const void* pNext;
4675 0u, // VkBufferCreateFlags flags;
4676 imageSizeBytes, // VkDeviceSize size;
4677 VK_BUFFER_USAGE_TRANSFER_DST_BIT, // VkBufferUsageFlags usage;
4678 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
4679 1u, // deUint32 queueFamilyCount;
4680 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices;
4682 const Unique<VkBuffer> readImageBuffer (createBuffer(vk, vkDevice, &readImageBufferParams));
4683 const UniquePtr<Allocation> readImageBufferMemory (context.getDefaultAllocator().allocate(getBufferMemoryRequirements(vk, vkDevice, *readImageBuffer), MemoryRequirement::HostVisible));
4685 VK_CHECK(vk.bindBufferMemory(vkDevice, *readImageBuffer, readImageBufferMemory->getMemory(), readImageBufferMemory->getOffset()));
4687 const VkImageCreateInfo imageParams =
4689 VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, // VkStructureType sType;
4690 DE_NULL, // const void* pNext;
4691 0u, // VkImageCreateFlags flags;
4692 VK_IMAGE_TYPE_2D, // VkImageType imageType;
4693 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format;
4694 { renderSize.x(), renderSize.y(), 1 }, // VkExtent3D extent;
4695 1u, // deUint32 mipLevels;
4696 1u, // deUint32 arraySize;
4697 VK_SAMPLE_COUNT_1_BIT, // deUint32 samples;
4698 VK_IMAGE_TILING_OPTIMAL, // VkImageTiling tiling;
4699 VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT|VK_IMAGE_USAGE_TRANSFER_SRC_BIT, // VkImageUsageFlags usage;
4700 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
4701 1u, // deUint32 queueFamilyCount;
4702 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices;
4703 VK_IMAGE_LAYOUT_UNDEFINED, // VkImageLayout initialLayout;
4706 const Unique<VkImage> image (createImage(vk, vkDevice, &imageParams));
4707 const UniquePtr<Allocation> imageMemory (context.getDefaultAllocator().allocate(getImageMemoryRequirements(vk, vkDevice, *image), MemoryRequirement::Any));
4709 VK_CHECK(vk.bindImageMemory(vkDevice, *image, imageMemory->getMemory(), imageMemory->getOffset()));
4711 const VkAttachmentDescription colorAttDesc =
4713 0u, // VkAttachmentDescriptionFlags flags;
4714 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format;
4715 VK_SAMPLE_COUNT_1_BIT, // deUint32 samples;
4716 VK_ATTACHMENT_LOAD_OP_CLEAR, // VkAttachmentLoadOp loadOp;
4717 VK_ATTACHMENT_STORE_OP_STORE, // VkAttachmentStoreOp storeOp;
4718 VK_ATTACHMENT_LOAD_OP_DONT_CARE, // VkAttachmentLoadOp stencilLoadOp;
4719 VK_ATTACHMENT_STORE_OP_DONT_CARE, // VkAttachmentStoreOp stencilStoreOp;
4720 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout initialLayout;
4721 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout finalLayout;
4723 const VkAttachmentReference colorAttRef =
4725 0u, // deUint32 attachment;
4726 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout layout;
4728 const VkSubpassDescription subpassDesc =
4730 0u, // VkSubpassDescriptionFlags flags;
4731 VK_PIPELINE_BIND_POINT_GRAPHICS, // VkPipelineBindPoint pipelineBindPoint;
4732 0u, // deUint32 inputCount;
4733 DE_NULL, // const VkAttachmentReference* pInputAttachments;
4734 1u, // deUint32 colorCount;
4735 &colorAttRef, // const VkAttachmentReference* pColorAttachments;
4736 DE_NULL, // const VkAttachmentReference* pResolveAttachments;
4737 DE_NULL, // const VkAttachmentReference* pDepthStencilAttachment;
4738 0u, // deUint32 preserveCount;
4739 DE_NULL, // const VkAttachmentReference* pPreserveAttachments;
4742 const VkRenderPassCreateInfo renderPassParams =
4744 VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO, // VkStructureType sType;
4745 DE_NULL, // const void* pNext;
4746 (VkRenderPassCreateFlags)0,
4747 1u, // deUint32 attachmentCount;
4748 &colorAttDesc, // const VkAttachmentDescription* pAttachments;
4749 1u, // deUint32 subpassCount;
4750 &subpassDesc, // const VkSubpassDescription* pSubpasses;
4751 0u, // deUint32 dependencyCount;
4752 DE_NULL, // const VkSubpassDependency* pDependencies;
4754 const Unique<VkRenderPass> renderPass (createRenderPass(vk, vkDevice, &renderPassParams));
4756 const VkImageViewCreateInfo colorAttViewParams =
4758 VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, // VkStructureType sType;
4759 DE_NULL, // const void* pNext;
4760 0u, // VkImageViewCreateFlags flags;
4761 *image, // VkImage image;
4762 VK_IMAGE_VIEW_TYPE_2D, // VkImageViewType viewType;
4763 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format;
4765 VK_COMPONENT_SWIZZLE_R,
4766 VK_COMPONENT_SWIZZLE_G,
4767 VK_COMPONENT_SWIZZLE_B,
4768 VK_COMPONENT_SWIZZLE_A
4769 }, // VkChannelMapping channels;
4771 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspectFlags aspectMask;
4772 0u, // deUint32 baseMipLevel;
4773 1u, // deUint32 mipLevels;
4774 0u, // deUint32 baseArrayLayer;
4775 1u, // deUint32 arraySize;
4776 }, // VkImageSubresourceRange subresourceRange;
4778 const Unique<VkImageView> colorAttView (createImageView(vk, vkDevice, &colorAttViewParams));
4782 const VkPipelineLayoutCreateInfo pipelineLayoutParams =
4784 VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, // VkStructureType sType;
4785 DE_NULL, // const void* pNext;
4786 (VkPipelineLayoutCreateFlags)0,
4787 0u, // deUint32 descriptorSetCount;
4788 DE_NULL, // const VkDescriptorSetLayout* pSetLayouts;
4789 0u, // deUint32 pushConstantRangeCount;
4790 DE_NULL, // const VkPushConstantRange* pPushConstantRanges;
4792 const Unique<VkPipelineLayout> pipelineLayout (createPipelineLayout(vk, vkDevice, &pipelineLayoutParams));
4795 vector<VkPipelineShaderStageCreateInfo> shaderStageParams;
4796 // We need these vectors to make sure that information about specialization constants for each stage can outlive createGraphicsPipeline().
4797 vector<vector<VkSpecializationMapEntry> > specConstantEntries;
4798 vector<VkSpecializationInfo> specializationInfos;
4799 createPipelineShaderStages(vk, vkDevice, instance, context, modules, shaderStageParams);
4801 // And we don't want the reallocation of these vectors to invalidate pointers pointing to their contents.
4802 specConstantEntries.reserve(shaderStageParams.size());
4803 specializationInfos.reserve(shaderStageParams.size());
4805 // Patch the specialization info field in PipelineShaderStageCreateInfos.
4806 for (vector<VkPipelineShaderStageCreateInfo>::iterator stageInfo = shaderStageParams.begin(); stageInfo != shaderStageParams.end(); ++stageInfo)
4808 const StageToSpecConstantMap::const_iterator stageIt = instance.specConstants.find(stageInfo->stage);
4810 if (stageIt != instance.specConstants.end())
4812 const size_t numSpecConstants = stageIt->second.size();
4813 vector<VkSpecializationMapEntry> entries;
4814 VkSpecializationInfo specInfo;
4816 entries.resize(numSpecConstants);
4818 // Only support 32-bit integers as spec constants now. And their constant IDs are numbered sequentially starting from 0.
4819 for (size_t ndx = 0; ndx < numSpecConstants; ++ndx)
4821 entries[ndx].constantID = (deUint32)ndx;
4822 entries[ndx].offset = deUint32(ndx * sizeof(deInt32));
4823 entries[ndx].size = sizeof(deInt32);
4826 specConstantEntries.push_back(entries);
4828 specInfo.mapEntryCount = (deUint32)numSpecConstants;
4829 specInfo.pMapEntries = specConstantEntries.back().data();
4830 specInfo.dataSize = numSpecConstants * sizeof(deInt32);
4831 specInfo.pData = stageIt->second.data();
4832 specializationInfos.push_back(specInfo);
4834 stageInfo->pSpecializationInfo = &specializationInfos.back();
4837 const VkPipelineDepthStencilStateCreateInfo depthStencilParams =
4839 VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO, // VkStructureType sType;
4840 DE_NULL, // const void* pNext;
4841 (VkPipelineDepthStencilStateCreateFlags)0,
4842 DE_FALSE, // deUint32 depthTestEnable;
4843 DE_FALSE, // deUint32 depthWriteEnable;
4844 VK_COMPARE_OP_ALWAYS, // VkCompareOp depthCompareOp;
4845 DE_FALSE, // deUint32 depthBoundsTestEnable;
4846 DE_FALSE, // deUint32 stencilTestEnable;
4848 VK_STENCIL_OP_KEEP, // VkStencilOp stencilFailOp;
4849 VK_STENCIL_OP_KEEP, // VkStencilOp stencilPassOp;
4850 VK_STENCIL_OP_KEEP, // VkStencilOp stencilDepthFailOp;
4851 VK_COMPARE_OP_ALWAYS, // VkCompareOp stencilCompareOp;
4852 0u, // deUint32 stencilCompareMask;
4853 0u, // deUint32 stencilWriteMask;
4854 0u, // deUint32 stencilReference;
4855 }, // VkStencilOpState front;
4857 VK_STENCIL_OP_KEEP, // VkStencilOp stencilFailOp;
4858 VK_STENCIL_OP_KEEP, // VkStencilOp stencilPassOp;
4859 VK_STENCIL_OP_KEEP, // VkStencilOp stencilDepthFailOp;
4860 VK_COMPARE_OP_ALWAYS, // VkCompareOp stencilCompareOp;
4861 0u, // deUint32 stencilCompareMask;
4862 0u, // deUint32 stencilWriteMask;
4863 0u, // deUint32 stencilReference;
4864 }, // VkStencilOpState back;
4865 -1.0f, // float minDepthBounds;
4866 +1.0f, // float maxDepthBounds;
4868 const VkViewport viewport0 =
4870 0.0f, // float originX;
4871 0.0f, // float originY;
4872 (float)renderSize.x(), // float width;
4873 (float)renderSize.y(), // float height;
4874 0.0f, // float minDepth;
4875 1.0f, // float maxDepth;
4877 const VkRect2D scissor0 =
4882 }, // VkOffset2D offset;
4884 renderSize.x(), // deInt32 width;
4885 renderSize.y(), // deInt32 height;
4886 }, // VkExtent2D extent;
4888 const VkPipelineViewportStateCreateInfo viewportParams =
4890 VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO, // VkStructureType sType;
4891 DE_NULL, // const void* pNext;
4892 (VkPipelineViewportStateCreateFlags)0,
4893 1u, // deUint32 viewportCount;
4898 const VkSampleMask sampleMask = ~0u;
4899 const VkPipelineMultisampleStateCreateInfo multisampleParams =
4901 VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO, // VkStructureType sType;
4902 DE_NULL, // const void* pNext;
4903 (VkPipelineMultisampleStateCreateFlags)0,
4904 VK_SAMPLE_COUNT_1_BIT, // VkSampleCountFlagBits rasterSamples;
4905 DE_FALSE, // deUint32 sampleShadingEnable;
4906 0.0f, // float minSampleShading;
4907 &sampleMask, // const VkSampleMask* pSampleMask;
4908 DE_FALSE, // VkBool32 alphaToCoverageEnable;
4909 DE_FALSE, // VkBool32 alphaToOneEnable;
4911 const VkPipelineRasterizationStateCreateInfo rasterParams =
4913 VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO, // VkStructureType sType;
4914 DE_NULL, // const void* pNext;
4915 (VkPipelineRasterizationStateCreateFlags)0,
4916 DE_TRUE, // deUint32 depthClipEnable;
4917 DE_FALSE, // deUint32 rasterizerDiscardEnable;
4918 VK_POLYGON_MODE_FILL, // VkFillMode fillMode;
4919 VK_CULL_MODE_NONE, // VkCullMode cullMode;
4920 VK_FRONT_FACE_COUNTER_CLOCKWISE, // VkFrontFace frontFace;
4921 VK_FALSE, // VkBool32 depthBiasEnable;
4922 0.0f, // float depthBias;
4923 0.0f, // float depthBiasClamp;
4924 0.0f, // float slopeScaledDepthBias;
4925 1.0f, // float lineWidth;
4927 const VkPrimitiveTopology topology = hasTessellation? VK_PRIMITIVE_TOPOLOGY_PATCH_LIST: VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
4928 const VkPipelineInputAssemblyStateCreateInfo inputAssemblyParams =
4930 VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO, // VkStructureType sType;
4931 DE_NULL, // const void* pNext;
4932 (VkPipelineInputAssemblyStateCreateFlags)0,
4933 topology, // VkPrimitiveTopology topology;
4934 DE_FALSE, // deUint32 primitiveRestartEnable;
4936 const VkVertexInputBindingDescription vertexBinding0 =
4938 0u, // deUint32 binding;
4939 deUint32(singleVertexDataSize), // deUint32 strideInBytes;
4940 VK_VERTEX_INPUT_RATE_VERTEX // VkVertexInputStepRate stepRate;
4942 const VkVertexInputAttributeDescription vertexAttrib0[2] =
4945 0u, // deUint32 location;
4946 0u, // deUint32 binding;
4947 VK_FORMAT_R32G32B32A32_SFLOAT, // VkFormat format;
4948 0u // deUint32 offsetInBytes;
4951 1u, // deUint32 location;
4952 0u, // deUint32 binding;
4953 VK_FORMAT_R32G32B32A32_SFLOAT, // VkFormat format;
4954 sizeof(Vec4), // deUint32 offsetInBytes;
4958 const VkPipelineVertexInputStateCreateInfo vertexInputStateParams =
4960 VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO, // VkStructureType sType;
4961 DE_NULL, // const void* pNext;
4962 (VkPipelineVertexInputStateCreateFlags)0,
4963 1u, // deUint32 bindingCount;
4964 &vertexBinding0, // const VkVertexInputBindingDescription* pVertexBindingDescriptions;
4965 2u, // deUint32 attributeCount;
4966 vertexAttrib0, // const VkVertexInputAttributeDescription* pVertexAttributeDescriptions;
4968 const VkPipelineColorBlendAttachmentState attBlendParams =
4970 DE_FALSE, // deUint32 blendEnable;
4971 VK_BLEND_FACTOR_ONE, // VkBlend srcBlendColor;
4972 VK_BLEND_FACTOR_ZERO, // VkBlend destBlendColor;
4973 VK_BLEND_OP_ADD, // VkBlendOp blendOpColor;
4974 VK_BLEND_FACTOR_ONE, // VkBlend srcBlendAlpha;
4975 VK_BLEND_FACTOR_ZERO, // VkBlend destBlendAlpha;
4976 VK_BLEND_OP_ADD, // VkBlendOp blendOpAlpha;
4977 (VK_COLOR_COMPONENT_R_BIT|
4978 VK_COLOR_COMPONENT_G_BIT|
4979 VK_COLOR_COMPONENT_B_BIT|
4980 VK_COLOR_COMPONENT_A_BIT), // VkChannelFlags channelWriteMask;
4982 const VkPipelineColorBlendStateCreateInfo blendParams =
4984 VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO, // VkStructureType sType;
4985 DE_NULL, // const void* pNext;
4986 (VkPipelineColorBlendStateCreateFlags)0,
4987 DE_FALSE, // VkBool32 logicOpEnable;
4988 VK_LOGIC_OP_COPY, // VkLogicOp logicOp;
4989 1u, // deUint32 attachmentCount;
4990 &attBlendParams, // const VkPipelineColorBlendAttachmentState* pAttachments;
4991 { 0.0f, 0.0f, 0.0f, 0.0f }, // float blendConst[4];
4993 const VkPipelineDynamicStateCreateInfo dynamicStateInfo =
4995 VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO, // VkStructureType sType;
4996 DE_NULL, // const void* pNext;
4997 (VkPipelineDynamicStateCreateFlags)0,
4998 0u, // deUint32 dynamicStateCount;
4999 DE_NULL // const VkDynamicState* pDynamicStates;
5002 const VkPipelineTessellationStateCreateInfo tessellationState =
5004 VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO,
5006 (VkPipelineTessellationStateCreateFlags)0,
5010 const VkPipelineTessellationStateCreateInfo* tessellationInfo = hasTessellation ? &tessellationState: DE_NULL;
5011 const VkGraphicsPipelineCreateInfo pipelineParams =
5013 VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, // VkStructureType sType;
5014 DE_NULL, // const void* pNext;
5015 0u, // VkPipelineCreateFlags flags;
5016 (deUint32)shaderStageParams.size(), // deUint32 stageCount;
5017 &shaderStageParams[0], // const VkPipelineShaderStageCreateInfo* pStages;
5018 &vertexInputStateParams, // const VkPipelineVertexInputStateCreateInfo* pVertexInputState;
5019 &inputAssemblyParams, // const VkPipelineInputAssemblyStateCreateInfo* pInputAssemblyState;
5020 tessellationInfo, // const VkPipelineTessellationStateCreateInfo* pTessellationState;
5021 &viewportParams, // const VkPipelineViewportStateCreateInfo* pViewportState;
5022 &rasterParams, // const VkPipelineRasterStateCreateInfo* pRasterState;
5023 &multisampleParams, // const VkPipelineMultisampleStateCreateInfo* pMultisampleState;
5024 &depthStencilParams, // const VkPipelineDepthStencilStateCreateInfo* pDepthStencilState;
5025 &blendParams, // const VkPipelineColorBlendStateCreateInfo* pColorBlendState;
5026 &dynamicStateInfo, // const VkPipelineDynamicStateCreateInfo* pDynamicState;
5027 *pipelineLayout, // VkPipelineLayout layout;
5028 *renderPass, // VkRenderPass renderPass;
5029 0u, // deUint32 subpass;
5030 DE_NULL, // VkPipeline basePipelineHandle;
5031 0u, // deInt32 basePipelineIndex;
5034 const Unique<VkPipeline> pipeline (createGraphicsPipeline(vk, vkDevice, DE_NULL, &pipelineParams));
5037 const VkFramebufferCreateInfo framebufferParams =
5039 VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO, // VkStructureType sType;
5040 DE_NULL, // const void* pNext;
5041 (VkFramebufferCreateFlags)0,
5042 *renderPass, // VkRenderPass renderPass;
5043 1u, // deUint32 attachmentCount;
5044 &*colorAttView, // const VkImageView* pAttachments;
5045 (deUint32)renderSize.x(), // deUint32 width;
5046 (deUint32)renderSize.y(), // deUint32 height;
5047 1u, // deUint32 layers;
5049 const Unique<VkFramebuffer> framebuffer (createFramebuffer(vk, vkDevice, &framebufferParams));
5051 const VkCommandPoolCreateInfo cmdPoolParams =
5053 VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO, // VkStructureType sType;
5054 DE_NULL, // const void* pNext;
5055 VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, // VkCmdPoolCreateFlags flags;
5056 queueFamilyIndex, // deUint32 queueFamilyIndex;
5058 const Unique<VkCommandPool> cmdPool (createCommandPool(vk, vkDevice, &cmdPoolParams));
5061 const VkCommandBufferAllocateInfo cmdBufParams =
5063 VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO, // VkStructureType sType;
5064 DE_NULL, // const void* pNext;
5065 *cmdPool, // VkCmdPool pool;
5066 VK_COMMAND_BUFFER_LEVEL_PRIMARY, // VkCmdBufferLevel level;
5067 1u, // deUint32 count;
5069 const Unique<VkCommandBuffer> cmdBuf (allocateCommandBuffer(vk, vkDevice, &cmdBufParams));
5071 const VkCommandBufferBeginInfo cmdBufBeginParams =
5073 VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, // VkStructureType sType;
5074 DE_NULL, // const void* pNext;
5075 (VkCommandBufferUsageFlags)0,
5076 (const VkCommandBufferInheritanceInfo*)DE_NULL,
5080 VK_CHECK(vk.beginCommandBuffer(*cmdBuf, &cmdBufBeginParams));
5083 const VkMemoryBarrier vertFlushBarrier =
5085 VK_STRUCTURE_TYPE_MEMORY_BARRIER, // VkStructureType sType;
5086 DE_NULL, // const void* pNext;
5087 VK_ACCESS_HOST_WRITE_BIT, // VkMemoryOutputFlags outputMask;
5088 VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT, // VkMemoryInputFlags inputMask;
5090 const VkImageMemoryBarrier colorAttBarrier =
5092 VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType;
5093 DE_NULL, // const void* pNext;
5094 0u, // VkMemoryOutputFlags outputMask;
5095 VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, // VkMemoryInputFlags inputMask;
5096 VK_IMAGE_LAYOUT_UNDEFINED, // VkImageLayout oldLayout;
5097 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout newLayout;
5098 queueFamilyIndex, // deUint32 srcQueueFamilyIndex;
5099 queueFamilyIndex, // deUint32 destQueueFamilyIndex;
5100 *image, // VkImage image;
5102 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspect aspect;
5103 0u, // deUint32 baseMipLevel;
5104 1u, // deUint32 mipLevels;
5105 0u, // deUint32 baseArraySlice;
5106 1u, // deUint32 arraySize;
5107 } // VkImageSubresourceRange subresourceRange;
5109 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);
5113 const VkClearValue clearValue = makeClearValueColorF32(0.125f, 0.25f, 0.75f, 1.0f);
5114 const VkRenderPassBeginInfo passBeginParams =
5116 VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO, // VkStructureType sType;
5117 DE_NULL, // const void* pNext;
5118 *renderPass, // VkRenderPass renderPass;
5119 *framebuffer, // VkFramebuffer framebuffer;
5120 { { 0, 0 }, { renderSize.x(), renderSize.y() } }, // VkRect2D renderArea;
5121 1u, // deUint32 clearValueCount;
5122 &clearValue, // const VkClearValue* pClearValues;
5124 vk.cmdBeginRenderPass(*cmdBuf, &passBeginParams, VK_SUBPASS_CONTENTS_INLINE);
5127 vk.cmdBindPipeline(*cmdBuf, VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline);
5129 const VkDeviceSize bindingOffset = 0;
5130 vk.cmdBindVertexBuffers(*cmdBuf, 0u, 1u, &vertexBuffer.get(), &bindingOffset);
5132 vk.cmdDraw(*cmdBuf, deUint32(vertexCount), 1u /*run pipeline once*/, 0u /*first vertex*/, 0u /*first instanceIndex*/);
5133 vk.cmdEndRenderPass(*cmdBuf);
5136 const VkImageMemoryBarrier renderFinishBarrier =
5138 VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType;
5139 DE_NULL, // const void* pNext;
5140 VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, // VkMemoryOutputFlags outputMask;
5141 VK_ACCESS_TRANSFER_READ_BIT, // VkMemoryInputFlags inputMask;
5142 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout oldLayout;
5143 VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, // VkImageLayout newLayout;
5144 queueFamilyIndex, // deUint32 srcQueueFamilyIndex;
5145 queueFamilyIndex, // deUint32 destQueueFamilyIndex;
5146 *image, // VkImage image;
5148 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspectFlags aspectMask;
5149 0u, // deUint32 baseMipLevel;
5150 1u, // deUint32 mipLevels;
5151 0u, // deUint32 baseArraySlice;
5152 1u, // deUint32 arraySize;
5153 } // VkImageSubresourceRange subresourceRange;
5155 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);
5159 const VkBufferImageCopy copyParams =
5161 (VkDeviceSize)0u, // VkDeviceSize bufferOffset;
5162 (deUint32)renderSize.x(), // deUint32 bufferRowLength;
5163 (deUint32)renderSize.y(), // deUint32 bufferImageHeight;
5165 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspect aspect;
5166 0u, // deUint32 mipLevel;
5167 0u, // deUint32 arrayLayer;
5168 1u, // deUint32 arraySize;
5169 }, // VkImageSubresourceCopy imageSubresource;
5170 { 0u, 0u, 0u }, // VkOffset3D imageOffset;
5171 { renderSize.x(), renderSize.y(), 1u } // VkExtent3D imageExtent;
5173 vk.cmdCopyImageToBuffer(*cmdBuf, *image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, *readImageBuffer, 1u, ©Params);
5177 const VkBufferMemoryBarrier copyFinishBarrier =
5179 VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, // VkStructureType sType;
5180 DE_NULL, // const void* pNext;
5181 VK_ACCESS_TRANSFER_WRITE_BIT, // VkMemoryOutputFlags outputMask;
5182 VK_ACCESS_HOST_READ_BIT, // VkMemoryInputFlags inputMask;
5183 queueFamilyIndex, // deUint32 srcQueueFamilyIndex;
5184 queueFamilyIndex, // deUint32 destQueueFamilyIndex;
5185 *readImageBuffer, // VkBuffer buffer;
5186 0u, // VkDeviceSize offset;
5187 imageSizeBytes // VkDeviceSize size;
5189 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);
5192 VK_CHECK(vk.endCommandBuffer(*cmdBuf));
5194 // Upload vertex data
5196 const VkMappedMemoryRange range =
5198 VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, // VkStructureType sType;
5199 DE_NULL, // const void* pNext;
5200 vertexBufferMemory->getMemory(), // VkDeviceMemory mem;
5201 0, // VkDeviceSize offset;
5202 (VkDeviceSize)sizeof(vertexData), // VkDeviceSize size;
5204 void* vertexBufPtr = vertexBufferMemory->getHostPtr();
5206 deMemcpy(vertexBufPtr, &vertexData[0], sizeof(vertexData));
5207 VK_CHECK(vk.flushMappedMemoryRanges(vkDevice, 1u, &range));
5210 // Submit & wait for completion
5212 const VkFenceCreateInfo fenceParams =
5214 VK_STRUCTURE_TYPE_FENCE_CREATE_INFO, // VkStructureType sType;
5215 DE_NULL, // const void* pNext;
5216 0u, // VkFenceCreateFlags flags;
5218 const Unique<VkFence> fence (createFence(vk, vkDevice, &fenceParams));
5219 const VkSubmitInfo submitInfo =
5221 VK_STRUCTURE_TYPE_SUBMIT_INFO,
5224 (const VkSemaphore*)DE_NULL,
5225 (const VkPipelineStageFlags*)DE_NULL,
5229 (const VkSemaphore*)DE_NULL,
5232 VK_CHECK(vk.queueSubmit(queue, 1u, &submitInfo, *fence));
5233 VK_CHECK(vk.waitForFences(vkDevice, 1u, &fence.get(), DE_TRUE, ~0ull));
5236 const void* imagePtr = readImageBufferMemory->getHostPtr();
5237 const tcu::ConstPixelBufferAccess pixelBuffer(tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNORM_INT8),
5238 renderSize.x(), renderSize.y(), 1, imagePtr);
5241 const VkMappedMemoryRange range =
5243 VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, // VkStructureType sType;
5244 DE_NULL, // const void* pNext;
5245 readImageBufferMemory->getMemory(), // VkDeviceMemory mem;
5246 0, // VkDeviceSize offset;
5247 imageSizeBytes, // VkDeviceSize size;
5250 VK_CHECK(vk.invalidateMappedMemoryRanges(vkDevice, 1u, &range));
5251 context.getTestContext().getLog() << TestLog::Image("Result", "Result", pixelBuffer);
5254 const RGBA threshold(1, 1, 1, 1);
5255 const RGBA upperLeft(pixelBuffer.getPixel(1, 1));
5256 if (!tcu::compareThreshold(upperLeft, instance.outputColors[0], threshold))
5257 return TestStatus::fail("Upper left corner mismatch");
5259 const RGBA upperRight(pixelBuffer.getPixel(pixelBuffer.getWidth() - 1, 1));
5260 if (!tcu::compareThreshold(upperRight, instance.outputColors[1], threshold))
5261 return TestStatus::fail("Upper right corner mismatch");
5263 const RGBA lowerLeft(pixelBuffer.getPixel(1, pixelBuffer.getHeight() - 1));
5264 if (!tcu::compareThreshold(lowerLeft, instance.outputColors[2], threshold))
5265 return TestStatus::fail("Lower left corner mismatch");
5267 const RGBA lowerRight(pixelBuffer.getPixel(pixelBuffer.getWidth() - 1, pixelBuffer.getHeight() - 1));
5268 if (!tcu::compareThreshold(lowerRight, instance.outputColors[3], threshold))
5269 return TestStatus::fail("Lower right corner mismatch");
5271 return TestStatus::pass("Rendered output matches input");
5274 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)
5276 const ShaderElement vertFragPipelineStages[] =
5278 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
5279 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
5282 const ShaderElement tessPipelineStages[] =
5284 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
5285 ShaderElement("tessc", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
5286 ShaderElement("tesse", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
5287 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
5290 const ShaderElement geomPipelineStages[] =
5292 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
5293 ShaderElement("geom", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
5294 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
5297 StageToSpecConstantMap specConstantMap;
5299 specConstantMap[VK_SHADER_STAGE_VERTEX_BIT] = specConstants;
5300 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_vert", "", addShaderCodeCustomVertex, runAndVerifyDefaultPipeline,
5301 createInstanceContext(vertFragPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5303 specConstantMap.clear();
5304 specConstantMap[VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT] = specConstants;
5305 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_tessc", "", addShaderCodeCustomTessControl, runAndVerifyDefaultPipeline,
5306 createInstanceContext(tessPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5308 specConstantMap.clear();
5309 specConstantMap[VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT] = specConstants;
5310 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_tesse", "", addShaderCodeCustomTessEval, runAndVerifyDefaultPipeline,
5311 createInstanceContext(tessPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5313 specConstantMap.clear();
5314 specConstantMap[VK_SHADER_STAGE_GEOMETRY_BIT] = specConstants;
5315 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_geom", "", addShaderCodeCustomGeometry, runAndVerifyDefaultPipeline,
5316 createInstanceContext(geomPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5318 specConstantMap.clear();
5319 specConstantMap[VK_SHADER_STAGE_FRAGMENT_BIT] = specConstants;
5320 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_frag", "", addShaderCodeCustomFragment, runAndVerifyDefaultPipeline,
5321 createInstanceContext(vertFragPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5324 inline void createTestsForAllStages (const std::string& name, const RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments, tcu::TestCaseGroup* tests)
5326 vector<deInt32> noSpecConstants;
5327 createTestsForAllStages(name, inputColors, outputColors, testCodeFragments, noSpecConstants, tests);
5332 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
5334 struct NameCodePair { string name, code; };
5335 RGBA defaultColors[4];
5336 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
5337 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
5338 map<string, string> fragments = passthruFragments();
5339 const NameCodePair tests[] =
5341 {"unknown", "OpSource Unknown 321"},
5342 {"essl", "OpSource ESSL 310"},
5343 {"glsl", "OpSource GLSL 450"},
5344 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
5345 {"opencl_c", "OpSource OpenCL_C 120"},
5346 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
5347 {"file", opsourceGLSLWithFile},
5348 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
5349 // Longest possible source string: SPIR-V limits instructions to 65535
5350 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
5351 // contain 65530 UTF8 characters (one word each) plus one last word
5352 // containing 3 ASCII characters and \0.
5353 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
5356 getDefaultColors(defaultColors);
5357 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5359 fragments["debug"] = tests[testNdx].code;
5360 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5363 return opSourceTests.release();
5366 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
5368 struct NameCodePair { string name, code; };
5369 RGBA defaultColors[4];
5370 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
5371 map<string, string> fragments = passthruFragments();
5372 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
5373 const NameCodePair tests[] =
5375 {"empty", opsource + "OpSourceContinued \"\""},
5376 {"short", opsource + "OpSourceContinued \"abcde\""},
5377 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
5378 // Longest possible source string: SPIR-V limits instructions to 65535
5379 // words, of which the first one is OpSourceContinued/length; the rest
5380 // will contain 65533 UTF8 characters (one word each) plus one last word
5381 // containing 3 ASCII characters and \0.
5382 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
5385 getDefaultColors(defaultColors);
5386 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5388 fragments["debug"] = tests[testNdx].code;
5389 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5392 return opSourceTests.release();
5395 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
5397 RGBA defaultColors[4];
5398 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
5399 map<string, string> fragments;
5400 getDefaultColors(defaultColors);
5401 fragments["debug"] =
5402 "%name = OpString \"name\"\n";
5404 fragments["pre_main"] =
5407 "OpLine %name 1 1\n"
5409 "OpLine %name 1 1\n"
5410 "OpLine %name 1 1\n"
5411 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5413 "OpLine %name 1 1\n"
5415 "OpLine %name 1 1\n"
5416 "OpLine %name 1 1\n"
5417 "%second_param1 = OpFunctionParameter %v4f32\n"
5420 "%label_secondfunction = OpLabel\n"
5422 "OpReturnValue %second_param1\n"
5427 fragments["testfun"] =
5428 // A %test_code function that returns its argument unchanged.
5431 "OpLine %name 1 1\n"
5432 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5434 "%param1 = OpFunctionParameter %v4f32\n"
5437 "%label_testfun = OpLabel\n"
5439 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5440 "OpReturnValue %val1\n"
5442 "OpLine %name 1 1\n"
5445 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
5447 return opLineTests.release();
5451 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
5453 RGBA defaultColors[4];
5454 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
5455 map<string, string> fragments;
5456 std::vector<std::pair<std::string, std::string> > problemStrings;
5458 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
5459 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
5460 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
5461 getDefaultColors(defaultColors);
5463 fragments["debug"] =
5464 "%other_name = OpString \"other_name\"\n";
5466 fragments["pre_main"] =
5467 "OpLine %file_name 32 0\n"
5468 "OpLine %file_name 32 32\n"
5469 "OpLine %file_name 32 40\n"
5470 "OpLine %other_name 32 40\n"
5471 "OpLine %other_name 0 100\n"
5472 "OpLine %other_name 0 4294967295\n"
5473 "OpLine %other_name 4294967295 0\n"
5474 "OpLine %other_name 32 40\n"
5475 "OpLine %file_name 0 0\n"
5476 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5477 "OpLine %file_name 1 0\n"
5478 "%second_param1 = OpFunctionParameter %v4f32\n"
5479 "OpLine %file_name 1 3\n"
5480 "OpLine %file_name 1 2\n"
5481 "%label_secondfunction = OpLabel\n"
5482 "OpLine %file_name 0 2\n"
5483 "OpReturnValue %second_param1\n"
5485 "OpLine %file_name 0 2\n"
5486 "OpLine %file_name 0 2\n";
5488 fragments["testfun"] =
5489 // A %test_code function that returns its argument unchanged.
5490 "OpLine %file_name 1 0\n"
5491 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5492 "OpLine %file_name 16 330\n"
5493 "%param1 = OpFunctionParameter %v4f32\n"
5494 "OpLine %file_name 14 442\n"
5495 "%label_testfun = OpLabel\n"
5496 "OpLine %file_name 11 1024\n"
5497 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5498 "OpLine %file_name 2 97\n"
5499 "OpReturnValue %val1\n"
5501 "OpLine %file_name 5 32\n";
5503 for (size_t i = 0; i < problemStrings.size(); ++i)
5505 map<string, string> testFragments = fragments;
5506 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
5507 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
5510 return opLineTests.release();
5513 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
5515 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
5519 const char functionStart[] =
5520 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5521 "%param1 = OpFunctionParameter %v4f32\n"
5524 const char functionEnd[] =
5525 "OpReturnValue %transformed_param\n"
5528 struct NameConstantsCode
5535 NameConstantsCode tests[] =
5539 "%cnull = OpConstantNull %v4f32\n",
5540 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
5544 "%cnull = OpConstantNull %f32\n",
5545 "%vp = OpVariable %fp_v4f32 Function\n"
5546 "%v = OpLoad %v4f32 %vp\n"
5547 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
5548 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
5549 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
5550 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
5551 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
5555 "%cnull = OpConstantNull %bool\n",
5556 "%v = OpVariable %fp_v4f32 Function\n"
5557 " OpStore %v %param1\n"
5558 " OpSelectionMerge %false_label None\n"
5559 " OpBranchConditional %cnull %true_label %false_label\n"
5560 "%true_label = OpLabel\n"
5561 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
5562 " OpBranch %false_label\n"
5563 "%false_label = OpLabel\n"
5564 "%transformed_param = OpLoad %v4f32 %v\n"
5568 "%cnull = OpConstantNull %i32\n",
5569 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
5570 "%b = OpIEqual %bool %cnull %c_i32_0\n"
5571 " OpSelectionMerge %false_label None\n"
5572 " OpBranchConditional %b %true_label %false_label\n"
5573 "%true_label = OpLabel\n"
5574 " OpStore %v %param1\n"
5575 " OpBranch %false_label\n"
5576 "%false_label = OpLabel\n"
5577 "%transformed_param = OpLoad %v4f32 %v\n"
5581 "%stype = OpTypeStruct %f32 %v4f32\n"
5582 "%fp_stype = OpTypePointer Function %stype\n"
5583 "%cnull = OpConstantNull %stype\n",
5584 "%v = OpVariable %fp_stype Function %cnull\n"
5585 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
5586 "%f_val = OpLoad %v4f32 %f\n"
5587 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
5591 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
5592 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
5593 "%cnull = OpConstantNull %a4_v4f32\n",
5594 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
5595 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5596 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
5597 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
5598 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
5599 "%f_val = OpLoad %v4f32 %f\n"
5600 "%f1_val = OpLoad %v4f32 %f1\n"
5601 "%f2_val = OpLoad %v4f32 %f2\n"
5602 "%f3_val = OpLoad %v4f32 %f3\n"
5603 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
5604 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
5605 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
5606 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
5610 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5611 "%cnull = OpConstantNull %mat4x4_f32\n",
5612 // Our null matrix * any vector should result in a zero vector.
5613 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
5614 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
5618 getHalfColorsFullAlpha(colors);
5620 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5622 map<string, string> fragments;
5623 fragments["pre_main"] = tests[testNdx].constants;
5624 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5625 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
5627 return opConstantNullTests.release();
5629 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
5631 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
5632 RGBA inputColors[4];
5633 RGBA outputColors[4];
5636 const char functionStart[] =
5637 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5638 "%param1 = OpFunctionParameter %v4f32\n"
5641 const char functionEnd[] =
5642 "OpReturnValue %transformed_param\n"
5645 struct NameConstantsCode
5652 NameConstantsCode tests[] =
5657 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
5658 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
5663 "%stype = OpTypeStruct %v4f32 %f32\n"
5664 "%fp_stype = OpTypePointer Function %stype\n"
5665 "%f32_n_1 = OpConstant %f32 -1.0\n"
5666 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5667 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
5668 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
5670 "%v = OpVariable %fp_stype Function %cval\n"
5671 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5672 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
5673 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
5674 "%f32_val = OpLoad %f32 %f32_ptr\n"
5675 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
5676 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
5677 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
5680 // [1|0|0|0.5] [x] = x + 0.5
5681 // [0|1|0|0.5] [y] = y + 0.5
5682 // [0|0|1|0.5] [z] = z + 0.5
5683 // [0|0|0|1 ] [1] = 1
5686 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5687 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
5688 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
5689 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
5690 "%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"
5691 "%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",
5693 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
5698 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5699 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5700 "%f32_n_1 = OpConstant %f32 -1.0\n"
5701 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5702 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
5704 "%v = OpVariable %fp_a4f32 Function %carr\n"
5705 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
5706 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
5707 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5708 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
5709 "%f_val = OpLoad %f32 %f\n"
5710 "%f1_val = OpLoad %f32 %f1\n"
5711 "%f2_val = OpLoad %f32 %f2\n"
5712 "%f3_val = OpLoad %f32 %f3\n"
5713 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
5714 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
5715 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
5716 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
5717 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5724 // [ 1.0, 1.0, 1.0, 1.0]
5728 // [ 0.0, 0.5, 0.0, 0.0]
5732 // [ 1.0, 1.0, 1.0, 1.0]
5735 "array_of_struct_of_array",
5737 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5738 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5739 "%stype = OpTypeStruct %f32 %a4f32\n"
5740 "%a3stype = OpTypeArray %stype %c_u32_3\n"
5741 "%fp_a3stype = OpTypePointer Function %a3stype\n"
5742 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
5743 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5744 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
5745 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
5746 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
5748 "%v = OpVariable %fp_a3stype Function %carr\n"
5749 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
5750 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f\n"
5751 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5755 getHalfColorsFullAlpha(inputColors);
5756 outputColors[0] = RGBA(255, 255, 255, 255);
5757 outputColors[1] = RGBA(255, 127, 127, 255);
5758 outputColors[2] = RGBA(127, 255, 127, 255);
5759 outputColors[3] = RGBA(127, 127, 255, 255);
5761 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5763 map<string, string> fragments;
5764 fragments["pre_main"] = tests[testNdx].constants;
5765 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5766 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
5768 return opConstantCompositeTests.release();
5771 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
5773 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
5774 RGBA inputColors[4];
5775 RGBA outputColors[4];
5776 map<string, string> fragments;
5778 // vec4 test_code(vec4 param) {
5779 // vec4 result = param;
5780 // for (int i = 0; i < 4; ++i) {
5781 // if (i == 0) result[i] = 0.;
5782 // else result[i] = 1. - result[i];
5786 const char function[] =
5787 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5788 "%param1 = OpFunctionParameter %v4f32\n"
5790 "%iptr = OpVariable %fp_i32 Function\n"
5791 " OpStore %iptr %c_i32_0\n"
5792 "%result = OpVariable %fp_v4f32 Function\n"
5793 " OpStore %result %param1\n"
5796 // Loop entry block.
5798 "%ival = OpLoad %i32 %iptr\n"
5799 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5800 " OpLoopMerge %exit %loop None\n"
5801 " OpBranchConditional %lt_4 %if_entry %exit\n"
5803 // Merge block for loop.
5805 "%ret = OpLoad %v4f32 %result\n"
5806 " OpReturnValue %ret\n"
5808 // If-statement entry block.
5809 "%if_entry = OpLabel\n"
5810 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5811 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
5812 " OpSelectionMerge %if_exit None\n"
5813 " OpBranchConditional %eq_0 %if_true %if_false\n"
5815 // False branch for if-statement.
5816 "%if_false = OpLabel\n"
5817 "%val = OpLoad %f32 %loc\n"
5818 "%sub = OpFSub %f32 %c_f32_1 %val\n"
5819 " OpStore %loc %sub\n"
5820 " OpBranch %if_exit\n"
5822 // Merge block for if-statement.
5823 "%if_exit = OpLabel\n"
5824 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5825 " OpStore %iptr %ival_next\n"
5828 // True branch for if-statement.
5829 "%if_true = OpLabel\n"
5830 " OpStore %loc %c_f32_0\n"
5831 " OpBranch %if_exit\n"
5835 fragments["testfun"] = function;
5837 inputColors[0] = RGBA(127, 127, 127, 0);
5838 inputColors[1] = RGBA(127, 0, 0, 0);
5839 inputColors[2] = RGBA(0, 127, 0, 0);
5840 inputColors[3] = RGBA(0, 0, 127, 0);
5842 outputColors[0] = RGBA(0, 128, 128, 255);
5843 outputColors[1] = RGBA(0, 255, 255, 255);
5844 outputColors[2] = RGBA(0, 128, 255, 255);
5845 outputColors[3] = RGBA(0, 255, 128, 255);
5847 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5849 return group.release();
5852 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
5854 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
5855 RGBA inputColors[4];
5856 RGBA outputColors[4];
5857 map<string, string> fragments;
5859 const char typesAndConstants[] =
5860 "%c_f32_p2 = OpConstant %f32 0.2\n"
5861 "%c_f32_p4 = OpConstant %f32 0.4\n"
5862 "%c_f32_p6 = OpConstant %f32 0.6\n"
5863 "%c_f32_p8 = OpConstant %f32 0.8\n";
5865 // vec4 test_code(vec4 param) {
5866 // vec4 result = param;
5867 // for (int i = 0; i < 4; ++i) {
5869 // case 0: result[i] += .2; break;
5870 // case 1: result[i] += .6; break;
5871 // case 2: result[i] += .4; break;
5872 // case 3: result[i] += .8; break;
5873 // default: break; // unreachable
5878 const char function[] =
5879 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5880 "%param1 = OpFunctionParameter %v4f32\n"
5882 "%iptr = OpVariable %fp_i32 Function\n"
5883 " OpStore %iptr %c_i32_0\n"
5884 "%result = OpVariable %fp_v4f32 Function\n"
5885 " OpStore %result %param1\n"
5888 // Loop entry block.
5890 "%ival = OpLoad %i32 %iptr\n"
5891 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5892 " OpLoopMerge %exit %loop None\n"
5893 " OpBranchConditional %lt_4 %switch_entry %exit\n"
5895 // Merge block for loop.
5897 "%ret = OpLoad %v4f32 %result\n"
5898 " OpReturnValue %ret\n"
5900 // Switch-statement entry block.
5901 "%switch_entry = OpLabel\n"
5902 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5903 "%val = OpLoad %f32 %loc\n"
5904 " OpSelectionMerge %switch_exit None\n"
5905 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5907 "%case2 = OpLabel\n"
5908 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
5909 " OpStore %loc %addp4\n"
5910 " OpBranch %switch_exit\n"
5912 "%switch_default = OpLabel\n"
5915 "%case3 = OpLabel\n"
5916 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
5917 " OpStore %loc %addp8\n"
5918 " OpBranch %switch_exit\n"
5920 "%case0 = OpLabel\n"
5921 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
5922 " OpStore %loc %addp2\n"
5923 " OpBranch %switch_exit\n"
5925 // Merge block for switch-statement.
5926 "%switch_exit = OpLabel\n"
5927 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5928 " OpStore %iptr %ival_next\n"
5931 "%case1 = OpLabel\n"
5932 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
5933 " OpStore %loc %addp6\n"
5934 " OpBranch %switch_exit\n"
5938 fragments["pre_main"] = typesAndConstants;
5939 fragments["testfun"] = function;
5941 inputColors[0] = RGBA(127, 27, 127, 51);
5942 inputColors[1] = RGBA(127, 0, 0, 51);
5943 inputColors[2] = RGBA(0, 27, 0, 51);
5944 inputColors[3] = RGBA(0, 0, 127, 51);
5946 outputColors[0] = RGBA(178, 180, 229, 255);
5947 outputColors[1] = RGBA(178, 153, 102, 255);
5948 outputColors[2] = RGBA(51, 180, 102, 255);
5949 outputColors[3] = RGBA(51, 153, 229, 255);
5951 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5953 return group.release();
5956 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
5958 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
5959 RGBA inputColors[4];
5960 RGBA outputColors[4];
5961 map<string, string> fragments;
5963 const char decorations[] =
5964 "OpDecorate %array_group ArrayStride 4\n"
5965 "OpDecorate %struct_member_group Offset 0\n"
5966 "%array_group = OpDecorationGroup\n"
5967 "%struct_member_group = OpDecorationGroup\n"
5969 "OpDecorate %group1 RelaxedPrecision\n"
5970 "OpDecorate %group3 RelaxedPrecision\n"
5971 "OpDecorate %group3 Invariant\n"
5972 "OpDecorate %group3 Restrict\n"
5973 "%group0 = OpDecorationGroup\n"
5974 "%group1 = OpDecorationGroup\n"
5975 "%group3 = OpDecorationGroup\n";
5977 const char typesAndConstants[] =
5978 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
5979 "%struct1 = OpTypeStruct %a3f32\n"
5980 "%struct2 = OpTypeStruct %a3f32\n"
5981 "%fp_struct1 = OpTypePointer Function %struct1\n"
5982 "%fp_struct2 = OpTypePointer Function %struct2\n"
5983 "%c_f32_2 = OpConstant %f32 2.\n"
5984 "%c_f32_n2 = OpConstant %f32 -2.\n"
5986 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
5987 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
5988 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
5989 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
5991 const char function[] =
5992 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5993 "%param = OpFunctionParameter %v4f32\n"
5994 "%entry = OpLabel\n"
5995 "%result = OpVariable %fp_v4f32 Function\n"
5996 " OpStore %result %param\n"
5997 "%v_struct1 = OpVariable %fp_struct1 Function\n"
5998 " OpStore %v_struct1 %c_struct1\n"
5999 "%v_struct2 = OpVariable %fp_struct2 Function\n"
6000 " OpStore %v_struct2 %c_struct2\n"
6001 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_1\n"
6002 "%val1 = OpLoad %f32 %ptr1\n"
6003 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
6004 "%val2 = OpLoad %f32 %ptr2\n"
6005 "%addvalues = OpFAdd %f32 %val1 %val2\n"
6006 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
6007 "%val = OpLoad %f32 %ptr\n"
6008 "%addresult = OpFAdd %f32 %addvalues %val\n"
6009 " OpStore %ptr %addresult\n"
6010 "%ret = OpLoad %v4f32 %result\n"
6011 " OpReturnValue %ret\n"
6014 struct CaseNameDecoration
6020 CaseNameDecoration tests[] =
6023 "same_decoration_group_on_multiple_types",
6024 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
6027 "empty_decoration_group",
6028 "OpGroupDecorate %group0 %a3f32\n"
6029 "OpGroupDecorate %group0 %result\n"
6032 "one_element_decoration_group",
6033 "OpGroupDecorate %array_group %a3f32\n"
6036 "multiple_elements_decoration_group",
6037 "OpGroupDecorate %group3 %v_struct1\n"
6040 "multiple_decoration_groups_on_same_variable",
6041 "OpGroupDecorate %group0 %v_struct2\n"
6042 "OpGroupDecorate %group1 %v_struct2\n"
6043 "OpGroupDecorate %group3 %v_struct2\n"
6046 "same_decoration_group_multiple_times",
6047 "OpGroupDecorate %group1 %addvalues\n"
6048 "OpGroupDecorate %group1 %addvalues\n"
6049 "OpGroupDecorate %group1 %addvalues\n"
6054 getHalfColorsFullAlpha(inputColors);
6055 getHalfColorsFullAlpha(outputColors);
6057 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
6059 fragments["decoration"] = decorations + tests[idx].decoration;
6060 fragments["pre_main"] = typesAndConstants;
6061 fragments["testfun"] = function;
6063 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
6066 return group.release();
6069 struct SpecConstantTwoIntGraphicsCase
6071 const char* caseName;
6072 const char* scDefinition0;
6073 const char* scDefinition1;
6074 const char* scResultType;
6075 const char* scOperation;
6076 deInt32 scActualValue0;
6077 deInt32 scActualValue1;
6078 const char* resultOperation;
6079 RGBA expectedColors[4];
6081 SpecConstantTwoIntGraphicsCase (const char* name,
6082 const char* definition0,
6083 const char* definition1,
6084 const char* resultType,
6085 const char* operation,
6088 const char* resultOp,
6089 const RGBA (&output)[4])
6091 , scDefinition0 (definition0)
6092 , scDefinition1 (definition1)
6093 , scResultType (resultType)
6094 , scOperation (operation)
6095 , scActualValue0 (value0)
6096 , scActualValue1 (value1)
6097 , resultOperation (resultOp)
6099 expectedColors[0] = output[0];
6100 expectedColors[1] = output[1];
6101 expectedColors[2] = output[2];
6102 expectedColors[3] = output[3];
6106 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
6108 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
6109 vector<SpecConstantTwoIntGraphicsCase> cases;
6110 RGBA inputColors[4];
6111 RGBA outputColors0[4];
6112 RGBA outputColors1[4];
6113 RGBA outputColors2[4];
6115 const char decorations1[] =
6116 "OpDecorate %sc_0 SpecId 0\n"
6117 "OpDecorate %sc_1 SpecId 1\n";
6119 const char typesAndConstants1[] =
6120 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
6121 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
6122 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
6124 const char function1[] =
6125 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6126 "%param = OpFunctionParameter %v4f32\n"
6127 "%label = OpLabel\n"
6128 "%result = OpVariable %fp_v4f32 Function\n"
6129 " OpStore %result %param\n"
6130 "%gen = ${GEN_RESULT}\n"
6131 "%index = OpIAdd %i32 %gen %c_i32_1\n"
6132 "%loc = OpAccessChain %fp_f32 %result %index\n"
6133 "%val = OpLoad %f32 %loc\n"
6134 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6135 " OpStore %loc %add\n"
6136 "%ret = OpLoad %v4f32 %result\n"
6137 " OpReturnValue %ret\n"
6140 inputColors[0] = RGBA(127, 127, 127, 255);
6141 inputColors[1] = RGBA(127, 0, 0, 255);
6142 inputColors[2] = RGBA(0, 127, 0, 255);
6143 inputColors[3] = RGBA(0, 0, 127, 255);
6145 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
6146 outputColors0[0] = RGBA(255, 127, 127, 255);
6147 outputColors0[1] = RGBA(255, 0, 0, 255);
6148 outputColors0[2] = RGBA(128, 127, 0, 255);
6149 outputColors0[3] = RGBA(128, 0, 127, 255);
6151 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
6152 outputColors1[0] = RGBA(127, 255, 127, 255);
6153 outputColors1[1] = RGBA(127, 128, 0, 255);
6154 outputColors1[2] = RGBA(0, 255, 0, 255);
6155 outputColors1[3] = RGBA(0, 128, 127, 255);
6157 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
6158 outputColors2[0] = RGBA(127, 127, 255, 255);
6159 outputColors2[1] = RGBA(127, 0, 128, 255);
6160 outputColors2[2] = RGBA(0, 127, 128, 255);
6161 outputColors2[3] = RGBA(0, 0, 255, 255);
6163 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
6164 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
6165 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
6167 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
6168 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
6169 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
6170 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
6171 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
6172 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6173 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6174 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
6175 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
6176 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
6177 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
6178 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
6179 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
6180 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
6181 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
6182 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
6183 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6184 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
6185 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
6186 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
6187 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6188 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
6189 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
6190 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6191 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6192 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6193 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6194 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
6195 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
6196 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
6197 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
6198 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
6199 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
6201 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
6203 map<string, string> specializations;
6204 map<string, string> fragments;
6205 vector<deInt32> specConstants;
6207 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
6208 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
6209 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
6210 specializations["SC_OP"] = cases[caseNdx].scOperation;
6211 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
6213 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
6214 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
6215 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
6217 specConstants.push_back(cases[caseNdx].scActualValue0);
6218 specConstants.push_back(cases[caseNdx].scActualValue1);
6220 createTestsForAllStages(cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants, group.get());
6223 const char decorations2[] =
6224 "OpDecorate %sc_0 SpecId 0\n"
6225 "OpDecorate %sc_1 SpecId 1\n"
6226 "OpDecorate %sc_2 SpecId 2\n";
6228 const char typesAndConstants2[] =
6229 "%v3i32 = OpTypeVector %i32 3\n"
6231 "%sc_0 = OpSpecConstant %i32 0\n"
6232 "%sc_1 = OpSpecConstant %i32 0\n"
6233 "%sc_2 = OpSpecConstant %i32 0\n"
6235 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
6236 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
6237 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
6238 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
6239 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %sc_vec3_1 1 0 4\n" // (0, sc_0, sc_1)
6240 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2 5 1 2\n" // (sc_2, sc_0, sc_1)
6241 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
6242 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
6243 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
6244 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
6245 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
6247 const char function2[] =
6248 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6249 "%param = OpFunctionParameter %v4f32\n"
6250 "%label = OpLabel\n"
6251 "%result = OpVariable %fp_v4f32 Function\n"
6252 " OpStore %result %param\n"
6253 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
6254 "%val = OpLoad %f32 %loc\n"
6255 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6256 " OpStore %loc %add\n"
6257 "%ret = OpLoad %v4f32 %result\n"
6258 " OpReturnValue %ret\n"
6261 map<string, string> fragments;
6262 vector<deInt32> specConstants;
6264 fragments["decoration"] = decorations2;
6265 fragments["pre_main"] = typesAndConstants2;
6266 fragments["testfun"] = function2;
6268 specConstants.push_back(56789);
6269 specConstants.push_back(-2);
6270 specConstants.push_back(56788);
6272 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
6274 return group.release();
6277 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
6279 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
6280 RGBA inputColors[4];
6281 RGBA outputColors1[4];
6282 RGBA outputColors2[4];
6283 RGBA outputColors3[4];
6284 map<string, string> fragments1;
6285 map<string, string> fragments2;
6286 map<string, string> fragments3;
6288 const char typesAndConstants1[] =
6289 "%c_f32_p2 = OpConstant %f32 0.2\n"
6290 "%c_f32_p4 = OpConstant %f32 0.4\n"
6291 "%c_f32_p6 = OpConstant %f32 0.6\n"
6292 "%c_f32_p8 = OpConstant %f32 0.8\n";
6294 // vec4 test_code(vec4 param) {
6295 // vec4 result = param;
6296 // for (int i = 0; i < 4; ++i) {
6299 // case 0: operand = .2; break;
6300 // case 1: operand = .6; break;
6301 // case 2: operand = .4; break;
6302 // case 3: operand = .0; break;
6303 // default: break; // unreachable
6305 // result[i] += operand;
6309 const char function1[] =
6310 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6311 "%param1 = OpFunctionParameter %v4f32\n"
6313 "%iptr = OpVariable %fp_i32 Function\n"
6314 " OpStore %iptr %c_i32_0\n"
6315 "%result = OpVariable %fp_v4f32 Function\n"
6316 " OpStore %result %param1\n"
6320 "%ival = OpLoad %i32 %iptr\n"
6321 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6322 " OpLoopMerge %exit %loop None\n"
6323 " OpBranchConditional %lt_4 %entry %exit\n"
6325 "%entry = OpLabel\n"
6326 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6327 "%val = OpLoad %f32 %loc\n"
6328 " OpSelectionMerge %phi None\n"
6329 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6331 "%case0 = OpLabel\n"
6333 "%case1 = OpLabel\n"
6335 "%case2 = OpLabel\n"
6337 "%case3 = OpLabel\n"
6340 "%default = OpLabel\n"
6344 "%operand = OpPhi %f32 %c_f32_p4 %case2 %c_f32_p6 %case1 %c_f32_p2 %case0 %c_f32_0 %case3\n" // not in the order of blocks
6345 "%add = OpFAdd %f32 %val %operand\n"
6346 " OpStore %loc %add\n"
6347 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6348 " OpStore %iptr %ival_next\n"
6352 "%ret = OpLoad %v4f32 %result\n"
6353 " OpReturnValue %ret\n"
6357 fragments1["pre_main"] = typesAndConstants1;
6358 fragments1["testfun"] = function1;
6360 getHalfColorsFullAlpha(inputColors);
6362 outputColors1[0] = RGBA(178, 180, 229, 255);
6363 outputColors1[1] = RGBA(178, 153, 102, 255);
6364 outputColors1[2] = RGBA(51, 180, 102, 255);
6365 outputColors1[3] = RGBA(51, 153, 229, 255);
6367 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
6369 const char typesAndConstants2[] =
6370 "%c_f32_p2 = OpConstant %f32 0.2\n";
6372 // Add .4 to the second element of the given parameter.
6373 const char function2[] =
6374 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6375 "%param = OpFunctionParameter %v4f32\n"
6376 "%entry = OpLabel\n"
6377 "%result = OpVariable %fp_v4f32 Function\n"
6378 " OpStore %result %param\n"
6379 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6380 "%val = OpLoad %f32 %loc\n"
6384 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
6385 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
6386 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
6387 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
6388 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
6389 " OpLoopMerge %exit %phi None\n"
6390 " OpBranchConditional %still_loop %phi %exit\n"
6393 " OpStore %loc %accum\n"
6394 "%ret = OpLoad %v4f32 %result\n"
6395 " OpReturnValue %ret\n"
6399 fragments2["pre_main"] = typesAndConstants2;
6400 fragments2["testfun"] = function2;
6402 outputColors2[0] = RGBA(127, 229, 127, 255);
6403 outputColors2[1] = RGBA(127, 102, 0, 255);
6404 outputColors2[2] = RGBA(0, 229, 0, 255);
6405 outputColors2[3] = RGBA(0, 102, 127, 255);
6407 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
6409 const char typesAndConstants3[] =
6410 "%true = OpConstantTrue %bool\n"
6411 "%false = OpConstantFalse %bool\n"
6412 "%c_f32_p2 = OpConstant %f32 0.2\n";
6414 // Swap the second and the third element of the given parameter.
6415 const char function3[] =
6416 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6417 "%param = OpFunctionParameter %v4f32\n"
6418 "%entry = OpLabel\n"
6419 "%result = OpVariable %fp_v4f32 Function\n"
6420 " OpStore %result %param\n"
6421 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6422 "%a_init = OpLoad %f32 %a_loc\n"
6423 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
6424 "%b_init = OpLoad %f32 %b_loc\n"
6428 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
6429 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
6430 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
6431 " OpLoopMerge %exit %phi None\n"
6432 " OpBranchConditional %still_loop %phi %exit\n"
6435 " OpStore %a_loc %a_next\n"
6436 " OpStore %b_loc %b_next\n"
6437 "%ret = OpLoad %v4f32 %result\n"
6438 " OpReturnValue %ret\n"
6442 fragments3["pre_main"] = typesAndConstants3;
6443 fragments3["testfun"] = function3;
6445 outputColors3[0] = RGBA(127, 127, 127, 255);
6446 outputColors3[1] = RGBA(127, 0, 0, 255);
6447 outputColors3[2] = RGBA(0, 0, 127, 255);
6448 outputColors3[3] = RGBA(0, 127, 0, 255);
6450 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
6452 return group.release();
6455 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
6457 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
6458 RGBA inputColors[4];
6459 RGBA outputColors[4];
6461 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
6462 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
6463 // only have 23-bit fraction.) So it will be rounded to 1. Then the final result is 0. On the contrary, the result will
6464 // be 2^-46, which is a normalized number perfectly representable as 32-bit float.
6465 const char constantsAndTypes[] =
6466 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
6467 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
6468 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
6469 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
6472 const char function[] =
6473 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6474 "%param = OpFunctionParameter %v4f32\n"
6475 "%label = OpLabel\n"
6476 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
6477 "%var2 = OpVariable %fp_f32 Function\n"
6478 "%red = OpCompositeExtract %f32 %param 0\n"
6479 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
6480 " OpStore %var2 %plus_red\n"
6481 "%val1 = OpLoad %f32 %var1\n"
6482 "%val2 = OpLoad %f32 %var2\n"
6483 "%mul = OpFMul %f32 %val1 %val2\n"
6484 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
6485 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
6486 "%ret = OpSelect %v4f32 %is0 %c_vec4_0 %c_vec4_1\n"
6487 " OpReturnValue %ret\n"
6490 struct CaseNameDecoration
6497 CaseNameDecoration tests[] = {
6498 {"multiplication", "OpDecorate %mul NoContraction"},
6499 {"addition", "OpDecorate %add NoContraction"},
6500 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
6503 getHalfColorsFullAlpha(inputColors);
6505 for (deUint8 idx = 0; idx < 4; ++idx)
6507 inputColors[idx].setRed(0);
6508 outputColors[idx] = RGBA(0, 0, 0, 255);
6511 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
6513 map<string, string> fragments;
6515 fragments["decoration"] = tests[testNdx].decoration;
6516 fragments["pre_main"] = constantsAndTypes;
6517 fragments["testfun"] = function;
6519 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
6522 return group.release();
6525 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
6527 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
6530 const char constantsAndTypes[] =
6531 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
6532 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
6533 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
6534 "%fp_stype = OpTypePointer Function %stype\n";
6536 const char function[] =
6537 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6538 "%param1 = OpFunctionParameter %v4f32\n"
6540 "%v1 = OpVariable %fp_v4f32 Function\n"
6541 " OpStore %v1 %c_v4f32_1_1_1_1\n"
6542 "%v2 = OpVariable %fp_a2f32 Function\n"
6543 " OpStore %v2 %c_a2f32_1\n"
6544 "%v3 = OpVariable %fp_f32 Function\n"
6545 " OpStore %v3 %c_f32_1\n"
6547 "%v = OpVariable %fp_stype Function\n"
6548 "%vv = OpVariable %fp_stype Function\n"
6549 "%vvv = OpVariable %fp_f32 Function\n"
6551 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
6552 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
6553 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
6554 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
6555 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
6556 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
6558 " OpStore %p_v4f32 %v1_v ${access_type}\n"
6559 " OpStore %p_a2f32 %v2_v ${access_type}\n"
6560 " OpStore %p_f32 %v3_v ${access_type}\n"
6562 " OpCopyMemory %vv %v ${access_type}\n"
6563 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
6565 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
6566 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
6567 "%v_f32_3 = OpLoad %f32 %vvv\n"
6569 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
6570 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
6571 " OpReturnValue %ret2\n"
6574 struct NameMemoryAccess
6581 NameMemoryAccess tests[] =
6584 { "volatile", "Volatile" },
6585 { "aligned", "Aligned 1" },
6586 { "volatile_aligned", "Volatile|Aligned 1" },
6587 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
6588 { "volatile_nontemporal", "Volatile|Nontemporal" },
6589 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
6592 getHalfColorsFullAlpha(colors);
6594 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
6596 map<string, string> fragments;
6597 map<string, string> memoryAccess;
6598 memoryAccess["access_type"] = tests[testNdx].accessType;
6600 fragments["pre_main"] = constantsAndTypes;
6601 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
6602 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
6604 return memoryAccessTests.release();
6606 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
6608 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
6609 RGBA defaultColors[4];
6610 map<string, string> fragments;
6611 getDefaultColors(defaultColors);
6613 // First, simple cases that don't do anything with the OpUndef result.
6614 fragments["testfun"] =
6615 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6616 "%param1 = OpFunctionParameter %v4f32\n"
6617 "%label_testfun = OpLabel\n"
6618 "%undef = OpUndef %type\n"
6619 "OpReturnValue %param1\n"
6622 struct NameCodePair { string name, code; };
6623 const NameCodePair tests[] =
6625 {"bool", "%type = OpTypeBool"},
6626 {"vec2uint32", "%type = OpTypeVector %u32 2"},
6627 {"image", "%type = OpTypeImage %f32 2D 0 0 0 0 Unknown"},
6628 {"sampler", "%type = OpTypeSampler"},
6629 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 0 Unknown\n" "%type = OpTypeSampledImage %img"},
6630 {"pointer", "%type = OpTypePointer Function %i32"},
6631 {"runtimearray", "%type = OpTypeRuntimeArray %f32"},
6632 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100"},
6633 {"struct", "%type = OpTypeStruct %f32 %i32 %u32"}};
6634 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
6636 fragments["pre_main"] = tests[testNdx].code;
6637 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
6641 fragments["testfun"] =
6642 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6643 "%param1 = OpFunctionParameter %v4f32\n"
6644 "%label_testfun = OpLabel\n"
6645 "%undef = OpUndef %f32\n"
6646 "%zero = OpFMul %f32 %undef %c_f32_0\n"
6647 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6648 "%b = OpFAdd %f32 %a %zero\n"
6649 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
6650 "OpReturnValue %ret\n"
6653 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6655 fragments["testfun"] =
6656 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6657 "%param1 = OpFunctionParameter %v4f32\n"
6658 "%label_testfun = OpLabel\n"
6659 "%undef = OpUndef %i32\n"
6660 "%zero = OpIMul %i32 %undef %c_i32_0\n"
6661 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6662 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6663 "OpReturnValue %ret\n"
6666 createTestsForAllStages("sint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6668 fragments["testfun"] =
6669 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6670 "%param1 = OpFunctionParameter %v4f32\n"
6671 "%label_testfun = OpLabel\n"
6672 "%undef = OpUndef %u32\n"
6673 "%zero = OpIMul %u32 %undef %c_i32_0\n"
6674 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6675 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6676 "OpReturnValue %ret\n"
6679 createTestsForAllStages("uint32", defaultColors, defaultColors, fragments, opUndefTests.get());
6681 fragments["testfun"] =
6682 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6683 "%param1 = OpFunctionParameter %v4f32\n"
6684 "%label_testfun = OpLabel\n"
6685 "%undef = OpUndef %v4f32\n"
6686 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
6687 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
6688 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
6689 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
6690 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
6691 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6692 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6693 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6694 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6695 "%sum_0 = OpFAdd %f32 %param1_0 %zero_0\n"
6696 "%sum_1 = OpFAdd %f32 %param1_1 %zero_1\n"
6697 "%sum_2 = OpFAdd %f32 %param1_2 %zero_2\n"
6698 "%sum_3 = OpFAdd %f32 %param1_3 %zero_3\n"
6699 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6700 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6701 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6702 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6703 "OpReturnValue %ret\n"
6706 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6708 fragments["pre_main"] =
6709 "%v2f32 = OpTypeVector %f32 2\n"
6710 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
6711 fragments["testfun"] =
6712 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6713 "%param1 = OpFunctionParameter %v4f32\n"
6714 "%label_testfun = OpLabel\n"
6715 "%undef = OpUndef %m2x2f32\n"
6716 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
6717 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
6718 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
6719 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
6720 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
6721 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6722 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6723 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6724 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6725 "%sum_0 = OpFAdd %f32 %param1_0 %zero_0\n"
6726 "%sum_1 = OpFAdd %f32 %param1_1 %zero_1\n"
6727 "%sum_2 = OpFAdd %f32 %param1_2 %zero_2\n"
6728 "%sum_3 = OpFAdd %f32 %param1_3 %zero_3\n"
6729 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6730 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6731 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6732 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6733 "OpReturnValue %ret\n"
6736 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
6738 return opUndefTests.release();
6741 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
6743 const RGBA inputColors[4] =
6746 RGBA(0, 0, 255, 255),
6747 RGBA(0, 255, 0, 255),
6748 RGBA(0, 255, 255, 255)
6751 const RGBA expectedColors[4] =
6753 RGBA(255, 0, 0, 255),
6754 RGBA(255, 0, 0, 255),
6755 RGBA(255, 0, 0, 255),
6756 RGBA(255, 0, 0, 255)
6759 const struct SingleFP16Possibility
6762 const char* constant; // Value to assign to %test_constant.
6764 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
6770 -constructNormalizedFloat(1, 0x300000),
6771 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6776 constructNormalizedFloat(7, 0x000000),
6777 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6779 // SPIR-V requires that OpQuantizeToF16 flushes
6780 // any numbers that would end up denormalized in F16 to zero.
6784 std::ldexp(1.5f, -140),
6785 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6790 -std::ldexp(1.5f, -140),
6791 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6796 std::ldexp(1.0f, -16),
6797 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6798 }, // too small positive
6800 "negative_too_small",
6802 -std::ldexp(1.0f, -32),
6803 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6804 }, // too small negative
6808 -std::ldexp(1.0f, 128),
6810 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6811 "%inf = OpIsInf %bool %c\n"
6812 "%cond = OpLogicalAnd %bool %gz %inf\n"
6817 std::ldexp(1.0f, 128),
6819 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6820 "%inf = OpIsInf %bool %c\n"
6821 "%cond = OpLogicalAnd %bool %gz %inf\n"
6824 "round_to_negative_inf",
6826 -std::ldexp(1.0f, 32),
6828 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6829 "%inf = OpIsInf %bool %c\n"
6830 "%cond = OpLogicalAnd %bool %gz %inf\n"
6835 std::ldexp(1.0f, 16),
6837 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6838 "%inf = OpIsInf %bool %c\n"
6839 "%cond = OpLogicalAnd %bool %gz %inf\n"
6844 std::numeric_limits<float>::quiet_NaN(),
6846 // Can't use %c, because NaN+0 isn't necessarily a NaN (Vulkan spec A.4).
6847 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6848 "%nan = OpIsNan %bool %direct_quant\n"
6849 "%as_int = OpBitcast %i32 %direct_quant\n"
6850 "%positive = OpSGreaterThan %bool %as_int %c_i32_0\n"
6851 "%cond = OpLogicalAnd %bool %nan %positive\n"
6856 std::numeric_limits<float>::quiet_NaN(),
6858 // Can't use %c, because NaN+0 isn't necessarily a NaN (Vulkan spec A.4).
6859 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6860 "%nan = OpIsNan %bool %direct_quant\n"
6861 "%as_int = OpBitcast %i32 %direct_quant\n"
6862 "%negative = OpSLessThan %bool %as_int %c_i32_0\n"
6863 "%cond = OpLogicalAnd %bool %nan %negative\n"
6866 const char* constants =
6867 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
6869 StringTemplate function (
6870 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6871 "%param1 = OpFunctionParameter %v4f32\n"
6872 "%label_testfun = OpLabel\n"
6873 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6874 "%b = OpFAdd %f32 %test_constant %a\n"
6875 "%c = OpQuantizeToF16 %f32 %b\n"
6877 "%retval = OpSelect %v4f32 %cond %c_v4f32_1_0_0_1 %param1"
6878 " OpReturnValue %retval\n"
6882 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
6883 const char* specConstants =
6884 "%test_constant = OpSpecConstant %f32 0.\n"
6885 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
6887 StringTemplate specConstantFunction(
6888 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6889 "%param1 = OpFunctionParameter %v4f32\n"
6890 "%label_testfun = OpLabel\n"
6892 "%retval = OpSelect %v4f32 %cond %c_v4f32_1_0_0_1 %param1"
6893 " OpReturnValue %retval\n"
6897 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6899 map<string, string> codeSpecialization;
6900 map<string, string> fragments;
6901 codeSpecialization["condition"] = tests[idx].condition;
6902 fragments["testfun"] = function.specialize(codeSpecialization);
6903 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
6904 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6907 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6909 map<string, string> codeSpecialization;
6910 map<string, string> fragments;
6911 vector<deInt32> passConstants;
6912 deInt32 specConstant;
6914 codeSpecialization["condition"] = tests[idx].condition;
6915 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
6916 fragments["decoration"] = specDecorations;
6917 fragments["pre_main"] = specConstants;
6919 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
6920 passConstants.push_back(specConstant);
6922 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6926 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
6928 RGBA inputColors[4] = {
6930 RGBA(0, 0, 255, 255),
6931 RGBA(0, 255, 0, 255),
6932 RGBA(0, 255, 255, 255)
6935 RGBA expectedColors[4] =
6937 RGBA(255, 0, 0, 255),
6938 RGBA(255, 0, 0, 255),
6939 RGBA(255, 0, 0, 255),
6940 RGBA(255, 0, 0, 255)
6943 struct DualFP16Possibility
6948 const char* possibleOutput1;
6949 const char* possibleOutput2;
6952 "positive_round_up_or_round_down",
6954 constructNormalizedFloat(8, 0x300300),
6959 "negative_round_up_or_round_down",
6961 -constructNormalizedFloat(-7, 0x600800),
6968 constructNormalizedFloat(2, 0x01e000),
6973 "carry_to_exponent",
6975 constructNormalizedFloat(1, 0xffe000),
6980 StringTemplate constants (
6981 "%input_const = OpConstant %f32 ${input}\n"
6982 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6983 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6986 StringTemplate specConstants (
6987 "%input_const = OpSpecConstant %f32 0.\n"
6988 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6989 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6992 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
6994 const char* function =
6995 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6996 "%param1 = OpFunctionParameter %v4f32\n"
6997 "%label_testfun = OpLabel\n"
6998 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6999 // For the purposes of this test we assume that 0.f will always get
7000 // faithfully passed through the pipeline stages.
7001 "%b = OpFAdd %f32 %input_const %a\n"
7002 "%c = OpQuantizeToF16 %f32 %b\n"
7003 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
7004 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
7005 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
7006 "%retval = OpSelect %v4f32 %cond %c_v4f32_1_0_0_1 %param1"
7007 " OpReturnValue %retval\n"
7010 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7011 map<string, string> fragments;
7012 map<string, string> constantSpecialization;
7014 constantSpecialization["input"] = tests[idx].input;
7015 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7016 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7017 fragments["testfun"] = function;
7018 fragments["pre_main"] = constants.specialize(constantSpecialization);
7019 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7022 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7023 map<string, string> fragments;
7024 map<string, string> constantSpecialization;
7025 vector<deInt32> passConstants;
7026 deInt32 specConstant;
7028 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7029 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7030 fragments["testfun"] = function;
7031 fragments["decoration"] = specDecorations;
7032 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
7034 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
7035 passConstants.push_back(specConstant);
7037 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7041 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
7043 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
7044 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
7045 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
7046 return opQuantizeTests.release();
7049 struct ShaderPermutation
7051 deUint8 vertexPermutation;
7052 deUint8 geometryPermutation;
7053 deUint8 tesscPermutation;
7054 deUint8 tessePermutation;
7055 deUint8 fragmentPermutation;
7058 ShaderPermutation getShaderPermutation(deUint8 inputValue)
7060 ShaderPermutation permutation =
7062 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
7063 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
7064 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
7065 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
7066 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
7071 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
7073 RGBA defaultColors[4];
7074 RGBA invertedColors[4];
7075 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
7077 const ShaderElement combinedPipeline[] =
7079 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
7080 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
7081 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7082 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7083 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
7086 getDefaultColors(defaultColors);
7087 getInvertedDefaultColors(invertedColors);
7088 addFunctionCaseWithPrograms<InstanceContext>(moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline, createInstanceContext(combinedPipeline, map<string, string>()));
7090 const char* numbers[] =
7095 for (deInt8 idx = 0; idx < 32; ++idx)
7097 ShaderPermutation permutation = getShaderPermutation(idx);
7098 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
7099 const ShaderElement pipeline[] =
7101 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
7102 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
7103 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7104 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7105 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
7108 // If there are an even number of swaps, then it should be no-op.
7109 // If there are an odd number, the color should be flipped.
7110 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
7112 addFunctionCaseWithPrograms<InstanceContext>(moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline, createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
7116 addFunctionCaseWithPrograms<InstanceContext>(moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline, createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
7119 return moduleTests.release();
7122 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
7124 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
7125 RGBA defaultColors[4];
7126 getDefaultColors(defaultColors);
7127 map<string, string> fragments;
7128 fragments["pre_main"] =
7129 "%c_f32_5 = OpConstant %f32 5.\n";
7131 // A loop with a single block. The Continue Target is the loop block
7132 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
7133 // -- the "continue construct" forms the entire loop.
7134 fragments["testfun"] =
7135 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7136 "%param1 = OpFunctionParameter %v4f32\n"
7138 "%entry = OpLabel\n"
7139 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7142 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7144 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7145 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
7146 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7147 "%val = OpFAdd %f32 %val1 %delta\n"
7148 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
7149 "%count__ = OpISub %i32 %count %c_i32_1\n"
7150 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7151 "OpLoopMerge %exit %loop None\n"
7152 "OpBranchConditional %again %loop %exit\n"
7155 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7156 "OpReturnValue %result\n"
7160 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
7162 // Body comprised of multiple basic blocks.
7163 const StringTemplate multiBlock(
7164 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7165 "%param1 = OpFunctionParameter %v4f32\n"
7167 "%entry = OpLabel\n"
7168 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7171 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7173 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
7174 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
7175 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
7176 // There are several possibilities for the Continue Target below. Each
7177 // will be specialized into a separate test case.
7178 "OpLoopMerge %exit ${continue_target} None\n"
7182 ";delta_next = (delta > 0) ? -1 : 1;\n"
7183 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
7184 "OpSelectionMerge %gather DontFlatten\n"
7185 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
7188 "OpBranch %gather\n"
7191 "OpBranch %gather\n"
7193 "%gather = OpLabel\n"
7194 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
7195 "%val = OpFAdd %f32 %val1 %delta\n"
7196 "%count__ = OpISub %i32 %count %c_i32_1\n"
7197 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7198 "OpBranchConditional %again %loop %exit\n"
7201 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7202 "OpReturnValue %result\n"
7206 map<string, string> continue_target;
7208 // The Continue Target is the loop block itself.
7209 continue_target["continue_target"] = "%loop";
7210 fragments["testfun"] = multiBlock.specialize(continue_target);
7211 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
7213 // The Continue Target is at the end of the loop.
7214 continue_target["continue_target"] = "%gather";
7215 fragments["testfun"] = multiBlock.specialize(continue_target);
7216 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
7218 // A loop with continue statement.
7219 fragments["testfun"] =
7220 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7221 "%param1 = OpFunctionParameter %v4f32\n"
7223 "%entry = OpLabel\n"
7224 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7227 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
7229 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7230 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
7231 "OpLoopMerge %exit %continue None\n"
7235 ";skip if %count==2\n"
7236 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
7237 "OpSelectionMerge %continue DontFlatten\n"
7238 "OpBranchConditional %eq2 %continue %body\n"
7241 "%fcount = OpConvertSToF %f32 %count\n"
7242 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7243 "OpBranch %continue\n"
7245 "%continue = OpLabel\n"
7246 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
7247 "%count__ = OpISub %i32 %count %c_i32_1\n"
7248 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7249 "OpBranchConditional %again %loop %exit\n"
7252 "%same = OpFSub %f32 %val %c_f32_8\n"
7253 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7254 "OpReturnValue %result\n"
7256 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
7258 // A loop with early exit. May be specialized with either break or return.
7259 StringTemplate earlyExitLoop(
7260 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7261 "%param1 = OpFunctionParameter %v4f32\n"
7263 "%entry = OpLabel\n"
7264 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7265 "%dot = OpDot %f32 %param1 %param1\n"
7266 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7267 "%zero = OpConvertFToU %u32 %div\n"
7268 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7269 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7272 ";adds 4 and 3 to %val0 (exits early)\n"
7274 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7275 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
7276 "OpLoopMerge %exit %continue None\n"
7280 ";end loop if %count==%two\n"
7281 "%above2 = OpSGreaterThan %bool %count %two\n"
7282 "OpSelectionMerge %continue DontFlatten\n"
7283 // This can either branch to %exit or to another block with OpReturnValue %param1.
7284 "OpBranchConditional %above2 %body ${branch_destination}\n"
7287 "%fcount = OpConvertSToF %f32 %count\n"
7288 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7289 "OpBranch %continue\n"
7291 "%continue = OpLabel\n"
7292 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
7293 "%count__ = OpISub %i32 %count %c_i32_1\n"
7294 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7295 "OpBranchConditional %again %loop %exit\n"
7298 "%same = OpFSub %f32 %val %c_f32_7\n"
7299 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7300 "OpReturnValue %result\n"
7303 map<string, string> branch_destination;
7305 // A loop with break.
7306 branch_destination["branch_destination"] = "%exit";
7307 fragments["testfun"] = earlyExitLoop.specialize(branch_destination);
7308 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
7310 // A loop with return.
7311 branch_destination["branch_destination"] = "%early_exit\n"
7312 "%early_exit = OpLabel\n"
7313 "OpReturnValue %param1\n";
7314 fragments["testfun"] = earlyExitLoop.specialize(branch_destination);
7315 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
7317 return testGroup.release();
7320 // Adds a new test to group using custom fragments for the tessellation-control
7321 // stage and passthrough fragments for all other stages. Uses default colors
7322 // for input and expected output.
7323 void addTessCtrlTest(tcu::TestCaseGroup* group, const char* name, const map<string, string>& fragments)
7325 RGBA defaultColors[4];
7326 getDefaultColors(defaultColors);
7327 const ShaderElement pipelineStages[] =
7329 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
7330 ShaderElement("tessc", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7331 ShaderElement("tesse", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7332 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
7335 addFunctionCaseWithPrograms<InstanceContext>(group, name, "", addShaderCodeCustomTessControl,
7336 runAndVerifyDefaultPipeline, createInstanceContext(
7337 pipelineStages, defaultColors, defaultColors, fragments, StageToSpecConstantMap()));
7340 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
7341 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
7343 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
7344 map<string, string> fragments;
7346 // A barrier inside a function body.
7347 fragments["pre_main"] =
7348 "%Workgroup = OpConstant %i32 2\n"
7349 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
7350 fragments["testfun"] =
7351 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7352 "%param1 = OpFunctionParameter %v4f32\n"
7353 "%label_testfun = OpLabel\n"
7354 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7355 "OpReturnValue %param1\n"
7357 addTessCtrlTest(testGroup.get(), "in_function", fragments);
7359 // Common setup code for the following tests.
7360 fragments["pre_main"] =
7361 "%Workgroup = OpConstant %i32 2\n"
7362 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7363 "%c_f32_5 = OpConstant %f32 5.\n";
7364 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
7365 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7366 "%param1 = OpFunctionParameter %v4f32\n"
7367 "%entry = OpLabel\n"
7368 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7369 "%dot = OpDot %f32 %param1 %param1\n"
7370 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7371 "%zero = OpConvertFToU %u32 %div\n";
7373 // Barriers inside OpSwitch branches.
7374 fragments["testfun"] =
7376 "OpSelectionMerge %switch_exit None\n"
7377 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
7379 "%case1 = OpLabel\n"
7380 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7381 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7382 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7383 "OpBranch %switch_exit\n"
7385 "%switch_default = OpLabel\n"
7386 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7387 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7388 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7389 "OpBranch %switch_exit\n"
7391 "%case0 = OpLabel\n"
7392 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7393 "OpBranch %switch_exit\n"
7395 "%switch_exit = OpLabel\n"
7396 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
7397 "OpReturnValue %ret\n"
7399 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
7401 // Barriers inside if-then-else.
7402 fragments["testfun"] =
7404 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
7405 "OpSelectionMerge %exit DontFlatten\n"
7406 "OpBranchConditional %eq0 %then %else\n"
7409 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7410 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7411 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7415 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7419 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
7420 "OpReturnValue %ret\n"
7422 addTessCtrlTest(testGroup.get(), "in_if", fragments);
7424 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
7425 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
7426 fragments["testfun"] =
7428 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
7429 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
7430 "OpSelectionMerge %exit DontFlatten\n"
7431 "OpBranchConditional %thread0 %then %else\n"
7434 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7438 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
7442 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
7443 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7444 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
7445 "OpReturnValue %ret\n"
7447 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
7449 // A barrier inside a loop.
7450 fragments["pre_main"] =
7451 "%Workgroup = OpConstant %i32 2\n"
7452 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7453 "%c_f32_10 = OpConstant %f32 10.\n";
7454 fragments["testfun"] =
7455 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7456 "%param1 = OpFunctionParameter %v4f32\n"
7457 "%entry = OpLabel\n"
7458 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7461 ";adds 4, 3, 2, and 1 to %val0\n"
7463 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7464 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7465 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7466 "%fcount = OpConvertSToF %f32 %count\n"
7467 "%val = OpFAdd %f32 %val1 %fcount\n"
7468 "%count__ = OpISub %i32 %count %c_i32_1\n"
7469 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7470 "OpLoopMerge %exit %loop None\n"
7471 "OpBranchConditional %again %loop %exit\n"
7474 "%same = OpFSub %f32 %val %c_f32_10\n"
7475 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7476 "OpReturnValue %ret\n"
7478 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
7480 return testGroup.release();
7483 // Test for the OpFRem instruction.
7484 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
7486 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
7487 map<string, string> fragments;
7488 RGBA inputColors[4];
7489 RGBA outputColors[4];
7491 fragments["pre_main"] =
7492 "%c_f32_3 = OpConstant %f32 3.0\n"
7493 "%c_f32_n3 = OpConstant %f32 -3.0\n"
7494 "%c_f32_4 = OpConstant %f32 4.0\n"
7495 "%c_f32_p75 = OpConstant %f32 0.75\n"
7496 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
7497 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
7498 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
7500 // The test does the following.
7501 // vec4 result = (param1 * 8.0) - 4.0;
7502 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
7503 fragments["testfun"] =
7504 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7505 "%param1 = OpFunctionParameter %v4f32\n"
7506 "%label_testfun = OpLabel\n"
7507 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
7508 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
7509 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
7510 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
7511 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
7512 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
7513 "OpReturnValue %xy_0_1\n"
7517 inputColors[0] = RGBA(16, 16, 0, 255);
7518 inputColors[1] = RGBA(232, 232, 0, 255);
7519 inputColors[2] = RGBA(232, 16, 0, 255);
7520 inputColors[3] = RGBA(16, 232, 0, 255);
7522 outputColors[0] = RGBA(64, 64, 0, 255);
7523 outputColors[1] = RGBA(255, 255, 0, 255);
7524 outputColors[2] = RGBA(255, 64, 0, 255);
7525 outputColors[3] = RGBA(64, 255, 0, 255);
7527 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
7528 return testGroup.release();
7531 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
7533 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
7534 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
7535 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
7537 computeTests->addChild(createOpNopGroup(testCtx));
7538 computeTests->addChild(createOpLineGroup(testCtx));
7539 computeTests->addChild(createOpNoLineGroup(testCtx));
7540 computeTests->addChild(createOpConstantNullGroup(testCtx));
7541 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
7542 computeTests->addChild(createOpConstantUsageGroup(testCtx));
7543 computeTests->addChild(createSpecConstantGroup(testCtx));
7544 computeTests->addChild(createOpSourceGroup(testCtx));
7545 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
7546 computeTests->addChild(createDecorationGroupGroup(testCtx));
7547 computeTests->addChild(createOpPhiGroup(testCtx));
7548 computeTests->addChild(createLoopControlGroup(testCtx));
7549 computeTests->addChild(createFunctionControlGroup(testCtx));
7550 computeTests->addChild(createSelectionControlGroup(testCtx));
7551 computeTests->addChild(createBlockOrderGroup(testCtx));
7552 computeTests->addChild(createMultipleShaderGroup(testCtx));
7553 computeTests->addChild(createMemoryAccessGroup(testCtx));
7554 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
7555 computeTests->addChild(createOpCopyObjectGroup(testCtx));
7556 computeTests->addChild(createNoContractionGroup(testCtx));
7557 computeTests->addChild(createOpUndefGroup(testCtx));
7558 computeTests->addChild(createOpUnreachableGroup(testCtx));
7559 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
7560 computeTests ->addChild(createOpFRemGroup(testCtx));
7562 RGBA defaultColors[4];
7563 getDefaultColors(defaultColors);
7565 de::MovePtr<tcu::TestCaseGroup> opnopTests (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
7566 map<string, string> opNopFragments;
7567 opNopFragments["testfun"] =
7568 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7569 "%param1 = OpFunctionParameter %v4f32\n"
7570 "%label_testfun = OpLabel\n"
7579 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7580 "%b = OpFAdd %f32 %a %a\n"
7582 "%c = OpFSub %f32 %b %a\n"
7583 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
7586 "OpReturnValue %ret\n"
7589 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, opnopTests.get());
7592 graphicsTests->addChild(opnopTests.release());
7593 graphicsTests->addChild(createOpSourceTests(testCtx));
7594 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
7595 graphicsTests->addChild(createOpLineTests(testCtx));
7596 graphicsTests->addChild(createOpNoLineTests(testCtx));
7597 graphicsTests->addChild(createOpConstantNullTests(testCtx));
7598 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
7599 graphicsTests->addChild(createMemoryAccessTests(testCtx));
7600 graphicsTests->addChild(createOpUndefTests(testCtx));
7601 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
7602 graphicsTests->addChild(createModuleTests(testCtx));
7603 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
7604 graphicsTests->addChild(createOpPhiTests(testCtx));
7605 graphicsTests->addChild(createNoContractionTests(testCtx));
7606 graphicsTests->addChild(createOpQuantizeTests(testCtx));
7607 graphicsTests->addChild(createLoopTests(testCtx));
7608 graphicsTests->addChild(createSpecConstantTests(testCtx));
7609 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
7610 graphicsTests->addChild(createBarrierTests(testCtx));
7611 graphicsTests->addChild(createDecorationGroupTests(testCtx));
7612 graphicsTests->addChild(createFRemTests(testCtx));
7614 instructionTests->addChild(computeTests.release());
7615 instructionTests->addChild(graphicsTests.release());
7617 return instructionTests.release();
projects / platform / upstream / VK-GL-CTS.git / blob