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 "%uvec3ptr = OpTypePointer Input %uvec3\n"
149 "%f32ptr = OpTypePointer Uniform %f32\n"
150 "%f32arr = OpTypeRuntimeArray %f32\n";
152 // Declares two uniform variables (indata, outdata) of type "struct { float[] }". Depends on type "f32arr" (for "float[]").
153 static const char* const s_InputOutputBuffer =
154 "%inbuf = OpTypeStruct %f32arr\n"
155 "%inbufptr = OpTypePointer Uniform %inbuf\n"
156 "%indata = OpVariable %inbufptr Uniform\n"
157 "%outbuf = OpTypeStruct %f32arr\n"
158 "%outbufptr = OpTypePointer Uniform %outbuf\n"
159 "%outdata = OpVariable %outbufptr Uniform\n";
161 // Declares buffer type and layout for uniform variables indata and outdata. Both of them are SSBO bounded to descriptor set 0.
162 // indata is at binding point 0, while outdata is at 1.
163 static const char* const s_InputOutputBufferTraits =
164 "OpDecorate %inbuf BufferBlock\n"
165 "OpDecorate %indata DescriptorSet 0\n"
166 "OpDecorate %indata Binding 0\n"
167 "OpDecorate %outbuf BufferBlock\n"
168 "OpDecorate %outdata DescriptorSet 0\n"
169 "OpDecorate %outdata Binding 1\n"
170 "OpDecorate %f32arr ArrayStride 4\n"
171 "OpMemberDecorate %inbuf 0 Offset 0\n"
172 "OpMemberDecorate %outbuf 0 Offset 0\n";
174 tcu::TestCaseGroup* createOpNopGroup (tcu::TestContext& testCtx)
176 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnop", "Test the OpNop instruction"));
177 ComputeShaderSpec spec;
178 de::Random rnd (deStringHash(group->getName()));
179 const int numElements = 100;
180 vector<float> positiveFloats (numElements, 0);
181 vector<float> negativeFloats (numElements, 0);
183 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
185 for (size_t ndx = 0; ndx < numElements; ++ndx)
186 negativeFloats[ndx] = -positiveFloats[ndx];
189 string(s_ShaderPreamble) +
191 "OpSource GLSL 430\n"
192 "OpName %main \"main\"\n"
193 "OpName %id \"gl_GlobalInvocationID\"\n"
195 "OpDecorate %id BuiltIn GlobalInvocationId\n"
197 + string(s_InputOutputBufferTraits) + string(s_CommonTypes)
199 + string(s_InputOutputBuffer) +
201 "%id = OpVariable %uvec3ptr Input\n"
202 "%zero = OpConstant %i32 0\n"
204 "%main = OpFunction %void None %voidf\n"
206 "%idval = OpLoad %uvec3 %id\n"
207 "%x = OpCompositeExtract %u32 %idval 0\n"
209 " OpNop\n" // Inside a function body
211 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
212 "%inval = OpLoad %f32 %inloc\n"
213 "%neg = OpFNegate %f32 %inval\n"
214 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
215 " OpStore %outloc %neg\n"
218 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
219 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
220 spec.numWorkGroups = IVec3(numElements, 1, 1);
222 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNop appearing at different places", spec));
224 return group.release();
227 tcu::TestCaseGroup* createOpLineGroup (tcu::TestContext& testCtx)
229 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opline", "Test the OpLine instruction"));
230 ComputeShaderSpec spec;
231 de::Random rnd (deStringHash(group->getName()));
232 const int numElements = 100;
233 vector<float> positiveFloats (numElements, 0);
234 vector<float> negativeFloats (numElements, 0);
236 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
238 for (size_t ndx = 0; ndx < numElements; ++ndx)
239 negativeFloats[ndx] = -positiveFloats[ndx];
242 string(s_ShaderPreamble) +
244 "%fname1 = OpString \"negateInputs.comp\"\n"
245 "%fname2 = OpString \"negateInputs\"\n"
247 "OpSource GLSL 430\n"
248 "OpName %main \"main\"\n"
249 "OpName %id \"gl_GlobalInvocationID\"\n"
251 "OpDecorate %id BuiltIn GlobalInvocationId\n"
253 + string(s_InputOutputBufferTraits) +
255 "OpLine %fname1 0 0\n" // At the earliest possible position
257 + string(s_CommonTypes) + string(s_InputOutputBuffer) +
259 "OpLine %fname1 0 1\n" // Multiple OpLines in sequence
260 "OpLine %fname2 1 0\n" // Different filenames
261 "OpLine %fname1 1000 100000\n"
263 "%id = OpVariable %uvec3ptr Input\n"
264 "%zero = OpConstant %i32 0\n"
266 "OpLine %fname1 1 1\n" // Before a function
268 "%main = OpFunction %void None %voidf\n"
271 "OpLine %fname1 1 1\n" // In a function
273 "%idval = OpLoad %uvec3 %id\n"
274 "%x = OpCompositeExtract %u32 %idval 0\n"
275 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
276 "%inval = OpLoad %f32 %inloc\n"
277 "%neg = OpFNegate %f32 %inval\n"
278 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
279 " OpStore %outloc %neg\n"
282 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
283 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
284 spec.numWorkGroups = IVec3(numElements, 1, 1);
286 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpLine appearing at different places", spec));
288 return group.release();
291 tcu::TestCaseGroup* createOpNoLineGroup (tcu::TestContext& testCtx)
293 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opnoline", "Test the OpNoLine instruction"));
294 ComputeShaderSpec spec;
295 de::Random rnd (deStringHash(group->getName()));
296 const int numElements = 100;
297 vector<float> positiveFloats (numElements, 0);
298 vector<float> negativeFloats (numElements, 0);
300 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
302 for (size_t ndx = 0; ndx < numElements; ++ndx)
303 negativeFloats[ndx] = -positiveFloats[ndx];
306 string(s_ShaderPreamble) +
308 "%fname = OpString \"negateInputs.comp\"\n"
310 "OpSource GLSL 430\n"
311 "OpName %main \"main\"\n"
312 "OpName %id \"gl_GlobalInvocationID\"\n"
314 "OpDecorate %id BuiltIn GlobalInvocationId\n"
316 + string(s_InputOutputBufferTraits) +
318 "OpNoLine\n" // At the earliest possible position, without preceding OpLine
320 + string(s_CommonTypes) + string(s_InputOutputBuffer) +
322 "OpLine %fname 0 1\n"
323 "OpNoLine\n" // Immediately following a preceding OpLine
325 "OpLine %fname 1000 1\n"
327 "%id = OpVariable %uvec3ptr Input\n"
328 "%zero = OpConstant %i32 0\n"
330 "OpNoLine\n" // Contents after the previous OpLine
332 "%main = OpFunction %void None %voidf\n"
334 "%idval = OpLoad %uvec3 %id\n"
335 "%x = OpCompositeExtract %u32 %idval 0\n"
337 "OpNoLine\n" // Multiple OpNoLine
341 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
342 "%inval = OpLoad %f32 %inloc\n"
343 "%neg = OpFNegate %f32 %inval\n"
344 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
345 " OpStore %outloc %neg\n"
348 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
349 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
350 spec.numWorkGroups = IVec3(numElements, 1, 1);
352 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpNoLine appearing at different places", spec));
354 return group.release();
357 tcu::TestCaseGroup* createNoContractionGroup (tcu::TestContext& testCtx)
359 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
360 vector<CaseParameter> cases;
361 const int numElements = 100;
362 vector<float> inputFloats1 (numElements, 0);
363 vector<float> inputFloats2 (numElements, 0);
364 vector<float> outputFloats (numElements, 0);
365 const StringTemplate shaderTemplate (
366 string(s_ShaderPreamble) +
368 "OpName %main \"main\"\n"
369 "OpName %id \"gl_GlobalInvocationID\"\n"
371 "OpDecorate %id BuiltIn GlobalInvocationId\n"
375 "OpDecorate %inbuf1 BufferBlock\n"
376 "OpDecorate %indata1 DescriptorSet 0\n"
377 "OpDecorate %indata1 Binding 0\n"
378 "OpDecorate %inbuf2 BufferBlock\n"
379 "OpDecorate %indata2 DescriptorSet 0\n"
380 "OpDecorate %indata2 Binding 1\n"
381 "OpDecorate %outbuf BufferBlock\n"
382 "OpDecorate %outdata DescriptorSet 0\n"
383 "OpDecorate %outdata Binding 2\n"
384 "OpDecorate %f32arr ArrayStride 4\n"
385 "OpMemberDecorate %inbuf1 0 Offset 0\n"
386 "OpMemberDecorate %inbuf2 0 Offset 0\n"
387 "OpMemberDecorate %outbuf 0 Offset 0\n"
389 + string(s_CommonTypes) +
391 "%inbuf1 = OpTypeStruct %f32arr\n"
392 "%inbufptr1 = OpTypePointer Uniform %inbuf1\n"
393 "%indata1 = OpVariable %inbufptr1 Uniform\n"
394 "%inbuf2 = OpTypeStruct %f32arr\n"
395 "%inbufptr2 = OpTypePointer Uniform %inbuf2\n"
396 "%indata2 = OpVariable %inbufptr2 Uniform\n"
397 "%outbuf = OpTypeStruct %f32arr\n"
398 "%outbufptr = OpTypePointer Uniform %outbuf\n"
399 "%outdata = OpVariable %outbufptr Uniform\n"
401 "%id = OpVariable %uvec3ptr Input\n"
402 "%zero = OpConstant %i32 0\n"
403 "%c_f_m1 = OpConstant %f32 -1.\n"
405 "%main = OpFunction %void None %voidf\n"
407 "%idval = OpLoad %uvec3 %id\n"
408 "%x = OpCompositeExtract %u32 %idval 0\n"
409 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
410 "%inval1 = OpLoad %f32 %inloc1\n"
411 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
412 "%inval2 = OpLoad %f32 %inloc2\n"
413 "%mul = OpFMul %f32 %inval1 %inval2\n"
414 "%add = OpFAdd %f32 %mul %c_f_m1\n"
415 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
416 " OpStore %outloc %add\n"
420 cases.push_back(CaseParameter("multiplication", "OpDecorate %mul NoContraction"));
421 cases.push_back(CaseParameter("addition", "OpDecorate %add NoContraction"));
422 cases.push_back(CaseParameter("both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"));
424 for (size_t ndx = 0; ndx < numElements; ++ndx)
426 inputFloats1[ndx] = 1.f + std::ldexp(1.f, -23); // 1 + 2^-23.
427 inputFloats2[ndx] = 1.f - std::ldexp(1.f, -23); // 1 - 2^-23.
428 // Result for (1 + 2^-23) * (1 - 2^-23) - 1. With NoContraction, the multiplication will be
429 // conducted separately and the result is rounded to 1. So the final result will be 0.f.
430 // If the operation is combined into a precise fused multiply-add, then the result would be
431 // 2^-46 (0xa8800000).
432 outputFloats[ndx] = 0.f;
435 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
437 map<string, string> specializations;
438 ComputeShaderSpec spec;
440 specializations["DECORATION"] = cases[caseNdx].param;
441 spec.assembly = shaderTemplate.specialize(specializations);
442 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
443 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
444 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
445 spec.numWorkGroups = IVec3(numElements, 1, 1);
447 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
449 return group.release();
452 tcu::TestCaseGroup* createOpFRemGroup (tcu::TestContext& testCtx)
454 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opfrem", "Test the OpFRem instruction"));
455 ComputeShaderSpec spec;
456 de::Random rnd (deStringHash(group->getName()));
457 const int numElements = 200;
458 vector<float> inputFloats1 (numElements, 0);
459 vector<float> inputFloats2 (numElements, 0);
460 vector<float> outputFloats (numElements, 0);
462 fillRandomScalars(rnd, -10000.f, 10000.f, &inputFloats1[0], numElements);
463 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats2[0], numElements);
465 for (size_t ndx = 0; ndx < numElements; ++ndx)
467 // Guard against divisors near zero.
468 if (std::fabs(inputFloats2[ndx]) < 1e-3)
469 inputFloats2[ndx] = 8.f;
471 // The return value of std::fmod() has the same sign as its first operand, which is how OpFRem spec'd.
472 outputFloats[ndx] = std::fmod(inputFloats1[ndx], inputFloats2[ndx]);
476 string(s_ShaderPreamble) +
478 "OpName %main \"main\"\n"
479 "OpName %id \"gl_GlobalInvocationID\"\n"
481 "OpDecorate %id BuiltIn GlobalInvocationId\n"
483 "OpDecorate %inbuf1 BufferBlock\n"
484 "OpDecorate %indata1 DescriptorSet 0\n"
485 "OpDecorate %indata1 Binding 0\n"
486 "OpDecorate %inbuf2 BufferBlock\n"
487 "OpDecorate %indata2 DescriptorSet 0\n"
488 "OpDecorate %indata2 Binding 1\n"
489 "OpDecorate %outbuf BufferBlock\n"
490 "OpDecorate %outdata DescriptorSet 0\n"
491 "OpDecorate %outdata Binding 2\n"
492 "OpDecorate %f32arr ArrayStride 4\n"
493 "OpMemberDecorate %inbuf1 0 Offset 0\n"
494 "OpMemberDecorate %inbuf2 0 Offset 0\n"
495 "OpMemberDecorate %outbuf 0 Offset 0\n"
497 + string(s_CommonTypes) +
499 "%inbuf1 = OpTypeStruct %f32arr\n"
500 "%inbufptr1 = OpTypePointer Uniform %inbuf1\n"
501 "%indata1 = OpVariable %inbufptr1 Uniform\n"
502 "%inbuf2 = OpTypeStruct %f32arr\n"
503 "%inbufptr2 = OpTypePointer Uniform %inbuf2\n"
504 "%indata2 = OpVariable %inbufptr2 Uniform\n"
505 "%outbuf = OpTypeStruct %f32arr\n"
506 "%outbufptr = OpTypePointer Uniform %outbuf\n"
507 "%outdata = OpVariable %outbufptr Uniform\n"
509 "%id = OpVariable %uvec3ptr Input\n"
510 "%zero = OpConstant %i32 0\n"
512 "%main = OpFunction %void None %voidf\n"
514 "%idval = OpLoad %uvec3 %id\n"
515 "%x = OpCompositeExtract %u32 %idval 0\n"
516 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
517 "%inval1 = OpLoad %f32 %inloc1\n"
518 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
519 "%inval2 = OpLoad %f32 %inloc2\n"
520 "%rem = OpFRem %f32 %inval1 %inval2\n"
521 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
522 " OpStore %outloc %rem\n"
526 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
527 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
528 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
529 spec.numWorkGroups = IVec3(numElements, 1, 1);
531 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "", spec));
533 return group.release();
536 // Copy contents in the input buffer to the output buffer.
537 tcu::TestCaseGroup* createOpCopyMemoryGroup (tcu::TestContext& testCtx)
539 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopymemory", "Test the OpCopyMemory instruction"));
540 de::Random rnd (deStringHash(group->getName()));
541 const int numElements = 100;
543 // 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.
544 ComputeShaderSpec spec1;
545 vector<Vec4> inputFloats1 (numElements);
546 vector<Vec4> outputFloats1 (numElements);
548 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats1[0], numElements * 4);
550 for (size_t ndx = 0; ndx < numElements; ++ndx)
551 outputFloats1[ndx] = inputFloats1[ndx] + Vec4(0.f, 0.5f, 1.5f, 2.5f);
554 string(s_ShaderPreamble) +
556 "OpName %main \"main\"\n"
557 "OpName %id \"gl_GlobalInvocationID\"\n"
559 "OpDecorate %id BuiltIn GlobalInvocationId\n"
561 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
563 "%vec4 = OpTypeVector %f32 4\n"
564 "%vec4ptr_u = OpTypePointer Uniform %vec4\n"
565 "%vec4ptr_f = OpTypePointer Function %vec4\n"
566 "%vec4arr = OpTypeRuntimeArray %vec4\n"
567 "%inbuf = OpTypeStruct %vec4arr\n"
568 "%inbufptr = OpTypePointer Uniform %inbuf\n"
569 "%indata = OpVariable %inbufptr Uniform\n"
570 "%outbuf = OpTypeStruct %vec4arr\n"
571 "%outbufptr = OpTypePointer Uniform %outbuf\n"
572 "%outdata = OpVariable %outbufptr Uniform\n"
574 "%id = OpVariable %uvec3ptr Input\n"
575 "%zero = OpConstant %i32 0\n"
576 "%c_f_0 = OpConstant %f32 0.\n"
577 "%c_f_0_5 = OpConstant %f32 0.5\n"
578 "%c_f_1_5 = OpConstant %f32 1.5\n"
579 "%c_f_2_5 = OpConstant %f32 2.5\n"
580 "%c_vec4 = OpConstantComposite %vec4 %c_f_0 %c_f_0_5 %c_f_1_5 %c_f_2_5\n"
582 "%main = OpFunction %void None %voidf\n"
584 "%v_vec4 = OpVariable %vec4ptr_f Function\n"
585 "%idval = OpLoad %uvec3 %id\n"
586 "%x = OpCompositeExtract %u32 %idval 0\n"
587 "%inloc = OpAccessChain %vec4ptr_u %indata %zero %x\n"
588 "%outloc = OpAccessChain %vec4ptr_u %outdata %zero %x\n"
589 " OpCopyMemory %v_vec4 %inloc\n"
590 "%v_vec4_val = OpLoad %vec4 %v_vec4\n"
591 "%add = OpFAdd %vec4 %v_vec4_val %c_vec4\n"
592 " OpStore %outloc %add\n"
596 spec1.inputs.push_back(BufferSp(new Vec4Buffer(inputFloats1)));
597 spec1.outputs.push_back(BufferSp(new Vec4Buffer(outputFloats1)));
598 spec1.numWorkGroups = IVec3(numElements, 1, 1);
600 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector", "OpCopyMemory elements of vector type", spec1));
602 // The following case copies a float[100] variable from the input buffer to the output buffer.
603 ComputeShaderSpec spec2;
604 vector<float> inputFloats2 (numElements);
605 vector<float> outputFloats2 (numElements);
607 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats2[0], numElements);
609 for (size_t ndx = 0; ndx < numElements; ++ndx)
610 outputFloats2[ndx] = inputFloats2[ndx];
613 string(s_ShaderPreamble) +
615 "OpName %main \"main\"\n"
616 "OpName %id \"gl_GlobalInvocationID\"\n"
618 "OpDecorate %id BuiltIn GlobalInvocationId\n"
620 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
622 "%hundred = OpConstant %u32 100\n"
623 "%f32arr100 = OpTypeArray %f32 %hundred\n"
624 "%f32arr100ptr_f = OpTypePointer Function %f32arr100\n"
625 "%f32arr100ptr_u = OpTypePointer Uniform %f32arr100\n"
626 "%inbuf = OpTypeStruct %f32arr100\n"
627 "%inbufptr = OpTypePointer Uniform %inbuf\n"
628 "%indata = OpVariable %inbufptr Uniform\n"
629 "%outbuf = OpTypeStruct %f32arr100\n"
630 "%outbufptr = OpTypePointer Uniform %outbuf\n"
631 "%outdata = OpVariable %outbufptr Uniform\n"
633 "%id = OpVariable %uvec3ptr Input\n"
634 "%zero = OpConstant %i32 0\n"
636 "%main = OpFunction %void None %voidf\n"
638 "%var = OpVariable %f32arr100ptr_f Function\n"
639 "%inarr = OpAccessChain %f32arr100ptr_u %indata %zero\n"
640 "%outarr = OpAccessChain %f32arr100ptr_u %outdata %zero\n"
641 " OpCopyMemory %var %inarr\n"
642 " OpCopyMemory %outarr %var\n"
646 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
647 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
648 spec2.numWorkGroups = IVec3(1, 1, 1);
650 group->addChild(new SpvAsmComputeShaderCase(testCtx, "array", "OpCopyMemory elements of array type", spec2));
652 // The following case copies a struct{vec4, vec4, vec4, vec4} variable from the input buffer to the output buffer.
653 ComputeShaderSpec spec3;
654 vector<float> inputFloats3 (16);
655 vector<float> outputFloats3 (16);
657 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats3[0], 16);
659 for (size_t ndx = 0; ndx < 16; ++ndx)
660 outputFloats3[ndx] = -inputFloats3[ndx];
663 string(s_ShaderPreamble) +
665 "OpName %main \"main\"\n"
666 "OpName %id \"gl_GlobalInvocationID\"\n"
668 "OpDecorate %id BuiltIn GlobalInvocationId\n"
670 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
672 "%vec4 = OpTypeVector %f32 4\n"
673 "%inbuf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
674 "%inbufptr = OpTypePointer Uniform %inbuf\n"
675 "%indata = OpVariable %inbufptr Uniform\n"
676 "%outbuf = OpTypeStruct %vec4 %vec4 %vec4 %vec4\n"
677 "%outbufptr = OpTypePointer Uniform %outbuf\n"
678 "%outdata = OpVariable %outbufptr Uniform\n"
679 "%vec4stptr = OpTypePointer Function %inbuf\n"
681 "%id = OpVariable %uvec3ptr Input\n"
682 "%zero = OpConstant %i32 0\n"
684 "%main = OpFunction %void None %voidf\n"
686 "%var = OpVariable %vec4stptr Function\n"
687 " OpCopyMemory %var %indata\n"
688 " OpCopyMemory %outdata %var\n"
692 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
693 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
694 spec3.numWorkGroups = IVec3(1, 1, 1);
696 group->addChild(new SpvAsmComputeShaderCase(testCtx, "struct", "OpCopyMemory elements of struct type", spec3));
698 // The following case negates multiple float variables from the input buffer and stores the results to the output buffer.
699 ComputeShaderSpec spec4;
700 vector<float> inputFloats4 (numElements);
701 vector<float> outputFloats4 (numElements);
703 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats4[0], numElements);
705 for (size_t ndx = 0; ndx < numElements; ++ndx)
706 outputFloats4[ndx] = -inputFloats4[ndx];
709 string(s_ShaderPreamble) +
711 "OpName %main \"main\"\n"
712 "OpName %id \"gl_GlobalInvocationID\"\n"
714 "OpDecorate %id BuiltIn GlobalInvocationId\n"
716 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
718 "%f32ptr_f = OpTypePointer Function %f32\n"
719 "%id = OpVariable %uvec3ptr Input\n"
720 "%zero = OpConstant %i32 0\n"
722 "%main = OpFunction %void None %voidf\n"
724 "%var = OpVariable %f32ptr_f Function\n"
725 "%idval = OpLoad %uvec3 %id\n"
726 "%x = OpCompositeExtract %u32 %idval 0\n"
727 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
728 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
729 " OpCopyMemory %var %inloc\n"
730 "%val = OpLoad %f32 %var\n"
731 "%neg = OpFNegate %f32 %val\n"
732 " OpStore %outloc %neg\n"
736 spec4.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
737 spec4.outputs.push_back(BufferSp(new Float32Buffer(outputFloats4)));
738 spec4.numWorkGroups = IVec3(numElements, 1, 1);
740 group->addChild(new SpvAsmComputeShaderCase(testCtx, "float", "OpCopyMemory elements of float type", spec4));
742 return group.release();
745 tcu::TestCaseGroup* createOpCopyObjectGroup (tcu::TestContext& testCtx)
747 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opcopyobject", "Test the OpCopyObject instruction"));
748 ComputeShaderSpec spec;
749 de::Random rnd (deStringHash(group->getName()));
750 const int numElements = 100;
751 vector<float> inputFloats (numElements, 0);
752 vector<float> outputFloats (numElements, 0);
754 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
756 for (size_t ndx = 0; ndx < numElements; ++ndx)
757 outputFloats[ndx] = inputFloats[ndx] + 7.5f;
760 string(s_ShaderPreamble) +
762 "OpName %main \"main\"\n"
763 "OpName %id \"gl_GlobalInvocationID\"\n"
765 "OpDecorate %id BuiltIn GlobalInvocationId\n"
767 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
769 "%fvec3 = OpTypeVector %f32 3\n"
770 "%fmat = OpTypeMatrix %fvec3 3\n"
771 "%three = OpConstant %u32 3\n"
772 "%farr = OpTypeArray %f32 %three\n"
773 "%fst = OpTypeStruct %f32 %f32\n"
775 + string(s_InputOutputBuffer) +
777 "%id = OpVariable %uvec3ptr Input\n"
778 "%zero = OpConstant %i32 0\n"
779 "%c_f = OpConstant %f32 1.5\n"
780 "%c_fvec3 = OpConstantComposite %fvec3 %c_f %c_f %c_f\n"
781 "%c_fmat = OpConstantComposite %fmat %c_fvec3 %c_fvec3 %c_fvec3\n"
782 "%c_farr = OpConstantComposite %farr %c_f %c_f %c_f\n"
783 "%c_fst = OpConstantComposite %fst %c_f %c_f\n"
785 "%main = OpFunction %void None %voidf\n"
787 "%c_f_copy = OpCopyObject %f32 %c_f\n"
788 "%c_fvec3_copy = OpCopyObject %fvec3 %c_fvec3\n"
789 "%c_fmat_copy = OpCopyObject %fmat %c_fmat\n"
790 "%c_farr_copy = OpCopyObject %farr %c_farr\n"
791 "%c_fst_copy = OpCopyObject %fst %c_fst\n"
792 "%fvec3_elem = OpCompositeExtract %f32 %c_fvec3_copy 0\n"
793 "%fmat_elem = OpCompositeExtract %f32 %c_fmat_copy 1 2\n"
794 "%farr_elem = OpCompositeExtract %f32 %c_fmat_copy 2\n"
795 "%fst_elem = OpCompositeExtract %f32 %c_fmat_copy 1\n"
796 // Add up. 1.5 * 5 = 7.5.
797 "%add1 = OpFAdd %f32 %c_f_copy %fvec3_elem\n"
798 "%add2 = OpFAdd %f32 %add1 %fmat_elem\n"
799 "%add3 = OpFAdd %f32 %add2 %farr_elem\n"
800 "%add4 = OpFAdd %f32 %add3 %fst_elem\n"
802 "%idval = OpLoad %uvec3 %id\n"
803 "%x = OpCompositeExtract %u32 %idval 0\n"
804 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
805 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
806 "%inval = OpLoad %f32 %inloc\n"
807 "%add = OpFAdd %f32 %add4 %inval\n"
808 " OpStore %outloc %add\n"
811 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
812 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
813 spec.numWorkGroups = IVec3(numElements, 1, 1);
815 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "OpCopyObject on different types", spec));
817 return group.release();
819 // Assembly code used for testing OpUnreachable is based on GLSL source code:
823 // layout(std140, set = 0, binding = 0) readonly buffer Input {
826 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
830 // void not_called_func() {
831 // // place OpUnreachable here
834 // uint modulo4(uint val) {
835 // switch (val % uint(4)) {
840 // default: return 100; // place OpUnreachable here
846 // // place OpUnreachable here
850 // uint x = gl_GlobalInvocationID.x;
851 // if (const5() > modulo4(1000)) {
852 // output_data.elements[x] = -input_data.elements[x];
854 // // place OpUnreachable here
855 // output_data.elements[x] = input_data.elements[x];
859 tcu::TestCaseGroup* createOpUnreachableGroup (tcu::TestContext& testCtx)
861 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opunreachable", "Test the OpUnreachable instruction"));
862 ComputeShaderSpec spec;
863 de::Random rnd (deStringHash(group->getName()));
864 const int numElements = 100;
865 vector<float> positiveFloats (numElements, 0);
866 vector<float> negativeFloats (numElements, 0);
868 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
870 for (size_t ndx = 0; ndx < numElements; ++ndx)
871 negativeFloats[ndx] = -positiveFloats[ndx];
874 string(s_ShaderPreamble) +
876 "OpSource GLSL 430\n"
877 "OpName %main \"main\"\n"
878 "OpName %func_not_called_func \"not_called_func(\"\n"
879 "OpName %func_modulo4 \"modulo4(u1;\"\n"
880 "OpName %func_const5 \"const5(\"\n"
881 "OpName %id \"gl_GlobalInvocationID\"\n"
883 "OpDecorate %id BuiltIn GlobalInvocationId\n"
885 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
887 "%u32ptr = OpTypePointer Function %u32\n"
888 "%uintfuint = OpTypeFunction %u32 %u32ptr\n"
889 "%unitf = OpTypeFunction %u32\n"
891 "%id = OpVariable %uvec3ptr Input\n"
892 "%zero = OpConstant %u32 0\n"
893 "%one = OpConstant %u32 1\n"
894 "%two = OpConstant %u32 2\n"
895 "%three = OpConstant %u32 3\n"
896 "%four = OpConstant %u32 4\n"
897 "%five = OpConstant %u32 5\n"
898 "%hundred = OpConstant %u32 100\n"
899 "%thousand = OpConstant %u32 1000\n"
901 + string(s_InputOutputBuffer) +
904 "%main = OpFunction %void None %voidf\n"
905 "%main_entry = OpLabel\n"
906 "%idval = OpLoad %uvec3 %id\n"
907 "%x = OpCompositeExtract %u32 %idval 0\n"
908 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
909 "%inval = OpLoad %f32 %inloc\n"
910 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
911 "%ret_const5 = OpFunctionCall %u32 %func_const5\n"
912 "%ret_modulo4 = OpFunctionCall %u32 %func_modulo4 %thousand\n"
913 "%cmp_gt = OpUGreaterThan %bool %ret_const5 %ret_modulo4\n"
914 " OpSelectionMerge %if_end None\n"
915 " OpBranchConditional %cmp_gt %if_true %if_false\n"
916 "%if_true = OpLabel\n"
917 "%negate = OpFNegate %f32 %inval\n"
918 " OpStore %outloc %negate\n"
919 " OpBranch %if_end\n"
920 "%if_false = OpLabel\n"
921 " OpUnreachable\n" // Unreachable else branch for if statement
922 "%if_end = OpLabel\n"
926 // not_called_function()
927 "%func_not_called_func = OpFunction %void None %voidf\n"
928 "%not_called_func_entry = OpLabel\n"
929 " OpUnreachable\n" // Unreachable entry block in not called static function
933 "%func_modulo4 = OpFunction %u32 None %uintfuint\n"
934 "%valptr = OpFunctionParameter %u32ptr\n"
935 "%modulo4_entry = OpLabel\n"
936 "%val = OpLoad %u32 %valptr\n"
937 "%modulo = OpUMod %u32 %val %four\n"
938 " OpSelectionMerge %switch_merge None\n"
939 " OpSwitch %modulo %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
941 " OpReturnValue %three\n"
943 " OpReturnValue %two\n"
945 " OpReturnValue %one\n"
947 " OpReturnValue %zero\n"
948 "%default = OpLabel\n"
949 " OpUnreachable\n" // Unreachable default case for switch statement
950 "%switch_merge = OpLabel\n"
951 " OpUnreachable\n" // Unreachable merge block for switch statement
955 "%func_const5 = OpFunction %u32 None %unitf\n"
956 "%const5_entry = OpLabel\n"
957 " OpReturnValue %five\n"
958 "%unreachable = OpLabel\n"
959 " OpUnreachable\n" // Unreachable block in function
961 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
962 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
963 spec.numWorkGroups = IVec3(numElements, 1, 1);
965 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "OpUnreachable appearing at different places", spec));
967 return group.release();
970 // Assembly code used for testing decoration group is based on GLSL source code:
974 // layout(std140, set = 0, binding = 0) readonly buffer Input0 {
977 // layout(std140, set = 0, binding = 1) readonly buffer Input1 {
980 // layout(std140, set = 0, binding = 2) readonly buffer Input2 {
983 // layout(std140, set = 0, binding = 3) readonly buffer Input3 {
986 // layout(std140, set = 0, binding = 4) readonly buffer Input4 {
989 // layout(std140, set = 0, binding = 5) writeonly buffer Output {
994 // uint x = gl_GlobalInvocationID.x;
995 // output_data.elements[x] = input_data0.elements[x] + input_data1.elements[x] + input_data2.elements[x] + input_data3.elements[x] + input_data4.elements[x];
997 tcu::TestCaseGroup* createDecorationGroupGroup (tcu::TestContext& testCtx)
999 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Test the OpDecorationGroup & OpGroupDecorate instruction"));
1000 ComputeShaderSpec spec;
1001 de::Random rnd (deStringHash(group->getName()));
1002 const int numElements = 100;
1003 vector<float> inputFloats0 (numElements, 0);
1004 vector<float> inputFloats1 (numElements, 0);
1005 vector<float> inputFloats2 (numElements, 0);
1006 vector<float> inputFloats3 (numElements, 0);
1007 vector<float> inputFloats4 (numElements, 0);
1008 vector<float> outputFloats (numElements, 0);
1010 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats0[0], numElements);
1011 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats1[0], numElements);
1012 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats2[0], numElements);
1013 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats3[0], numElements);
1014 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats4[0], numElements);
1016 for (size_t ndx = 0; ndx < numElements; ++ndx)
1017 outputFloats[ndx] = inputFloats0[ndx] + inputFloats1[ndx] + inputFloats2[ndx] + inputFloats3[ndx] + inputFloats4[ndx];
1020 string(s_ShaderPreamble) +
1022 "OpSource GLSL 430\n"
1023 "OpName %main \"main\"\n"
1024 "OpName %id \"gl_GlobalInvocationID\"\n"
1026 // Not using group decoration on variable.
1027 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1028 // Not using group decoration on type.
1029 "OpDecorate %f32arr ArrayStride 4\n"
1031 "OpDecorate %groups BufferBlock\n"
1032 "OpDecorate %groupm Offset 0\n"
1033 "%groups = OpDecorationGroup\n"
1034 "%groupm = OpDecorationGroup\n"
1036 // Group decoration on multiple structs.
1037 "OpGroupDecorate %groups %outbuf %inbuf0 %inbuf1 %inbuf2 %inbuf3 %inbuf4\n"
1038 // Group decoration on multiple struct members.
1039 "OpGroupMemberDecorate %groupm %outbuf 0 %inbuf0 0 %inbuf1 0 %inbuf2 0 %inbuf3 0 %inbuf4 0\n"
1041 "OpDecorate %group1 DescriptorSet 0\n"
1042 "OpDecorate %group3 DescriptorSet 0\n"
1043 "OpDecorate %group3 NonWritable\n"
1044 "OpDecorate %group3 Restrict\n"
1045 "%group0 = OpDecorationGroup\n"
1046 "%group1 = OpDecorationGroup\n"
1047 "%group3 = OpDecorationGroup\n"
1049 // Applying the same decoration group multiple times.
1050 "OpGroupDecorate %group1 %outdata\n"
1051 "OpGroupDecorate %group1 %outdata\n"
1052 "OpGroupDecorate %group1 %outdata\n"
1053 "OpDecorate %outdata DescriptorSet 0\n"
1054 "OpDecorate %outdata Binding 5\n"
1055 // Applying decoration group containing nothing.
1056 "OpGroupDecorate %group0 %indata0\n"
1057 "OpDecorate %indata0 DescriptorSet 0\n"
1058 "OpDecorate %indata0 Binding 0\n"
1059 // Applying decoration group containing one decoration.
1060 "OpGroupDecorate %group1 %indata1\n"
1061 "OpDecorate %indata1 Binding 1\n"
1062 // Applying decoration group containing multiple decorations.
1063 "OpGroupDecorate %group3 %indata2 %indata3\n"
1064 "OpDecorate %indata2 Binding 2\n"
1065 "OpDecorate %indata3 Binding 3\n"
1066 // Applying multiple decoration groups (with overlapping).
1067 "OpGroupDecorate %group0 %indata4\n"
1068 "OpGroupDecorate %group1 %indata4\n"
1069 "OpGroupDecorate %group3 %indata4\n"
1070 "OpDecorate %indata4 Binding 4\n"
1072 + string(s_CommonTypes) +
1074 "%id = OpVariable %uvec3ptr Input\n"
1075 "%zero = OpConstant %i32 0\n"
1077 "%outbuf = OpTypeStruct %f32arr\n"
1078 "%outbufptr = OpTypePointer Uniform %outbuf\n"
1079 "%outdata = OpVariable %outbufptr Uniform\n"
1080 "%inbuf0 = OpTypeStruct %f32arr\n"
1081 "%inbuf0ptr = OpTypePointer Uniform %inbuf0\n"
1082 "%indata0 = OpVariable %inbuf0ptr Uniform\n"
1083 "%inbuf1 = OpTypeStruct %f32arr\n"
1084 "%inbuf1ptr = OpTypePointer Uniform %inbuf1\n"
1085 "%indata1 = OpVariable %inbuf1ptr Uniform\n"
1086 "%inbuf2 = OpTypeStruct %f32arr\n"
1087 "%inbuf2ptr = OpTypePointer Uniform %inbuf2\n"
1088 "%indata2 = OpVariable %inbuf2ptr Uniform\n"
1089 "%inbuf3 = OpTypeStruct %f32arr\n"
1090 "%inbuf3ptr = OpTypePointer Uniform %inbuf3\n"
1091 "%indata3 = OpVariable %inbuf3ptr Uniform\n"
1092 "%inbuf4 = OpTypeStruct %f32arr\n"
1093 "%inbufptr = OpTypePointer Uniform %inbuf4\n"
1094 "%indata4 = OpVariable %inbufptr Uniform\n"
1096 "%main = OpFunction %void None %voidf\n"
1097 "%label = OpLabel\n"
1098 "%idval = OpLoad %uvec3 %id\n"
1099 "%x = OpCompositeExtract %u32 %idval 0\n"
1100 "%inloc0 = OpAccessChain %f32ptr %indata0 %zero %x\n"
1101 "%inloc1 = OpAccessChain %f32ptr %indata1 %zero %x\n"
1102 "%inloc2 = OpAccessChain %f32ptr %indata2 %zero %x\n"
1103 "%inloc3 = OpAccessChain %f32ptr %indata3 %zero %x\n"
1104 "%inloc4 = OpAccessChain %f32ptr %indata4 %zero %x\n"
1105 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1106 "%inval0 = OpLoad %f32 %inloc0\n"
1107 "%inval1 = OpLoad %f32 %inloc1\n"
1108 "%inval2 = OpLoad %f32 %inloc2\n"
1109 "%inval3 = OpLoad %f32 %inloc3\n"
1110 "%inval4 = OpLoad %f32 %inloc4\n"
1111 "%add0 = OpFAdd %f32 %inval0 %inval1\n"
1112 "%add1 = OpFAdd %f32 %add0 %inval2\n"
1113 "%add2 = OpFAdd %f32 %add1 %inval3\n"
1114 "%add = OpFAdd %f32 %add2 %inval4\n"
1115 " OpStore %outloc %add\n"
1118 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats0)));
1119 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats1)));
1120 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats2)));
1121 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats3)));
1122 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats4)));
1123 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1124 spec.numWorkGroups = IVec3(numElements, 1, 1);
1126 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "decoration group cases", spec));
1128 return group.release();
1131 struct SpecConstantTwoIntCase
1133 const char* caseName;
1134 const char* scDefinition0;
1135 const char* scDefinition1;
1136 const char* scResultType;
1137 const char* scOperation;
1138 deInt32 scActualValue0;
1139 deInt32 scActualValue1;
1140 const char* resultOperation;
1141 vector<deInt32> expectedOutput;
1143 SpecConstantTwoIntCase (const char* name,
1144 const char* definition0,
1145 const char* definition1,
1146 const char* resultType,
1147 const char* operation,
1150 const char* resultOp,
1151 const vector<deInt32>& output)
1153 , scDefinition0 (definition0)
1154 , scDefinition1 (definition1)
1155 , scResultType (resultType)
1156 , scOperation (operation)
1157 , scActualValue0 (value0)
1158 , scActualValue1 (value1)
1159 , resultOperation (resultOp)
1160 , expectedOutput (output) {}
1163 tcu::TestCaseGroup* createSpecConstantGroup (tcu::TestContext& testCtx)
1165 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
1166 vector<SpecConstantTwoIntCase> cases;
1167 de::Random rnd (deStringHash(group->getName()));
1168 const int numElements = 100;
1169 vector<deInt32> inputInts (numElements, 0);
1170 vector<deInt32> outputInts1 (numElements, 0);
1171 vector<deInt32> outputInts2 (numElements, 0);
1172 vector<deInt32> outputInts3 (numElements, 0);
1173 vector<deInt32> outputInts4 (numElements, 0);
1174 const StringTemplate shaderTemplate (
1175 string(s_ShaderPreamble) +
1177 "OpName %main \"main\"\n"
1178 "OpName %id \"gl_GlobalInvocationID\"\n"
1180 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1181 "OpDecorate %sc_0 SpecId 0\n"
1182 "OpDecorate %sc_1 SpecId 1\n"
1184 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1186 "%i32ptr = OpTypePointer Uniform %i32\n"
1187 "%i32arr = OpTypeRuntimeArray %i32\n"
1188 "%boolptr = OpTypePointer Uniform %bool\n"
1189 "%boolarr = OpTypeRuntimeArray %bool\n"
1190 "%inbuf = OpTypeStruct %i32arr\n"
1191 "%inbufptr = OpTypePointer Uniform %inbuf\n"
1192 "%indata = OpVariable %inbufptr Uniform\n"
1193 "%outbuf = OpTypeStruct %i32arr\n"
1194 "%outbufptr = OpTypePointer Uniform %outbuf\n"
1195 "%outdata = OpVariable %outbufptr Uniform\n"
1197 "%id = OpVariable %uvec3ptr Input\n"
1198 "%zero = OpConstant %i32 0\n"
1200 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
1201 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
1202 "%sc_final = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n"
1204 "%main = OpFunction %void None %voidf\n"
1205 "%label = OpLabel\n"
1206 "%idval = OpLoad %uvec3 %id\n"
1207 "%x = OpCompositeExtract %u32 %idval 0\n"
1208 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
1209 "%inval = OpLoad %i32 %inloc\n"
1210 "%final = ${GEN_RESULT}\n"
1211 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1212 " OpStore %outloc %final\n"
1214 " OpFunctionEnd\n");
1216 fillRandomScalars(rnd, -65536, 65536, &inputInts[0], numElements);
1218 for (size_t ndx = 0; ndx < numElements; ++ndx)
1220 outputInts1[ndx] = inputInts[ndx] + 42;
1221 outputInts2[ndx] = inputInts[ndx];
1222 outputInts3[ndx] = inputInts[ndx] - 11200;
1223 outputInts4[ndx] = inputInts[ndx] + 1;
1226 const char addScToInput[] = "OpIAdd %i32 %inval %sc_final";
1227 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_final %inval %zero";
1228 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_final %zero %inval";
1230 cases.push_back(SpecConstantTwoIntCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 62, -20, addScToInput, outputInts1));
1231 cases.push_back(SpecConstantTwoIntCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 100, 58, addScToInput, outputInts1));
1232 cases.push_back(SpecConstantTwoIntCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -2, -21, addScToInput, outputInts1));
1233 cases.push_back(SpecConstantTwoIntCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, -3, addScToInput, outputInts1));
1234 cases.push_back(SpecConstantTwoIntCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 3, addScToInput, outputInts1));
1235 cases.push_back(SpecConstantTwoIntCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
1236 cases.push_back(SpecConstantTwoIntCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 7, 3, addScToInput, outputInts4));
1237 cases.push_back(SpecConstantTwoIntCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 342, 50, addScToInput, outputInts1));
1238 cases.push_back(SpecConstantTwoIntCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 42, 63, addScToInput, outputInts1));
1239 cases.push_back(SpecConstantTwoIntCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 34, 8, addScToInput, outputInts1));
1240 cases.push_back(SpecConstantTwoIntCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 18, 56, addScToInput, outputInts1));
1241 cases.push_back(SpecConstantTwoIntCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
1242 cases.push_back(SpecConstantTwoIntCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", 168, 2, addScToInput, outputInts1));
1243 cases.push_back(SpecConstantTwoIntCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 21, 1, addScToInput, outputInts1));
1244 cases.push_back(SpecConstantTwoIntCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputInts2));
1245 cases.push_back(SpecConstantTwoIntCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputInts2));
1246 cases.push_back(SpecConstantTwoIntCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
1247 cases.push_back(SpecConstantTwoIntCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputInts2));
1248 cases.push_back(SpecConstantTwoIntCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputInts2));
1249 cases.push_back(SpecConstantTwoIntCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputInts2));
1250 cases.push_back(SpecConstantTwoIntCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputInts2));
1251 cases.push_back(SpecConstantTwoIntCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputInts2));
1252 cases.push_back(SpecConstantTwoIntCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputInts2));
1253 cases.push_back(SpecConstantTwoIntCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
1254 cases.push_back(SpecConstantTwoIntCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
1255 cases.push_back(SpecConstantTwoIntCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputInts2));
1256 cases.push_back(SpecConstantTwoIntCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputInts2));
1257 cases.push_back(SpecConstantTwoIntCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -42, 0, addScToInput, outputInts1));
1258 cases.push_back(SpecConstantTwoIntCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -43, 0, addScToInput, outputInts1));
1259 cases.push_back(SpecConstantTwoIntCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputInts2));
1260 cases.push_back(SpecConstantTwoIntCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %zero", 1, 42, addScToInput, outputInts1));
1261 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
1263 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
1265 map<string, string> specializations;
1266 ComputeShaderSpec spec;
1268 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
1269 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
1270 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
1271 specializations["SC_OP"] = cases[caseNdx].scOperation;
1272 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
1274 spec.assembly = shaderTemplate.specialize(specializations);
1275 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
1276 spec.outputs.push_back(BufferSp(new Int32Buffer(cases[caseNdx].expectedOutput)));
1277 spec.numWorkGroups = IVec3(numElements, 1, 1);
1278 spec.specConstants.push_back(cases[caseNdx].scActualValue0);
1279 spec.specConstants.push_back(cases[caseNdx].scActualValue1);
1281 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].caseName, cases[caseNdx].caseName, spec));
1284 ComputeShaderSpec spec;
1287 string(s_ShaderPreamble) +
1289 "OpName %main \"main\"\n"
1290 "OpName %id \"gl_GlobalInvocationID\"\n"
1292 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1293 "OpDecorate %sc_0 SpecId 0\n"
1294 "OpDecorate %sc_1 SpecId 1\n"
1295 "OpDecorate %sc_2 SpecId 2\n"
1297 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1299 "%ivec3 = OpTypeVector %i32 3\n"
1300 "%i32ptr = OpTypePointer Uniform %i32\n"
1301 "%i32arr = OpTypeRuntimeArray %i32\n"
1302 "%boolptr = OpTypePointer Uniform %bool\n"
1303 "%boolarr = OpTypeRuntimeArray %bool\n"
1304 "%inbuf = OpTypeStruct %i32arr\n"
1305 "%inbufptr = OpTypePointer Uniform %inbuf\n"
1306 "%indata = OpVariable %inbufptr Uniform\n"
1307 "%outbuf = OpTypeStruct %i32arr\n"
1308 "%outbufptr = OpTypePointer Uniform %outbuf\n"
1309 "%outdata = OpVariable %outbufptr Uniform\n"
1311 "%id = OpVariable %uvec3ptr Input\n"
1312 "%zero = OpConstant %i32 0\n"
1313 "%ivec3_0 = OpConstantComposite %ivec3 %zero %zero %zero\n"
1315 "%sc_0 = OpSpecConstant %i32 0\n"
1316 "%sc_1 = OpSpecConstant %i32 0\n"
1317 "%sc_2 = OpSpecConstant %i32 0\n"
1318 "%sc_vec3_0 = OpSpecConstantOp %ivec3 CompositeInsert %sc_0 %ivec3_0 0\n" // (sc_0, 0, 0)
1319 "%sc_vec3_1 = OpSpecConstantOp %ivec3 CompositeInsert %sc_1 %ivec3_0 1\n" // (0, sc_1, 0)
1320 "%sc_vec3_2 = OpSpecConstantOp %ivec3 CompositeInsert %sc_2 %ivec3_0 2\n" // (0, 0, sc_2)
1321 "%sc_vec3_01 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_0 %sc_vec3_1 1 0 4\n" // (0, sc_0, sc_1)
1322 "%sc_vec3_012 = OpSpecConstantOp %ivec3 VectorShuffle %sc_vec3_01 %sc_vec3_2 5 1 2\n" // (sc_2, sc_0, sc_1)
1323 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
1324 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
1325 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
1326 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
1327 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n" // (sc_2 - sc_0) * sc_1
1329 "%main = OpFunction %void None %voidf\n"
1330 "%label = OpLabel\n"
1331 "%idval = OpLoad %uvec3 %id\n"
1332 "%x = OpCompositeExtract %u32 %idval 0\n"
1333 "%inloc = OpAccessChain %i32ptr %indata %zero %x\n"
1334 "%inval = OpLoad %i32 %inloc\n"
1335 "%final = OpIAdd %i32 %inval %sc_final\n"
1336 "%outloc = OpAccessChain %i32ptr %outdata %zero %x\n"
1337 " OpStore %outloc %final\n"
1340 spec.inputs.push_back(BufferSp(new Int32Buffer(inputInts)));
1341 spec.outputs.push_back(BufferSp(new Int32Buffer(outputInts3)));
1342 spec.numWorkGroups = IVec3(numElements, 1, 1);
1343 spec.specConstants.push_back(123);
1344 spec.specConstants.push_back(56);
1345 spec.specConstants.push_back(-77);
1347 group->addChild(new SpvAsmComputeShaderCase(testCtx, "vector_related", "VectorShuffle, CompositeExtract, & CompositeInsert", spec));
1349 return group.release();
1352 tcu::TestCaseGroup* createOpPhiGroup (tcu::TestContext& testCtx)
1354 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
1355 ComputeShaderSpec spec1;
1356 ComputeShaderSpec spec2;
1357 ComputeShaderSpec spec3;
1358 de::Random rnd (deStringHash(group->getName()));
1359 const int numElements = 100;
1360 vector<float> inputFloats (numElements, 0);
1361 vector<float> outputFloats1 (numElements, 0);
1362 vector<float> outputFloats2 (numElements, 0);
1363 vector<float> outputFloats3 (numElements, 0);
1365 fillRandomScalars(rnd, -300.f, 300.f, &inputFloats[0], numElements);
1367 for (size_t ndx = 0; ndx < numElements; ++ndx)
1371 case 0: outputFloats1[ndx] = inputFloats[ndx] + 5.5f; break;
1372 case 1: outputFloats1[ndx] = inputFloats[ndx] + 20.5f; break;
1373 case 2: outputFloats1[ndx] = inputFloats[ndx] + 1.75f; break;
1376 outputFloats2[ndx] = inputFloats[ndx] + 6.5f * 3;
1377 outputFloats3[ndx] = 8.5f - inputFloats[ndx];
1381 string(s_ShaderPreamble) +
1383 "OpSource GLSL 430\n"
1384 "OpName %main \"main\"\n"
1385 "OpName %id \"gl_GlobalInvocationID\"\n"
1387 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1389 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1391 "%id = OpVariable %uvec3ptr Input\n"
1392 "%zero = OpConstant %i32 0\n"
1393 "%three = OpConstant %u32 3\n"
1394 "%constf5p5 = OpConstant %f32 5.5\n"
1395 "%constf20p5 = OpConstant %f32 20.5\n"
1396 "%constf1p75 = OpConstant %f32 1.75\n"
1397 "%constf8p5 = OpConstant %f32 8.5\n"
1398 "%constf6p5 = OpConstant %f32 6.5\n"
1400 "%main = OpFunction %void None %voidf\n"
1401 "%entry = OpLabel\n"
1402 "%idval = OpLoad %uvec3 %id\n"
1403 "%x = OpCompositeExtract %u32 %idval 0\n"
1404 "%selector = OpUMod %u32 %x %three\n"
1405 " OpSelectionMerge %phi None\n"
1406 " OpSwitch %selector %default 0 %case0 1 %case1 2 %case2\n"
1408 // Case 1 before OpPhi.
1409 "%case1 = OpLabel\n"
1412 "%default = OpLabel\n"
1416 "%operand = OpPhi %f32 %constf1p75 %case2 %constf20p5 %case1 %constf5p5 %case0\n" // not in the order of blocks
1417 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1418 "%inval = OpLoad %f32 %inloc\n"
1419 "%add = OpFAdd %f32 %inval %operand\n"
1420 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1421 " OpStore %outloc %add\n"
1424 // Case 0 after OpPhi.
1425 "%case0 = OpLabel\n"
1429 // Case 2 after OpPhi.
1430 "%case2 = OpLabel\n"
1434 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1435 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
1436 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1438 group->addChild(new SpvAsmComputeShaderCase(testCtx, "block", "out-of-order and unreachable blocks for OpPhi", spec1));
1441 string(s_ShaderPreamble) +
1443 "OpName %main \"main\"\n"
1444 "OpName %id \"gl_GlobalInvocationID\"\n"
1446 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1448 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1450 "%id = OpVariable %uvec3ptr Input\n"
1451 "%zero = OpConstant %i32 0\n"
1452 "%one = OpConstant %i32 1\n"
1453 "%three = OpConstant %i32 3\n"
1454 "%constf6p5 = OpConstant %f32 6.5\n"
1456 "%main = OpFunction %void None %voidf\n"
1457 "%entry = OpLabel\n"
1458 "%idval = OpLoad %uvec3 %id\n"
1459 "%x = OpCompositeExtract %u32 %idval 0\n"
1460 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1461 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1462 "%inval = OpLoad %f32 %inloc\n"
1466 "%step = OpPhi %i32 %zero %entry %step_next %phi\n"
1467 "%accum = OpPhi %f32 %inval %entry %accum_next %phi\n"
1468 "%step_next = OpIAdd %i32 %step %one\n"
1469 "%accum_next = OpFAdd %f32 %accum %constf6p5\n"
1470 "%still_loop = OpSLessThan %bool %step %three\n"
1471 " OpLoopMerge %exit %phi None\n"
1472 " OpBranchConditional %still_loop %phi %exit\n"
1475 " OpStore %outloc %accum\n"
1478 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1479 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1480 spec2.numWorkGroups = IVec3(numElements, 1, 1);
1482 group->addChild(new SpvAsmComputeShaderCase(testCtx, "induction", "The usual way induction variables are handled in LLVM IR", spec2));
1485 string(s_ShaderPreamble) +
1487 "OpName %main \"main\"\n"
1488 "OpName %id \"gl_GlobalInvocationID\"\n"
1490 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1492 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1494 "%f32ptr_f = OpTypePointer Function %f32\n"
1495 "%id = OpVariable %uvec3ptr Input\n"
1496 "%true = OpConstantTrue %bool\n"
1497 "%false = OpConstantFalse %bool\n"
1498 "%zero = OpConstant %i32 0\n"
1499 "%constf8p5 = OpConstant %f32 8.5\n"
1501 "%main = OpFunction %void None %voidf\n"
1502 "%entry = OpLabel\n"
1503 "%b = OpVariable %f32ptr_f Function %constf8p5\n"
1504 "%idval = OpLoad %uvec3 %id\n"
1505 "%x = OpCompositeExtract %u32 %idval 0\n"
1506 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1507 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1508 "%a_init = OpLoad %f32 %inloc\n"
1509 "%b_init = OpLoad %f32 %b\n"
1513 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
1514 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
1515 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
1516 " OpLoopMerge %exit %phi None\n"
1517 " OpBranchConditional %still_loop %phi %exit\n"
1520 "%sub = OpFSub %f32 %a_next %b_next\n"
1521 " OpStore %outloc %sub\n"
1524 spec3.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1525 spec3.outputs.push_back(BufferSp(new Float32Buffer(outputFloats3)));
1526 spec3.numWorkGroups = IVec3(numElements, 1, 1);
1528 group->addChild(new SpvAsmComputeShaderCase(testCtx, "swap", "Swap the values of two variables using OpPhi", spec3));
1530 return group.release();
1533 // Assembly code used for testing block order is based on GLSL source code:
1537 // layout(std140, set = 0, binding = 0) readonly buffer Input {
1538 // float elements[];
1540 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
1541 // float elements[];
1545 // uint x = gl_GlobalInvocationID.x;
1546 // output_data.elements[x] = input_data.elements[x];
1547 // if (x > uint(50)) {
1548 // switch (x % uint(3)) {
1549 // case 0: output_data.elements[x] += 1.5f; break;
1550 // case 1: output_data.elements[x] += 42.f; break;
1551 // case 2: output_data.elements[x] -= 27.f; break;
1555 // output_data.elements[x] = -input_data.elements[x];
1558 tcu::TestCaseGroup* createBlockOrderGroup (tcu::TestContext& testCtx)
1560 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "block_order", "Test block orders"));
1561 ComputeShaderSpec spec;
1562 de::Random rnd (deStringHash(group->getName()));
1563 const int numElements = 100;
1564 vector<float> inputFloats (numElements, 0);
1565 vector<float> outputFloats (numElements, 0);
1567 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
1569 for (size_t ndx = 0; ndx <= 50; ++ndx)
1570 outputFloats[ndx] = -inputFloats[ndx];
1572 for (size_t ndx = 51; ndx < numElements; ++ndx)
1576 case 0: outputFloats[ndx] = inputFloats[ndx] + 1.5f; break;
1577 case 1: outputFloats[ndx] = inputFloats[ndx] + 42.f; break;
1578 case 2: outputFloats[ndx] = inputFloats[ndx] - 27.f; break;
1584 string(s_ShaderPreamble) +
1586 "OpSource GLSL 430\n"
1587 "OpName %main \"main\"\n"
1588 "OpName %id \"gl_GlobalInvocationID\"\n"
1590 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1592 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
1594 "%u32ptr = OpTypePointer Function %u32\n"
1595 "%u32ptr_input = OpTypePointer Input %u32\n"
1597 + string(s_InputOutputBuffer) +
1599 "%id = OpVariable %uvec3ptr Input\n"
1600 "%zero = OpConstant %i32 0\n"
1601 "%const3 = OpConstant %u32 3\n"
1602 "%const50 = OpConstant %u32 50\n"
1603 "%constf1p5 = OpConstant %f32 1.5\n"
1604 "%constf27 = OpConstant %f32 27.0\n"
1605 "%constf42 = OpConstant %f32 42.0\n"
1607 "%main = OpFunction %void None %voidf\n"
1610 "%entry = OpLabel\n"
1612 // Create a temporary variable to hold the value of gl_GlobalInvocationID.x.
1613 "%xvar = OpVariable %u32ptr Function\n"
1614 "%xptr = OpAccessChain %u32ptr_input %id %zero\n"
1615 "%x = OpLoad %u32 %xptr\n"
1616 " OpStore %xvar %x\n"
1618 "%cmp = OpUGreaterThan %bool %x %const50\n"
1619 " OpSelectionMerge %if_merge None\n"
1620 " OpBranchConditional %cmp %if_true %if_false\n"
1622 // Merge block for switch-statement: placed at the beginning.
1623 "%switch_merge = OpLabel\n"
1624 " OpBranch %if_merge\n"
1626 // Case 1 for switch-statement.
1627 "%case1 = OpLabel\n"
1628 "%x_1 = OpLoad %u32 %xvar\n"
1629 "%inloc_1 = OpAccessChain %f32ptr %indata %zero %x_1\n"
1630 "%inval_1 = OpLoad %f32 %inloc_1\n"
1631 "%addf42 = OpFAdd %f32 %inval_1 %constf42\n"
1632 "%outloc_1 = OpAccessChain %f32ptr %outdata %zero %x_1\n"
1633 " OpStore %outloc_1 %addf42\n"
1634 " OpBranch %switch_merge\n"
1636 // False branch for if-statement: placed in the middle of switch cases and before true branch.
1637 "%if_false = OpLabel\n"
1638 "%x_f = OpLoad %u32 %xvar\n"
1639 "%inloc_f = OpAccessChain %f32ptr %indata %zero %x_f\n"
1640 "%inval_f = OpLoad %f32 %inloc_f\n"
1641 "%negate = OpFNegate %f32 %inval_f\n"
1642 "%outloc_f = OpAccessChain %f32ptr %outdata %zero %x_f\n"
1643 " OpStore %outloc_f %negate\n"
1644 " OpBranch %if_merge\n"
1646 // Merge block for if-statement: placed in the middle of true and false branch.
1647 "%if_merge = OpLabel\n"
1650 // True branch for if-statement: placed in the middle of swtich cases and after the false branch.
1651 "%if_true = OpLabel\n"
1652 "%xval_t = OpLoad %u32 %xvar\n"
1653 "%mod = OpUMod %u32 %xval_t %const3\n"
1654 " OpSelectionMerge %switch_merge None\n"
1655 " OpSwitch %mod %default 0 %case0 1 %case1 2 %case2\n"
1657 // Case 2 for switch-statement.
1658 "%case2 = OpLabel\n"
1659 "%x_2 = OpLoad %u32 %xvar\n"
1660 "%inloc_2 = OpAccessChain %f32ptr %indata %zero %x_2\n"
1661 "%inval_2 = OpLoad %f32 %inloc_2\n"
1662 "%subf27 = OpFSub %f32 %inval_2 %constf27\n"
1663 "%outloc_2 = OpAccessChain %f32ptr %outdata %zero %x_2\n"
1664 " OpStore %outloc_2 %subf27\n"
1665 " OpBranch %switch_merge\n"
1667 // Default case for switch-statement: placed in the middle of normal cases.
1668 "%default = OpLabel\n"
1669 " OpBranch %switch_merge\n"
1671 // Case 0 for switch-statement: out of order.
1672 "%case0 = OpLabel\n"
1673 "%x_0 = OpLoad %u32 %xvar\n"
1674 "%inloc_0 = OpAccessChain %f32ptr %indata %zero %x_0\n"
1675 "%inval_0 = OpLoad %f32 %inloc_0\n"
1676 "%addf1p5 = OpFAdd %f32 %inval_0 %constf1p5\n"
1677 "%outloc_0 = OpAccessChain %f32ptr %outdata %zero %x_0\n"
1678 " OpStore %outloc_0 %addf1p5\n"
1679 " OpBranch %switch_merge\n"
1682 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1683 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1684 spec.numWorkGroups = IVec3(numElements, 1, 1);
1686 group->addChild(new SpvAsmComputeShaderCase(testCtx, "all", "various out-of-order blocks", spec));
1688 return group.release();
1691 tcu::TestCaseGroup* createMultipleShaderGroup (tcu::TestContext& testCtx)
1693 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "multiple_shaders", "Test multiple shaders in the same module"));
1694 ComputeShaderSpec spec1;
1695 ComputeShaderSpec spec2;
1696 de::Random rnd (deStringHash(group->getName()));
1697 const int numElements = 100;
1698 vector<float> inputFloats (numElements, 0);
1699 vector<float> outputFloats1 (numElements, 0);
1700 vector<float> outputFloats2 (numElements, 0);
1701 fillRandomScalars(rnd, -500.f, 500.f, &inputFloats[0], numElements);
1703 for (size_t ndx = 0; ndx < numElements; ++ndx)
1705 outputFloats1[ndx] = inputFloats[ndx] + inputFloats[ndx];
1706 outputFloats2[ndx] = -inputFloats[ndx];
1709 const string assembly(
1710 "OpCapability Shader\n"
1711 "OpMemoryModel Logical GLSL450\n"
1712 "OpEntryPoint GLCompute %comp_main1 \"entrypoint1\" %id\n"
1713 "OpEntryPoint GLCompute %comp_main2 \"entrypoint2\" %id\n"
1714 // A module cannot have two OpEntryPoint instructions with the same Execution Model and the same Name string.
1715 "OpEntryPoint Vertex %vert_main \"entrypoint2\" %vert_builtins %vertexID %instanceID\n"
1716 "OpExecutionMode %vert_main LocalSize 1 1 1\n"
1718 "OpName %comp_main1 \"entrypoint1\"\n"
1719 "OpName %comp_main2 \"entrypoint2\"\n"
1720 "OpName %vert_main \"entrypoint2\"\n"
1721 "OpName %id \"gl_GlobalInvocationID\"\n"
1722 "OpName %vert_builtin_st \"gl_PerVertex\"\n"
1723 "OpName %vertexID \"gl_VertexIndex\"\n"
1724 "OpName %instanceID \"gl_InstanceIndex\"\n"
1725 "OpMemberName %vert_builtin_st 0 \"gl_Position\"\n"
1726 "OpMemberName %vert_builtin_st 1 \"gl_PointSize\"\n"
1727 "OpMemberName %vert_builtin_st 2 \"gl_ClipDistance\"\n"
1729 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1730 "OpDecorate %vertexID BuiltIn VertexIndex\n"
1731 "OpDecorate %instanceID BuiltIn InstanceIndex\n"
1732 "OpDecorate %vert_builtin_st Block\n"
1733 "OpMemberDecorate %vert_builtin_st 0 BuiltIn Position\n"
1734 "OpMemberDecorate %vert_builtin_st 1 BuiltIn PointSize\n"
1735 "OpMemberDecorate %vert_builtin_st 2 BuiltIn ClipDistance\n"
1737 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1739 "%i32ptr = OpTypePointer Input %i32\n"
1740 "%vec4 = OpTypeVector %f32 4\n"
1741 "%vec4ptr = OpTypePointer Output %vec4\n"
1742 "%f32arr1 = OpTypeArray %f32 %one\n"
1743 "%vert_builtin_st = OpTypeStruct %vec4 %f32 %f32arr1\n"
1744 "%vert_builtin_st_ptr = OpTypePointer Output %vert_builtin_st\n"
1745 "%vert_builtins = OpVariable %vert_builtin_st_ptr Output\n"
1747 "%id = OpVariable %uvec3ptr Input\n"
1748 "%vertexID = OpVariable %i32ptr Input\n"
1749 "%instanceID = OpVariable %i32ptr Input\n"
1750 "%zero = OpConstant %i32 0\n"
1751 "%one = OpConstant %u32 1\n"
1752 "%c_f32_1 = OpConstant %f32 1\n"
1753 "%c_vec4_1 = OpConstantComposite %vec4 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
1755 // gl_Position = vec4(1.);
1756 "%vert_main = OpFunction %void None %voidf\n"
1757 "%vert_entry = OpLabel\n"
1758 "%position = OpAccessChain %vec4ptr %vert_builtins %zero\n"
1759 " OpStore %position %c_vec4_1\n"
1764 "%comp_main1 = OpFunction %void None %voidf\n"
1765 "%comp1_entry = OpLabel\n"
1766 "%idval1 = OpLoad %uvec3 %id\n"
1767 "%x1 = OpCompositeExtract %u32 %idval1 0\n"
1768 "%inloc1 = OpAccessChain %f32ptr %indata %zero %x1\n"
1769 "%inval1 = OpLoad %f32 %inloc1\n"
1770 "%add = OpFAdd %f32 %inval1 %inval1\n"
1771 "%outloc1 = OpAccessChain %f32ptr %outdata %zero %x1\n"
1772 " OpStore %outloc1 %add\n"
1777 "%comp_main2 = OpFunction %void None %voidf\n"
1778 "%comp2_entry = OpLabel\n"
1779 "%idval2 = OpLoad %uvec3 %id\n"
1780 "%x2 = OpCompositeExtract %u32 %idval2 0\n"
1781 "%inloc2 = OpAccessChain %f32ptr %indata %zero %x2\n"
1782 "%inval2 = OpLoad %f32 %inloc2\n"
1783 "%neg = OpFNegate %f32 %inval2\n"
1784 "%outloc2 = OpAccessChain %f32ptr %outdata %zero %x2\n"
1785 " OpStore %outloc2 %neg\n"
1787 " OpFunctionEnd\n");
1789 spec1.assembly = assembly;
1790 spec1.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1791 spec1.outputs.push_back(BufferSp(new Float32Buffer(outputFloats1)));
1792 spec1.numWorkGroups = IVec3(numElements, 1, 1);
1793 spec1.entryPoint = "entrypoint1";
1795 spec2.assembly = assembly;
1796 spec2.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1797 spec2.outputs.push_back(BufferSp(new Float32Buffer(outputFloats2)));
1798 spec2.numWorkGroups = IVec3(numElements, 1, 1);
1799 spec2.entryPoint = "entrypoint2";
1801 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader1", "multiple shaders in the same module", spec1));
1802 group->addChild(new SpvAsmComputeShaderCase(testCtx, "shader2", "multiple shaders in the same module", spec2));
1804 return group.release();
1807 inline std::string makeLongUTF8String (size_t num4ByteChars)
1809 // An example of a longest valid UTF-8 character. Be explicit about the
1810 // character type because Microsoft compilers can otherwise interpret the
1811 // character string as being over wide (16-bit) characters. Ideally, we
1812 // would just use a C++11 UTF-8 string literal, but we want to support older
1813 // Microsoft compilers.
1814 const std::basic_string<char> earthAfrica("\xF0\x9F\x8C\x8D");
1815 std::string longString;
1816 longString.reserve(num4ByteChars * 4);
1817 for (size_t count = 0; count < num4ByteChars; count++)
1819 longString += earthAfrica;
1824 tcu::TestCaseGroup* createOpSourceGroup (tcu::TestContext& testCtx)
1826 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsource", "Tests the OpSource & OpSourceContinued instruction"));
1827 vector<CaseParameter> cases;
1828 de::Random rnd (deStringHash(group->getName()));
1829 const int numElements = 100;
1830 vector<float> positiveFloats (numElements, 0);
1831 vector<float> negativeFloats (numElements, 0);
1832 const StringTemplate shaderTemplate (
1833 "OpCapability Shader\n"
1834 "OpMemoryModel Logical GLSL450\n"
1836 "OpEntryPoint GLCompute %main \"main\" %id\n"
1837 "OpExecutionMode %main LocalSize 1 1 1\n"
1841 "OpName %main \"main\"\n"
1842 "OpName %id \"gl_GlobalInvocationID\"\n"
1844 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1846 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1848 "%id = OpVariable %uvec3ptr Input\n"
1849 "%zero = OpConstant %i32 0\n"
1851 "%main = OpFunction %void None %voidf\n"
1852 "%label = OpLabel\n"
1853 "%idval = OpLoad %uvec3 %id\n"
1854 "%x = OpCompositeExtract %u32 %idval 0\n"
1855 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1856 "%inval = OpLoad %f32 %inloc\n"
1857 "%neg = OpFNegate %f32 %inval\n"
1858 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1859 " OpStore %outloc %neg\n"
1861 " OpFunctionEnd\n");
1863 cases.push_back(CaseParameter("unknown_source", "OpSource Unknown 0"));
1864 cases.push_back(CaseParameter("wrong_source", "OpSource OpenCL_C 210"));
1865 cases.push_back(CaseParameter("normal_filename", "%fname = OpString \"filename\"\n"
1866 "OpSource GLSL 430 %fname"));
1867 cases.push_back(CaseParameter("empty_filename", "%fname = OpString \"\"\n"
1868 "OpSource GLSL 430 %fname"));
1869 cases.push_back(CaseParameter("normal_source_code", "%fname = OpString \"filename\"\n"
1870 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\""));
1871 cases.push_back(CaseParameter("empty_source_code", "%fname = OpString \"filename\"\n"
1872 "OpSource GLSL 430 %fname \"\""));
1873 cases.push_back(CaseParameter("long_source_code", "%fname = OpString \"filename\"\n"
1874 "OpSource GLSL 430 %fname \"" + makeLongUTF8String(65530) + "ccc\"")); // word count: 65535
1875 cases.push_back(CaseParameter("utf8_source_code", "%fname = OpString \"filename\"\n"
1876 "OpSource GLSL 430 %fname \"\xE2\x98\x82\xE2\x98\x85\"")); // umbrella & black star symbol
1877 cases.push_back(CaseParameter("normal_sourcecontinued", "%fname = OpString \"filename\"\n"
1878 "OpSource GLSL 430 %fname \"#version 430\nvo\"\n"
1879 "OpSourceContinued \"id main() {}\""));
1880 cases.push_back(CaseParameter("empty_sourcecontinued", "%fname = OpString \"filename\"\n"
1881 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
1882 "OpSourceContinued \"\""));
1883 cases.push_back(CaseParameter("long_sourcecontinued", "%fname = OpString \"filename\"\n"
1884 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
1885 "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\"")); // word count: 65535
1886 cases.push_back(CaseParameter("utf8_sourcecontinued", "%fname = OpString \"filename\"\n"
1887 "OpSource GLSL 430 %fname \"#version 430\nvoid main() {}\"\n"
1888 "OpSourceContinued \"\xE2\x98\x8E\xE2\x9A\x91\"")); // white telephone & black flag symbol
1889 cases.push_back(CaseParameter("multi_sourcecontinued", "%fname = OpString \"filename\"\n"
1890 "OpSource GLSL 430 %fname \"#version 430\n\"\n"
1891 "OpSourceContinued \"void\"\n"
1892 "OpSourceContinued \"main()\"\n"
1893 "OpSourceContinued \"{}\""));
1894 cases.push_back(CaseParameter("empty_source_before_sourcecontinued", "%fname = OpString \"filename\"\n"
1895 "OpSource GLSL 430 %fname \"\"\n"
1896 "OpSourceContinued \"#version 430\nvoid main() {}\""));
1898 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
1900 for (size_t ndx = 0; ndx < numElements; ++ndx)
1901 negativeFloats[ndx] = -positiveFloats[ndx];
1903 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
1905 map<string, string> specializations;
1906 ComputeShaderSpec spec;
1908 specializations["SOURCE"] = cases[caseNdx].param;
1909 spec.assembly = shaderTemplate.specialize(specializations);
1910 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
1911 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
1912 spec.numWorkGroups = IVec3(numElements, 1, 1);
1914 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
1917 return group.release();
1920 tcu::TestCaseGroup* createOpSourceExtensionGroup (tcu::TestContext& testCtx)
1922 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opsourceextension", "Tests the OpSource instruction"));
1923 vector<CaseParameter> cases;
1924 de::Random rnd (deStringHash(group->getName()));
1925 const int numElements = 100;
1926 vector<float> inputFloats (numElements, 0);
1927 vector<float> outputFloats (numElements, 0);
1928 const StringTemplate shaderTemplate (
1929 string(s_ShaderPreamble) +
1931 "OpSourceExtension \"${EXTENSION}\"\n"
1933 "OpName %main \"main\"\n"
1934 "OpName %id \"gl_GlobalInvocationID\"\n"
1936 "OpDecorate %id BuiltIn GlobalInvocationId\n"
1938 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
1940 "%id = OpVariable %uvec3ptr Input\n"
1941 "%zero = OpConstant %i32 0\n"
1943 "%main = OpFunction %void None %voidf\n"
1944 "%label = OpLabel\n"
1945 "%idval = OpLoad %uvec3 %id\n"
1946 "%x = OpCompositeExtract %u32 %idval 0\n"
1947 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
1948 "%inval = OpLoad %f32 %inloc\n"
1949 "%neg = OpFNegate %f32 %inval\n"
1950 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
1951 " OpStore %outloc %neg\n"
1953 " OpFunctionEnd\n");
1955 cases.push_back(CaseParameter("empty_extension", ""));
1956 cases.push_back(CaseParameter("real_extension", "GL_ARB_texture_rectangle"));
1957 cases.push_back(CaseParameter("fake_extension", "GL_ARB_im_the_ultimate_extension"));
1958 cases.push_back(CaseParameter("utf8_extension", "GL_ARB_\xE2\x98\x82\xE2\x98\x85"));
1959 cases.push_back(CaseParameter("long_extension", makeLongUTF8String(65533) + "ccc")); // word count: 65535
1961 fillRandomScalars(rnd, -200.f, 200.f, &inputFloats[0], numElements);
1963 for (size_t ndx = 0; ndx < numElements; ++ndx)
1964 outputFloats[ndx] = -inputFloats[ndx];
1966 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
1968 map<string, string> specializations;
1969 ComputeShaderSpec spec;
1971 specializations["EXTENSION"] = cases[caseNdx].param;
1972 spec.assembly = shaderTemplate.specialize(specializations);
1973 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
1974 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
1975 spec.numWorkGroups = IVec3(numElements, 1, 1);
1977 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
1980 return group.release();
1983 // Checks that a compute shader can generate a constant null value of various types, without exercising a computation on it.
1984 tcu::TestCaseGroup* createOpConstantNullGroup (tcu::TestContext& testCtx)
1986 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnull", "Tests the OpConstantNull instruction"));
1987 vector<CaseParameter> cases;
1988 de::Random rnd (deStringHash(group->getName()));
1989 const int numElements = 100;
1990 vector<float> positiveFloats (numElements, 0);
1991 vector<float> negativeFloats (numElements, 0);
1992 const StringTemplate shaderTemplate (
1993 string(s_ShaderPreamble) +
1995 "OpSource GLSL 430\n"
1996 "OpName %main \"main\"\n"
1997 "OpName %id \"gl_GlobalInvocationID\"\n"
1999 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2001 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2004 "%null = OpConstantNull %type\n"
2006 "%id = OpVariable %uvec3ptr Input\n"
2007 "%zero = OpConstant %i32 0\n"
2009 "%main = OpFunction %void None %voidf\n"
2010 "%label = OpLabel\n"
2011 "%idval = OpLoad %uvec3 %id\n"
2012 "%x = OpCompositeExtract %u32 %idval 0\n"
2013 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2014 "%inval = OpLoad %f32 %inloc\n"
2015 "%neg = OpFNegate %f32 %inval\n"
2016 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2017 " OpStore %outloc %neg\n"
2019 " OpFunctionEnd\n");
2021 cases.push_back(CaseParameter("bool", "%type = OpTypeBool"));
2022 cases.push_back(CaseParameter("sint32", "%type = OpTypeInt 32 1"));
2023 cases.push_back(CaseParameter("uint32", "%type = OpTypeInt 32 0"));
2024 cases.push_back(CaseParameter("float32", "%type = OpTypeFloat 32"));
2025 cases.push_back(CaseParameter("vec4float32", "%type = OpTypeVector %f32 4"));
2026 cases.push_back(CaseParameter("vec3bool", "%type = OpTypeVector %bool 3"));
2027 cases.push_back(CaseParameter("vec2uint32", "%type = OpTypeVector %u32 2"));
2028 cases.push_back(CaseParameter("matrix", "%type = OpTypeMatrix %uvec3 3"));
2029 cases.push_back(CaseParameter("array", "%100 = OpConstant %u32 100\n"
2030 "%type = OpTypeArray %i32 %100"));
2031 cases.push_back(CaseParameter("runtimearray", "%type = OpTypeRuntimeArray %f32"));
2032 cases.push_back(CaseParameter("struct", "%type = OpTypeStruct %f32 %i32 %u32"));
2033 cases.push_back(CaseParameter("pointer", "%type = OpTypePointer Function %i32"));
2035 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2037 for (size_t ndx = 0; ndx < numElements; ++ndx)
2038 negativeFloats[ndx] = -positiveFloats[ndx];
2040 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2042 map<string, string> specializations;
2043 ComputeShaderSpec spec;
2045 specializations["TYPE"] = cases[caseNdx].param;
2046 spec.assembly = shaderTemplate.specialize(specializations);
2047 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2048 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2049 spec.numWorkGroups = IVec3(numElements, 1, 1);
2051 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2054 return group.release();
2057 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
2058 tcu::TestCaseGroup* createOpConstantCompositeGroup (tcu::TestContext& testCtx)
2060 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "Tests the OpConstantComposite instruction"));
2061 vector<CaseParameter> cases;
2062 de::Random rnd (deStringHash(group->getName()));
2063 const int numElements = 100;
2064 vector<float> positiveFloats (numElements, 0);
2065 vector<float> negativeFloats (numElements, 0);
2066 const StringTemplate shaderTemplate (
2067 string(s_ShaderPreamble) +
2069 "OpSource GLSL 430\n"
2070 "OpName %main \"main\"\n"
2071 "OpName %id \"gl_GlobalInvocationID\"\n"
2073 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2075 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2077 "%id = OpVariable %uvec3ptr Input\n"
2078 "%zero = OpConstant %i32 0\n"
2082 "%main = OpFunction %void None %voidf\n"
2083 "%label = OpLabel\n"
2084 "%idval = OpLoad %uvec3 %id\n"
2085 "%x = OpCompositeExtract %u32 %idval 0\n"
2086 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2087 "%inval = OpLoad %f32 %inloc\n"
2088 "%neg = OpFNegate %f32 %inval\n"
2089 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2090 " OpStore %outloc %neg\n"
2092 " OpFunctionEnd\n");
2094 cases.push_back(CaseParameter("vector", "%five = OpConstant %u32 5\n"
2095 "%const = OpConstantComposite %uvec3 %five %zero %five"));
2096 cases.push_back(CaseParameter("matrix", "%m3uvec3 = OpTypeMatrix %uvec3 3\n"
2097 "%ten = OpConstant %u32 10\n"
2098 "%vec = OpConstantComposite %uvec3 %ten %zero %ten\n"
2099 "%mat = OpConstantComposite %m3uvec3 %vec %vec %vec"));
2100 cases.push_back(CaseParameter("struct", "%m2vec3 = OpTypeMatrix %uvec3 2\n"
2101 "%struct = OpTypeStruct %u32 %f32 %uvec3 %m2vec3\n"
2102 "%one = OpConstant %u32 1\n"
2103 "%point5 = OpConstant %f32 0.5\n"
2104 "%vec = OpConstantComposite %uvec3 %one %one %zero\n"
2105 "%mat = OpConstantComposite %m2vec3 %vec %vec\n"
2106 "%const = OpConstantComposite %struct %one %point5 %vec %mat"));
2107 cases.push_back(CaseParameter("nested_struct", "%st1 = OpTypeStruct %u32 %f32\n"
2108 "%st2 = OpTypeStruct %i32 %i32\n"
2109 "%struct = OpTypeStruct %st1 %st2\n"
2110 "%point5 = OpConstant %f32 0.5\n"
2111 "%one = OpConstant %u32 1\n"
2112 "%ten = OpConstant %i32 10\n"
2113 "%st1val = OpConstantComposite %st1 %one %point5\n"
2114 "%st2val = OpConstantComposite %st2 %ten %ten\n"
2115 "%const = OpConstantComposite %struct %st1val %st2val"));
2117 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2119 for (size_t ndx = 0; ndx < numElements; ++ndx)
2120 negativeFloats[ndx] = -positiveFloats[ndx];
2122 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2124 map<string, string> specializations;
2125 ComputeShaderSpec spec;
2127 specializations["CONSTANT"] = cases[caseNdx].param;
2128 spec.assembly = shaderTemplate.specialize(specializations);
2129 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2130 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2131 spec.numWorkGroups = IVec3(numElements, 1, 1);
2133 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2136 return group.release();
2139 // Creates a floating point number with the given exponent, and significand
2140 // bits set. It can only create normalized numbers. Only the least significant
2141 // 24 bits of the significand will be examined. The final bit of the
2142 // significand will also be ignored. This allows alignment to be written
2143 // similarly to C99 hex-floats.
2144 // For example if you wanted to write 0x1.7f34p-12 you would call
2145 // constructNormalizedFloat(-12, 0x7f3400)
2146 float constructNormalizedFloat (deInt32 exponent, deUint32 significand)
2150 for (deInt32 idx = 0; idx < 23; ++idx)
2152 f += ((significand & 0x800000) == 0) ? 0.f : std::ldexp(1.0f, -idx);
2156 return std::ldexp(f, exponent);
2159 // Compare instruction for the OpQuantizeF16 compute exact case.
2160 // Returns true if the output is what is expected from the test case.
2161 bool compareOpQuantizeF16ComputeExactCase (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs)
2163 if (outputAllocs.size() != 1)
2166 // We really just need this for size because we cannot compare Nans.
2167 const BufferSp& expectedOutput = expectedOutputs[0];
2168 const float* outputAsFloat = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
2170 if (expectedOutput->getNumBytes() != 4*sizeof(float)) {
2174 if (*outputAsFloat != constructNormalizedFloat(8, 0x304000) &&
2175 *outputAsFloat != constructNormalizedFloat(8, 0x300000)) {
2179 if (*outputAsFloat != -constructNormalizedFloat(-7, 0x600000) &&
2180 *outputAsFloat != -constructNormalizedFloat(-7, 0x604000)) {
2184 if (*outputAsFloat != constructNormalizedFloat(2, 0x01C000) &&
2185 *outputAsFloat != constructNormalizedFloat(2, 0x020000)) {
2189 if (*outputAsFloat != constructNormalizedFloat(1, 0xFFC000) &&
2190 *outputAsFloat != constructNormalizedFloat(2, 0x000000)) {
2197 // Checks that every output from a test-case is a float NaN.
2198 bool compareNan (const std::vector<BufferSp>&, const vector<AllocationSp>& outputAllocs, const std::vector<BufferSp>& expectedOutputs)
2200 if (outputAllocs.size() != 1)
2203 // We really just need this for size because we cannot compare Nans.
2204 const BufferSp& expectedOutput = expectedOutputs[0];
2205 const float* output_as_float = static_cast<const float*>(outputAllocs[0]->getHostPtr());;
2207 for (size_t idx = 0; idx < expectedOutput->getNumBytes() / sizeof(float); ++idx)
2209 if (!isnan(output_as_float[idx]))
2218 // Checks that a compute shader can generate a constant composite value of various types, without exercising a computation on it.
2219 tcu::TestCaseGroup* createOpQuantizeToF16Group (tcu::TestContext& testCtx)
2221 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opquantize", "Tests the OpQuantizeToF16 instruction"));
2223 const std::string shader (
2224 string(s_ShaderPreamble) +
2226 "OpSource GLSL 430\n"
2227 "OpName %main \"main\"\n"
2228 "OpName %id \"gl_GlobalInvocationID\"\n"
2230 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2232 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2234 "%id = OpVariable %uvec3ptr Input\n"
2235 "%zero = OpConstant %i32 0\n"
2237 "%main = OpFunction %void None %voidf\n"
2238 "%label = OpLabel\n"
2239 "%idval = OpLoad %uvec3 %id\n"
2240 "%x = OpCompositeExtract %u32 %idval 0\n"
2241 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2242 "%inval = OpLoad %f32 %inloc\n"
2243 "%quant = OpQuantizeToF16 %f32 %inval\n"
2244 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2245 " OpStore %outloc %quant\n"
2247 " OpFunctionEnd\n");
2250 ComputeShaderSpec spec;
2251 const deUint32 numElements = 100;
2252 vector<float> infinities;
2253 vector<float> results;
2255 infinities.reserve(numElements);
2256 results.reserve(numElements);
2258 for (size_t idx = 0; idx < numElements; ++idx)
2263 infinities.push_back(std::numeric_limits<float>::infinity());
2264 results.push_back(std::numeric_limits<float>::infinity());
2267 infinities.push_back(-std::numeric_limits<float>::infinity());
2268 results.push_back(-std::numeric_limits<float>::infinity());
2271 infinities.push_back(std::ldexp(1.0f, 16));
2272 results.push_back(std::numeric_limits<float>::infinity());
2275 infinities.push_back(std::ldexp(-1.0f, 32));
2276 results.push_back(-std::numeric_limits<float>::infinity());
2281 spec.assembly = shader;
2282 spec.inputs.push_back(BufferSp(new Float32Buffer(infinities)));
2283 spec.outputs.push_back(BufferSp(new Float32Buffer(results)));
2284 spec.numWorkGroups = IVec3(numElements, 1, 1);
2286 group->addChild(new SpvAsmComputeShaderCase(
2287 testCtx, "infinities", "Check that infinities propagated and created", spec));
2291 ComputeShaderSpec spec;
2293 const deUint32 numElements = 100;
2295 nans.reserve(numElements);
2297 for (size_t idx = 0; idx < numElements; ++idx)
2301 nans.push_back(std::numeric_limits<float>::quiet_NaN());
2305 nans.push_back(-std::numeric_limits<float>::quiet_NaN());
2309 spec.assembly = shader;
2310 spec.inputs.push_back(BufferSp(new Float32Buffer(nans)));
2311 spec.outputs.push_back(BufferSp(new Float32Buffer(nans)));
2312 spec.numWorkGroups = IVec3(numElements, 1, 1);
2313 spec.verifyIO = &compareNan;
2315 group->addChild(new SpvAsmComputeShaderCase(
2316 testCtx, "propagated_nans", "Check that nans are propagated", spec));
2320 ComputeShaderSpec spec;
2321 vector<float> small;
2322 vector<float> zeros;
2323 const deUint32 numElements = 100;
2325 small.reserve(numElements);
2326 zeros.reserve(numElements);
2328 for (size_t idx = 0; idx < numElements; ++idx)
2333 small.push_back(0.f);
2334 zeros.push_back(0.f);
2337 small.push_back(-0.f);
2338 zeros.push_back(-0.f);
2341 small.push_back(std::ldexp(1.0f, -16));
2342 zeros.push_back(0.f);
2345 small.push_back(std::ldexp(-1.0f, -32));
2346 zeros.push_back(-0.f);
2349 small.push_back(std::ldexp(1.0f, -127));
2350 zeros.push_back(0.f);
2353 small.push_back(-std::ldexp(1.0f, -128));
2354 zeros.push_back(-0.f);
2359 spec.assembly = shader;
2360 spec.inputs.push_back(BufferSp(new Float32Buffer(small)));
2361 spec.outputs.push_back(BufferSp(new Float32Buffer(zeros)));
2362 spec.numWorkGroups = IVec3(numElements, 1, 1);
2364 group->addChild(new SpvAsmComputeShaderCase(
2365 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
2369 ComputeShaderSpec spec;
2370 vector<float> exact;
2371 const deUint32 numElements = 200;
2373 exact.reserve(numElements);
2375 for (size_t idx = 0; idx < numElements; ++idx)
2376 exact.push_back(static_cast<float>(idx - 100));
2378 spec.assembly = shader;
2379 spec.inputs.push_back(BufferSp(new Float32Buffer(exact)));
2380 spec.outputs.push_back(BufferSp(new Float32Buffer(exact)));
2381 spec.numWorkGroups = IVec3(numElements, 1, 1);
2383 group->addChild(new SpvAsmComputeShaderCase(
2384 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
2388 ComputeShaderSpec spec;
2389 vector<float> inputs;
2390 const deUint32 numElements = 4;
2392 inputs.push_back(constructNormalizedFloat(8, 0x300300));
2393 inputs.push_back(-constructNormalizedFloat(-7, 0x600800));
2394 inputs.push_back(constructNormalizedFloat(2, 0x01E000));
2395 inputs.push_back(constructNormalizedFloat(1, 0xFFE000));
2397 spec.assembly = shader;
2398 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
2399 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2400 spec.outputs.push_back(BufferSp(new Float32Buffer(inputs)));
2401 spec.numWorkGroups = IVec3(numElements, 1, 1);
2403 group->addChild(new SpvAsmComputeShaderCase(
2404 testCtx, "rounded", "Check that are rounded when needed", spec));
2407 return group.release();
2410 // Performs a bitwise copy of source to the destination type Dest.
2411 template <typename Dest, typename Src>
2412 Dest bitwiseCast(Src source)
2415 DE_STATIC_ASSERT(sizeof(source) == sizeof(dest));
2416 deMemcpy(&dest, &source, sizeof(dest));
2420 tcu::TestCaseGroup* createSpecConstantOpQuantizeToF16Group (tcu::TestContext& testCtx)
2422 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop_opquantize", "Tests the OpQuantizeToF16 opcode for the OpSpecConstantOp instruction"));
2424 const std::string shader (
2425 string(s_ShaderPreamble) +
2427 "OpName %main \"main\"\n"
2428 "OpName %id \"gl_GlobalInvocationID\"\n"
2430 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2432 "OpDecorate %sc_0 SpecId 0\n"
2433 "OpDecorate %sc_1 SpecId 1\n"
2434 "OpDecorate %sc_2 SpecId 2\n"
2435 "OpDecorate %sc_3 SpecId 3\n"
2436 "OpDecorate %sc_4 SpecId 4\n"
2437 "OpDecorate %sc_5 SpecId 5\n"
2439 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2441 "%id = OpVariable %uvec3ptr Input\n"
2442 "%zero = OpConstant %i32 0\n"
2443 "%c_u32_6 = OpConstant %u32 6\n"
2445 "%sc_0 = OpSpecConstant %f32 0.\n"
2446 "%sc_1 = OpSpecConstant %f32 0.\n"
2447 "%sc_2 = OpSpecConstant %f32 0.\n"
2448 "%sc_3 = OpSpecConstant %f32 0.\n"
2449 "%sc_4 = OpSpecConstant %f32 0.\n"
2450 "%sc_5 = OpSpecConstant %f32 0.\n"
2452 "%sc_0_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_0\n"
2453 "%sc_1_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_1\n"
2454 "%sc_2_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_2\n"
2455 "%sc_3_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_3\n"
2456 "%sc_4_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_4\n"
2457 "%sc_5_quant = OpSpecConstantOp %f32 QuantizeToF16 %sc_5\n"
2459 "%main = OpFunction %void None %voidf\n"
2460 "%label = OpLabel\n"
2461 "%idval = OpLoad %uvec3 %id\n"
2462 "%x = OpCompositeExtract %u32 %idval 0\n"
2463 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2464 "%selector = OpUMod %u32 %x %c_u32_6\n"
2465 " OpSelectionMerge %exit None\n"
2466 " OpSwitch %selector %exit 0 %case0 1 %case1 2 %case2 3 %case3 4 %case4 5 %case5\n"
2468 "%case0 = OpLabel\n"
2469 " OpStore %outloc %sc_0_quant\n"
2472 "%case1 = OpLabel\n"
2473 " OpStore %outloc %sc_1_quant\n"
2476 "%case2 = OpLabel\n"
2477 " OpStore %outloc %sc_2_quant\n"
2480 "%case3 = OpLabel\n"
2481 " OpStore %outloc %sc_3_quant\n"
2484 "%case4 = OpLabel\n"
2485 " OpStore %outloc %sc_4_quant\n"
2488 "%case5 = OpLabel\n"
2489 " OpStore %outloc %sc_5_quant\n"
2495 " OpFunctionEnd\n");
2498 ComputeShaderSpec spec;
2499 const deUint8 numCases = 4;
2500 vector<float> inputs (numCases, 0.f);
2501 vector<float> outputs;
2503 spec.numWorkGroups = IVec3(numCases, 1, 1);
2505 spec.specConstants.push_back(bitwiseCast<deUint32>(std::numeric_limits<float>::infinity()));
2506 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::numeric_limits<float>::infinity()));
2507 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, 16)));
2508 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, 32)));
2510 outputs.push_back(std::numeric_limits<float>::infinity());
2511 outputs.push_back(-std::numeric_limits<float>::infinity());
2512 outputs.push_back(std::numeric_limits<float>::infinity());
2513 outputs.push_back(-std::numeric_limits<float>::infinity());
2515 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2516 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2518 group->addChild(new SpvAsmComputeShaderCase(
2519 testCtx, "infinities", "Check that infinities propagated and created", spec));
2523 ComputeShaderSpec spec;
2524 const deUint8 numCases = 2;
2525 vector<float> inputs (numCases, 0.f);
2526 vector<float> outputs;
2528 spec.numWorkGroups = IVec3(numCases, 1, 1);
2529 spec.verifyIO = &compareNan;
2531 outputs.push_back(std::numeric_limits<float>::quiet_NaN());
2532 outputs.push_back(-std::numeric_limits<float>::quiet_NaN());
2534 for (deUint8 idx = 0; idx < numCases; ++idx)
2535 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
2537 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2538 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2540 group->addChild(new SpvAsmComputeShaderCase(
2541 testCtx, "propagated_nans", "Check that nans are propagated", spec));
2545 ComputeShaderSpec spec;
2546 const deUint8 numCases = 6;
2547 vector<float> inputs (numCases, 0.f);
2548 vector<float> outputs;
2550 spec.numWorkGroups = IVec3(numCases, 1, 1);
2552 spec.specConstants.push_back(bitwiseCast<deUint32>(0.f));
2553 spec.specConstants.push_back(bitwiseCast<deUint32>(-0.f));
2554 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -16)));
2555 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(-1.0f, -32)));
2556 spec.specConstants.push_back(bitwiseCast<deUint32>(std::ldexp(1.0f, -127)));
2557 spec.specConstants.push_back(bitwiseCast<deUint32>(-std::ldexp(1.0f, -128)));
2559 outputs.push_back(0.f);
2560 outputs.push_back(-0.f);
2561 outputs.push_back(0.f);
2562 outputs.push_back(-0.f);
2563 outputs.push_back(0.f);
2564 outputs.push_back(-0.f);
2566 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2567 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2569 group->addChild(new SpvAsmComputeShaderCase(
2570 testCtx, "flush_to_zero", "Check that values are zeroed correctly", spec));
2574 ComputeShaderSpec spec;
2575 const deUint8 numCases = 6;
2576 vector<float> inputs (numCases, 0.f);
2577 vector<float> outputs;
2579 spec.numWorkGroups = IVec3(numCases, 1, 1);
2581 for (deUint8 idx = 0; idx < 6; ++idx)
2583 const float f = static_cast<float>(idx * 10 - 30) / 4.f;
2584 spec.specConstants.push_back(bitwiseCast<deUint32>(f));
2585 outputs.push_back(f);
2588 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2589 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2591 group->addChild(new SpvAsmComputeShaderCase(
2592 testCtx, "exact", "Check that values exactly preserved where appropriate", spec));
2596 ComputeShaderSpec spec;
2597 const deUint8 numCases = 4;
2598 vector<float> inputs (numCases, 0.f);
2599 vector<float> outputs;
2601 spec.numWorkGroups = IVec3(numCases, 1, 1);
2602 spec.verifyIO = &compareOpQuantizeF16ComputeExactCase;
2604 outputs.push_back(constructNormalizedFloat(8, 0x300300));
2605 outputs.push_back(-constructNormalizedFloat(-7, 0x600800));
2606 outputs.push_back(constructNormalizedFloat(2, 0x01E000));
2607 outputs.push_back(constructNormalizedFloat(1, 0xFFE000));
2609 for (deUint8 idx = 0; idx < numCases; ++idx)
2610 spec.specConstants.push_back(bitwiseCast<deUint32>(outputs[idx]));
2612 spec.inputs.push_back(BufferSp(new Float32Buffer(inputs)));
2613 spec.outputs.push_back(BufferSp(new Float32Buffer(outputs)));
2615 group->addChild(new SpvAsmComputeShaderCase(
2616 testCtx, "rounded", "Check that are rounded when needed", spec));
2619 return group.release();
2622 // Checks that constant null/composite values can be used in computation.
2623 tcu::TestCaseGroup* createOpConstantUsageGroup (tcu::TestContext& testCtx)
2625 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opconstantnullcomposite", "Spotcheck the OpConstantNull & OpConstantComposite instruction"));
2626 ComputeShaderSpec spec;
2627 de::Random rnd (deStringHash(group->getName()));
2628 const int numElements = 100;
2629 vector<float> positiveFloats (numElements, 0);
2630 vector<float> negativeFloats (numElements, 0);
2632 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
2634 for (size_t ndx = 0; ndx < numElements; ++ndx)
2635 negativeFloats[ndx] = -positiveFloats[ndx];
2638 "OpCapability Shader\n"
2639 "%std450 = OpExtInstImport \"GLSL.std.450\"\n"
2640 "OpMemoryModel Logical GLSL450\n"
2641 "OpEntryPoint GLCompute %main \"main\" %id\n"
2642 "OpExecutionMode %main LocalSize 1 1 1\n"
2644 "OpSource GLSL 430\n"
2645 "OpName %main \"main\"\n"
2646 "OpName %id \"gl_GlobalInvocationID\"\n"
2648 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2650 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) +
2652 "%fvec3 = OpTypeVector %f32 3\n"
2653 "%fmat = OpTypeMatrix %fvec3 3\n"
2654 "%ten = OpConstant %u32 10\n"
2655 "%f32arr10 = OpTypeArray %f32 %ten\n"
2656 "%fst = OpTypeStruct %f32 %f32\n"
2658 + string(s_InputOutputBuffer) +
2660 "%id = OpVariable %uvec3ptr Input\n"
2661 "%zero = OpConstant %i32 0\n"
2663 // Create a bunch of null values
2664 "%unull = OpConstantNull %u32\n"
2665 "%fnull = OpConstantNull %f32\n"
2666 "%vnull = OpConstantNull %fvec3\n"
2667 "%mnull = OpConstantNull %fmat\n"
2668 "%anull = OpConstantNull %f32arr10\n"
2669 "%snull = OpConstantComposite %fst %fnull %fnull\n"
2671 "%main = OpFunction %void None %voidf\n"
2672 "%label = OpLabel\n"
2673 "%idval = OpLoad %uvec3 %id\n"
2674 "%x = OpCompositeExtract %u32 %idval 0\n"
2675 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2676 "%inval = OpLoad %f32 %inloc\n"
2677 "%neg = OpFNegate %f32 %inval\n"
2679 // Get the abs() of (a certain element of) those null values
2680 "%unull_cov = OpConvertUToF %f32 %unull\n"
2681 "%unull_abs = OpExtInst %f32 %std450 FAbs %unull_cov\n"
2682 "%fnull_abs = OpExtInst %f32 %std450 FAbs %fnull\n"
2683 "%vnull_0 = OpCompositeExtract %f32 %vnull 0\n"
2684 "%vnull_abs = OpExtInst %f32 %std450 FAbs %vnull_0\n"
2685 "%mnull_12 = OpCompositeExtract %f32 %mnull 1 2\n"
2686 "%mnull_abs = OpExtInst %f32 %std450 FAbs %mnull_12\n"
2687 "%anull_3 = OpCompositeExtract %f32 %anull 3\n"
2688 "%anull_abs = OpExtInst %f32 %std450 FAbs %anull_3\n"
2689 "%snull_1 = OpCompositeExtract %f32 %snull 1\n"
2690 "%snull_abs = OpExtInst %f32 %std450 FAbs %snull_1\n"
2693 "%add1 = OpFAdd %f32 %neg %unull_abs\n"
2694 "%add2 = OpFAdd %f32 %add1 %fnull_abs\n"
2695 "%add3 = OpFAdd %f32 %add2 %vnull_abs\n"
2696 "%add4 = OpFAdd %f32 %add3 %mnull_abs\n"
2697 "%add5 = OpFAdd %f32 %add4 %anull_abs\n"
2698 "%final = OpFAdd %f32 %add5 %snull_abs\n"
2700 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2701 " OpStore %outloc %final\n" // write to output
2704 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
2705 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
2706 spec.numWorkGroups = IVec3(numElements, 1, 1);
2708 group->addChild(new SpvAsmComputeShaderCase(testCtx, "spotcheck", "Check that values constructed via OpConstantNull & OpConstantComposite can be used", spec));
2710 return group.release();
2713 // Assembly code used for testing loop control is based on GLSL source code:
2716 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2717 // float elements[];
2719 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2720 // float elements[];
2724 // uint x = gl_GlobalInvocationID.x;
2725 // output_data.elements[x] = input_data.elements[x];
2726 // for (uint i = 0; i < 4; ++i)
2727 // output_data.elements[x] += 1.f;
2729 tcu::TestCaseGroup* createLoopControlGroup (tcu::TestContext& testCtx)
2731 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "loop_control", "Tests loop control cases"));
2732 vector<CaseParameter> cases;
2733 de::Random rnd (deStringHash(group->getName()));
2734 const int numElements = 100;
2735 vector<float> inputFloats (numElements, 0);
2736 vector<float> outputFloats (numElements, 0);
2737 const StringTemplate shaderTemplate (
2738 string(s_ShaderPreamble) +
2740 "OpSource GLSL 430\n"
2741 "OpName %main \"main\"\n"
2742 "OpName %id \"gl_GlobalInvocationID\"\n"
2744 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2746 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2748 "%u32ptr = OpTypePointer Function %u32\n"
2750 "%id = OpVariable %uvec3ptr Input\n"
2751 "%zero = OpConstant %i32 0\n"
2752 "%uzero = OpConstant %u32 0\n"
2753 "%one = OpConstant %i32 1\n"
2754 "%constf1 = OpConstant %f32 1.0\n"
2755 "%four = OpConstant %u32 4\n"
2757 "%main = OpFunction %void None %voidf\n"
2758 "%entry = OpLabel\n"
2759 "%i = OpVariable %u32ptr Function\n"
2760 " OpStore %i %uzero\n"
2762 "%idval = OpLoad %uvec3 %id\n"
2763 "%x = OpCompositeExtract %u32 %idval 0\n"
2764 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2765 "%inval = OpLoad %f32 %inloc\n"
2766 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2767 " OpStore %outloc %inval\n"
2768 " OpBranch %loop_entry\n"
2770 "%loop_entry = OpLabel\n"
2771 "%i_val = OpLoad %u32 %i\n"
2772 "%cmp_lt = OpULessThan %bool %i_val %four\n"
2773 " OpLoopMerge %loop_merge %loop_entry ${CONTROL}\n"
2774 " OpBranchConditional %cmp_lt %loop_body %loop_merge\n"
2775 "%loop_body = OpLabel\n"
2776 "%outval = OpLoad %f32 %outloc\n"
2777 "%addf1 = OpFAdd %f32 %outval %constf1\n"
2778 " OpStore %outloc %addf1\n"
2779 "%new_i = OpIAdd %u32 %i_val %one\n"
2780 " OpStore %i %new_i\n"
2781 " OpBranch %loop_entry\n"
2782 "%loop_merge = OpLabel\n"
2784 " OpFunctionEnd\n");
2786 cases.push_back(CaseParameter("none", "None"));
2787 cases.push_back(CaseParameter("unroll", "Unroll"));
2788 cases.push_back(CaseParameter("dont_unroll", "DontUnroll"));
2789 cases.push_back(CaseParameter("unroll_dont_unroll", "Unroll|DontUnroll"));
2791 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2793 for (size_t ndx = 0; ndx < numElements; ++ndx)
2794 outputFloats[ndx] = inputFloats[ndx] + 4.f;
2796 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2798 map<string, string> specializations;
2799 ComputeShaderSpec spec;
2801 specializations["CONTROL"] = cases[caseNdx].param;
2802 spec.assembly = shaderTemplate.specialize(specializations);
2803 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2804 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2805 spec.numWorkGroups = IVec3(numElements, 1, 1);
2807 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2810 return group.release();
2813 // Assembly code used for testing selection control is based on GLSL source code:
2816 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2817 // float elements[];
2819 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2820 // float elements[];
2824 // uint x = gl_GlobalInvocationID.x;
2825 // float val = input_data.elements[x];
2827 // output_data.elements[x] = val + 1.f;
2829 // output_data.elements[x] = val - 1.f;
2831 tcu::TestCaseGroup* createSelectionControlGroup (tcu::TestContext& testCtx)
2833 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_control", "Tests selection control cases"));
2834 vector<CaseParameter> cases;
2835 de::Random rnd (deStringHash(group->getName()));
2836 const int numElements = 100;
2837 vector<float> inputFloats (numElements, 0);
2838 vector<float> outputFloats (numElements, 0);
2839 const StringTemplate shaderTemplate (
2840 string(s_ShaderPreamble) +
2842 "OpSource GLSL 430\n"
2843 "OpName %main \"main\"\n"
2844 "OpName %id \"gl_GlobalInvocationID\"\n"
2846 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2848 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2850 "%id = OpVariable %uvec3ptr Input\n"
2851 "%zero = OpConstant %i32 0\n"
2852 "%constf1 = OpConstant %f32 1.0\n"
2853 "%constf10 = OpConstant %f32 10.0\n"
2855 "%main = OpFunction %void None %voidf\n"
2856 "%entry = OpLabel\n"
2857 "%idval = OpLoad %uvec3 %id\n"
2858 "%x = OpCompositeExtract %u32 %idval 0\n"
2859 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2860 "%inval = OpLoad %f32 %inloc\n"
2861 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2862 "%cmp_gt = OpFOrdGreaterThan %bool %inval %constf10\n"
2864 " OpSelectionMerge %if_end ${CONTROL}\n"
2865 " OpBranchConditional %cmp_gt %if_true %if_false\n"
2866 "%if_true = OpLabel\n"
2867 "%addf1 = OpFAdd %f32 %inval %constf1\n"
2868 " OpStore %outloc %addf1\n"
2869 " OpBranch %if_end\n"
2870 "%if_false = OpLabel\n"
2871 "%subf1 = OpFSub %f32 %inval %constf1\n"
2872 " OpStore %outloc %subf1\n"
2873 " OpBranch %if_end\n"
2874 "%if_end = OpLabel\n"
2876 " OpFunctionEnd\n");
2878 cases.push_back(CaseParameter("none", "None"));
2879 cases.push_back(CaseParameter("flatten", "Flatten"));
2880 cases.push_back(CaseParameter("dont_flatten", "DontFlatten"));
2881 cases.push_back(CaseParameter("flatten_dont_flatten", "DontFlatten|Flatten"));
2883 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2885 for (size_t ndx = 0; ndx < numElements; ++ndx)
2886 outputFloats[ndx] = inputFloats[ndx] + (inputFloats[ndx] > 10.f ? 1.f : -1.f);
2888 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2890 map<string, string> specializations;
2891 ComputeShaderSpec spec;
2893 specializations["CONTROL"] = cases[caseNdx].param;
2894 spec.assembly = shaderTemplate.specialize(specializations);
2895 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2896 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2897 spec.numWorkGroups = IVec3(numElements, 1, 1);
2899 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2902 return group.release();
2905 // Assembly code used for testing function control is based on GLSL source code:
2909 // layout(std140, set = 0, binding = 0) readonly buffer Input {
2910 // float elements[];
2912 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
2913 // float elements[];
2916 // float const10() { return 10.f; }
2919 // uint x = gl_GlobalInvocationID.x;
2920 // output_data.elements[x] = input_data.elements[x] + const10();
2922 tcu::TestCaseGroup* createFunctionControlGroup (tcu::TestContext& testCtx)
2924 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "function_control", "Tests function control cases"));
2925 vector<CaseParameter> cases;
2926 de::Random rnd (deStringHash(group->getName()));
2927 const int numElements = 100;
2928 vector<float> inputFloats (numElements, 0);
2929 vector<float> outputFloats (numElements, 0);
2930 const StringTemplate shaderTemplate (
2931 string(s_ShaderPreamble) +
2933 "OpSource GLSL 430\n"
2934 "OpName %main \"main\"\n"
2935 "OpName %func_const10 \"const10(\"\n"
2936 "OpName %id \"gl_GlobalInvocationID\"\n"
2938 "OpDecorate %id BuiltIn GlobalInvocationId\n"
2940 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
2942 "%f32f = OpTypeFunction %f32\n"
2943 "%id = OpVariable %uvec3ptr Input\n"
2944 "%zero = OpConstant %i32 0\n"
2945 "%constf10 = OpConstant %f32 10.0\n"
2947 "%main = OpFunction %void None %voidf\n"
2948 "%entry = OpLabel\n"
2949 "%idval = OpLoad %uvec3 %id\n"
2950 "%x = OpCompositeExtract %u32 %idval 0\n"
2951 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
2952 "%inval = OpLoad %f32 %inloc\n"
2953 "%ret_10 = OpFunctionCall %f32 %func_const10\n"
2954 "%fadd = OpFAdd %f32 %inval %ret_10\n"
2955 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
2956 " OpStore %outloc %fadd\n"
2960 "%func_const10 = OpFunction %f32 ${CONTROL} %f32f\n"
2961 "%label = OpLabel\n"
2962 " OpReturnValue %constf10\n"
2963 " OpFunctionEnd\n");
2965 cases.push_back(CaseParameter("none", "None"));
2966 cases.push_back(CaseParameter("inline", "Inline"));
2967 cases.push_back(CaseParameter("dont_inline", "DontInline"));
2968 cases.push_back(CaseParameter("pure", "Pure"));
2969 cases.push_back(CaseParameter("const", "Const"));
2970 cases.push_back(CaseParameter("inline_pure", "Inline|Pure"));
2971 cases.push_back(CaseParameter("const_dont_inline", "Const|DontInline"));
2972 cases.push_back(CaseParameter("inline_dont_inline", "Inline|DontInline"));
2973 cases.push_back(CaseParameter("pure_inline_dont_inline", "Pure|Inline|DontInline"));
2975 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
2977 for (size_t ndx = 0; ndx < numElements; ++ndx)
2978 outputFloats[ndx] = inputFloats[ndx] + 10.f;
2980 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
2982 map<string, string> specializations;
2983 ComputeShaderSpec spec;
2985 specializations["CONTROL"] = cases[caseNdx].param;
2986 spec.assembly = shaderTemplate.specialize(specializations);
2987 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
2988 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
2989 spec.numWorkGroups = IVec3(numElements, 1, 1);
2991 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
2994 return group.release();
2997 tcu::TestCaseGroup* createMemoryAccessGroup (tcu::TestContext& testCtx)
2999 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "memory_access", "Tests memory access cases"));
3000 vector<CaseParameter> cases;
3001 de::Random rnd (deStringHash(group->getName()));
3002 const int numElements = 100;
3003 vector<float> inputFloats (numElements, 0);
3004 vector<float> outputFloats (numElements, 0);
3005 const StringTemplate shaderTemplate (
3006 string(s_ShaderPreamble) +
3008 "OpSource GLSL 430\n"
3009 "OpName %main \"main\"\n"
3010 "OpName %id \"gl_GlobalInvocationID\"\n"
3012 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3014 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
3016 "%f32ptr_f = OpTypePointer Function %f32\n"
3018 "%id = OpVariable %uvec3ptr Input\n"
3019 "%zero = OpConstant %i32 0\n"
3020 "%four = OpConstant %i32 4\n"
3022 "%main = OpFunction %void None %voidf\n"
3023 "%label = OpLabel\n"
3024 "%copy = OpVariable %f32ptr_f Function\n"
3025 "%idval = OpLoad %uvec3 %id ${ACCESS}\n"
3026 "%x = OpCompositeExtract %u32 %idval 0\n"
3027 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3028 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3029 " OpCopyMemory %copy %inloc ${ACCESS}\n"
3030 "%val1 = OpLoad %f32 %copy\n"
3031 "%val2 = OpLoad %f32 %inloc\n"
3032 "%add = OpFAdd %f32 %val1 %val2\n"
3033 " OpStore %outloc %add ${ACCESS}\n"
3035 " OpFunctionEnd\n");
3037 cases.push_back(CaseParameter("null", ""));
3038 cases.push_back(CaseParameter("none", "None"));
3039 cases.push_back(CaseParameter("volatile", "Volatile"));
3040 cases.push_back(CaseParameter("aligned", "Aligned 4"));
3041 cases.push_back(CaseParameter("nontemporal", "Nontemporal"));
3042 cases.push_back(CaseParameter("aligned_nontemporal", "Aligned|Nontemporal 4"));
3043 cases.push_back(CaseParameter("aligned_volatile", "Volatile|Aligned 4"));
3045 fillRandomScalars(rnd, -100.f, 100.f, &inputFloats[0], numElements);
3047 for (size_t ndx = 0; ndx < numElements; ++ndx)
3048 outputFloats[ndx] = inputFloats[ndx] + inputFloats[ndx];
3050 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3052 map<string, string> specializations;
3053 ComputeShaderSpec spec;
3055 specializations["ACCESS"] = cases[caseNdx].param;
3056 spec.assembly = shaderTemplate.specialize(specializations);
3057 spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
3058 spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
3059 spec.numWorkGroups = IVec3(numElements, 1, 1);
3061 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3064 return group.release();
3067 // Checks that we can get undefined values for various types, without exercising a computation with it.
3068 tcu::TestCaseGroup* createOpUndefGroup (tcu::TestContext& testCtx)
3070 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opundef", "Tests the OpUndef instruction"));
3071 vector<CaseParameter> cases;
3072 de::Random rnd (deStringHash(group->getName()));
3073 const int numElements = 100;
3074 vector<float> positiveFloats (numElements, 0);
3075 vector<float> negativeFloats (numElements, 0);
3076 const StringTemplate shaderTemplate (
3077 string(s_ShaderPreamble) +
3079 "OpSource GLSL 430\n"
3080 "OpName %main \"main\"\n"
3081 "OpName %id \"gl_GlobalInvocationID\"\n"
3083 "OpDecorate %id BuiltIn GlobalInvocationId\n"
3085 + string(s_InputOutputBufferTraits) + string(s_CommonTypes) + string(s_InputOutputBuffer) +
3089 "%id = OpVariable %uvec3ptr Input\n"
3090 "%zero = OpConstant %i32 0\n"
3092 "%main = OpFunction %void None %voidf\n"
3093 "%label = OpLabel\n"
3095 "%undef = OpUndef %type\n"
3097 "%idval = OpLoad %uvec3 %id\n"
3098 "%x = OpCompositeExtract %u32 %idval 0\n"
3100 "%inloc = OpAccessChain %f32ptr %indata %zero %x\n"
3101 "%inval = OpLoad %f32 %inloc\n"
3102 "%neg = OpFNegate %f32 %inval\n"
3103 "%outloc = OpAccessChain %f32ptr %outdata %zero %x\n"
3104 " OpStore %outloc %neg\n"
3106 " OpFunctionEnd\n");
3108 cases.push_back(CaseParameter("bool", "%type = OpTypeBool"));
3109 cases.push_back(CaseParameter("sint32", "%type = OpTypeInt 32 1"));
3110 cases.push_back(CaseParameter("uint32", "%type = OpTypeInt 32 0"));
3111 cases.push_back(CaseParameter("float32", "%type = OpTypeFloat 32"));
3112 cases.push_back(CaseParameter("vec4float32", "%type = OpTypeVector %f32 4"));
3113 cases.push_back(CaseParameter("vec2uint32", "%type = OpTypeVector %u32 2"));
3114 cases.push_back(CaseParameter("matrix", "%type = OpTypeMatrix %uvec3 3"));
3115 cases.push_back(CaseParameter("image", "%type = OpTypeImage %f32 2D 0 0 0 0 Unknown"));
3116 cases.push_back(CaseParameter("sampler", "%type = OpTypeSampler"));
3117 cases.push_back(CaseParameter("sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 0 Unknown\n"
3118 "%type = OpTypeSampledImage %img"));
3119 cases.push_back(CaseParameter("array", "%100 = OpConstant %u32 100\n"
3120 "%type = OpTypeArray %i32 %100"));
3121 cases.push_back(CaseParameter("runtimearray", "%type = OpTypeRuntimeArray %f32"));
3122 cases.push_back(CaseParameter("struct", "%type = OpTypeStruct %f32 %i32 %u32"));
3123 cases.push_back(CaseParameter("pointer", "%type = OpTypePointer Function %i32"));
3125 fillRandomScalars(rnd, 1.f, 100.f, &positiveFloats[0], numElements);
3127 for (size_t ndx = 0; ndx < numElements; ++ndx)
3128 negativeFloats[ndx] = -positiveFloats[ndx];
3130 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
3132 map<string, string> specializations;
3133 ComputeShaderSpec spec;
3135 specializations["TYPE"] = cases[caseNdx].param;
3136 spec.assembly = shaderTemplate.specialize(specializations);
3137 spec.inputs.push_back(BufferSp(new Float32Buffer(positiveFloats)));
3138 spec.outputs.push_back(BufferSp(new Float32Buffer(negativeFloats)));
3139 spec.numWorkGroups = IVec3(numElements, 1, 1);
3141 group->addChild(new SpvAsmComputeShaderCase(testCtx, cases[caseNdx].name, cases[caseNdx].name, spec));
3144 return group.release();
3146 typedef std::pair<std::string, VkShaderStageFlagBits> EntryToStage;
3147 typedef map<string, vector<EntryToStage> > ModuleMap;
3148 typedef map<VkShaderStageFlagBits, vector<deInt32> > StageToSpecConstantMap;
3150 // Context for a specific test instantiation. For example, an instantiation
3151 // may test colors yellow/magenta/cyan/mauve in a tesselation shader
3152 // with an entry point named 'main_to_the_main'
3153 struct InstanceContext
3155 // Map of modules to what entry_points we care to use from those modules.
3156 ModuleMap moduleMap;
3157 RGBA inputColors[4];
3158 RGBA outputColors[4];
3159 // Concrete SPIR-V code to test via boilerplate specialization.
3160 map<string, string> testCodeFragments;
3161 StageToSpecConstantMap specConstants;
3162 bool hasTessellation;
3163 VkShaderStageFlagBits requiredStages;
3165 InstanceContext (const RGBA (&inputs)[4], const RGBA (&outputs)[4], const map<string, string>& testCodeFragments_, const StageToSpecConstantMap& specConstants_)
3166 : testCodeFragments (testCodeFragments_)
3167 , specConstants (specConstants_)
3168 , hasTessellation (false)
3169 , requiredStages (static_cast<VkShaderStageFlagBits>(0))
3171 inputColors[0] = inputs[0];
3172 inputColors[1] = inputs[1];
3173 inputColors[2] = inputs[2];
3174 inputColors[3] = inputs[3];
3176 outputColors[0] = outputs[0];
3177 outputColors[1] = outputs[1];
3178 outputColors[2] = outputs[2];
3179 outputColors[3] = outputs[3];
3182 InstanceContext (const InstanceContext& other)
3183 : moduleMap (other.moduleMap)
3184 , testCodeFragments (other.testCodeFragments)
3185 , specConstants (other.specConstants)
3186 , hasTessellation (other.hasTessellation)
3187 , requiredStages (other.requiredStages)
3189 inputColors[0] = other.inputColors[0];
3190 inputColors[1] = other.inputColors[1];
3191 inputColors[2] = other.inputColors[2];
3192 inputColors[3] = other.inputColors[3];
3194 outputColors[0] = other.outputColors[0];
3195 outputColors[1] = other.outputColors[1];
3196 outputColors[2] = other.outputColors[2];
3197 outputColors[3] = other.outputColors[3];
3201 // A description of a shader to be used for a single stage of the graphics pipeline.
3202 struct ShaderElement
3204 // The module that contains this shader entrypoint.
3207 // The name of the entrypoint.
3210 // Which shader stage this entry point represents.
3211 VkShaderStageFlagBits stage;
3213 ShaderElement (const string& moduleName_, const string& entryPoint_, VkShaderStageFlagBits shaderStage_)
3214 : moduleName(moduleName_)
3215 , entryName(entryPoint_)
3216 , stage(shaderStage_)
3221 void getDefaultColors (RGBA (&colors)[4])
3223 colors[0] = RGBA::white();
3224 colors[1] = RGBA::red();
3225 colors[2] = RGBA::green();
3226 colors[3] = RGBA::blue();
3229 void getHalfColorsFullAlpha (RGBA (&colors)[4])
3231 colors[0] = RGBA(127, 127, 127, 255);
3232 colors[1] = RGBA(127, 0, 0, 255);
3233 colors[2] = RGBA(0, 127, 0, 255);
3234 colors[3] = RGBA(0, 0, 127, 255);
3237 void getInvertedDefaultColors (RGBA (&colors)[4])
3239 colors[0] = RGBA(0, 0, 0, 255);
3240 colors[1] = RGBA(0, 255, 255, 255);
3241 colors[2] = RGBA(255, 0, 255, 255);
3242 colors[3] = RGBA(255, 255, 0, 255);
3245 // Turns a statically sized array of ShaderElements into an instance-context
3246 // by setting up the mapping of modules to their contained shaders and stages.
3247 // The inputs and expected outputs are given by inputColors and outputColors
3249 InstanceContext createInstanceContext (const ShaderElement (&elements)[N], const RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments, const StageToSpecConstantMap& specConstants)
3251 InstanceContext ctx (inputColors, outputColors, testCodeFragments, specConstants);
3252 for (size_t i = 0; i < N; ++i)
3254 ctx.moduleMap[elements[i].moduleName].push_back(std::make_pair(elements[i].entryName, elements[i].stage));
3255 ctx.requiredStages = static_cast<VkShaderStageFlagBits>(ctx.requiredStages | elements[i].stage);
3261 inline InstanceContext createInstanceContext (const ShaderElement (&elements)[N], RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments)
3263 return createInstanceContext(elements, inputColors, outputColors, testCodeFragments, StageToSpecConstantMap());
3266 // The same as createInstanceContext above, but with default colors.
3268 InstanceContext createInstanceContext (const ShaderElement (&elements)[N], const map<string, string>& testCodeFragments)
3270 RGBA defaultColors[4];
3271 getDefaultColors(defaultColors);
3272 return createInstanceContext(elements, defaultColors, defaultColors, testCodeFragments);
3275 // For the current InstanceContext, constructs the required modules and shader stage create infos.
3276 void createPipelineShaderStages (const DeviceInterface& vk, const VkDevice vkDevice, InstanceContext& instance, Context& context, vector<ModuleHandleSp>& modules, vector<VkPipelineShaderStageCreateInfo>& createInfos)
3278 for (ModuleMap::const_iterator moduleNdx = instance.moduleMap.begin(); moduleNdx != instance.moduleMap.end(); ++moduleNdx)
3280 const ModuleHandleSp mod(new Unique<VkShaderModule>(createShaderModule(vk, vkDevice, context.getBinaryCollection().get(moduleNdx->first), 0)));
3281 modules.push_back(ModuleHandleSp(mod));
3282 for (vector<EntryToStage>::const_iterator shaderNdx = moduleNdx->second.begin(); shaderNdx != moduleNdx->second.end(); ++shaderNdx)
3284 const EntryToStage& stage = *shaderNdx;
3285 const VkPipelineShaderStageCreateInfo shaderParam =
3287 VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, // VkStructureType sType;
3288 DE_NULL, // const void* pNext;
3289 (VkPipelineShaderStageCreateFlags)0,
3290 stage.second, // VkShaderStageFlagBits stage;
3291 **modules.back(), // VkShaderModule module;
3292 stage.first.c_str(), // const char* pName;
3293 (const VkSpecializationInfo*)DE_NULL,
3295 createInfos.push_back(shaderParam);
3300 #define SPIRV_ASSEMBLY_TYPES \
3301 "%void = OpTypeVoid\n" \
3302 "%bool = OpTypeBool\n" \
3304 "%i32 = OpTypeInt 32 1\n" \
3305 "%u32 = OpTypeInt 32 0\n" \
3307 "%f32 = OpTypeFloat 32\n" \
3308 "%v3f32 = OpTypeVector %f32 3\n" \
3309 "%v4f32 = OpTypeVector %f32 4\n" \
3311 "%v4f32_function = OpTypeFunction %v4f32 %v4f32\n" \
3312 "%fun = OpTypeFunction %void\n" \
3314 "%ip_f32 = OpTypePointer Input %f32\n" \
3315 "%ip_i32 = OpTypePointer Input %i32\n" \
3316 "%ip_v3f32 = OpTypePointer Input %v3f32\n" \
3317 "%ip_v4f32 = OpTypePointer Input %v4f32\n" \
3319 "%op_f32 = OpTypePointer Output %f32\n" \
3320 "%op_v4f32 = OpTypePointer Output %v4f32\n" \
3322 "%fp_f32 = OpTypePointer Function %f32\n" \
3323 "%fp_i32 = OpTypePointer Function %i32\n" \
3324 "%fp_v4f32 = OpTypePointer Function %v4f32\n"
3326 #define SPIRV_ASSEMBLY_CONSTANTS \
3327 "%c_f32_1 = OpConstant %f32 1.0\n" \
3328 "%c_f32_0 = OpConstant %f32 0.0\n" \
3329 "%c_f32_0_5 = OpConstant %f32 0.5\n" \
3330 "%c_f32_n1 = OpConstant %f32 -1.\n" \
3331 "%c_f32_7 = OpConstant %f32 7.0\n" \
3332 "%c_f32_8 = OpConstant %f32 8.0\n" \
3333 "%c_i32_0 = OpConstant %i32 0\n" \
3334 "%c_i32_1 = OpConstant %i32 1\n" \
3335 "%c_i32_2 = OpConstant %i32 2\n" \
3336 "%c_i32_3 = OpConstant %i32 3\n" \
3337 "%c_i32_4 = OpConstant %i32 4\n" \
3338 "%c_u32_0 = OpConstant %u32 0\n" \
3339 "%c_u32_1 = OpConstant %u32 1\n" \
3340 "%c_u32_2 = OpConstant %u32 2\n" \
3341 "%c_u32_3 = OpConstant %u32 3\n" \
3342 "%c_u32_32 = OpConstant %u32 32\n" \
3343 "%c_u32_4 = OpConstant %u32 4\n" \
3344 "%c_u32_31_bits = OpConstant %u32 0x7FFFFFFF\n" \
3345 "%c_v4f32_1_1_1_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n" \
3346 "%c_v4f32_1_0_0_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_1\n" \
3347 "%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"
3349 #define SPIRV_ASSEMBLY_ARRAYS \
3350 "%a1f32 = OpTypeArray %f32 %c_u32_1\n" \
3351 "%a2f32 = OpTypeArray %f32 %c_u32_2\n" \
3352 "%a3v4f32 = OpTypeArray %v4f32 %c_u32_3\n" \
3353 "%a4f32 = OpTypeArray %f32 %c_u32_4\n" \
3354 "%a32v4f32 = OpTypeArray %v4f32 %c_u32_32\n" \
3355 "%ip_a3v4f32 = OpTypePointer Input %a3v4f32\n" \
3356 "%ip_a32v4f32 = OpTypePointer Input %a32v4f32\n" \
3357 "%op_a2f32 = OpTypePointer Output %a2f32\n" \
3358 "%op_a3v4f32 = OpTypePointer Output %a3v4f32\n" \
3359 "%op_a4f32 = OpTypePointer Output %a4f32\n"
3361 // Creates vertex-shader assembly by specializing a boilerplate StringTemplate
3362 // on fragments, which must (at least) map "testfun" to an OpFunction definition
3363 // for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed
3364 // with "BP_" to avoid collisions with fragments.
3366 // It corresponds roughly to this GLSL:
3368 // layout(location = 0) in vec4 position;
3369 // layout(location = 1) in vec4 color;
3370 // layout(location = 1) out highp vec4 vtxColor;
3371 // void main (void) { gl_Position = position; vtxColor = test_func(color); }
3372 string makeVertexShaderAssembly(const map<string, string>& fragments)
3374 // \todo [2015-11-23 awoloszyn] Remove OpName once these have stabalized
3375 static const char vertexShaderBoilerplate[] =
3376 "OpCapability Shader\n"
3377 "OpMemoryModel Logical GLSL450\n"
3378 "OpEntryPoint Vertex %main \"main\" %BP_stream %BP_position %BP_vtx_color %BP_color %BP_gl_VertexIndex %BP_gl_InstanceIndex\n"
3380 "OpName %main \"main\"\n"
3381 "OpName %BP_gl_PerVertex \"gl_PerVertex\"\n"
3382 "OpMemberName %BP_gl_PerVertex 0 \"gl_Position\"\n"
3383 "OpMemberName %BP_gl_PerVertex 1 \"gl_PointSize\"\n"
3384 "OpMemberName %BP_gl_PerVertex 2 \"gl_ClipDistance\"\n"
3385 "OpMemberName %BP_gl_PerVertex 3 \"gl_CullDistance\"\n"
3386 "OpName %test_code \"testfun(vf4;\"\n"
3387 "OpName %BP_stream \"\"\n"
3388 "OpName %BP_position \"position\"\n"
3389 "OpName %BP_vtx_color \"vtxColor\"\n"
3390 "OpName %BP_color \"color\"\n"
3391 "OpName %BP_gl_VertexIndex \"gl_VertexIndex\"\n"
3392 "OpName %BP_gl_InstanceIndex \"gl_InstanceIndex\"\n"
3393 "OpMemberDecorate %BP_gl_PerVertex 0 BuiltIn Position\n"
3394 "OpMemberDecorate %BP_gl_PerVertex 1 BuiltIn PointSize\n"
3395 "OpMemberDecorate %BP_gl_PerVertex 2 BuiltIn ClipDistance\n"
3396 "OpMemberDecorate %BP_gl_PerVertex 3 BuiltIn CullDistance\n"
3397 "OpDecorate %BP_gl_PerVertex Block\n"
3398 "OpDecorate %BP_position Location 0\n"
3399 "OpDecorate %BP_vtx_color Location 1\n"
3400 "OpDecorate %BP_color Location 1\n"
3401 "OpDecorate %BP_gl_VertexIndex BuiltIn VertexIndex\n"
3402 "OpDecorate %BP_gl_InstanceIndex BuiltIn InstanceIndex\n"
3403 "${decoration:opt}\n"
3404 SPIRV_ASSEMBLY_TYPES
3405 SPIRV_ASSEMBLY_CONSTANTS
3406 SPIRV_ASSEMBLY_ARRAYS
3407 "%BP_gl_PerVertex = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3408 "%BP_op_gl_PerVertex = OpTypePointer Output %BP_gl_PerVertex\n"
3409 "%BP_stream = OpVariable %BP_op_gl_PerVertex Output\n"
3410 "%BP_position = OpVariable %ip_v4f32 Input\n"
3411 "%BP_vtx_color = OpVariable %op_v4f32 Output\n"
3412 "%BP_color = OpVariable %ip_v4f32 Input\n"
3413 "%BP_gl_VertexIndex = OpVariable %ip_i32 Input\n"
3414 "%BP_gl_InstanceIndex = OpVariable %ip_i32 Input\n"
3416 "%main = OpFunction %void None %fun\n"
3417 "%BP_label = OpLabel\n"
3418 "%BP_pos = OpLoad %v4f32 %BP_position\n"
3419 "%BP_gl_pos = OpAccessChain %op_v4f32 %BP_stream %c_i32_0\n"
3420 "OpStore %BP_gl_pos %BP_pos\n"
3421 "%BP_col = OpLoad %v4f32 %BP_color\n"
3422 "%BP_col_transformed = OpFunctionCall %v4f32 %test_code %BP_col\n"
3423 "OpStore %BP_vtx_color %BP_col_transformed\n"
3427 return tcu::StringTemplate(vertexShaderBoilerplate).specialize(fragments);
3430 // Creates tess-control-shader assembly by specializing a boilerplate
3431 // StringTemplate on fragments, which must (at least) map "testfun" to an
3432 // OpFunction definition for %test_code that takes and returns a %v4f32.
3433 // Boilerplate IDs are prefixed with "BP_" to avoid collisions with fragments.
3435 // It roughly corresponds to the following GLSL.
3438 // layout(vertices = 3) out;
3439 // layout(location = 1) in vec4 in_color[];
3440 // layout(location = 1) out vec4 out_color[];
3443 // out_color[gl_InvocationID] = testfun(in_color[gl_InvocationID]);
3444 // gl_out[gl_InvocationID].gl_Position = gl_in[gl_InvocationID].gl_Position;
3445 // if (gl_InvocationID == 0) {
3446 // gl_TessLevelOuter[0] = 1.0;
3447 // gl_TessLevelOuter[1] = 1.0;
3448 // gl_TessLevelOuter[2] = 1.0;
3449 // gl_TessLevelInner[0] = 1.0;
3452 string makeTessControlShaderAssembly (const map<string, string>& fragments)
3454 static const char tessControlShaderBoilerplate[] =
3455 "OpCapability Tessellation\n"
3456 "OpMemoryModel Logical GLSL450\n"
3457 "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"
3458 "OpExecutionMode %BP_main OutputVertices 3\n"
3460 "OpName %BP_main \"main\"\n"
3461 "OpName %test_code \"testfun(vf4;\"\n"
3462 "OpName %BP_out_color \"out_color\"\n"
3463 "OpName %BP_gl_InvocationID \"gl_InvocationID\"\n"
3464 "OpName %BP_in_color \"in_color\"\n"
3465 "OpName %BP_gl_PerVertex \"gl_PerVertex\"\n"
3466 "OpMemberName %BP_gl_PerVertex 0 \"gl_Position\"\n"
3467 "OpMemberName %BP_gl_PerVertex 1 \"gl_PointSize\"\n"
3468 "OpMemberName %BP_gl_PerVertex 2 \"gl_ClipDistance\"\n"
3469 "OpMemberName %BP_gl_PerVertex 3 \"gl_CullDistance\"\n"
3470 "OpName %BP_gl_out \"gl_out\"\n"
3471 "OpName %BP_gl_PVOut \"gl_PerVertex\"\n"
3472 "OpMemberName %BP_gl_PVOut 0 \"gl_Position\"\n"
3473 "OpMemberName %BP_gl_PVOut 1 \"gl_PointSize\"\n"
3474 "OpMemberName %BP_gl_PVOut 2 \"gl_ClipDistance\"\n"
3475 "OpMemberName %BP_gl_PVOut 3 \"gl_CullDistance\"\n"
3476 "OpName %BP_gl_in \"gl_in\"\n"
3477 "OpName %BP_gl_TessLevelOuter \"gl_TessLevelOuter\"\n"
3478 "OpName %BP_gl_TessLevelInner \"gl_TessLevelInner\"\n"
3479 "OpDecorate %BP_out_color Location 1\n"
3480 "OpDecorate %BP_gl_InvocationID BuiltIn InvocationId\n"
3481 "OpDecorate %BP_in_color Location 1\n"
3482 "OpMemberDecorate %BP_gl_PerVertex 0 BuiltIn Position\n"
3483 "OpMemberDecorate %BP_gl_PerVertex 1 BuiltIn PointSize\n"
3484 "OpMemberDecorate %BP_gl_PerVertex 2 BuiltIn ClipDistance\n"
3485 "OpMemberDecorate %BP_gl_PerVertex 3 BuiltIn CullDistance\n"
3486 "OpDecorate %BP_gl_PerVertex Block\n"
3487 "OpMemberDecorate %BP_gl_PVOut 0 BuiltIn Position\n"
3488 "OpMemberDecorate %BP_gl_PVOut 1 BuiltIn PointSize\n"
3489 "OpMemberDecorate %BP_gl_PVOut 2 BuiltIn ClipDistance\n"
3490 "OpMemberDecorate %BP_gl_PVOut 3 BuiltIn CullDistance\n"
3491 "OpDecorate %BP_gl_PVOut Block\n"
3492 "OpDecorate %BP_gl_TessLevelOuter Patch\n"
3493 "OpDecorate %BP_gl_TessLevelOuter BuiltIn TessLevelOuter\n"
3494 "OpDecorate %BP_gl_TessLevelInner Patch\n"
3495 "OpDecorate %BP_gl_TessLevelInner BuiltIn TessLevelInner\n"
3496 "${decoration:opt}\n"
3497 SPIRV_ASSEMBLY_TYPES
3498 SPIRV_ASSEMBLY_CONSTANTS
3499 SPIRV_ASSEMBLY_ARRAYS
3500 "%BP_out_color = OpVariable %op_a3v4f32 Output\n"
3501 "%BP_gl_InvocationID = OpVariable %ip_i32 Input\n"
3502 "%BP_in_color = OpVariable %ip_a32v4f32 Input\n"
3503 "%BP_gl_PerVertex = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3504 "%BP_a3_gl_PerVertex = OpTypeArray %BP_gl_PerVertex %c_u32_3\n"
3505 "%BP_op_a3_gl_PerVertex = OpTypePointer Output %BP_a3_gl_PerVertex\n"
3506 "%BP_gl_out = OpVariable %BP_op_a3_gl_PerVertex Output\n"
3507 "%BP_gl_PVOut = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3508 "%BP_a32_gl_PVOut = OpTypeArray %BP_gl_PVOut %c_u32_32\n"
3509 "%BP_ip_a32_gl_PVOut = OpTypePointer Input %BP_a32_gl_PVOut\n"
3510 "%BP_gl_in = OpVariable %BP_ip_a32_gl_PVOut Input\n"
3511 "%BP_gl_TessLevelOuter = OpVariable %op_a4f32 Output\n"
3512 "%BP_gl_TessLevelInner = OpVariable %op_a2f32 Output\n"
3515 "%BP_main = OpFunction %void None %fun\n"
3516 "%BP_label = OpLabel\n"
3518 "%BP_gl_Invoc = OpLoad %i32 %BP_gl_InvocationID\n"
3520 "%BP_in_col_loc = OpAccessChain %ip_v4f32 %BP_in_color %BP_gl_Invoc\n"
3521 "%BP_out_col_loc = OpAccessChain %op_v4f32 %BP_out_color %BP_gl_Invoc\n"
3522 "%BP_in_col_val = OpLoad %v4f32 %BP_in_col_loc\n"
3523 "%BP_clr_transformed = OpFunctionCall %v4f32 %test_code %BP_in_col_val\n"
3524 "OpStore %BP_out_col_loc %BP_clr_transformed\n"
3526 "%BP_in_pos_loc = OpAccessChain %ip_v4f32 %BP_gl_in %BP_gl_Invoc %c_i32_0\n"
3527 "%BP_out_pos_loc = OpAccessChain %op_v4f32 %BP_gl_out %BP_gl_Invoc %c_i32_0\n"
3528 "%BP_in_pos_val = OpLoad %v4f32 %BP_in_pos_loc\n"
3529 "OpStore %BP_out_pos_loc %BP_in_pos_val\n"
3531 "%BP_cmp = OpIEqual %bool %BP_gl_Invoc %c_i32_0\n"
3532 "OpSelectionMerge %BP_merge_label None\n"
3533 "OpBranchConditional %BP_cmp %BP_if_label %BP_merge_label\n"
3534 "%BP_if_label = OpLabel\n"
3535 "%BP_gl_TessLevelOuterPos_0 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_0\n"
3536 "%BP_gl_TessLevelOuterPos_1 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_1\n"
3537 "%BP_gl_TessLevelOuterPos_2 = OpAccessChain %op_f32 %BP_gl_TessLevelOuter %c_i32_2\n"
3538 "%BP_gl_TessLevelInnerPos_0 = OpAccessChain %op_f32 %BP_gl_TessLevelInner %c_i32_0\n"
3539 "OpStore %BP_gl_TessLevelOuterPos_0 %c_f32_1\n"
3540 "OpStore %BP_gl_TessLevelOuterPos_1 %c_f32_1\n"
3541 "OpStore %BP_gl_TessLevelOuterPos_2 %c_f32_1\n"
3542 "OpStore %BP_gl_TessLevelInnerPos_0 %c_f32_1\n"
3543 "OpBranch %BP_merge_label\n"
3544 "%BP_merge_label = OpLabel\n"
3548 return tcu::StringTemplate(tessControlShaderBoilerplate).specialize(fragments);
3551 // Creates tess-evaluation-shader assembly by specializing a boilerplate
3552 // StringTemplate on fragments, which must (at least) map "testfun" to an
3553 // OpFunction definition for %test_code that takes and returns a %v4f32.
3554 // Boilerplate IDs are prefixed with "BP_" to avoid collisions with fragments.
3556 // It roughly corresponds to the following glsl.
3560 // layout(triangles, equal_spacing, ccw) in;
3561 // layout(location = 1) in vec4 in_color[];
3562 // layout(location = 1) out vec4 out_color;
3564 // #define interpolate(val)
3565 // vec4(gl_TessCoord.x) * val[0] + vec4(gl_TessCoord.y) * val[1] +
3566 // vec4(gl_TessCoord.z) * val[2]
3569 // gl_Position = vec4(gl_TessCoord.x) * gl_in[0].gl_Position +
3570 // vec4(gl_TessCoord.y) * gl_in[1].gl_Position +
3571 // vec4(gl_TessCoord.z) * gl_in[2].gl_Position;
3572 // out_color = testfun(interpolate(in_color));
3574 string makeTessEvalShaderAssembly(const map<string, string>& fragments)
3576 static const char tessEvalBoilerplate[] =
3577 "OpCapability Tessellation\n"
3578 "OpMemoryModel Logical GLSL450\n"
3579 "OpEntryPoint TessellationEvaluation %BP_main \"main\" %BP_stream %BP_gl_TessCoord %BP_gl_in %BP_out_color %BP_in_color\n"
3580 "OpExecutionMode %BP_main Triangles\n"
3581 "OpExecutionMode %BP_main SpacingEqual\n"
3582 "OpExecutionMode %BP_main VertexOrderCcw\n"
3584 "OpName %BP_main \"main\"\n"
3585 "OpName %test_code \"testfun(vf4;\"\n"
3586 "OpName %BP_gl_PerVertexOut \"gl_PerVertex\"\n"
3587 "OpMemberName %BP_gl_PerVertexOut 0 \"gl_Position\"\n"
3588 "OpMemberName %BP_gl_PerVertexOut 1 \"gl_PointSize\"\n"
3589 "OpMemberName %BP_gl_PerVertexOut 2 \"gl_ClipDistance\"\n"
3590 "OpMemberName %BP_gl_PerVertexOut 3 \"gl_CullDistance\"\n"
3591 "OpName %BP_stream \"\"\n"
3592 "OpName %BP_gl_TessCoord \"gl_TessCoord\"\n"
3593 "OpName %BP_gl_PerVertexIn \"gl_PerVertex\"\n"
3594 "OpMemberName %BP_gl_PerVertexIn 0 \"gl_Position\"\n"
3595 "OpMemberName %BP_gl_PerVertexIn 1 \"gl_PointSize\"\n"
3596 "OpMemberName %BP_gl_PerVertexIn 2 \"gl_ClipDistance\"\n"
3597 "OpMemberName %BP_gl_PerVertexIn 3 \"gl_CullDistance\"\n"
3598 "OpName %BP_gl_in \"gl_in\"\n"
3599 "OpName %BP_out_color \"out_color\"\n"
3600 "OpName %BP_in_color \"in_color\"\n"
3601 "OpMemberDecorate %BP_gl_PerVertexOut 0 BuiltIn Position\n"
3602 "OpMemberDecorate %BP_gl_PerVertexOut 1 BuiltIn PointSize\n"
3603 "OpMemberDecorate %BP_gl_PerVertexOut 2 BuiltIn ClipDistance\n"
3604 "OpMemberDecorate %BP_gl_PerVertexOut 3 BuiltIn CullDistance\n"
3605 "OpDecorate %BP_gl_PerVertexOut Block\n"
3606 "OpDecorate %BP_gl_TessCoord BuiltIn TessCoord\n"
3607 "OpMemberDecorate %BP_gl_PerVertexIn 0 BuiltIn Position\n"
3608 "OpMemberDecorate %BP_gl_PerVertexIn 1 BuiltIn PointSize\n"
3609 "OpMemberDecorate %BP_gl_PerVertexIn 2 BuiltIn ClipDistance\n"
3610 "OpMemberDecorate %BP_gl_PerVertexIn 3 BuiltIn CullDistance\n"
3611 "OpDecorate %BP_gl_PerVertexIn Block\n"
3612 "OpDecorate %BP_out_color Location 1\n"
3613 "OpDecorate %BP_in_color Location 1\n"
3614 "${decoration:opt}\n"
3615 SPIRV_ASSEMBLY_TYPES
3616 SPIRV_ASSEMBLY_CONSTANTS
3617 SPIRV_ASSEMBLY_ARRAYS
3618 "%BP_gl_PerVertexOut = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3619 "%BP_op_gl_PerVertexOut = OpTypePointer Output %BP_gl_PerVertexOut\n"
3620 "%BP_stream = OpVariable %BP_op_gl_PerVertexOut Output\n"
3621 "%BP_gl_TessCoord = OpVariable %ip_v3f32 Input\n"
3622 "%BP_gl_PerVertexIn = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3623 "%BP_a32_gl_PerVertexIn = OpTypeArray %BP_gl_PerVertexIn %c_u32_32\n"
3624 "%BP_ip_a32_gl_PerVertexIn = OpTypePointer Input %BP_a32_gl_PerVertexIn\n"
3625 "%BP_gl_in = OpVariable %BP_ip_a32_gl_PerVertexIn Input\n"
3626 "%BP_out_color = OpVariable %op_v4f32 Output\n"
3627 "%BP_in_color = OpVariable %ip_a32v4f32 Input\n"
3629 "%BP_main = OpFunction %void None %fun\n"
3630 "%BP_label = OpLabel\n"
3631 "%BP_gl_TC_0 = OpAccessChain %ip_f32 %BP_gl_TessCoord %c_u32_0\n"
3632 "%BP_gl_TC_1 = OpAccessChain %ip_f32 %BP_gl_TessCoord %c_u32_1\n"
3633 "%BP_gl_TC_2 = OpAccessChain %ip_f32 %BP_gl_TessCoord %c_u32_2\n"
3634 "%BP_gl_in_gl_Pos_0 = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_0 %c_i32_0\n"
3635 "%BP_gl_in_gl_Pos_1 = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_1 %c_i32_0\n"
3636 "%BP_gl_in_gl_Pos_2 = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_2 %c_i32_0\n"
3638 "%BP_gl_OPos = OpAccessChain %op_v4f32 %BP_stream %c_i32_0\n"
3639 "%BP_in_color_0 = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_0\n"
3640 "%BP_in_color_1 = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_1\n"
3641 "%BP_in_color_2 = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_2\n"
3643 "%BP_TC_W_0 = OpLoad %f32 %BP_gl_TC_0\n"
3644 "%BP_TC_W_1 = OpLoad %f32 %BP_gl_TC_1\n"
3645 "%BP_TC_W_2 = OpLoad %f32 %BP_gl_TC_2\n"
3646 "%BP_v4f32_TC_0 = OpCompositeConstruct %v4f32 %BP_TC_W_0 %BP_TC_W_0 %BP_TC_W_0 %BP_TC_W_0\n"
3647 "%BP_v4f32_TC_1 = OpCompositeConstruct %v4f32 %BP_TC_W_1 %BP_TC_W_1 %BP_TC_W_1 %BP_TC_W_1\n"
3648 "%BP_v4f32_TC_2 = OpCompositeConstruct %v4f32 %BP_TC_W_2 %BP_TC_W_2 %BP_TC_W_2 %BP_TC_W_2\n"
3650 "%BP_gl_IP_0 = OpLoad %v4f32 %BP_gl_in_gl_Pos_0\n"
3651 "%BP_gl_IP_1 = OpLoad %v4f32 %BP_gl_in_gl_Pos_1\n"
3652 "%BP_gl_IP_2 = OpLoad %v4f32 %BP_gl_in_gl_Pos_2\n"
3654 "%BP_IP_W_0 = OpFMul %v4f32 %BP_v4f32_TC_0 %BP_gl_IP_0\n"
3655 "%BP_IP_W_1 = OpFMul %v4f32 %BP_v4f32_TC_1 %BP_gl_IP_1\n"
3656 "%BP_IP_W_2 = OpFMul %v4f32 %BP_v4f32_TC_2 %BP_gl_IP_2\n"
3658 "%BP_pos_sum_0 = OpFAdd %v4f32 %BP_IP_W_0 %BP_IP_W_1\n"
3659 "%BP_pos_sum_1 = OpFAdd %v4f32 %BP_pos_sum_0 %BP_IP_W_2\n"
3661 "OpStore %BP_gl_OPos %BP_pos_sum_1\n"
3663 "%BP_IC_0 = OpLoad %v4f32 %BP_in_color_0\n"
3664 "%BP_IC_1 = OpLoad %v4f32 %BP_in_color_1\n"
3665 "%BP_IC_2 = OpLoad %v4f32 %BP_in_color_2\n"
3667 "%BP_IC_W_0 = OpFMul %v4f32 %BP_v4f32_TC_0 %BP_IC_0\n"
3668 "%BP_IC_W_1 = OpFMul %v4f32 %BP_v4f32_TC_1 %BP_IC_1\n"
3669 "%BP_IC_W_2 = OpFMul %v4f32 %BP_v4f32_TC_2 %BP_IC_2\n"
3671 "%BP_col_sum_0 = OpFAdd %v4f32 %BP_IC_W_0 %BP_IC_W_1\n"
3672 "%BP_col_sum_1 = OpFAdd %v4f32 %BP_col_sum_0 %BP_IC_W_2\n"
3674 "%BP_clr_transformed = OpFunctionCall %v4f32 %test_code %BP_col_sum_1\n"
3676 "OpStore %BP_out_color %BP_clr_transformed\n"
3680 return tcu::StringTemplate(tessEvalBoilerplate).specialize(fragments);
3683 // Creates geometry-shader assembly by specializing a boilerplate StringTemplate
3684 // on fragments, which must (at least) map "testfun" to an OpFunction definition
3685 // for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed
3686 // with "BP_" to avoid collisions with fragments.
3688 // Derived from this GLSL:
3691 // layout(triangles) in;
3692 // layout(triangle_strip, max_vertices = 3) out;
3694 // layout(location = 1) in vec4 in_color[];
3695 // layout(location = 1) out vec4 out_color;
3698 // gl_Position = gl_in[0].gl_Position;
3699 // out_color = test_fun(in_color[0]);
3701 // gl_Position = gl_in[1].gl_Position;
3702 // out_color = test_fun(in_color[1]);
3704 // gl_Position = gl_in[2].gl_Position;
3705 // out_color = test_fun(in_color[2]);
3709 string makeGeometryShaderAssembly(const map<string, string>& fragments)
3711 static const char geometryShaderBoilerplate[] =
3712 "OpCapability Geometry\n"
3713 "OpMemoryModel Logical GLSL450\n"
3714 "OpEntryPoint Geometry %BP_main \"main\" %BP_out_gl_position %BP_gl_in %BP_out_color %BP_in_color\n"
3715 "OpExecutionMode %BP_main Triangles\n"
3716 "OpExecutionMode %BP_main Invocations 0\n"
3717 "OpExecutionMode %BP_main OutputTriangleStrip\n"
3718 "OpExecutionMode %BP_main OutputVertices 3\n"
3720 "OpName %BP_main \"main\"\n"
3721 "OpName %BP_per_vertex_in \"gl_PerVertex\"\n"
3722 "OpMemberName %BP_per_vertex_in 0 \"gl_Position\"\n"
3723 "OpMemberName %BP_per_vertex_in 1 \"gl_PointSize\"\n"
3724 "OpMemberName %BP_per_vertex_in 2 \"gl_ClipDistance\"\n"
3725 "OpMemberName %BP_per_vertex_in 3 \"gl_CullDistance\"\n"
3726 "OpName %BP_gl_in \"gl_in\"\n"
3727 "OpName %BP_out_color \"out_color\"\n"
3728 "OpName %BP_in_color \"in_color\"\n"
3729 "OpName %test_code \"testfun(vf4;\"\n"
3730 "OpDecorate %BP_out_gl_position BuiltIn Position\n"
3731 "OpMemberDecorate %BP_per_vertex_in 0 BuiltIn Position\n"
3732 "OpMemberDecorate %BP_per_vertex_in 1 BuiltIn PointSize\n"
3733 "OpMemberDecorate %BP_per_vertex_in 2 BuiltIn ClipDistance\n"
3734 "OpMemberDecorate %BP_per_vertex_in 3 BuiltIn CullDistance\n"
3735 "OpDecorate %BP_per_vertex_in Block\n"
3736 "OpDecorate %BP_out_color Location 1\n"
3737 "OpDecorate %BP_out_color Stream 0\n"
3738 "OpDecorate %BP_in_color Location 1\n"
3739 "${decoration:opt}\n"
3740 SPIRV_ASSEMBLY_TYPES
3741 SPIRV_ASSEMBLY_CONSTANTS
3742 SPIRV_ASSEMBLY_ARRAYS
3743 "%BP_per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
3744 "%BP_a3_per_vertex_in = OpTypeArray %BP_per_vertex_in %c_u32_3\n"
3745 "%BP_ip_a3_per_vertex_in = OpTypePointer Input %BP_a3_per_vertex_in\n"
3747 "%BP_gl_in = OpVariable %BP_ip_a3_per_vertex_in Input\n"
3748 "%BP_out_color = OpVariable %op_v4f32 Output\n"
3749 "%BP_in_color = OpVariable %ip_a3v4f32 Input\n"
3750 "%BP_out_gl_position = OpVariable %op_v4f32 Output\n"
3753 "%BP_main = OpFunction %void None %fun\n"
3754 "%BP_label = OpLabel\n"
3755 "%BP_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_0 %c_i32_0\n"
3756 "%BP_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_1 %c_i32_0\n"
3757 "%BP_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %BP_gl_in %c_i32_2 %c_i32_0\n"
3759 "%BP_in_position_0 = OpLoad %v4f32 %BP_gl_in_0_gl_position\n"
3760 "%BP_in_position_1 = OpLoad %v4f32 %BP_gl_in_1_gl_position\n"
3761 "%BP_in_position_2 = OpLoad %v4f32 %BP_gl_in_2_gl_position \n"
3763 "%BP_in_color_0_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_0\n"
3764 "%BP_in_color_1_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_1\n"
3765 "%BP_in_color_2_ptr = OpAccessChain %ip_v4f32 %BP_in_color %c_i32_2\n"
3767 "%BP_in_color_0 = OpLoad %v4f32 %BP_in_color_0_ptr\n"
3768 "%BP_in_color_1 = OpLoad %v4f32 %BP_in_color_1_ptr\n"
3769 "%BP_in_color_2 = OpLoad %v4f32 %BP_in_color_2_ptr\n"
3771 "%BP_transformed_in_color_0 = OpFunctionCall %v4f32 %test_code %BP_in_color_0\n"
3772 "%BP_transformed_in_color_1 = OpFunctionCall %v4f32 %test_code %BP_in_color_1\n"
3773 "%BP_transformed_in_color_2 = OpFunctionCall %v4f32 %test_code %BP_in_color_2\n"
3776 "OpStore %BP_out_gl_position %BP_in_position_0\n"
3777 "OpStore %BP_out_color %BP_transformed_in_color_0\n"
3780 "OpStore %BP_out_gl_position %BP_in_position_1\n"
3781 "OpStore %BP_out_color %BP_transformed_in_color_1\n"
3784 "OpStore %BP_out_gl_position %BP_in_position_2\n"
3785 "OpStore %BP_out_color %BP_transformed_in_color_2\n"
3792 return tcu::StringTemplate(geometryShaderBoilerplate).specialize(fragments);
3795 // Creates fragment-shader assembly by specializing a boilerplate StringTemplate
3796 // on fragments, which must (at least) map "testfun" to an OpFunction definition
3797 // for %test_code that takes and returns a %v4f32. Boilerplate IDs are prefixed
3798 // with "BP_" to avoid collisions with fragments.
3800 // Derived from this GLSL:
3802 // layout(location = 1) in highp vec4 vtxColor;
3803 // layout(location = 0) out highp vec4 fragColor;
3804 // highp vec4 testfun(highp vec4 x) { return x; }
3805 // void main(void) { fragColor = testfun(vtxColor); }
3807 // with modifications including passing vtxColor by value and ripping out
3808 // testfun() definition.
3809 string makeFragmentShaderAssembly(const map<string, string>& fragments)
3811 static const char fragmentShaderBoilerplate[] =
3812 "OpCapability Shader\n"
3813 "OpMemoryModel Logical GLSL450\n"
3814 "OpEntryPoint Fragment %BP_main \"main\" %BP_vtxColor %BP_fragColor\n"
3815 "OpExecutionMode %BP_main OriginUpperLeft\n"
3817 "OpName %BP_main \"main\"\n"
3818 "OpName %BP_fragColor \"fragColor\"\n"
3819 "OpName %BP_vtxColor \"vtxColor\"\n"
3820 "OpName %test_code \"testfun(vf4;\"\n"
3821 "OpDecorate %BP_fragColor Location 0\n"
3822 "OpDecorate %BP_vtxColor Location 1\n"
3823 "${decoration:opt}\n"
3824 SPIRV_ASSEMBLY_TYPES
3825 SPIRV_ASSEMBLY_CONSTANTS
3826 SPIRV_ASSEMBLY_ARRAYS
3827 "%BP_fragColor = OpVariable %op_v4f32 Output\n"
3828 "%BP_vtxColor = OpVariable %ip_v4f32 Input\n"
3830 "%BP_main = OpFunction %void None %fun\n"
3831 "%BP_label_main = OpLabel\n"
3832 "%BP_tmp1 = OpLoad %v4f32 %BP_vtxColor\n"
3833 "%BP_tmp2 = OpFunctionCall %v4f32 %test_code %BP_tmp1\n"
3834 "OpStore %BP_fragColor %BP_tmp2\n"
3838 return tcu::StringTemplate(fragmentShaderBoilerplate).specialize(fragments);
3841 // Creates fragments that specialize into a simple pass-through shader (of any kind).
3842 map<string, string> passthruFragments(void)
3844 map<string, string> fragments;
3845 fragments["testfun"] =
3846 // A %test_code function that returns its argument unchanged.
3847 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
3848 "%param1 = OpFunctionParameter %v4f32\n"
3849 "%label_testfun = OpLabel\n"
3850 "OpReturnValue %param1\n"
3855 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3856 // Vertex shader gets custom code from context, the rest are pass-through.
3857 void addShaderCodeCustomVertex(vk::SourceCollections& dst, InstanceContext context)
3859 map<string, string> passthru = passthruFragments();
3860 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(context.testCodeFragments);
3861 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
3864 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3865 // Tessellation control shader gets custom code from context, the rest are
3867 void addShaderCodeCustomTessControl(vk::SourceCollections& dst, InstanceContext context)
3869 map<string, string> passthru = passthruFragments();
3870 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
3871 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(context.testCodeFragments);
3872 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(passthru);
3873 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
3876 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3877 // Tessellation evaluation shader gets custom code from context, the rest are
3879 void addShaderCodeCustomTessEval(vk::SourceCollections& dst, InstanceContext context)
3881 map<string, string> passthru = passthruFragments();
3882 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
3883 dst.spirvAsmSources.add("tessc") << makeTessControlShaderAssembly(passthru);
3884 dst.spirvAsmSources.add("tesse") << makeTessEvalShaderAssembly(context.testCodeFragments);
3885 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
3888 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3889 // Geometry shader gets custom code from context, the rest are pass-through.
3890 void addShaderCodeCustomGeometry(vk::SourceCollections& dst, InstanceContext context)
3892 map<string, string> passthru = passthruFragments();
3893 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
3894 dst.spirvAsmSources.add("geom") << makeGeometryShaderAssembly(context.testCodeFragments);
3895 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(passthru);
3898 // Adds shader assembly text to dst.spirvAsmSources for all shader kinds.
3899 // Fragment shader gets custom code from context, the rest are pass-through.
3900 void addShaderCodeCustomFragment(vk::SourceCollections& dst, InstanceContext context)
3902 map<string, string> passthru = passthruFragments();
3903 dst.spirvAsmSources.add("vert") << makeVertexShaderAssembly(passthru);
3904 dst.spirvAsmSources.add("frag") << makeFragmentShaderAssembly(context.testCodeFragments);
3907 void createCombinedModule(vk::SourceCollections& dst, InstanceContext)
3909 // \todo [2015-12-07 awoloszyn] Make tessellation / geometry conditional
3910 // \todo [2015-12-07 awoloszyn] Remove OpName and OpMemberName at some point
3911 dst.spirvAsmSources.add("module") <<
3912 "OpCapability Shader\n"
3913 "OpCapability Geometry\n"
3914 "OpCapability Tessellation\n"
3915 "OpMemoryModel Logical GLSL450\n"
3917 "OpEntryPoint Vertex %vert_main \"main\" %vert_Position %vert_vtxColor %vert_color %vert_vtxPosition %vert_vertex_id %vert_instance_id\n"
3918 "OpEntryPoint Geometry %geom_main \"main\" %geom_out_gl_position %geom_gl_in %geom_out_color %geom_in_color\n"
3919 "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"
3920 "OpEntryPoint TessellationEvaluation %tesse_main \"main\" %tesse_stream %tesse_gl_tessCoord %tesse_in_position %tesse_out_color %tesse_in_color \n"
3921 "OpEntryPoint Fragment %frag_main \"main\" %frag_vtxColor %frag_fragColor\n"
3923 "OpExecutionMode %geom_main Triangles\n"
3924 "OpExecutionMode %geom_main Invocations 0\n"
3925 "OpExecutionMode %geom_main OutputTriangleStrip\n"
3926 "OpExecutionMode %geom_main OutputVertices 3\n"
3928 "OpExecutionMode %tessc_main OutputVertices 3\n"
3930 "OpExecutionMode %tesse_main Triangles\n"
3932 "OpExecutionMode %frag_main OriginUpperLeft\n"
3934 "; Vertex decorations\n"
3935 "OpName %vert_main \"main\"\n"
3936 "OpName %vert_vtxPosition \"vtxPosition\"\n"
3937 "OpName %vert_Position \"position\"\n"
3938 "OpName %vert_vtxColor \"vtxColor\"\n"
3939 "OpName %vert_color \"color\"\n"
3940 "OpName %vert_vertex_id \"gl_VertexIndex\"\n"
3941 "OpName %vert_instance_id \"gl_InstanceIndex\"\n"
3942 "OpDecorate %vert_vtxPosition Location 2\n"
3943 "OpDecorate %vert_Position Location 0\n"
3944 "OpDecorate %vert_vtxColor Location 1\n"
3945 "OpDecorate %vert_color Location 1\n"
3946 "OpDecorate %vert_vertex_id BuiltIn VertexIndex\n"
3947 "OpDecorate %vert_instance_id BuiltIn InstanceIndex\n"
3949 "; Geometry decorations\n"
3950 "OpName %geom_main \"main\"\n"
3951 "OpName %geom_per_vertex_in \"gl_PerVertex\"\n"
3952 "OpMemberName %geom_per_vertex_in 0 \"gl_Position\"\n"
3953 "OpMemberName %geom_per_vertex_in 1 \"gl_PointSize\"\n"
3954 "OpMemberName %geom_per_vertex_in 2 \"gl_ClipDistance\"\n"
3955 "OpMemberName %geom_per_vertex_in 3 \"gl_CullDistance\"\n"
3956 "OpName %geom_gl_in \"gl_in\"\n"
3957 "OpName %geom_out_color \"out_color\"\n"
3958 "OpName %geom_in_color \"in_color\"\n"
3959 "OpDecorate %geom_out_gl_position BuiltIn Position\n"
3960 "OpMemberDecorate %geom_per_vertex_in 0 BuiltIn Position\n"
3961 "OpMemberDecorate %geom_per_vertex_in 1 BuiltIn PointSize\n"
3962 "OpMemberDecorate %geom_per_vertex_in 2 BuiltIn ClipDistance\n"
3963 "OpMemberDecorate %geom_per_vertex_in 3 BuiltIn CullDistance\n"
3964 "OpDecorate %geom_per_vertex_in Block\n"
3965 "OpDecorate %geom_out_color Location 1\n"
3966 "OpDecorate %geom_out_color Stream 0\n"
3967 "OpDecorate %geom_in_color Location 1\n"
3969 "; Tessellation Control decorations\n"
3970 "OpName %tessc_main \"main\"\n"
3971 "OpName %tessc_out_color \"out_color\"\n"
3972 "OpName %tessc_gl_InvocationID \"gl_InvocationID\"\n"
3973 "OpName %tessc_in_color \"in_color\"\n"
3974 "OpName %tessc_out_position \"out_position\"\n"
3975 "OpName %tessc_in_position \"in_position\"\n"
3976 "OpName %tessc_gl_TessLevelOuter \"gl_TessLevelOuter\"\n"
3977 "OpName %tessc_gl_TessLevelInner \"gl_TessLevelInner\"\n"
3978 "OpDecorate %tessc_out_color Location 1\n"
3979 "OpDecorate %tessc_gl_InvocationID BuiltIn InvocationId\n"
3980 "OpDecorate %tessc_in_color Location 1\n"
3981 "OpDecorate %tessc_out_position Location 2\n"
3982 "OpDecorate %tessc_in_position Location 2\n"
3983 "OpDecorate %tessc_gl_TessLevelOuter Patch\n"
3984 "OpDecorate %tessc_gl_TessLevelOuter BuiltIn TessLevelOuter\n"
3985 "OpDecorate %tessc_gl_TessLevelInner Patch\n"
3986 "OpDecorate %tessc_gl_TessLevelInner BuiltIn TessLevelInner\n"
3988 "; Tessellation Evaluation decorations\n"
3989 "OpName %tesse_main \"main\"\n"
3990 "OpName %tesse_per_vertex_out \"gl_PerVertex\"\n"
3991 "OpMemberName %tesse_per_vertex_out 0 \"gl_Position\"\n"
3992 "OpMemberName %tesse_per_vertex_out 1 \"gl_PointSize\"\n"
3993 "OpMemberName %tesse_per_vertex_out 2 \"gl_ClipDistance\"\n"
3994 "OpMemberName %tesse_per_vertex_out 3 \"gl_CullDistance\"\n"
3995 "OpName %tesse_stream \"\"\n"
3996 "OpName %tesse_gl_tessCoord \"gl_TessCoord\"\n"
3997 "OpName %tesse_in_position \"in_position\"\n"
3998 "OpName %tesse_out_color \"out_color\"\n"
3999 "OpName %tesse_in_color \"in_color\"\n"
4000 "OpMemberDecorate %tesse_per_vertex_out 0 BuiltIn Position\n"
4001 "OpMemberDecorate %tesse_per_vertex_out 1 BuiltIn PointSize\n"
4002 "OpMemberDecorate %tesse_per_vertex_out 2 BuiltIn ClipDistance\n"
4003 "OpMemberDecorate %tesse_per_vertex_out 3 BuiltIn CullDistance\n"
4004 "OpDecorate %tesse_per_vertex_out Block\n"
4005 "OpDecorate %tesse_gl_tessCoord BuiltIn TessCoord\n"
4006 "OpDecorate %tesse_in_position Location 2\n"
4007 "OpDecorate %tesse_out_color Location 1\n"
4008 "OpDecorate %tesse_in_color Location 1\n"
4010 "; Fragment decorations\n"
4011 "OpName %frag_main \"main\"\n"
4012 "OpName %frag_fragColor \"fragColor\"\n"
4013 "OpName %frag_vtxColor \"vtxColor\"\n"
4014 "OpDecorate %frag_fragColor Location 0\n"
4015 "OpDecorate %frag_vtxColor Location 1\n"
4017 SPIRV_ASSEMBLY_TYPES
4018 SPIRV_ASSEMBLY_CONSTANTS
4019 SPIRV_ASSEMBLY_ARRAYS
4021 "; Vertex Variables\n"
4022 "%vert_vtxPosition = OpVariable %op_v4f32 Output\n"
4023 "%vert_Position = OpVariable %ip_v4f32 Input\n"
4024 "%vert_vtxColor = OpVariable %op_v4f32 Output\n"
4025 "%vert_color = OpVariable %ip_v4f32 Input\n"
4026 "%vert_vertex_id = OpVariable %ip_i32 Input\n"
4027 "%vert_instance_id = OpVariable %ip_i32 Input\n"
4029 "; Geometry Variables\n"
4030 "%geom_per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4031 "%geom_a3_per_vertex_in = OpTypeArray %geom_per_vertex_in %c_u32_3\n"
4032 "%geom_ip_a3_per_vertex_in = OpTypePointer Input %geom_a3_per_vertex_in\n"
4033 "%geom_gl_in = OpVariable %geom_ip_a3_per_vertex_in Input\n"
4034 "%geom_out_color = OpVariable %op_v4f32 Output\n"
4035 "%geom_in_color = OpVariable %ip_a3v4f32 Input\n"
4036 "%geom_out_gl_position = OpVariable %op_v4f32 Output\n"
4038 "; Tessellation Control Variables\n"
4039 "%tessc_out_color = OpVariable %op_a3v4f32 Output\n"
4040 "%tessc_gl_InvocationID = OpVariable %ip_i32 Input\n"
4041 "%tessc_in_color = OpVariable %ip_a32v4f32 Input\n"
4042 "%tessc_out_position = OpVariable %op_a3v4f32 Output\n"
4043 "%tessc_in_position = OpVariable %ip_a32v4f32 Input\n"
4044 "%tessc_gl_TessLevelOuter = OpVariable %op_a4f32 Output\n"
4045 "%tessc_gl_TessLevelInner = OpVariable %op_a2f32 Output\n"
4047 "; Tessellation Evaluation Decorations\n"
4048 "%tesse_per_vertex_out = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4049 "%tesse_op_per_vertex_out = OpTypePointer Output %tesse_per_vertex_out\n"
4050 "%tesse_stream = OpVariable %tesse_op_per_vertex_out Output\n"
4051 "%tesse_gl_tessCoord = OpVariable %ip_v3f32 Input\n"
4052 "%tesse_in_position = OpVariable %ip_a32v4f32 Input\n"
4053 "%tesse_out_color = OpVariable %op_v4f32 Output\n"
4054 "%tesse_in_color = OpVariable %ip_a32v4f32 Input\n"
4056 "; Fragment Variables\n"
4057 "%frag_fragColor = OpVariable %op_v4f32 Output\n"
4058 "%frag_vtxColor = OpVariable %ip_v4f32 Input\n"
4061 "%vert_main = OpFunction %void None %fun\n"
4062 "%vert_label = OpLabel\n"
4063 "%vert_tmp_position = OpLoad %v4f32 %vert_Position\n"
4064 "OpStore %vert_vtxPosition %vert_tmp_position\n"
4065 "%vert_tmp_color = OpLoad %v4f32 %vert_color\n"
4066 "OpStore %vert_vtxColor %vert_tmp_color\n"
4070 "; Geometry Entry\n"
4071 "%geom_main = OpFunction %void None %fun\n"
4072 "%geom_label = OpLabel\n"
4073 "%geom_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %geom_gl_in %c_i32_0 %c_i32_0\n"
4074 "%geom_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %geom_gl_in %c_i32_1 %c_i32_0\n"
4075 "%geom_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %geom_gl_in %c_i32_2 %c_i32_0\n"
4076 "%geom_in_position_0 = OpLoad %v4f32 %geom_gl_in_0_gl_position\n"
4077 "%geom_in_position_1 = OpLoad %v4f32 %geom_gl_in_1_gl_position\n"
4078 "%geom_in_position_2 = OpLoad %v4f32 %geom_gl_in_2_gl_position \n"
4079 "%geom_in_color_0_ptr = OpAccessChain %ip_v4f32 %geom_in_color %c_i32_0\n"
4080 "%geom_in_color_1_ptr = OpAccessChain %ip_v4f32 %geom_in_color %c_i32_1\n"
4081 "%geom_in_color_2_ptr = OpAccessChain %ip_v4f32 %geom_in_color %c_i32_2\n"
4082 "%geom_in_color_0 = OpLoad %v4f32 %geom_in_color_0_ptr\n"
4083 "%geom_in_color_1 = OpLoad %v4f32 %geom_in_color_1_ptr\n"
4084 "%geom_in_color_2 = OpLoad %v4f32 %geom_in_color_2_ptr\n"
4085 "OpStore %geom_out_gl_position %geom_in_position_0\n"
4086 "OpStore %geom_out_color %geom_in_color_0\n"
4088 "OpStore %geom_out_gl_position %geom_in_position_1\n"
4089 "OpStore %geom_out_color %geom_in_color_1\n"
4091 "OpStore %geom_out_gl_position %geom_in_position_2\n"
4092 "OpStore %geom_out_color %geom_in_color_2\n"
4098 "; Tessellation Control Entry\n"
4099 "%tessc_main = OpFunction %void None %fun\n"
4100 "%tessc_label = OpLabel\n"
4101 "%tessc_invocation_id = OpLoad %i32 %tessc_gl_InvocationID\n"
4102 "%tessc_in_color_ptr = OpAccessChain %ip_v4f32 %tessc_in_color %tessc_invocation_id\n"
4103 "%tessc_in_position_ptr = OpAccessChain %ip_v4f32 %tessc_in_position %tessc_invocation_id\n"
4104 "%tessc_in_color_val = OpLoad %v4f32 %tessc_in_color_ptr\n"
4105 "%tessc_in_position_val = OpLoad %v4f32 %tessc_in_position_ptr\n"
4106 "%tessc_out_color_ptr = OpAccessChain %op_v4f32 %tessc_out_color %tessc_invocation_id\n"
4107 "%tessc_out_position_ptr = OpAccessChain %op_v4f32 %tessc_out_position %tessc_invocation_id\n"
4108 "OpStore %tessc_out_color_ptr %tessc_in_color_val\n"
4109 "OpStore %tessc_out_position_ptr %tessc_in_position_val\n"
4110 "%tessc_is_first_invocation = OpIEqual %bool %tessc_invocation_id %c_i32_0\n"
4111 "OpSelectionMerge %tessc_merge_label None\n"
4112 "OpBranchConditional %tessc_is_first_invocation %tessc_first_invocation %tessc_merge_label\n"
4113 "%tessc_first_invocation = OpLabel\n"
4114 "%tessc_tess_outer_0 = OpAccessChain %op_f32 %tessc_gl_TessLevelOuter %c_i32_0\n"
4115 "%tessc_tess_outer_1 = OpAccessChain %op_f32 %tessc_gl_TessLevelOuter %c_i32_1\n"
4116 "%tessc_tess_outer_2 = OpAccessChain %op_f32 %tessc_gl_TessLevelOuter %c_i32_2\n"
4117 "%tessc_tess_inner = OpAccessChain %op_f32 %tessc_gl_TessLevelInner %c_i32_0\n"
4118 "OpStore %tessc_tess_outer_0 %c_f32_1\n"
4119 "OpStore %tessc_tess_outer_1 %c_f32_1\n"
4120 "OpStore %tessc_tess_outer_2 %c_f32_1\n"
4121 "OpStore %tessc_tess_inner %c_f32_1\n"
4122 "OpBranch %tessc_merge_label\n"
4123 "%tessc_merge_label = OpLabel\n"
4127 "; Tessellation Evaluation Entry\n"
4128 "%tesse_main = OpFunction %void None %fun\n"
4129 "%tesse_label = OpLabel\n"
4130 "%tesse_tc_0_ptr = OpAccessChain %ip_f32 %tesse_gl_tessCoord %c_u32_0\n"
4131 "%tesse_tc_1_ptr = OpAccessChain %ip_f32 %tesse_gl_tessCoord %c_u32_1\n"
4132 "%tesse_tc_2_ptr = OpAccessChain %ip_f32 %tesse_gl_tessCoord %c_u32_2\n"
4133 "%tesse_tc_0 = OpLoad %f32 %tesse_tc_0_ptr\n"
4134 "%tesse_tc_1 = OpLoad %f32 %tesse_tc_1_ptr\n"
4135 "%tesse_tc_2 = OpLoad %f32 %tesse_tc_2_ptr\n"
4136 "%tesse_in_pos_0_ptr = OpAccessChain %ip_v4f32 %tesse_in_position %c_i32_0\n"
4137 "%tesse_in_pos_1_ptr = OpAccessChain %ip_v4f32 %tesse_in_position %c_i32_1\n"
4138 "%tesse_in_pos_2_ptr = OpAccessChain %ip_v4f32 %tesse_in_position %c_i32_2\n"
4139 "%tesse_in_pos_0 = OpLoad %v4f32 %tesse_in_pos_0_ptr\n"
4140 "%tesse_in_pos_1 = OpLoad %v4f32 %tesse_in_pos_1_ptr\n"
4141 "%tesse_in_pos_2 = OpLoad %v4f32 %tesse_in_pos_2_ptr\n"
4142 "%tesse_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_0 %tesse_in_pos_0\n"
4143 "%tesse_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_1 %tesse_in_pos_1\n"
4144 "%tesse_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_2 %tesse_in_pos_2\n"
4145 "%tesse_out_pos_ptr = OpAccessChain %op_v4f32 %tesse_stream %c_i32_0\n"
4146 "%tesse_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %tesse_in_pos_0_weighted %tesse_in_pos_1_weighted\n"
4147 "%tesse_computed_out = OpFAdd %v4f32 %tesse_in_pos_0_plus_pos_1 %tesse_in_pos_2_weighted\n"
4148 "OpStore %tesse_out_pos_ptr %tesse_computed_out\n"
4149 "%tesse_in_clr_0_ptr = OpAccessChain %ip_v4f32 %tesse_in_color %c_i32_0\n"
4150 "%tesse_in_clr_1_ptr = OpAccessChain %ip_v4f32 %tesse_in_color %c_i32_1\n"
4151 "%tesse_in_clr_2_ptr = OpAccessChain %ip_v4f32 %tesse_in_color %c_i32_2\n"
4152 "%tesse_in_clr_0 = OpLoad %v4f32 %tesse_in_clr_0_ptr\n"
4153 "%tesse_in_clr_1 = OpLoad %v4f32 %tesse_in_clr_1_ptr\n"
4154 "%tesse_in_clr_2 = OpLoad %v4f32 %tesse_in_clr_2_ptr\n"
4155 "%tesse_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_0 %tesse_in_clr_0\n"
4156 "%tesse_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_1 %tesse_in_clr_1\n"
4157 "%tesse_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %tesse_tc_2 %tesse_in_clr_2\n"
4158 "%tesse_in_clr_0_plus_col_1 = OpFAdd %v4f32 %tesse_in_clr_0_weighted %tesse_in_clr_1_weighted\n"
4159 "%tesse_computed_clr = OpFAdd %v4f32 %tesse_in_clr_0_plus_col_1 %tesse_in_clr_2_weighted\n"
4160 "OpStore %tesse_out_color %tesse_computed_clr\n"
4164 "; Fragment Entry\n"
4165 "%frag_main = OpFunction %void None %fun\n"
4166 "%frag_label_main = OpLabel\n"
4167 "%frag_tmp1 = OpLoad %v4f32 %frag_vtxColor\n"
4168 "OpStore %frag_fragColor %frag_tmp1\n"
4173 // This has two shaders of each stage. The first
4174 // is a passthrough, the second inverts the color.
4175 void createMultipleEntries(vk::SourceCollections& dst, InstanceContext)
4177 dst.spirvAsmSources.add("vert") <<
4178 // This module contains 2 vertex shaders. One that is a passthrough
4179 // and a second that inverts the color of the output (1.0 - color).
4180 "OpCapability Shader\n"
4181 "OpMemoryModel Logical GLSL450\n"
4182 "OpEntryPoint Vertex %main \"vert1\" %Position %vtxColor %color %vtxPosition %vertex_id %instance_id\n"
4183 "OpEntryPoint Vertex %main2 \"vert2\" %Position %vtxColor %color %vtxPosition %vertex_id %instance_id\n"
4185 "OpName %main \"frag1\"\n"
4186 "OpName %main2 \"frag2\"\n"
4187 "OpName %vtxPosition \"vtxPosition\"\n"
4188 "OpName %Position \"position\"\n"
4189 "OpName %vtxColor \"vtxColor\"\n"
4190 "OpName %color \"color\"\n"
4191 "OpName %vertex_id \"gl_VertexIndex\"\n"
4192 "OpName %instance_id \"gl_InstanceIndex\"\n"
4194 "OpDecorate %vtxPosition Location 2\n"
4195 "OpDecorate %Position Location 0\n"
4196 "OpDecorate %vtxColor Location 1\n"
4197 "OpDecorate %color Location 1\n"
4198 "OpDecorate %vertex_id BuiltIn VertexIndex\n"
4199 "OpDecorate %instance_id BuiltIn InstanceIndex\n"
4200 SPIRV_ASSEMBLY_TYPES
4201 SPIRV_ASSEMBLY_CONSTANTS
4202 SPIRV_ASSEMBLY_ARRAYS
4203 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4204 "%vtxPosition = OpVariable %op_v4f32 Output\n"
4205 "%Position = OpVariable %ip_v4f32 Input\n"
4206 "%vtxColor = OpVariable %op_v4f32 Output\n"
4207 "%color = OpVariable %ip_v4f32 Input\n"
4208 "%vertex_id = OpVariable %ip_i32 Input\n"
4209 "%instance_id = OpVariable %ip_i32 Input\n"
4211 "%main = OpFunction %void None %fun\n"
4212 "%label = OpLabel\n"
4213 "%tmp_position = OpLoad %v4f32 %Position\n"
4214 "OpStore %vtxPosition %tmp_position\n"
4215 "%tmp_color = OpLoad %v4f32 %color\n"
4216 "OpStore %vtxColor %tmp_color\n"
4220 "%main2 = OpFunction %void None %fun\n"
4221 "%label2 = OpLabel\n"
4222 "%tmp_position2 = OpLoad %v4f32 %Position\n"
4223 "OpStore %vtxPosition %tmp_position2\n"
4224 "%tmp_color2 = OpLoad %v4f32 %color\n"
4225 "%tmp_color3 = OpFSub %v4f32 %cval %tmp_color2\n"
4226 "OpStore %vtxColor %tmp_color3\n"
4230 dst.spirvAsmSources.add("frag") <<
4231 // This is a single module that contains 2 fragment shaders.
4232 // One that passes color through and the other that inverts the output
4233 // color (1.0 - color).
4234 "OpCapability Shader\n"
4235 "OpMemoryModel Logical GLSL450\n"
4236 "OpEntryPoint Fragment %main \"frag1\" %vtxColor %fragColor\n"
4237 "OpEntryPoint Fragment %main2 \"frag2\" %vtxColor %fragColor\n"
4238 "OpExecutionMode %main OriginUpperLeft\n"
4239 "OpExecutionMode %main2 OriginUpperLeft\n"
4241 "OpName %main \"frag1\"\n"
4242 "OpName %main2 \"frag2\"\n"
4243 "OpName %fragColor \"fragColor\"\n"
4244 "OpName %vtxColor \"vtxColor\"\n"
4245 "OpDecorate %fragColor Location 0\n"
4246 "OpDecorate %vtxColor Location 1\n"
4247 SPIRV_ASSEMBLY_TYPES
4248 SPIRV_ASSEMBLY_CONSTANTS
4249 SPIRV_ASSEMBLY_ARRAYS
4250 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4251 "%fragColor = OpVariable %op_v4f32 Output\n"
4252 "%vtxColor = OpVariable %ip_v4f32 Input\n"
4254 "%main = OpFunction %void None %fun\n"
4255 "%label_main = OpLabel\n"
4256 "%tmp1 = OpLoad %v4f32 %vtxColor\n"
4257 "OpStore %fragColor %tmp1\n"
4261 "%main2 = OpFunction %void None %fun\n"
4262 "%label_main2 = OpLabel\n"
4263 "%tmp2 = OpLoad %v4f32 %vtxColor\n"
4264 "%tmp3 = OpFSub %v4f32 %cval %tmp2\n"
4265 "OpStore %fragColor %tmp3\n"
4269 dst.spirvAsmSources.add("geom") <<
4270 "OpCapability Geometry\n"
4271 "OpMemoryModel Logical GLSL450\n"
4272 "OpEntryPoint Geometry %geom1_main \"geom1\" %out_gl_position %gl_in %out_color %in_color\n"
4273 "OpEntryPoint Geometry %geom2_main \"geom2\" %out_gl_position %gl_in %out_color %in_color\n"
4274 "OpExecutionMode %geom1_main Triangles\n"
4275 "OpExecutionMode %geom2_main Triangles\n"
4276 "OpExecutionMode %geom1_main Invocations 0\n"
4277 "OpExecutionMode %geom2_main Invocations 0\n"
4278 "OpExecutionMode %geom1_main OutputTriangleStrip\n"
4279 "OpExecutionMode %geom2_main OutputTriangleStrip\n"
4280 "OpExecutionMode %geom1_main OutputVertices 3\n"
4281 "OpExecutionMode %geom2_main OutputVertices 3\n"
4282 "OpName %geom1_main \"geom1\"\n"
4283 "OpName %geom2_main \"geom2\"\n"
4284 "OpName %per_vertex_in \"gl_PerVertex\"\n"
4285 "OpMemberName %per_vertex_in 0 \"gl_Position\"\n"
4286 "OpMemberName %per_vertex_in 1 \"gl_PointSize\"\n"
4287 "OpMemberName %per_vertex_in 2 \"gl_ClipDistance\"\n"
4288 "OpMemberName %per_vertex_in 3 \"gl_CullDistance\"\n"
4289 "OpName %gl_in \"gl_in\"\n"
4290 "OpName %out_color \"out_color\"\n"
4291 "OpName %in_color \"in_color\"\n"
4292 "OpDecorate %out_gl_position BuiltIn Position\n"
4293 "OpMemberDecorate %per_vertex_in 0 BuiltIn Position\n"
4294 "OpMemberDecorate %per_vertex_in 1 BuiltIn PointSize\n"
4295 "OpMemberDecorate %per_vertex_in 2 BuiltIn ClipDistance\n"
4296 "OpMemberDecorate %per_vertex_in 3 BuiltIn CullDistance\n"
4297 "OpDecorate %per_vertex_in Block\n"
4298 "OpDecorate %out_color Location 1\n"
4299 "OpDecorate %out_color Stream 0\n"
4300 "OpDecorate %in_color Location 1\n"
4301 SPIRV_ASSEMBLY_TYPES
4302 SPIRV_ASSEMBLY_CONSTANTS
4303 SPIRV_ASSEMBLY_ARRAYS
4304 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4305 "%per_vertex_in = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4306 "%a3_per_vertex_in = OpTypeArray %per_vertex_in %c_u32_3\n"
4307 "%ip_a3_per_vertex_in = OpTypePointer Input %a3_per_vertex_in\n"
4308 "%gl_in = OpVariable %ip_a3_per_vertex_in Input\n"
4309 "%out_color = OpVariable %op_v4f32 Output\n"
4310 "%in_color = OpVariable %ip_a3v4f32 Input\n"
4311 "%out_gl_position = OpVariable %op_v4f32 Output\n"
4313 "%geom1_main = OpFunction %void None %fun\n"
4314 "%geom1_label = OpLabel\n"
4315 "%geom1_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_0 %c_i32_0\n"
4316 "%geom1_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_1 %c_i32_0\n"
4317 "%geom1_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_2 %c_i32_0\n"
4318 "%geom1_in_position_0 = OpLoad %v4f32 %geom1_gl_in_0_gl_position\n"
4319 "%geom1_in_position_1 = OpLoad %v4f32 %geom1_gl_in_1_gl_position\n"
4320 "%geom1_in_position_2 = OpLoad %v4f32 %geom1_gl_in_2_gl_position \n"
4321 "%geom1_in_color_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4322 "%geom1_in_color_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4323 "%geom1_in_color_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4324 "%geom1_in_color_0 = OpLoad %v4f32 %geom1_in_color_0_ptr\n"
4325 "%geom1_in_color_1 = OpLoad %v4f32 %geom1_in_color_1_ptr\n"
4326 "%geom1_in_color_2 = OpLoad %v4f32 %geom1_in_color_2_ptr\n"
4327 "OpStore %out_gl_position %geom1_in_position_0\n"
4328 "OpStore %out_color %geom1_in_color_0\n"
4330 "OpStore %out_gl_position %geom1_in_position_1\n"
4331 "OpStore %out_color %geom1_in_color_1\n"
4333 "OpStore %out_gl_position %geom1_in_position_2\n"
4334 "OpStore %out_color %geom1_in_color_2\n"
4340 "%geom2_main = OpFunction %void None %fun\n"
4341 "%geom2_label = OpLabel\n"
4342 "%geom2_gl_in_0_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_0 %c_i32_0\n"
4343 "%geom2_gl_in_1_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_1 %c_i32_0\n"
4344 "%geom2_gl_in_2_gl_position = OpAccessChain %ip_v4f32 %gl_in %c_i32_2 %c_i32_0\n"
4345 "%geom2_in_position_0 = OpLoad %v4f32 %geom2_gl_in_0_gl_position\n"
4346 "%geom2_in_position_1 = OpLoad %v4f32 %geom2_gl_in_1_gl_position\n"
4347 "%geom2_in_position_2 = OpLoad %v4f32 %geom2_gl_in_2_gl_position \n"
4348 "%geom2_in_color_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4349 "%geom2_in_color_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4350 "%geom2_in_color_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4351 "%geom2_in_color_0 = OpLoad %v4f32 %geom2_in_color_0_ptr\n"
4352 "%geom2_in_color_1 = OpLoad %v4f32 %geom2_in_color_1_ptr\n"
4353 "%geom2_in_color_2 = OpLoad %v4f32 %geom2_in_color_2_ptr\n"
4354 "%geom2_transformed_in_color_0 = OpFSub %v4f32 %cval %geom2_in_color_0\n"
4355 "%geom2_transformed_in_color_1 = OpFSub %v4f32 %cval %geom2_in_color_1\n"
4356 "%geom2_transformed_in_color_2 = OpFSub %v4f32 %cval %geom2_in_color_2\n"
4357 "OpStore %out_gl_position %geom2_in_position_0\n"
4358 "OpStore %out_color %geom2_transformed_in_color_0\n"
4360 "OpStore %out_gl_position %geom2_in_position_1\n"
4361 "OpStore %out_color %geom2_transformed_in_color_1\n"
4363 "OpStore %out_gl_position %geom2_in_position_2\n"
4364 "OpStore %out_color %geom2_transformed_in_color_2\n"
4370 dst.spirvAsmSources.add("tessc") <<
4371 "OpCapability Tessellation\n"
4372 "OpMemoryModel Logical GLSL450\n"
4373 "OpEntryPoint TessellationControl %tessc1_main \"tessc1\" %out_color %gl_InvocationID %in_color %out_position %in_position %gl_TessLevelOuter %gl_TessLevelInner\n"
4374 "OpEntryPoint TessellationControl %tessc2_main \"tessc2\" %out_color %gl_InvocationID %in_color %out_position %in_position %gl_TessLevelOuter %gl_TessLevelInner\n"
4375 "OpExecutionMode %tessc1_main OutputVertices 3\n"
4376 "OpExecutionMode %tessc2_main OutputVertices 3\n"
4377 "OpName %tessc1_main \"tessc1\"\n"
4378 "OpName %tessc2_main \"tessc2\"\n"
4379 "OpName %out_color \"out_color\"\n"
4380 "OpName %gl_InvocationID \"gl_InvocationID\"\n"
4381 "OpName %in_color \"in_color\"\n"
4382 "OpName %out_position \"out_position\"\n"
4383 "OpName %in_position \"in_position\"\n"
4384 "OpName %gl_TessLevelOuter \"gl_TessLevelOuter\"\n"
4385 "OpName %gl_TessLevelInner \"gl_TessLevelInner\"\n"
4386 "OpDecorate %out_color Location 1\n"
4387 "OpDecorate %gl_InvocationID BuiltIn InvocationId\n"
4388 "OpDecorate %in_color Location 1\n"
4389 "OpDecorate %out_position Location 2\n"
4390 "OpDecorate %in_position Location 2\n"
4391 "OpDecorate %gl_TessLevelOuter Patch\n"
4392 "OpDecorate %gl_TessLevelOuter BuiltIn TessLevelOuter\n"
4393 "OpDecorate %gl_TessLevelInner Patch\n"
4394 "OpDecorate %gl_TessLevelInner BuiltIn TessLevelInner\n"
4395 SPIRV_ASSEMBLY_TYPES
4396 SPIRV_ASSEMBLY_CONSTANTS
4397 SPIRV_ASSEMBLY_ARRAYS
4398 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4399 "%out_color = OpVariable %op_a3v4f32 Output\n"
4400 "%gl_InvocationID = OpVariable %ip_i32 Input\n"
4401 "%in_color = OpVariable %ip_a32v4f32 Input\n"
4402 "%out_position = OpVariable %op_a3v4f32 Output\n"
4403 "%in_position = OpVariable %ip_a32v4f32 Input\n"
4404 "%gl_TessLevelOuter = OpVariable %op_a4f32 Output\n"
4405 "%gl_TessLevelInner = OpVariable %op_a2f32 Output\n"
4407 "%tessc1_main = OpFunction %void None %fun\n"
4408 "%tessc1_label = OpLabel\n"
4409 "%tessc1_invocation_id = OpLoad %i32 %gl_InvocationID\n"
4410 "%tessc1_in_color_ptr = OpAccessChain %ip_v4f32 %in_color %tessc1_invocation_id\n"
4411 "%tessc1_in_position_ptr = OpAccessChain %ip_v4f32 %in_position %tessc1_invocation_id\n"
4412 "%tessc1_in_color_val = OpLoad %v4f32 %tessc1_in_color_ptr\n"
4413 "%tessc1_in_position_val = OpLoad %v4f32 %tessc1_in_position_ptr\n"
4414 "%tessc1_out_color_ptr = OpAccessChain %op_v4f32 %out_color %tessc1_invocation_id\n"
4415 "%tessc1_out_position_ptr = OpAccessChain %op_v4f32 %out_position %tessc1_invocation_id\n"
4416 "OpStore %tessc1_out_color_ptr %tessc1_in_color_val\n"
4417 "OpStore %tessc1_out_position_ptr %tessc1_in_position_val\n"
4418 "%tessc1_is_first_invocation = OpIEqual %bool %tessc1_invocation_id %c_i32_0\n"
4419 "OpSelectionMerge %tessc1_merge_label None\n"
4420 "OpBranchConditional %tessc1_is_first_invocation %tessc1_first_invocation %tessc1_merge_label\n"
4421 "%tessc1_first_invocation = OpLabel\n"
4422 "%tessc1_tess_outer_0 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_0\n"
4423 "%tessc1_tess_outer_1 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_1\n"
4424 "%tessc1_tess_outer_2 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_2\n"
4425 "%tessc1_tess_inner = OpAccessChain %op_f32 %gl_TessLevelInner %c_i32_0\n"
4426 "OpStore %tessc1_tess_outer_0 %c_f32_1\n"
4427 "OpStore %tessc1_tess_outer_1 %c_f32_1\n"
4428 "OpStore %tessc1_tess_outer_2 %c_f32_1\n"
4429 "OpStore %tessc1_tess_inner %c_f32_1\n"
4430 "OpBranch %tessc1_merge_label\n"
4431 "%tessc1_merge_label = OpLabel\n"
4435 "%tessc2_main = OpFunction %void None %fun\n"
4436 "%tessc2_label = OpLabel\n"
4437 "%tessc2_invocation_id = OpLoad %i32 %gl_InvocationID\n"
4438 "%tessc2_in_color_ptr = OpAccessChain %ip_v4f32 %in_color %tessc2_invocation_id\n"
4439 "%tessc2_in_position_ptr = OpAccessChain %ip_v4f32 %in_position %tessc2_invocation_id\n"
4440 "%tessc2_in_color_val = OpLoad %v4f32 %tessc2_in_color_ptr\n"
4441 "%tessc2_in_position_val = OpLoad %v4f32 %tessc2_in_position_ptr\n"
4442 "%tessc2_out_color_ptr = OpAccessChain %op_v4f32 %out_color %tessc2_invocation_id\n"
4443 "%tessc2_out_position_ptr = OpAccessChain %op_v4f32 %out_position %tessc2_invocation_id\n"
4444 "%tessc2_transformed_color = OpFSub %v4f32 %cval %tessc2_in_color_val\n"
4445 "OpStore %tessc2_out_color_ptr %tessc2_transformed_color\n"
4446 "OpStore %tessc2_out_position_ptr %tessc2_in_position_val\n"
4447 "%tessc2_is_first_invocation = OpIEqual %bool %tessc2_invocation_id %c_i32_0\n"
4448 "OpSelectionMerge %tessc2_merge_label None\n"
4449 "OpBranchConditional %tessc2_is_first_invocation %tessc2_first_invocation %tessc2_merge_label\n"
4450 "%tessc2_first_invocation = OpLabel\n"
4451 "%tessc2_tess_outer_0 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_0\n"
4452 "%tessc2_tess_outer_1 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_1\n"
4453 "%tessc2_tess_outer_2 = OpAccessChain %op_f32 %gl_TessLevelOuter %c_i32_2\n"
4454 "%tessc2_tess_inner = OpAccessChain %op_f32 %gl_TessLevelInner %c_i32_0\n"
4455 "OpStore %tessc2_tess_outer_0 %c_f32_1\n"
4456 "OpStore %tessc2_tess_outer_1 %c_f32_1\n"
4457 "OpStore %tessc2_tess_outer_2 %c_f32_1\n"
4458 "OpStore %tessc2_tess_inner %c_f32_1\n"
4459 "OpBranch %tessc2_merge_label\n"
4460 "%tessc2_merge_label = OpLabel\n"
4464 dst.spirvAsmSources.add("tesse") <<
4465 "OpCapability Tessellation\n"
4466 "OpMemoryModel Logical GLSL450\n"
4467 "OpEntryPoint TessellationEvaluation %tesse1_main \"tesse1\" %stream %gl_tessCoord %in_position %out_color %in_color \n"
4468 "OpEntryPoint TessellationEvaluation %tesse2_main \"tesse2\" %stream %gl_tessCoord %in_position %out_color %in_color \n"
4469 "OpExecutionMode %tesse1_main Triangles\n"
4470 "OpExecutionMode %tesse2_main Triangles\n"
4471 "OpName %tesse1_main \"tesse1\"\n"
4472 "OpName %tesse2_main \"tesse2\"\n"
4473 "OpName %per_vertex_out \"gl_PerVertex\"\n"
4474 "OpMemberName %per_vertex_out 0 \"gl_Position\"\n"
4475 "OpMemberName %per_vertex_out 1 \"gl_PointSize\"\n"
4476 "OpMemberName %per_vertex_out 2 \"gl_ClipDistance\"\n"
4477 "OpMemberName %per_vertex_out 3 \"gl_CullDistance\"\n"
4478 "OpName %stream \"\"\n"
4479 "OpName %gl_tessCoord \"gl_TessCoord\"\n"
4480 "OpName %in_position \"in_position\"\n"
4481 "OpName %out_color \"out_color\"\n"
4482 "OpName %in_color \"in_color\"\n"
4483 "OpMemberDecorate %per_vertex_out 0 BuiltIn Position\n"
4484 "OpMemberDecorate %per_vertex_out 1 BuiltIn PointSize\n"
4485 "OpMemberDecorate %per_vertex_out 2 BuiltIn ClipDistance\n"
4486 "OpMemberDecorate %per_vertex_out 3 BuiltIn CullDistance\n"
4487 "OpDecorate %per_vertex_out Block\n"
4488 "OpDecorate %gl_tessCoord BuiltIn TessCoord\n"
4489 "OpDecorate %in_position Location 2\n"
4490 "OpDecorate %out_color Location 1\n"
4491 "OpDecorate %in_color Location 1\n"
4492 SPIRV_ASSEMBLY_TYPES
4493 SPIRV_ASSEMBLY_CONSTANTS
4494 SPIRV_ASSEMBLY_ARRAYS
4495 "%cval = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
4496 "%per_vertex_out = OpTypeStruct %v4f32 %f32 %a1f32 %a1f32\n"
4497 "%op_per_vertex_out = OpTypePointer Output %per_vertex_out\n"
4498 "%stream = OpVariable %op_per_vertex_out Output\n"
4499 "%gl_tessCoord = OpVariable %ip_v3f32 Input\n"
4500 "%in_position = OpVariable %ip_a32v4f32 Input\n"
4501 "%out_color = OpVariable %op_v4f32 Output\n"
4502 "%in_color = OpVariable %ip_a32v4f32 Input\n"
4504 "%tesse1_main = OpFunction %void None %fun\n"
4505 "%tesse1_label = OpLabel\n"
4506 "%tesse1_tc_0_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_0\n"
4507 "%tesse1_tc_1_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_1\n"
4508 "%tesse1_tc_2_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_2\n"
4509 "%tesse1_tc_0 = OpLoad %f32 %tesse1_tc_0_ptr\n"
4510 "%tesse1_tc_1 = OpLoad %f32 %tesse1_tc_1_ptr\n"
4511 "%tesse1_tc_2 = OpLoad %f32 %tesse1_tc_2_ptr\n"
4512 "%tesse1_in_pos_0_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_0\n"
4513 "%tesse1_in_pos_1_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_1\n"
4514 "%tesse1_in_pos_2_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_2\n"
4515 "%tesse1_in_pos_0 = OpLoad %v4f32 %tesse1_in_pos_0_ptr\n"
4516 "%tesse1_in_pos_1 = OpLoad %v4f32 %tesse1_in_pos_1_ptr\n"
4517 "%tesse1_in_pos_2 = OpLoad %v4f32 %tesse1_in_pos_2_ptr\n"
4518 "%tesse1_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_0 %tesse1_in_pos_0\n"
4519 "%tesse1_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_1 %tesse1_in_pos_1\n"
4520 "%tesse1_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_2 %tesse1_in_pos_2\n"
4521 "%tesse1_out_pos_ptr = OpAccessChain %op_v4f32 %stream %c_i32_0\n"
4522 "%tesse1_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %tesse1_in_pos_0_weighted %tesse1_in_pos_1_weighted\n"
4523 "%tesse1_computed_out = OpFAdd %v4f32 %tesse1_in_pos_0_plus_pos_1 %tesse1_in_pos_2_weighted\n"
4524 "OpStore %tesse1_out_pos_ptr %tesse1_computed_out\n"
4525 "%tesse1_in_clr_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4526 "%tesse1_in_clr_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4527 "%tesse1_in_clr_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4528 "%tesse1_in_clr_0 = OpLoad %v4f32 %tesse1_in_clr_0_ptr\n"
4529 "%tesse1_in_clr_1 = OpLoad %v4f32 %tesse1_in_clr_1_ptr\n"
4530 "%tesse1_in_clr_2 = OpLoad %v4f32 %tesse1_in_clr_2_ptr\n"
4531 "%tesse1_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_0 %tesse1_in_clr_0\n"
4532 "%tesse1_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_1 %tesse1_in_clr_1\n"
4533 "%tesse1_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %tesse1_tc_2 %tesse1_in_clr_2\n"
4534 "%tesse1_in_clr_0_plus_col_1 = OpFAdd %v4f32 %tesse1_in_clr_0_weighted %tesse1_in_clr_1_weighted\n"
4535 "%tesse1_computed_clr = OpFAdd %v4f32 %tesse1_in_clr_0_plus_col_1 %tesse1_in_clr_2_weighted\n"
4536 "OpStore %out_color %tesse1_computed_clr\n"
4540 "%tesse2_main = OpFunction %void None %fun\n"
4541 "%tesse2_label = OpLabel\n"
4542 "%tesse2_tc_0_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_0\n"
4543 "%tesse2_tc_1_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_1\n"
4544 "%tesse2_tc_2_ptr = OpAccessChain %ip_f32 %gl_tessCoord %c_u32_2\n"
4545 "%tesse2_tc_0 = OpLoad %f32 %tesse2_tc_0_ptr\n"
4546 "%tesse2_tc_1 = OpLoad %f32 %tesse2_tc_1_ptr\n"
4547 "%tesse2_tc_2 = OpLoad %f32 %tesse2_tc_2_ptr\n"
4548 "%tesse2_in_pos_0_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_0\n"
4549 "%tesse2_in_pos_1_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_1\n"
4550 "%tesse2_in_pos_2_ptr = OpAccessChain %ip_v4f32 %in_position %c_i32_2\n"
4551 "%tesse2_in_pos_0 = OpLoad %v4f32 %tesse2_in_pos_0_ptr\n"
4552 "%tesse2_in_pos_1 = OpLoad %v4f32 %tesse2_in_pos_1_ptr\n"
4553 "%tesse2_in_pos_2 = OpLoad %v4f32 %tesse2_in_pos_2_ptr\n"
4554 "%tesse2_in_pos_0_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_0 %tesse2_in_pos_0\n"
4555 "%tesse2_in_pos_1_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_1 %tesse2_in_pos_1\n"
4556 "%tesse2_in_pos_2_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_2 %tesse2_in_pos_2\n"
4557 "%tesse2_out_pos_ptr = OpAccessChain %op_v4f32 %stream %c_i32_0\n"
4558 "%tesse2_in_pos_0_plus_pos_1 = OpFAdd %v4f32 %tesse2_in_pos_0_weighted %tesse2_in_pos_1_weighted\n"
4559 "%tesse2_computed_out = OpFAdd %v4f32 %tesse2_in_pos_0_plus_pos_1 %tesse2_in_pos_2_weighted\n"
4560 "OpStore %tesse2_out_pos_ptr %tesse2_computed_out\n"
4561 "%tesse2_in_clr_0_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_0\n"
4562 "%tesse2_in_clr_1_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_1\n"
4563 "%tesse2_in_clr_2_ptr = OpAccessChain %ip_v4f32 %in_color %c_i32_2\n"
4564 "%tesse2_in_clr_0 = OpLoad %v4f32 %tesse2_in_clr_0_ptr\n"
4565 "%tesse2_in_clr_1 = OpLoad %v4f32 %tesse2_in_clr_1_ptr\n"
4566 "%tesse2_in_clr_2 = OpLoad %v4f32 %tesse2_in_clr_2_ptr\n"
4567 "%tesse2_in_clr_0_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_0 %tesse2_in_clr_0\n"
4568 "%tesse2_in_clr_1_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_1 %tesse2_in_clr_1\n"
4569 "%tesse2_in_clr_2_weighted = OpVectorTimesScalar %v4f32 %tesse2_tc_2 %tesse2_in_clr_2\n"
4570 "%tesse2_in_clr_0_plus_col_1 = OpFAdd %v4f32 %tesse2_in_clr_0_weighted %tesse2_in_clr_1_weighted\n"
4571 "%tesse2_computed_clr = OpFAdd %v4f32 %tesse2_in_clr_0_plus_col_1 %tesse2_in_clr_2_weighted\n"
4572 "%tesse2_clr_transformed = OpFSub %v4f32 %cval %tesse2_computed_clr\n"
4573 "OpStore %out_color %tesse2_clr_transformed\n"
4578 // Sets up and runs a Vulkan pipeline, then spot-checks the resulting image.
4579 // Feeds the pipeline a set of colored triangles, which then must occur in the
4580 // rendered image. The surface is cleared before executing the pipeline, so
4581 // whatever the shaders draw can be directly spot-checked.
4582 TestStatus runAndVerifyDefaultPipeline (Context& context, InstanceContext instance)
4584 const VkDevice vkDevice = context.getDevice();
4585 const DeviceInterface& vk = context.getDeviceInterface();
4586 const VkQueue queue = context.getUniversalQueue();
4587 const deUint32 queueFamilyIndex = context.getUniversalQueueFamilyIndex();
4588 const tcu::UVec2 renderSize (256, 256);
4589 vector<ModuleHandleSp> modules;
4590 map<VkShaderStageFlagBits, VkShaderModule> moduleByStage;
4591 const int testSpecificSeed = 31354125;
4592 const int seed = context.getTestContext().getCommandLine().getBaseSeed() ^ testSpecificSeed;
4593 bool supportsGeometry = false;
4594 bool supportsTessellation = false;
4595 bool hasTessellation = false;
4597 const VkPhysicalDeviceFeatures& features = context.getDeviceFeatures();
4598 supportsGeometry = features.geometryShader == VK_TRUE;
4599 supportsTessellation = features.tessellationShader == VK_TRUE;
4600 hasTessellation = (instance.requiredStages & VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT) ||
4601 (instance.requiredStages & VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT);
4603 if (hasTessellation && !supportsTessellation)
4605 throw tcu::NotSupportedError(std::string("Tessellation not supported"));
4608 if ((instance.requiredStages & VK_SHADER_STAGE_GEOMETRY_BIT) &&
4611 throw tcu::NotSupportedError(std::string("Geometry not supported"));
4614 de::Random(seed).shuffle(instance.inputColors, instance.inputColors+4);
4615 de::Random(seed).shuffle(instance.outputColors, instance.outputColors+4);
4616 const Vec4 vertexData[] =
4618 // Upper left corner:
4619 Vec4(-1.0f, -1.0f, 0.0f, 1.0f), instance.inputColors[0].toVec(),
4620 Vec4(-0.5f, -1.0f, 0.0f, 1.0f), instance.inputColors[0].toVec(),
4621 Vec4(-1.0f, -0.5f, 0.0f, 1.0f), instance.inputColors[0].toVec(),
4623 // Upper right corner:
4624 Vec4(+0.5f, -1.0f, 0.0f, 1.0f), instance.inputColors[1].toVec(),
4625 Vec4(+1.0f, -1.0f, 0.0f, 1.0f), instance.inputColors[1].toVec(),
4626 Vec4(+1.0f, -0.5f, 0.0f, 1.0f), instance.inputColors[1].toVec(),
4628 // Lower left corner:
4629 Vec4(-1.0f, +0.5f, 0.0f, 1.0f), instance.inputColors[2].toVec(),
4630 Vec4(-0.5f, +1.0f, 0.0f, 1.0f), instance.inputColors[2].toVec(),
4631 Vec4(-1.0f, +1.0f, 0.0f, 1.0f), instance.inputColors[2].toVec(),
4633 // Lower right corner:
4634 Vec4(+1.0f, +0.5f, 0.0f, 1.0f), instance.inputColors[3].toVec(),
4635 Vec4(+1.0f, +1.0f, 0.0f, 1.0f), instance.inputColors[3].toVec(),
4636 Vec4(+0.5f, +1.0f, 0.0f, 1.0f), instance.inputColors[3].toVec()
4638 const size_t singleVertexDataSize = 2 * sizeof(Vec4);
4639 const size_t vertexCount = sizeof(vertexData) / singleVertexDataSize;
4641 const VkBufferCreateInfo vertexBufferParams =
4643 VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType;
4644 DE_NULL, // const void* pNext;
4645 0u, // VkBufferCreateFlags flags;
4646 (VkDeviceSize)sizeof(vertexData), // VkDeviceSize size;
4647 VK_BUFFER_USAGE_VERTEX_BUFFER_BIT, // VkBufferUsageFlags usage;
4648 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
4649 1u, // deUint32 queueFamilyCount;
4650 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices;
4652 const Unique<VkBuffer> vertexBuffer (createBuffer(vk, vkDevice, &vertexBufferParams));
4653 const UniquePtr<Allocation> vertexBufferMemory (context.getDefaultAllocator().allocate(getBufferMemoryRequirements(vk, vkDevice, *vertexBuffer), MemoryRequirement::HostVisible));
4655 VK_CHECK(vk.bindBufferMemory(vkDevice, *vertexBuffer, vertexBufferMemory->getMemory(), vertexBufferMemory->getOffset()));
4657 const VkDeviceSize imageSizeBytes = (VkDeviceSize)(sizeof(deUint32)*renderSize.x()*renderSize.y());
4658 const VkBufferCreateInfo readImageBufferParams =
4660 VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType;
4661 DE_NULL, // const void* pNext;
4662 0u, // VkBufferCreateFlags flags;
4663 imageSizeBytes, // VkDeviceSize size;
4664 VK_BUFFER_USAGE_TRANSFER_DST_BIT, // VkBufferUsageFlags usage;
4665 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
4666 1u, // deUint32 queueFamilyCount;
4667 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices;
4669 const Unique<VkBuffer> readImageBuffer (createBuffer(vk, vkDevice, &readImageBufferParams));
4670 const UniquePtr<Allocation> readImageBufferMemory (context.getDefaultAllocator().allocate(getBufferMemoryRequirements(vk, vkDevice, *readImageBuffer), MemoryRequirement::HostVisible));
4672 VK_CHECK(vk.bindBufferMemory(vkDevice, *readImageBuffer, readImageBufferMemory->getMemory(), readImageBufferMemory->getOffset()));
4674 const VkImageCreateInfo imageParams =
4676 VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, // VkStructureType sType;
4677 DE_NULL, // const void* pNext;
4678 0u, // VkImageCreateFlags flags;
4679 VK_IMAGE_TYPE_2D, // VkImageType imageType;
4680 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format;
4681 { renderSize.x(), renderSize.y(), 1 }, // VkExtent3D extent;
4682 1u, // deUint32 mipLevels;
4683 1u, // deUint32 arraySize;
4684 VK_SAMPLE_COUNT_1_BIT, // deUint32 samples;
4685 VK_IMAGE_TILING_OPTIMAL, // VkImageTiling tiling;
4686 VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT|VK_IMAGE_USAGE_TRANSFER_SRC_BIT, // VkImageUsageFlags usage;
4687 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
4688 1u, // deUint32 queueFamilyCount;
4689 &queueFamilyIndex, // const deUint32* pQueueFamilyIndices;
4690 VK_IMAGE_LAYOUT_UNDEFINED, // VkImageLayout initialLayout;
4693 const Unique<VkImage> image (createImage(vk, vkDevice, &imageParams));
4694 const UniquePtr<Allocation> imageMemory (context.getDefaultAllocator().allocate(getImageMemoryRequirements(vk, vkDevice, *image), MemoryRequirement::Any));
4696 VK_CHECK(vk.bindImageMemory(vkDevice, *image, imageMemory->getMemory(), imageMemory->getOffset()));
4698 const VkAttachmentDescription colorAttDesc =
4700 0u, // VkAttachmentDescriptionFlags flags;
4701 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format;
4702 VK_SAMPLE_COUNT_1_BIT, // deUint32 samples;
4703 VK_ATTACHMENT_LOAD_OP_CLEAR, // VkAttachmentLoadOp loadOp;
4704 VK_ATTACHMENT_STORE_OP_STORE, // VkAttachmentStoreOp storeOp;
4705 VK_ATTACHMENT_LOAD_OP_DONT_CARE, // VkAttachmentLoadOp stencilLoadOp;
4706 VK_ATTACHMENT_STORE_OP_DONT_CARE, // VkAttachmentStoreOp stencilStoreOp;
4707 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout initialLayout;
4708 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout finalLayout;
4710 const VkAttachmentReference colorAttRef =
4712 0u, // deUint32 attachment;
4713 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout layout;
4715 const VkSubpassDescription subpassDesc =
4717 0u, // VkSubpassDescriptionFlags flags;
4718 VK_PIPELINE_BIND_POINT_GRAPHICS, // VkPipelineBindPoint pipelineBindPoint;
4719 0u, // deUint32 inputCount;
4720 DE_NULL, // const VkAttachmentReference* pInputAttachments;
4721 1u, // deUint32 colorCount;
4722 &colorAttRef, // const VkAttachmentReference* pColorAttachments;
4723 DE_NULL, // const VkAttachmentReference* pResolveAttachments;
4724 DE_NULL, // const VkAttachmentReference* pDepthStencilAttachment;
4725 0u, // deUint32 preserveCount;
4726 DE_NULL, // const VkAttachmentReference* pPreserveAttachments;
4729 const VkRenderPassCreateInfo renderPassParams =
4731 VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO, // VkStructureType sType;
4732 DE_NULL, // const void* pNext;
4733 (VkRenderPassCreateFlags)0,
4734 1u, // deUint32 attachmentCount;
4735 &colorAttDesc, // const VkAttachmentDescription* pAttachments;
4736 1u, // deUint32 subpassCount;
4737 &subpassDesc, // const VkSubpassDescription* pSubpasses;
4738 0u, // deUint32 dependencyCount;
4739 DE_NULL, // const VkSubpassDependency* pDependencies;
4741 const Unique<VkRenderPass> renderPass (createRenderPass(vk, vkDevice, &renderPassParams));
4743 const VkImageViewCreateInfo colorAttViewParams =
4745 VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, // VkStructureType sType;
4746 DE_NULL, // const void* pNext;
4747 0u, // VkImageViewCreateFlags flags;
4748 *image, // VkImage image;
4749 VK_IMAGE_VIEW_TYPE_2D, // VkImageViewType viewType;
4750 VK_FORMAT_R8G8B8A8_UNORM, // VkFormat format;
4752 VK_COMPONENT_SWIZZLE_R,
4753 VK_COMPONENT_SWIZZLE_G,
4754 VK_COMPONENT_SWIZZLE_B,
4755 VK_COMPONENT_SWIZZLE_A
4756 }, // VkChannelMapping channels;
4758 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspectFlags aspectMask;
4759 0u, // deUint32 baseMipLevel;
4760 1u, // deUint32 mipLevels;
4761 0u, // deUint32 baseArrayLayer;
4762 1u, // deUint32 arraySize;
4763 }, // VkImageSubresourceRange subresourceRange;
4765 const Unique<VkImageView> colorAttView (createImageView(vk, vkDevice, &colorAttViewParams));
4769 const VkPipelineLayoutCreateInfo pipelineLayoutParams =
4771 VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, // VkStructureType sType;
4772 DE_NULL, // const void* pNext;
4773 (VkPipelineLayoutCreateFlags)0,
4774 0u, // deUint32 descriptorSetCount;
4775 DE_NULL, // const VkDescriptorSetLayout* pSetLayouts;
4776 0u, // deUint32 pushConstantRangeCount;
4777 DE_NULL, // const VkPushConstantRange* pPushConstantRanges;
4779 const Unique<VkPipelineLayout> pipelineLayout (createPipelineLayout(vk, vkDevice, &pipelineLayoutParams));
4782 vector<VkPipelineShaderStageCreateInfo> shaderStageParams;
4783 // We need these vectors to make sure that information about specialization constants for each stage can outlive createGraphicsPipeline().
4784 vector<vector<VkSpecializationMapEntry> > specConstantEntries;
4785 vector<VkSpecializationInfo> specializationInfos;
4786 createPipelineShaderStages(vk, vkDevice, instance, context, modules, shaderStageParams);
4788 // And we don't want the reallocation of these vectors to invalidate pointers pointing to their contents.
4789 specConstantEntries.reserve(shaderStageParams.size());
4790 specializationInfos.reserve(shaderStageParams.size());
4792 // Patch the specialization info field in PipelineShaderStageCreateInfos.
4793 for (vector<VkPipelineShaderStageCreateInfo>::iterator stageInfo = shaderStageParams.begin(); stageInfo != shaderStageParams.end(); ++stageInfo)
4795 const StageToSpecConstantMap::const_iterator stageIt = instance.specConstants.find(stageInfo->stage);
4797 if (stageIt != instance.specConstants.end())
4799 const size_t numSpecConstants = stageIt->second.size();
4800 vector<VkSpecializationMapEntry> entries;
4801 VkSpecializationInfo specInfo;
4803 entries.resize(numSpecConstants);
4805 // Only support 32-bit integers as spec constants now. And their constant IDs are numbered sequentially starting from 0.
4806 for (size_t ndx = 0; ndx < numSpecConstants; ++ndx)
4808 entries[ndx].constantID = (deUint32)ndx;
4809 entries[ndx].offset = deUint32(ndx * sizeof(deInt32));
4810 entries[ndx].size = sizeof(deInt32);
4813 specConstantEntries.push_back(entries);
4815 specInfo.mapEntryCount = (deUint32)numSpecConstants;
4816 specInfo.pMapEntries = specConstantEntries.back().data();
4817 specInfo.dataSize = numSpecConstants * sizeof(deInt32);
4818 specInfo.pData = stageIt->second.data();
4819 specializationInfos.push_back(specInfo);
4821 stageInfo->pSpecializationInfo = &specializationInfos.back();
4824 const VkPipelineDepthStencilStateCreateInfo depthStencilParams =
4826 VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO, // VkStructureType sType;
4827 DE_NULL, // const void* pNext;
4828 (VkPipelineDepthStencilStateCreateFlags)0,
4829 DE_FALSE, // deUint32 depthTestEnable;
4830 DE_FALSE, // deUint32 depthWriteEnable;
4831 VK_COMPARE_OP_ALWAYS, // VkCompareOp depthCompareOp;
4832 DE_FALSE, // deUint32 depthBoundsTestEnable;
4833 DE_FALSE, // deUint32 stencilTestEnable;
4835 VK_STENCIL_OP_KEEP, // VkStencilOp stencilFailOp;
4836 VK_STENCIL_OP_KEEP, // VkStencilOp stencilPassOp;
4837 VK_STENCIL_OP_KEEP, // VkStencilOp stencilDepthFailOp;
4838 VK_COMPARE_OP_ALWAYS, // VkCompareOp stencilCompareOp;
4839 0u, // deUint32 stencilCompareMask;
4840 0u, // deUint32 stencilWriteMask;
4841 0u, // deUint32 stencilReference;
4842 }, // VkStencilOpState front;
4844 VK_STENCIL_OP_KEEP, // VkStencilOp stencilFailOp;
4845 VK_STENCIL_OP_KEEP, // VkStencilOp stencilPassOp;
4846 VK_STENCIL_OP_KEEP, // VkStencilOp stencilDepthFailOp;
4847 VK_COMPARE_OP_ALWAYS, // VkCompareOp stencilCompareOp;
4848 0u, // deUint32 stencilCompareMask;
4849 0u, // deUint32 stencilWriteMask;
4850 0u, // deUint32 stencilReference;
4851 }, // VkStencilOpState back;
4852 -1.0f, // float minDepthBounds;
4853 +1.0f, // float maxDepthBounds;
4855 const VkViewport viewport0 =
4857 0.0f, // float originX;
4858 0.0f, // float originY;
4859 (float)renderSize.x(), // float width;
4860 (float)renderSize.y(), // float height;
4861 0.0f, // float minDepth;
4862 1.0f, // float maxDepth;
4864 const VkRect2D scissor0 =
4869 }, // VkOffset2D offset;
4871 renderSize.x(), // deInt32 width;
4872 renderSize.y(), // deInt32 height;
4873 }, // VkExtent2D extent;
4875 const VkPipelineViewportStateCreateInfo viewportParams =
4877 VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO, // VkStructureType sType;
4878 DE_NULL, // const void* pNext;
4879 (VkPipelineViewportStateCreateFlags)0,
4880 1u, // deUint32 viewportCount;
4885 const VkSampleMask sampleMask = ~0u;
4886 const VkPipelineMultisampleStateCreateInfo multisampleParams =
4888 VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO, // VkStructureType sType;
4889 DE_NULL, // const void* pNext;
4890 (VkPipelineMultisampleStateCreateFlags)0,
4891 VK_SAMPLE_COUNT_1_BIT, // VkSampleCountFlagBits rasterSamples;
4892 DE_FALSE, // deUint32 sampleShadingEnable;
4893 0.0f, // float minSampleShading;
4894 &sampleMask, // const VkSampleMask* pSampleMask;
4895 DE_FALSE, // VkBool32 alphaToCoverageEnable;
4896 DE_FALSE, // VkBool32 alphaToOneEnable;
4898 const VkPipelineRasterizationStateCreateInfo rasterParams =
4900 VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO, // VkStructureType sType;
4901 DE_NULL, // const void* pNext;
4902 (VkPipelineRasterizationStateCreateFlags)0,
4903 DE_TRUE, // deUint32 depthClipEnable;
4904 DE_FALSE, // deUint32 rasterizerDiscardEnable;
4905 VK_POLYGON_MODE_FILL, // VkFillMode fillMode;
4906 VK_CULL_MODE_NONE, // VkCullMode cullMode;
4907 VK_FRONT_FACE_COUNTER_CLOCKWISE, // VkFrontFace frontFace;
4908 VK_FALSE, // VkBool32 depthBiasEnable;
4909 0.0f, // float depthBias;
4910 0.0f, // float depthBiasClamp;
4911 0.0f, // float slopeScaledDepthBias;
4912 1.0f, // float lineWidth;
4914 const VkPrimitiveTopology topology = hasTessellation? VK_PRIMITIVE_TOPOLOGY_PATCH_LIST: VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
4915 const VkPipelineInputAssemblyStateCreateInfo inputAssemblyParams =
4917 VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO, // VkStructureType sType;
4918 DE_NULL, // const void* pNext;
4919 (VkPipelineInputAssemblyStateCreateFlags)0,
4920 topology, // VkPrimitiveTopology topology;
4921 DE_FALSE, // deUint32 primitiveRestartEnable;
4923 const VkVertexInputBindingDescription vertexBinding0 =
4925 0u, // deUint32 binding;
4926 deUint32(singleVertexDataSize), // deUint32 strideInBytes;
4927 VK_VERTEX_INPUT_RATE_VERTEX // VkVertexInputStepRate stepRate;
4929 const VkVertexInputAttributeDescription vertexAttrib0[2] =
4932 0u, // deUint32 location;
4933 0u, // deUint32 binding;
4934 VK_FORMAT_R32G32B32A32_SFLOAT, // VkFormat format;
4935 0u // deUint32 offsetInBytes;
4938 1u, // deUint32 location;
4939 0u, // deUint32 binding;
4940 VK_FORMAT_R32G32B32A32_SFLOAT, // VkFormat format;
4941 sizeof(Vec4), // deUint32 offsetInBytes;
4945 const VkPipelineVertexInputStateCreateInfo vertexInputStateParams =
4947 VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO, // VkStructureType sType;
4948 DE_NULL, // const void* pNext;
4949 (VkPipelineVertexInputStateCreateFlags)0,
4950 1u, // deUint32 bindingCount;
4951 &vertexBinding0, // const VkVertexInputBindingDescription* pVertexBindingDescriptions;
4952 2u, // deUint32 attributeCount;
4953 vertexAttrib0, // const VkVertexInputAttributeDescription* pVertexAttributeDescriptions;
4955 const VkPipelineColorBlendAttachmentState attBlendParams =
4957 DE_FALSE, // deUint32 blendEnable;
4958 VK_BLEND_FACTOR_ONE, // VkBlend srcBlendColor;
4959 VK_BLEND_FACTOR_ZERO, // VkBlend destBlendColor;
4960 VK_BLEND_OP_ADD, // VkBlendOp blendOpColor;
4961 VK_BLEND_FACTOR_ONE, // VkBlend srcBlendAlpha;
4962 VK_BLEND_FACTOR_ZERO, // VkBlend destBlendAlpha;
4963 VK_BLEND_OP_ADD, // VkBlendOp blendOpAlpha;
4964 (VK_COLOR_COMPONENT_R_BIT|
4965 VK_COLOR_COMPONENT_G_BIT|
4966 VK_COLOR_COMPONENT_B_BIT|
4967 VK_COLOR_COMPONENT_A_BIT), // VkChannelFlags channelWriteMask;
4969 const VkPipelineColorBlendStateCreateInfo blendParams =
4971 VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO, // VkStructureType sType;
4972 DE_NULL, // const void* pNext;
4973 (VkPipelineColorBlendStateCreateFlags)0,
4974 DE_FALSE, // VkBool32 logicOpEnable;
4975 VK_LOGIC_OP_COPY, // VkLogicOp logicOp;
4976 1u, // deUint32 attachmentCount;
4977 &attBlendParams, // const VkPipelineColorBlendAttachmentState* pAttachments;
4978 { 0.0f, 0.0f, 0.0f, 0.0f }, // float blendConst[4];
4980 const VkPipelineDynamicStateCreateInfo dynamicStateInfo =
4982 VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO, // VkStructureType sType;
4983 DE_NULL, // const void* pNext;
4984 (VkPipelineDynamicStateCreateFlags)0,
4985 0u, // deUint32 dynamicStateCount;
4986 DE_NULL // const VkDynamicState* pDynamicStates;
4989 const VkPipelineTessellationStateCreateInfo tessellationState =
4991 VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO,
4993 (VkPipelineTessellationStateCreateFlags)0,
4997 const VkPipelineTessellationStateCreateInfo* tessellationInfo = hasTessellation ? &tessellationState: DE_NULL;
4998 const VkGraphicsPipelineCreateInfo pipelineParams =
5000 VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, // VkStructureType sType;
5001 DE_NULL, // const void* pNext;
5002 0u, // VkPipelineCreateFlags flags;
5003 (deUint32)shaderStageParams.size(), // deUint32 stageCount;
5004 &shaderStageParams[0], // const VkPipelineShaderStageCreateInfo* pStages;
5005 &vertexInputStateParams, // const VkPipelineVertexInputStateCreateInfo* pVertexInputState;
5006 &inputAssemblyParams, // const VkPipelineInputAssemblyStateCreateInfo* pInputAssemblyState;
5007 tessellationInfo, // const VkPipelineTessellationStateCreateInfo* pTessellationState;
5008 &viewportParams, // const VkPipelineViewportStateCreateInfo* pViewportState;
5009 &rasterParams, // const VkPipelineRasterStateCreateInfo* pRasterState;
5010 &multisampleParams, // const VkPipelineMultisampleStateCreateInfo* pMultisampleState;
5011 &depthStencilParams, // const VkPipelineDepthStencilStateCreateInfo* pDepthStencilState;
5012 &blendParams, // const VkPipelineColorBlendStateCreateInfo* pColorBlendState;
5013 &dynamicStateInfo, // const VkPipelineDynamicStateCreateInfo* pDynamicState;
5014 *pipelineLayout, // VkPipelineLayout layout;
5015 *renderPass, // VkRenderPass renderPass;
5016 0u, // deUint32 subpass;
5017 DE_NULL, // VkPipeline basePipelineHandle;
5018 0u, // deInt32 basePipelineIndex;
5021 const Unique<VkPipeline> pipeline (createGraphicsPipeline(vk, vkDevice, DE_NULL, &pipelineParams));
5024 const VkFramebufferCreateInfo framebufferParams =
5026 VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO, // VkStructureType sType;
5027 DE_NULL, // const void* pNext;
5028 (VkFramebufferCreateFlags)0,
5029 *renderPass, // VkRenderPass renderPass;
5030 1u, // deUint32 attachmentCount;
5031 &*colorAttView, // const VkImageView* pAttachments;
5032 (deUint32)renderSize.x(), // deUint32 width;
5033 (deUint32)renderSize.y(), // deUint32 height;
5034 1u, // deUint32 layers;
5036 const Unique<VkFramebuffer> framebuffer (createFramebuffer(vk, vkDevice, &framebufferParams));
5038 const VkCommandPoolCreateInfo cmdPoolParams =
5040 VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO, // VkStructureType sType;
5041 DE_NULL, // const void* pNext;
5042 VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, // VkCmdPoolCreateFlags flags;
5043 queueFamilyIndex, // deUint32 queueFamilyIndex;
5045 const Unique<VkCommandPool> cmdPool (createCommandPool(vk, vkDevice, &cmdPoolParams));
5048 const VkCommandBufferAllocateInfo cmdBufParams =
5050 VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO, // VkStructureType sType;
5051 DE_NULL, // const void* pNext;
5052 *cmdPool, // VkCmdPool pool;
5053 VK_COMMAND_BUFFER_LEVEL_PRIMARY, // VkCmdBufferLevel level;
5054 1u, // deUint32 count;
5056 const Unique<VkCommandBuffer> cmdBuf (allocateCommandBuffer(vk, vkDevice, &cmdBufParams));
5058 const VkCommandBufferBeginInfo cmdBufBeginParams =
5060 VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, // VkStructureType sType;
5061 DE_NULL, // const void* pNext;
5062 (VkCommandBufferUsageFlags)0,
5063 (const VkCommandBufferInheritanceInfo*)DE_NULL,
5067 VK_CHECK(vk.beginCommandBuffer(*cmdBuf, &cmdBufBeginParams));
5070 const VkMemoryBarrier vertFlushBarrier =
5072 VK_STRUCTURE_TYPE_MEMORY_BARRIER, // VkStructureType sType;
5073 DE_NULL, // const void* pNext;
5074 VK_ACCESS_HOST_WRITE_BIT, // VkMemoryOutputFlags outputMask;
5075 VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT, // VkMemoryInputFlags inputMask;
5077 const VkImageMemoryBarrier colorAttBarrier =
5079 VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType;
5080 DE_NULL, // const void* pNext;
5081 0u, // VkMemoryOutputFlags outputMask;
5082 VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, // VkMemoryInputFlags inputMask;
5083 VK_IMAGE_LAYOUT_UNDEFINED, // VkImageLayout oldLayout;
5084 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout newLayout;
5085 queueFamilyIndex, // deUint32 srcQueueFamilyIndex;
5086 queueFamilyIndex, // deUint32 destQueueFamilyIndex;
5087 *image, // VkImage image;
5089 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspect aspect;
5090 0u, // deUint32 baseMipLevel;
5091 1u, // deUint32 mipLevels;
5092 0u, // deUint32 baseArraySlice;
5093 1u, // deUint32 arraySize;
5094 } // VkImageSubresourceRange subresourceRange;
5096 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);
5100 const VkClearValue clearValue = makeClearValueColorF32(0.125f, 0.25f, 0.75f, 1.0f);
5101 const VkRenderPassBeginInfo passBeginParams =
5103 VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO, // VkStructureType sType;
5104 DE_NULL, // const void* pNext;
5105 *renderPass, // VkRenderPass renderPass;
5106 *framebuffer, // VkFramebuffer framebuffer;
5107 { { 0, 0 }, { renderSize.x(), renderSize.y() } }, // VkRect2D renderArea;
5108 1u, // deUint32 clearValueCount;
5109 &clearValue, // const VkClearValue* pClearValues;
5111 vk.cmdBeginRenderPass(*cmdBuf, &passBeginParams, VK_SUBPASS_CONTENTS_INLINE);
5114 vk.cmdBindPipeline(*cmdBuf, VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline);
5116 const VkDeviceSize bindingOffset = 0;
5117 vk.cmdBindVertexBuffers(*cmdBuf, 0u, 1u, &vertexBuffer.get(), &bindingOffset);
5119 vk.cmdDraw(*cmdBuf, deUint32(vertexCount), 1u /*run pipeline once*/, 0u /*first vertex*/, 0u /*first instanceIndex*/);
5120 vk.cmdEndRenderPass(*cmdBuf);
5123 const VkImageMemoryBarrier renderFinishBarrier =
5125 VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType;
5126 DE_NULL, // const void* pNext;
5127 VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, // VkMemoryOutputFlags outputMask;
5128 VK_ACCESS_TRANSFER_READ_BIT, // VkMemoryInputFlags inputMask;
5129 VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, // VkImageLayout oldLayout;
5130 VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, // VkImageLayout newLayout;
5131 queueFamilyIndex, // deUint32 srcQueueFamilyIndex;
5132 queueFamilyIndex, // deUint32 destQueueFamilyIndex;
5133 *image, // VkImage image;
5135 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspectFlags aspectMask;
5136 0u, // deUint32 baseMipLevel;
5137 1u, // deUint32 mipLevels;
5138 0u, // deUint32 baseArraySlice;
5139 1u, // deUint32 arraySize;
5140 } // VkImageSubresourceRange subresourceRange;
5142 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);
5146 const VkBufferImageCopy copyParams =
5148 (VkDeviceSize)0u, // VkDeviceSize bufferOffset;
5149 (deUint32)renderSize.x(), // deUint32 bufferRowLength;
5150 (deUint32)renderSize.y(), // deUint32 bufferImageHeight;
5152 VK_IMAGE_ASPECT_COLOR_BIT, // VkImageAspect aspect;
5153 0u, // deUint32 mipLevel;
5154 0u, // deUint32 arrayLayer;
5155 1u, // deUint32 arraySize;
5156 }, // VkImageSubresourceCopy imageSubresource;
5157 { 0u, 0u, 0u }, // VkOffset3D imageOffset;
5158 { renderSize.x(), renderSize.y(), 1u } // VkExtent3D imageExtent;
5160 vk.cmdCopyImageToBuffer(*cmdBuf, *image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, *readImageBuffer, 1u, ©Params);
5164 const VkBufferMemoryBarrier copyFinishBarrier =
5166 VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, // VkStructureType sType;
5167 DE_NULL, // const void* pNext;
5168 VK_ACCESS_TRANSFER_WRITE_BIT, // VkMemoryOutputFlags outputMask;
5169 VK_ACCESS_HOST_READ_BIT, // VkMemoryInputFlags inputMask;
5170 queueFamilyIndex, // deUint32 srcQueueFamilyIndex;
5171 queueFamilyIndex, // deUint32 destQueueFamilyIndex;
5172 *readImageBuffer, // VkBuffer buffer;
5173 0u, // VkDeviceSize offset;
5174 imageSizeBytes // VkDeviceSize size;
5176 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);
5179 VK_CHECK(vk.endCommandBuffer(*cmdBuf));
5181 // Upload vertex data
5183 const VkMappedMemoryRange range =
5185 VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, // VkStructureType sType;
5186 DE_NULL, // const void* pNext;
5187 vertexBufferMemory->getMemory(), // VkDeviceMemory mem;
5188 0, // VkDeviceSize offset;
5189 (VkDeviceSize)sizeof(vertexData), // VkDeviceSize size;
5191 void* vertexBufPtr = vertexBufferMemory->getHostPtr();
5193 deMemcpy(vertexBufPtr, &vertexData[0], sizeof(vertexData));
5194 VK_CHECK(vk.flushMappedMemoryRanges(vkDevice, 1u, &range));
5197 // Submit & wait for completion
5199 const VkFenceCreateInfo fenceParams =
5201 VK_STRUCTURE_TYPE_FENCE_CREATE_INFO, // VkStructureType sType;
5202 DE_NULL, // const void* pNext;
5203 0u, // VkFenceCreateFlags flags;
5205 const Unique<VkFence> fence (createFence(vk, vkDevice, &fenceParams));
5206 const VkSubmitInfo submitInfo =
5208 VK_STRUCTURE_TYPE_SUBMIT_INFO,
5211 (const VkSemaphore*)DE_NULL,
5212 (const VkPipelineStageFlags*)DE_NULL,
5216 (const VkSemaphore*)DE_NULL,
5219 VK_CHECK(vk.queueSubmit(queue, 1u, &submitInfo, *fence));
5220 VK_CHECK(vk.waitForFences(vkDevice, 1u, &fence.get(), DE_TRUE, ~0ull));
5223 const void* imagePtr = readImageBufferMemory->getHostPtr();
5224 const tcu::ConstPixelBufferAccess pixelBuffer(tcu::TextureFormat(tcu::TextureFormat::RGBA, tcu::TextureFormat::UNORM_INT8),
5225 renderSize.x(), renderSize.y(), 1, imagePtr);
5228 const VkMappedMemoryRange range =
5230 VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, // VkStructureType sType;
5231 DE_NULL, // const void* pNext;
5232 readImageBufferMemory->getMemory(), // VkDeviceMemory mem;
5233 0, // VkDeviceSize offset;
5234 imageSizeBytes, // VkDeviceSize size;
5237 VK_CHECK(vk.invalidateMappedMemoryRanges(vkDevice, 1u, &range));
5238 context.getTestContext().getLog() << TestLog::Image("Result", "Result", pixelBuffer);
5241 const RGBA threshold(1, 1, 1, 1);
5242 const RGBA upperLeft(pixelBuffer.getPixel(1, 1));
5243 if (!tcu::compareThreshold(upperLeft, instance.outputColors[0], threshold))
5244 return TestStatus::fail("Upper left corner mismatch");
5246 const RGBA upperRight(pixelBuffer.getPixel(pixelBuffer.getWidth() - 1, 1));
5247 if (!tcu::compareThreshold(upperRight, instance.outputColors[1], threshold))
5248 return TestStatus::fail("Upper right corner mismatch");
5250 const RGBA lowerLeft(pixelBuffer.getPixel(1, pixelBuffer.getHeight() - 1));
5251 if (!tcu::compareThreshold(lowerLeft, instance.outputColors[2], threshold))
5252 return TestStatus::fail("Lower left corner mismatch");
5254 const RGBA lowerRight(pixelBuffer.getPixel(pixelBuffer.getWidth() - 1, pixelBuffer.getHeight() - 1));
5255 if (!tcu::compareThreshold(lowerRight, instance.outputColors[3], threshold))
5256 return TestStatus::fail("Lower right corner mismatch");
5258 return TestStatus::pass("Rendered output matches input");
5261 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)
5263 const ShaderElement vertFragPipelineStages[] =
5265 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
5266 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
5269 const ShaderElement tessPipelineStages[] =
5271 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
5272 ShaderElement("tessc", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
5273 ShaderElement("tesse", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
5274 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
5277 const ShaderElement geomPipelineStages[] =
5279 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
5280 ShaderElement("geom", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
5281 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
5284 StageToSpecConstantMap specConstantMap;
5286 specConstantMap[VK_SHADER_STAGE_VERTEX_BIT] = specConstants;
5287 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_vert", "", addShaderCodeCustomVertex, runAndVerifyDefaultPipeline,
5288 createInstanceContext(vertFragPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5290 specConstantMap.clear();
5291 specConstantMap[VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT] = specConstants;
5292 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_tessc", "", addShaderCodeCustomTessControl, runAndVerifyDefaultPipeline,
5293 createInstanceContext(tessPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5295 specConstantMap.clear();
5296 specConstantMap[VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT] = specConstants;
5297 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_tesse", "", addShaderCodeCustomTessEval, runAndVerifyDefaultPipeline,
5298 createInstanceContext(tessPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5300 specConstantMap.clear();
5301 specConstantMap[VK_SHADER_STAGE_GEOMETRY_BIT] = specConstants;
5302 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_geom", "", addShaderCodeCustomGeometry, runAndVerifyDefaultPipeline,
5303 createInstanceContext(geomPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5305 specConstantMap.clear();
5306 specConstantMap[VK_SHADER_STAGE_FRAGMENT_BIT] = specConstants;
5307 addFunctionCaseWithPrograms<InstanceContext>(tests, name + "_frag", "", addShaderCodeCustomFragment, runAndVerifyDefaultPipeline,
5308 createInstanceContext(vertFragPipelineStages, inputColors, outputColors, testCodeFragments, specConstantMap));
5311 inline void createTestsForAllStages (const std::string& name, const RGBA (&inputColors)[4], const RGBA (&outputColors)[4], const map<string, string>& testCodeFragments, tcu::TestCaseGroup* tests)
5313 vector<deInt32> noSpecConstants;
5314 createTestsForAllStages(name, inputColors, outputColors, testCodeFragments, noSpecConstants, tests);
5319 tcu::TestCaseGroup* createOpSourceTests (tcu::TestContext& testCtx)
5321 struct NameCodePair { string name, code; };
5322 RGBA defaultColors[4];
5323 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsource", "OpSource instruction"));
5324 const std::string opsourceGLSLWithFile = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile ";
5325 map<string, string> fragments = passthruFragments();
5326 const NameCodePair tests[] =
5328 {"unknown", "OpSource Unknown 321"},
5329 {"essl", "OpSource ESSL 310"},
5330 {"glsl", "OpSource GLSL 450"},
5331 {"opencl_cpp", "OpSource OpenCL_CPP 120"},
5332 {"opencl_c", "OpSource OpenCL_C 120"},
5333 {"multiple", "OpSource GLSL 450\nOpSource GLSL 450"},
5334 {"file", opsourceGLSLWithFile},
5335 {"source", opsourceGLSLWithFile + "\"void main(){}\""},
5336 // Longest possible source string: SPIR-V limits instructions to 65535
5337 // words, of which the first 4 are opsourceGLSLWithFile; the rest will
5338 // contain 65530 UTF8 characters (one word each) plus one last word
5339 // containing 3 ASCII characters and \0.
5340 {"longsource", opsourceGLSLWithFile + '"' + makeLongUTF8String(65530) + "ccc" + '"'}
5343 getDefaultColors(defaultColors);
5344 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5346 fragments["debug"] = tests[testNdx].code;
5347 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5350 return opSourceTests.release();
5353 tcu::TestCaseGroup* createOpSourceContinuedTests (tcu::TestContext& testCtx)
5355 struct NameCodePair { string name, code; };
5356 RGBA defaultColors[4];
5357 de::MovePtr<tcu::TestCaseGroup> opSourceTests (new tcu::TestCaseGroup(testCtx, "opsourcecontinued", "OpSourceContinued instruction"));
5358 map<string, string> fragments = passthruFragments();
5359 const std::string opsource = "%opsrcfile = OpString \"foo.vert\"\nOpSource GLSL 450 %opsrcfile \"void main(){}\"\n";
5360 const NameCodePair tests[] =
5362 {"empty", opsource + "OpSourceContinued \"\""},
5363 {"short", opsource + "OpSourceContinued \"abcde\""},
5364 {"multiple", opsource + "OpSourceContinued \"abcde\"\nOpSourceContinued \"fghij\""},
5365 // Longest possible source string: SPIR-V limits instructions to 65535
5366 // words, of which the first one is OpSourceContinued/length; the rest
5367 // will contain 65533 UTF8 characters (one word each) plus one last word
5368 // containing 3 ASCII characters and \0.
5369 {"long", opsource + "OpSourceContinued \"" + makeLongUTF8String(65533) + "ccc\""}
5372 getDefaultColors(defaultColors);
5373 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
5375 fragments["debug"] = tests[testNdx].code;
5376 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opSourceTests.get());
5379 return opSourceTests.release();
5382 tcu::TestCaseGroup* createOpNoLineTests(tcu::TestContext& testCtx)
5384 RGBA defaultColors[4];
5385 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opnoline", "OpNoLine instruction"));
5386 map<string, string> fragments;
5387 getDefaultColors(defaultColors);
5388 fragments["debug"] =
5389 "%name = OpString \"name\"\n";
5391 fragments["pre_main"] =
5394 "OpLine %name 1 1\n"
5396 "OpLine %name 1 1\n"
5397 "OpLine %name 1 1\n"
5398 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5400 "OpLine %name 1 1\n"
5402 "OpLine %name 1 1\n"
5403 "OpLine %name 1 1\n"
5404 "%second_param1 = OpFunctionParameter %v4f32\n"
5407 "%label_secondfunction = OpLabel\n"
5409 "OpReturnValue %second_param1\n"
5414 fragments["testfun"] =
5415 // A %test_code function that returns its argument unchanged.
5418 "OpLine %name 1 1\n"
5419 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5421 "%param1 = OpFunctionParameter %v4f32\n"
5424 "%label_testfun = OpLabel\n"
5426 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5427 "OpReturnValue %val1\n"
5429 "OpLine %name 1 1\n"
5432 createTestsForAllStages("opnoline", defaultColors, defaultColors, fragments, opLineTests.get());
5434 return opLineTests.release();
5438 tcu::TestCaseGroup* createOpLineTests(tcu::TestContext& testCtx)
5440 RGBA defaultColors[4];
5441 de::MovePtr<tcu::TestCaseGroup> opLineTests (new tcu::TestCaseGroup(testCtx, "opline", "OpLine instruction"));
5442 map<string, string> fragments;
5443 std::vector<std::pair<std::string, std::string> > problemStrings;
5445 problemStrings.push_back(std::make_pair<std::string, std::string>("empty_name", ""));
5446 problemStrings.push_back(std::make_pair<std::string, std::string>("short_name", "short_name"));
5447 problemStrings.push_back(std::make_pair<std::string, std::string>("long_name", makeLongUTF8String(65530) + "ccc"));
5448 getDefaultColors(defaultColors);
5450 fragments["debug"] =
5451 "%other_name = OpString \"other_name\"\n";
5453 fragments["pre_main"] =
5454 "OpLine %file_name 32 0\n"
5455 "OpLine %file_name 32 32\n"
5456 "OpLine %file_name 32 40\n"
5457 "OpLine %other_name 32 40\n"
5458 "OpLine %other_name 0 100\n"
5459 "OpLine %other_name 0 4294967295\n"
5460 "OpLine %other_name 4294967295 0\n"
5461 "OpLine %other_name 32 40\n"
5462 "OpLine %file_name 0 0\n"
5463 "%second_function = OpFunction %v4f32 None %v4f32_function\n"
5464 "OpLine %file_name 1 0\n"
5465 "%second_param1 = OpFunctionParameter %v4f32\n"
5466 "OpLine %file_name 1 3\n"
5467 "OpLine %file_name 1 2\n"
5468 "%label_secondfunction = OpLabel\n"
5469 "OpLine %file_name 0 2\n"
5470 "OpReturnValue %second_param1\n"
5472 "OpLine %file_name 0 2\n"
5473 "OpLine %file_name 0 2\n";
5475 fragments["testfun"] =
5476 // A %test_code function that returns its argument unchanged.
5477 "OpLine %file_name 1 0\n"
5478 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5479 "OpLine %file_name 16 330\n"
5480 "%param1 = OpFunctionParameter %v4f32\n"
5481 "OpLine %file_name 14 442\n"
5482 "%label_testfun = OpLabel\n"
5483 "OpLine %file_name 11 1024\n"
5484 "%val1 = OpFunctionCall %v4f32 %second_function %param1\n"
5485 "OpLine %file_name 2 97\n"
5486 "OpReturnValue %val1\n"
5488 "OpLine %file_name 5 32\n";
5490 for (size_t i = 0; i < problemStrings.size(); ++i)
5492 map<string, string> testFragments = fragments;
5493 testFragments["debug"] += "%file_name = OpString \"" + problemStrings[i].second + "\"\n";
5494 createTestsForAllStages(string("opline") + "_" + problemStrings[i].first, defaultColors, defaultColors, testFragments, opLineTests.get());
5497 return opLineTests.release();
5500 tcu::TestCaseGroup* createOpConstantNullTests(tcu::TestContext& testCtx)
5502 de::MovePtr<tcu::TestCaseGroup> opConstantNullTests (new tcu::TestCaseGroup(testCtx, "opconstantnull", "OpConstantNull instruction"));
5506 const char functionStart[] =
5507 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5508 "%param1 = OpFunctionParameter %v4f32\n"
5511 const char functionEnd[] =
5512 "OpReturnValue %transformed_param\n"
5515 struct NameConstantsCode
5522 NameConstantsCode tests[] =
5526 "%cnull = OpConstantNull %v4f32\n",
5527 "%transformed_param = OpFAdd %v4f32 %param1 %cnull\n"
5531 "%cnull = OpConstantNull %f32\n",
5532 "%vp = OpVariable %fp_v4f32 Function\n"
5533 "%v = OpLoad %v4f32 %vp\n"
5534 "%v0 = OpVectorInsertDynamic %v4f32 %v %cnull %c_i32_0\n"
5535 "%v1 = OpVectorInsertDynamic %v4f32 %v0 %cnull %c_i32_1\n"
5536 "%v2 = OpVectorInsertDynamic %v4f32 %v1 %cnull %c_i32_2\n"
5537 "%v3 = OpVectorInsertDynamic %v4f32 %v2 %cnull %c_i32_3\n"
5538 "%transformed_param = OpFAdd %v4f32 %param1 %v3\n"
5542 "%cnull = OpConstantNull %bool\n",
5543 "%v = OpVariable %fp_v4f32 Function\n"
5544 " OpStore %v %param1\n"
5545 " OpSelectionMerge %false_label None\n"
5546 " OpBranchConditional %cnull %true_label %false_label\n"
5547 "%true_label = OpLabel\n"
5548 " OpStore %v %c_v4f32_0_5_0_5_0_5_0_5\n"
5549 " OpBranch %false_label\n"
5550 "%false_label = OpLabel\n"
5551 "%transformed_param = OpLoad %v4f32 %v\n"
5555 "%cnull = OpConstantNull %i32\n",
5556 "%v = OpVariable %fp_v4f32 Function %c_v4f32_0_5_0_5_0_5_0_5\n"
5557 "%b = OpIEqual %bool %cnull %c_i32_0\n"
5558 " OpSelectionMerge %false_label None\n"
5559 " OpBranchConditional %b %true_label %false_label\n"
5560 "%true_label = OpLabel\n"
5561 " OpStore %v %param1\n"
5562 " OpBranch %false_label\n"
5563 "%false_label = OpLabel\n"
5564 "%transformed_param = OpLoad %v4f32 %v\n"
5568 "%stype = OpTypeStruct %f32 %v4f32\n"
5569 "%fp_stype = OpTypePointer Function %stype\n"
5570 "%cnull = OpConstantNull %stype\n",
5571 "%v = OpVariable %fp_stype Function %cnull\n"
5572 "%f = OpAccessChain %fp_v4f32 %v %c_i32_1\n"
5573 "%f_val = OpLoad %v4f32 %f\n"
5574 "%transformed_param = OpFAdd %v4f32 %param1 %f_val\n"
5578 "%a4_v4f32 = OpTypeArray %v4f32 %c_u32_4\n"
5579 "%fp_a4_v4f32 = OpTypePointer Function %a4_v4f32\n"
5580 "%cnull = OpConstantNull %a4_v4f32\n",
5581 "%v = OpVariable %fp_a4_v4f32 Function %cnull\n"
5582 "%f = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5583 "%f1 = OpAccessChain %fp_v4f32 %v %c_u32_1\n"
5584 "%f2 = OpAccessChain %fp_v4f32 %v %c_u32_2\n"
5585 "%f3 = OpAccessChain %fp_v4f32 %v %c_u32_3\n"
5586 "%f_val = OpLoad %v4f32 %f\n"
5587 "%f1_val = OpLoad %v4f32 %f1\n"
5588 "%f2_val = OpLoad %v4f32 %f2\n"
5589 "%f3_val = OpLoad %v4f32 %f3\n"
5590 "%t0 = OpFAdd %v4f32 %param1 %f_val\n"
5591 "%t1 = OpFAdd %v4f32 %t0 %f1_val\n"
5592 "%t2 = OpFAdd %v4f32 %t1 %f2_val\n"
5593 "%transformed_param = OpFAdd %v4f32 %t2 %f3_val\n"
5597 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5598 "%cnull = OpConstantNull %mat4x4_f32\n",
5599 // Our null matrix * any vector should result in a zero vector.
5600 "%v = OpVectorTimesMatrix %v4f32 %param1 %cnull\n"
5601 "%transformed_param = OpFAdd %v4f32 %param1 %v\n"
5605 getHalfColorsFullAlpha(colors);
5607 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5609 map<string, string> fragments;
5610 fragments["pre_main"] = tests[testNdx].constants;
5611 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5612 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, opConstantNullTests.get());
5614 return opConstantNullTests.release();
5616 tcu::TestCaseGroup* createOpConstantCompositeTests(tcu::TestContext& testCtx)
5618 de::MovePtr<tcu::TestCaseGroup> opConstantCompositeTests (new tcu::TestCaseGroup(testCtx, "opconstantcomposite", "OpConstantComposite instruction"));
5619 RGBA inputColors[4];
5620 RGBA outputColors[4];
5623 const char functionStart[] =
5624 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5625 "%param1 = OpFunctionParameter %v4f32\n"
5628 const char functionEnd[] =
5629 "OpReturnValue %transformed_param\n"
5632 struct NameConstantsCode
5639 NameConstantsCode tests[] =
5644 "%cval = OpConstantComposite %v4f32 %c_f32_0_5 %c_f32_0_5 %c_f32_0_5 %c_f32_0\n",
5645 "%transformed_param = OpFAdd %v4f32 %param1 %cval\n"
5650 "%stype = OpTypeStruct %v4f32 %f32\n"
5651 "%fp_stype = OpTypePointer Function %stype\n"
5652 "%f32_n_1 = OpConstant %f32 -1.0\n"
5653 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5654 "%cvec = OpConstantComposite %v4f32 %f32_1_5 %f32_1_5 %f32_1_5 %c_f32_1\n"
5655 "%cval = OpConstantComposite %stype %cvec %f32_n_1\n",
5657 "%v = OpVariable %fp_stype Function %cval\n"
5658 "%vec_ptr = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
5659 "%f32_ptr = OpAccessChain %fp_f32 %v %c_u32_1\n"
5660 "%vec_val = OpLoad %v4f32 %vec_ptr\n"
5661 "%f32_val = OpLoad %f32 %f32_ptr\n"
5662 "%tmp1 = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_1 %f32_val\n" // vec4(-1)
5663 "%tmp2 = OpFAdd %v4f32 %tmp1 %param1\n" // param1 + vec4(-1)
5664 "%transformed_param = OpFAdd %v4f32 %tmp2 %vec_val\n" // param1 + vec4(-1) + vec4(1.5, 1.5, 1.5, 1.0)
5667 // [1|0|0|0.5] [x] = x + 0.5
5668 // [0|1|0|0.5] [y] = y + 0.5
5669 // [0|0|1|0.5] [z] = z + 0.5
5670 // [0|0|0|1 ] [1] = 1
5673 "%mat4x4_f32 = OpTypeMatrix %v4f32 4\n"
5674 "%v4f32_1_0_0_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_0 %c_f32_0 %c_f32_0\n"
5675 "%v4f32_0_1_0_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_1 %c_f32_0 %c_f32_0\n"
5676 "%v4f32_0_0_1_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_1 %c_f32_0\n"
5677 "%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"
5678 "%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",
5680 "%transformed_param = OpMatrixTimesVector %v4f32 %cval %param1\n"
5685 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5686 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5687 "%f32_n_1 = OpConstant %f32 -1.0\n"
5688 "%f32_1_5 = OpConstant %f32 !0x3fc00000\n" // +1.5
5689 "%carr = OpConstantComposite %a4f32 %c_f32_0 %f32_n_1 %f32_1_5 %c_f32_0\n",
5691 "%v = OpVariable %fp_a4f32 Function %carr\n"
5692 "%f = OpAccessChain %fp_f32 %v %c_u32_0\n"
5693 "%f1 = OpAccessChain %fp_f32 %v %c_u32_1\n"
5694 "%f2 = OpAccessChain %fp_f32 %v %c_u32_2\n"
5695 "%f3 = OpAccessChain %fp_f32 %v %c_u32_3\n"
5696 "%f_val = OpLoad %f32 %f\n"
5697 "%f1_val = OpLoad %f32 %f1\n"
5698 "%f2_val = OpLoad %f32 %f2\n"
5699 "%f3_val = OpLoad %f32 %f3\n"
5700 "%ftot1 = OpFAdd %f32 %f_val %f1_val\n"
5701 "%ftot2 = OpFAdd %f32 %ftot1 %f2_val\n"
5702 "%ftot3 = OpFAdd %f32 %ftot2 %f3_val\n" // 0 - 1 + 1.5 + 0
5703 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %ftot3\n"
5704 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5711 // [ 1.0, 1.0, 1.0, 1.0]
5715 // [ 0.0, 0.5, 0.0, 0.0]
5719 // [ 1.0, 1.0, 1.0, 1.0]
5722 "array_of_struct_of_array",
5724 "%c_v4f32_1_1_1_0 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_0\n"
5725 "%fp_a4f32 = OpTypePointer Function %a4f32\n"
5726 "%stype = OpTypeStruct %f32 %a4f32\n"
5727 "%a3stype = OpTypeArray %stype %c_u32_3\n"
5728 "%fp_a3stype = OpTypePointer Function %a3stype\n"
5729 "%ca4f32_0 = OpConstantComposite %a4f32 %c_f32_0 %c_f32_0_5 %c_f32_0 %c_f32_0\n"
5730 "%ca4f32_1 = OpConstantComposite %a4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
5731 "%cstype1 = OpConstantComposite %stype %c_f32_0 %ca4f32_1\n"
5732 "%cstype2 = OpConstantComposite %stype %c_f32_1 %ca4f32_0\n"
5733 "%carr = OpConstantComposite %a3stype %cstype1 %cstype2 %cstype1",
5735 "%v = OpVariable %fp_a3stype Function %carr\n"
5736 "%f = OpAccessChain %fp_f32 %v %c_u32_1 %c_u32_1 %c_u32_1\n"
5737 "%add_vec = OpVectorTimesScalar %v4f32 %c_v4f32_1_1_1_0 %f\n"
5738 "%transformed_param = OpFAdd %v4f32 %param1 %add_vec\n"
5742 getHalfColorsFullAlpha(inputColors);
5743 outputColors[0] = RGBA(255, 255, 255, 255);
5744 outputColors[1] = RGBA(255, 127, 127, 255);
5745 outputColors[2] = RGBA(127, 255, 127, 255);
5746 outputColors[3] = RGBA(127, 127, 255, 255);
5748 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameConstantsCode); ++testNdx)
5750 map<string, string> fragments;
5751 fragments["pre_main"] = tests[testNdx].constants;
5752 fragments["testfun"] = string(functionStart) + tests[testNdx].code + functionEnd;
5753 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, opConstantCompositeTests.get());
5755 return opConstantCompositeTests.release();
5758 tcu::TestCaseGroup* createSelectionBlockOrderTests(tcu::TestContext& testCtx)
5760 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "selection_block_order", "Out-of-order blocks for selection"));
5761 RGBA inputColors[4];
5762 RGBA outputColors[4];
5763 map<string, string> fragments;
5765 // vec4 test_code(vec4 param) {
5766 // vec4 result = param;
5767 // for (int i = 0; i < 4; ++i) {
5768 // if (i == 0) result[i] = 0.;
5769 // else result[i] = 1. - result[i];
5773 const char function[] =
5774 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5775 "%param1 = OpFunctionParameter %v4f32\n"
5777 "%iptr = OpVariable %fp_i32 Function\n"
5778 " OpStore %iptr %c_i32_0\n"
5779 "%result = OpVariable %fp_v4f32 Function\n"
5780 " OpStore %result %param1\n"
5783 // Loop entry block.
5785 "%ival = OpLoad %i32 %iptr\n"
5786 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5787 " OpLoopMerge %exit %loop None\n"
5788 " OpBranchConditional %lt_4 %if_entry %exit\n"
5790 // Merge block for loop.
5792 "%ret = OpLoad %v4f32 %result\n"
5793 " OpReturnValue %ret\n"
5795 // If-statement entry block.
5796 "%if_entry = OpLabel\n"
5797 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5798 "%eq_0 = OpIEqual %bool %ival %c_i32_0\n"
5799 " OpSelectionMerge %if_exit None\n"
5800 " OpBranchConditional %eq_0 %if_true %if_false\n"
5802 // False branch for if-statement.
5803 "%if_false = OpLabel\n"
5804 "%val = OpLoad %f32 %loc\n"
5805 "%sub = OpFSub %f32 %c_f32_1 %val\n"
5806 " OpStore %loc %sub\n"
5807 " OpBranch %if_exit\n"
5809 // Merge block for if-statement.
5810 "%if_exit = OpLabel\n"
5811 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5812 " OpStore %iptr %ival_next\n"
5815 // True branch for if-statement.
5816 "%if_true = OpLabel\n"
5817 " OpStore %loc %c_f32_0\n"
5818 " OpBranch %if_exit\n"
5822 fragments["testfun"] = function;
5824 inputColors[0] = RGBA(127, 127, 127, 0);
5825 inputColors[1] = RGBA(127, 0, 0, 0);
5826 inputColors[2] = RGBA(0, 127, 0, 0);
5827 inputColors[3] = RGBA(0, 0, 127, 0);
5829 outputColors[0] = RGBA(0, 128, 128, 255);
5830 outputColors[1] = RGBA(0, 255, 255, 255);
5831 outputColors[2] = RGBA(0, 128, 255, 255);
5832 outputColors[3] = RGBA(0, 255, 128, 255);
5834 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5836 return group.release();
5839 tcu::TestCaseGroup* createSwitchBlockOrderTests(tcu::TestContext& testCtx)
5841 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "switch_block_order", "Out-of-order blocks for switch"));
5842 RGBA inputColors[4];
5843 RGBA outputColors[4];
5844 map<string, string> fragments;
5846 const char typesAndConstants[] =
5847 "%c_f32_p2 = OpConstant %f32 0.2\n"
5848 "%c_f32_p4 = OpConstant %f32 0.4\n"
5849 "%c_f32_p6 = OpConstant %f32 0.6\n"
5850 "%c_f32_p8 = OpConstant %f32 0.8\n";
5852 // vec4 test_code(vec4 param) {
5853 // vec4 result = param;
5854 // for (int i = 0; i < 4; ++i) {
5856 // case 0: result[i] += .2; break;
5857 // case 1: result[i] += .6; break;
5858 // case 2: result[i] += .4; break;
5859 // case 3: result[i] += .8; break;
5860 // default: break; // unreachable
5865 const char function[] =
5866 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5867 "%param1 = OpFunctionParameter %v4f32\n"
5869 "%iptr = OpVariable %fp_i32 Function\n"
5870 " OpStore %iptr %c_i32_0\n"
5871 "%result = OpVariable %fp_v4f32 Function\n"
5872 " OpStore %result %param1\n"
5875 // Loop entry block.
5877 "%ival = OpLoad %i32 %iptr\n"
5878 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
5879 " OpLoopMerge %exit %loop None\n"
5880 " OpBranchConditional %lt_4 %switch_entry %exit\n"
5882 // Merge block for loop.
5884 "%ret = OpLoad %v4f32 %result\n"
5885 " OpReturnValue %ret\n"
5887 // Switch-statement entry block.
5888 "%switch_entry = OpLabel\n"
5889 "%loc = OpAccessChain %fp_f32 %result %ival\n"
5890 "%val = OpLoad %f32 %loc\n"
5891 " OpSelectionMerge %switch_exit None\n"
5892 " OpSwitch %ival %switch_default 0 %case0 1 %case1 2 %case2 3 %case3\n"
5894 "%case2 = OpLabel\n"
5895 "%addp4 = OpFAdd %f32 %val %c_f32_p4\n"
5896 " OpStore %loc %addp4\n"
5897 " OpBranch %switch_exit\n"
5899 "%switch_default = OpLabel\n"
5902 "%case3 = OpLabel\n"
5903 "%addp8 = OpFAdd %f32 %val %c_f32_p8\n"
5904 " OpStore %loc %addp8\n"
5905 " OpBranch %switch_exit\n"
5907 "%case0 = OpLabel\n"
5908 "%addp2 = OpFAdd %f32 %val %c_f32_p2\n"
5909 " OpStore %loc %addp2\n"
5910 " OpBranch %switch_exit\n"
5912 // Merge block for switch-statement.
5913 "%switch_exit = OpLabel\n"
5914 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
5915 " OpStore %iptr %ival_next\n"
5918 "%case1 = OpLabel\n"
5919 "%addp6 = OpFAdd %f32 %val %c_f32_p6\n"
5920 " OpStore %loc %addp6\n"
5921 " OpBranch %switch_exit\n"
5925 fragments["pre_main"] = typesAndConstants;
5926 fragments["testfun"] = function;
5928 inputColors[0] = RGBA(127, 27, 127, 51);
5929 inputColors[1] = RGBA(127, 0, 0, 51);
5930 inputColors[2] = RGBA(0, 27, 0, 51);
5931 inputColors[3] = RGBA(0, 0, 127, 51);
5933 outputColors[0] = RGBA(178, 180, 229, 255);
5934 outputColors[1] = RGBA(178, 153, 102, 255);
5935 outputColors[2] = RGBA(51, 180, 102, 255);
5936 outputColors[3] = RGBA(51, 153, 229, 255);
5938 createTestsForAllStages("out_of_order", inputColors, outputColors, fragments, group.get());
5940 return group.release();
5943 tcu::TestCaseGroup* createDecorationGroupTests(tcu::TestContext& testCtx)
5945 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "decoration_group", "Decoration group tests"));
5946 RGBA inputColors[4];
5947 RGBA outputColors[4];
5948 map<string, string> fragments;
5950 const char decorations[] =
5951 "OpDecorate %array_group ArrayStride 4\n"
5952 "OpDecorate %struct_member_group Offset 0\n"
5953 "%array_group = OpDecorationGroup\n"
5954 "%struct_member_group = OpDecorationGroup\n"
5956 "OpDecorate %group1 RelaxedPrecision\n"
5957 "OpDecorate %group3 RelaxedPrecision\n"
5958 "OpDecorate %group3 Invariant\n"
5959 "OpDecorate %group3 Restrict\n"
5960 "%group0 = OpDecorationGroup\n"
5961 "%group1 = OpDecorationGroup\n"
5962 "%group3 = OpDecorationGroup\n";
5964 const char typesAndConstants[] =
5965 "%a3f32 = OpTypeArray %f32 %c_u32_3\n"
5966 "%struct1 = OpTypeStruct %a3f32\n"
5967 "%struct2 = OpTypeStruct %a3f32\n"
5968 "%fp_struct1 = OpTypePointer Function %struct1\n"
5969 "%fp_struct2 = OpTypePointer Function %struct2\n"
5970 "%c_f32_2 = OpConstant %f32 2.\n"
5971 "%c_f32_n2 = OpConstant %f32 -2.\n"
5973 "%c_a3f32_1 = OpConstantComposite %a3f32 %c_f32_1 %c_f32_2 %c_f32_1\n"
5974 "%c_a3f32_2 = OpConstantComposite %a3f32 %c_f32_n1 %c_f32_n2 %c_f32_n1\n"
5975 "%c_struct1 = OpConstantComposite %struct1 %c_a3f32_1\n"
5976 "%c_struct2 = OpConstantComposite %struct2 %c_a3f32_2\n";
5978 const char function[] =
5979 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
5980 "%param = OpFunctionParameter %v4f32\n"
5981 "%entry = OpLabel\n"
5982 "%result = OpVariable %fp_v4f32 Function\n"
5983 " OpStore %result %param\n"
5984 "%v_struct1 = OpVariable %fp_struct1 Function\n"
5985 " OpStore %v_struct1 %c_struct1\n"
5986 "%v_struct2 = OpVariable %fp_struct2 Function\n"
5987 " OpStore %v_struct2 %c_struct2\n"
5988 "%ptr1 = OpAccessChain %fp_f32 %v_struct1 %c_i32_0 %c_i32_1\n"
5989 "%val1 = OpLoad %f32 %ptr1\n"
5990 "%ptr2 = OpAccessChain %fp_f32 %v_struct2 %c_i32_0 %c_i32_2\n"
5991 "%val2 = OpLoad %f32 %ptr2\n"
5992 "%addvalues = OpFAdd %f32 %val1 %val2\n"
5993 "%ptr = OpAccessChain %fp_f32 %result %c_i32_1\n"
5994 "%val = OpLoad %f32 %ptr\n"
5995 "%addresult = OpFAdd %f32 %addvalues %val\n"
5996 " OpStore %ptr %addresult\n"
5997 "%ret = OpLoad %v4f32 %result\n"
5998 " OpReturnValue %ret\n"
6001 struct CaseNameDecoration
6007 CaseNameDecoration tests[] =
6010 "same_decoration_group_on_multiple_types",
6011 "OpGroupMemberDecorate %struct_member_group %struct1 0 %struct2 0\n"
6014 "empty_decoration_group",
6015 "OpGroupDecorate %group0 %a3f32\n"
6016 "OpGroupDecorate %group0 %result\n"
6019 "one_element_decoration_group",
6020 "OpGroupDecorate %array_group %a3f32\n"
6023 "multiple_elements_decoration_group",
6024 "OpGroupDecorate %group3 %v_struct1\n"
6027 "multiple_decoration_groups_on_same_variable",
6028 "OpGroupDecorate %group0 %v_struct2\n"
6029 "OpGroupDecorate %group1 %v_struct2\n"
6030 "OpGroupDecorate %group3 %v_struct2\n"
6033 "same_decoration_group_multiple_times",
6034 "OpGroupDecorate %group1 %addvalues\n"
6035 "OpGroupDecorate %group1 %addvalues\n"
6036 "OpGroupDecorate %group1 %addvalues\n"
6041 getHalfColorsFullAlpha(inputColors);
6042 getHalfColorsFullAlpha(outputColors);
6044 for (size_t idx = 0; idx < (sizeof(tests) / sizeof(tests[0])); ++idx)
6046 fragments["decoration"] = decorations + tests[idx].decoration;
6047 fragments["pre_main"] = typesAndConstants;
6048 fragments["testfun"] = function;
6050 createTestsForAllStages(tests[idx].name, inputColors, outputColors, fragments, group.get());
6053 return group.release();
6056 struct SpecConstantTwoIntGraphicsCase
6058 const char* caseName;
6059 const char* scDefinition0;
6060 const char* scDefinition1;
6061 const char* scResultType;
6062 const char* scOperation;
6063 deInt32 scActualValue0;
6064 deInt32 scActualValue1;
6065 const char* resultOperation;
6066 RGBA expectedColors[4];
6068 SpecConstantTwoIntGraphicsCase (const char* name,
6069 const char* definition0,
6070 const char* definition1,
6071 const char* resultType,
6072 const char* operation,
6075 const char* resultOp,
6076 const RGBA (&output)[4])
6078 , scDefinition0 (definition0)
6079 , scDefinition1 (definition1)
6080 , scResultType (resultType)
6081 , scOperation (operation)
6082 , scActualValue0 (value0)
6083 , scActualValue1 (value1)
6084 , resultOperation (resultOp)
6086 expectedColors[0] = output[0];
6087 expectedColors[1] = output[1];
6088 expectedColors[2] = output[2];
6089 expectedColors[3] = output[3];
6093 tcu::TestCaseGroup* createSpecConstantTests (tcu::TestContext& testCtx)
6095 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opspecconstantop", "Test the OpSpecConstantOp instruction"));
6096 vector<SpecConstantTwoIntGraphicsCase> cases;
6097 RGBA inputColors[4];
6098 RGBA outputColors0[4];
6099 RGBA outputColors1[4];
6100 RGBA outputColors2[4];
6102 const char decorations1[] =
6103 "OpDecorate %sc_0 SpecId 0\n"
6104 "OpDecorate %sc_1 SpecId 1\n";
6106 const char typesAndConstants1[] =
6107 "%sc_0 = OpSpecConstant${SC_DEF0}\n"
6108 "%sc_1 = OpSpecConstant${SC_DEF1}\n"
6109 "%sc_op = OpSpecConstantOp ${SC_RESULT_TYPE} ${SC_OP}\n";
6111 const char function1[] =
6112 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6113 "%param = OpFunctionParameter %v4f32\n"
6114 "%label = OpLabel\n"
6115 "%result = OpVariable %fp_v4f32 Function\n"
6116 " OpStore %result %param\n"
6117 "%gen = ${GEN_RESULT}\n"
6118 "%index = OpIAdd %i32 %gen %c_i32_1\n"
6119 "%loc = OpAccessChain %fp_f32 %result %index\n"
6120 "%val = OpLoad %f32 %loc\n"
6121 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6122 " OpStore %loc %add\n"
6123 "%ret = OpLoad %v4f32 %result\n"
6124 " OpReturnValue %ret\n"
6127 inputColors[0] = RGBA(127, 127, 127, 255);
6128 inputColors[1] = RGBA(127, 0, 0, 255);
6129 inputColors[2] = RGBA(0, 127, 0, 255);
6130 inputColors[3] = RGBA(0, 0, 127, 255);
6132 // Derived from inputColors[x] by adding 128 to inputColors[x][0].
6133 outputColors0[0] = RGBA(255, 127, 127, 255);
6134 outputColors0[1] = RGBA(255, 0, 0, 255);
6135 outputColors0[2] = RGBA(128, 127, 0, 255);
6136 outputColors0[3] = RGBA(128, 0, 127, 255);
6138 // Derived from inputColors[x] by adding 128 to inputColors[x][1].
6139 outputColors1[0] = RGBA(127, 255, 127, 255);
6140 outputColors1[1] = RGBA(127, 128, 0, 255);
6141 outputColors1[2] = RGBA(0, 255, 0, 255);
6142 outputColors1[3] = RGBA(0, 128, 127, 255);
6144 // Derived from inputColors[x] by adding 128 to inputColors[x][2].
6145 outputColors2[0] = RGBA(127, 127, 255, 255);
6146 outputColors2[1] = RGBA(127, 0, 128, 255);
6147 outputColors2[2] = RGBA(0, 127, 128, 255);
6148 outputColors2[3] = RGBA(0, 0, 255, 255);
6150 const char addZeroToSc[] = "OpIAdd %i32 %c_i32_0 %sc_op";
6151 const char selectTrueUsingSc[] = "OpSelect %i32 %sc_op %c_i32_1 %c_i32_0";
6152 const char selectFalseUsingSc[] = "OpSelect %i32 %sc_op %c_i32_0 %c_i32_1";
6154 cases.push_back(SpecConstantTwoIntGraphicsCase("iadd", " %i32 0", " %i32 0", "%i32", "IAdd %sc_0 %sc_1", 19, -20, addZeroToSc, outputColors0));
6155 cases.push_back(SpecConstantTwoIntGraphicsCase("isub", " %i32 0", " %i32 0", "%i32", "ISub %sc_0 %sc_1", 19, 20, addZeroToSc, outputColors0));
6156 cases.push_back(SpecConstantTwoIntGraphicsCase("imul", " %i32 0", " %i32 0", "%i32", "IMul %sc_0 %sc_1", -1, -1, addZeroToSc, outputColors2));
6157 cases.push_back(SpecConstantTwoIntGraphicsCase("sdiv", " %i32 0", " %i32 0", "%i32", "SDiv %sc_0 %sc_1", -126, 126, addZeroToSc, outputColors0));
6158 cases.push_back(SpecConstantTwoIntGraphicsCase("udiv", " %i32 0", " %i32 0", "%i32", "UDiv %sc_0 %sc_1", 126, 126, addZeroToSc, outputColors2));
6159 cases.push_back(SpecConstantTwoIntGraphicsCase("srem", " %i32 0", " %i32 0", "%i32", "SRem %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6160 cases.push_back(SpecConstantTwoIntGraphicsCase("smod", " %i32 0", " %i32 0", "%i32", "SMod %sc_0 %sc_1", 3, 2, addZeroToSc, outputColors2));
6161 cases.push_back(SpecConstantTwoIntGraphicsCase("umod", " %i32 0", " %i32 0", "%i32", "UMod %sc_0 %sc_1", 1001, 500, addZeroToSc, outputColors2));
6162 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseand", " %i32 0", " %i32 0", "%i32", "BitwiseAnd %sc_0 %sc_1", 0x33, 0x0d, addZeroToSc, outputColors2));
6163 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwiseor", " %i32 0", " %i32 0", "%i32", "BitwiseOr %sc_0 %sc_1", 0, 1, addZeroToSc, outputColors2));
6164 cases.push_back(SpecConstantTwoIntGraphicsCase("bitwisexor", " %i32 0", " %i32 0", "%i32", "BitwiseXor %sc_0 %sc_1", 0x2e, 0x2f, addZeroToSc, outputColors2));
6165 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightlogical", " %i32 0", " %i32 0", "%i32", "ShiftRightLogical %sc_0 %sc_1", 2, 1, addZeroToSc, outputColors2));
6166 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftrightarithmetic", " %i32 0", " %i32 0", "%i32", "ShiftRightArithmetic %sc_0 %sc_1", -4, 2, addZeroToSc, outputColors0));
6167 cases.push_back(SpecConstantTwoIntGraphicsCase("shiftleftlogical", " %i32 0", " %i32 0", "%i32", "ShiftLeftLogical %sc_0 %sc_1", 1, 0, addZeroToSc, outputColors2));
6168 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthan", " %i32 0", " %i32 0", "%bool", "SLessThan %sc_0 %sc_1", -20, -10, selectTrueUsingSc, outputColors2));
6169 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthan", " %i32 0", " %i32 0", "%bool", "ULessThan %sc_0 %sc_1", 10, 20, selectTrueUsingSc, outputColors2));
6170 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthan", " %i32 0", " %i32 0", "%bool", "SGreaterThan %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6171 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthan", " %i32 0", " %i32 0", "%bool", "UGreaterThan %sc_0 %sc_1", 10, 5, selectTrueUsingSc, outputColors2));
6172 cases.push_back(SpecConstantTwoIntGraphicsCase("slessthanequal", " %i32 0", " %i32 0", "%bool", "SLessThanEqual %sc_0 %sc_1", -10, -10, selectTrueUsingSc, outputColors2));
6173 cases.push_back(SpecConstantTwoIntGraphicsCase("ulessthanequal", " %i32 0", " %i32 0", "%bool", "ULessThanEqual %sc_0 %sc_1", 50, 100, selectTrueUsingSc, outputColors2));
6174 cases.push_back(SpecConstantTwoIntGraphicsCase("sgreaterthanequal", " %i32 0", " %i32 0", "%bool", "SGreaterThanEqual %sc_0 %sc_1", -1000, 50, selectFalseUsingSc, outputColors2));
6175 cases.push_back(SpecConstantTwoIntGraphicsCase("ugreaterthanequal", " %i32 0", " %i32 0", "%bool", "UGreaterThanEqual %sc_0 %sc_1", 10, 10, selectTrueUsingSc, outputColors2));
6176 cases.push_back(SpecConstantTwoIntGraphicsCase("iequal", " %i32 0", " %i32 0", "%bool", "IEqual %sc_0 %sc_1", 42, 24, selectFalseUsingSc, outputColors2));
6177 cases.push_back(SpecConstantTwoIntGraphicsCase("logicaland", "True %bool", "True %bool", "%bool", "LogicalAnd %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6178 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalor", "False %bool", "False %bool", "%bool", "LogicalOr %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6179 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalequal", "True %bool", "True %bool", "%bool", "LogicalEqual %sc_0 %sc_1", 0, 1, selectFalseUsingSc, outputColors2));
6180 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnotequal", "False %bool", "False %bool", "%bool", "LogicalNotEqual %sc_0 %sc_1", 1, 0, selectTrueUsingSc, outputColors2));
6181 cases.push_back(SpecConstantTwoIntGraphicsCase("snegate", " %i32 0", " %i32 0", "%i32", "SNegate %sc_0", -1, 0, addZeroToSc, outputColors2));
6182 cases.push_back(SpecConstantTwoIntGraphicsCase("not", " %i32 0", " %i32 0", "%i32", "Not %sc_0", -2, 0, addZeroToSc, outputColors2));
6183 cases.push_back(SpecConstantTwoIntGraphicsCase("logicalnot", "False %bool", "False %bool", "%bool", "LogicalNot %sc_0", 1, 0, selectFalseUsingSc, outputColors2));
6184 cases.push_back(SpecConstantTwoIntGraphicsCase("select", "False %bool", " %i32 0", "%i32", "Select %sc_0 %sc_1 %c_i32_0", 1, 1, addZeroToSc, outputColors2));
6185 // OpSConvert, OpFConvert: these two instructions involve ints/floats of different bitwidths.
6186 // \todo[2015-12-1 antiagainst] OpQuantizeToF16
6188 for (size_t caseNdx = 0; caseNdx < cases.size(); ++caseNdx)
6190 map<string, string> specializations;
6191 map<string, string> fragments;
6192 vector<deInt32> specConstants;
6194 specializations["SC_DEF0"] = cases[caseNdx].scDefinition0;
6195 specializations["SC_DEF1"] = cases[caseNdx].scDefinition1;
6196 specializations["SC_RESULT_TYPE"] = cases[caseNdx].scResultType;
6197 specializations["SC_OP"] = cases[caseNdx].scOperation;
6198 specializations["GEN_RESULT"] = cases[caseNdx].resultOperation;
6200 fragments["decoration"] = tcu::StringTemplate(decorations1).specialize(specializations);
6201 fragments["pre_main"] = tcu::StringTemplate(typesAndConstants1).specialize(specializations);
6202 fragments["testfun"] = tcu::StringTemplate(function1).specialize(specializations);
6204 specConstants.push_back(cases[caseNdx].scActualValue0);
6205 specConstants.push_back(cases[caseNdx].scActualValue1);
6207 createTestsForAllStages(cases[caseNdx].caseName, inputColors, cases[caseNdx].expectedColors, fragments, specConstants, group.get());
6210 const char decorations2[] =
6211 "OpDecorate %sc_0 SpecId 0\n"
6212 "OpDecorate %sc_1 SpecId 1\n"
6213 "OpDecorate %sc_2 SpecId 2\n";
6215 const char typesAndConstants2[] =
6216 "%v3i32 = OpTypeVector %i32 3\n"
6218 "%sc_0 = OpSpecConstant %i32 0\n"
6219 "%sc_1 = OpSpecConstant %i32 0\n"
6220 "%sc_2 = OpSpecConstant %i32 0\n"
6222 "%vec3_0 = OpConstantComposite %v3i32 %c_i32_0 %c_i32_0 %c_i32_0\n"
6223 "%sc_vec3_0 = OpSpecConstantOp %v3i32 CompositeInsert %sc_0 %vec3_0 0\n" // (sc_0, 0, 0)
6224 "%sc_vec3_1 = OpSpecConstantOp %v3i32 CompositeInsert %sc_1 %vec3_0 1\n" // (0, sc_1, 0)
6225 "%sc_vec3_2 = OpSpecConstantOp %v3i32 CompositeInsert %sc_2 %vec3_0 2\n" // (0, 0, sc_2)
6226 "%sc_vec3_01 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_0 %sc_vec3_1 1 0 4\n" // (0, sc_0, sc_1)
6227 "%sc_vec3_012 = OpSpecConstantOp %v3i32 VectorShuffle %sc_vec3_01 %sc_vec3_2 5 1 2\n" // (sc_2, sc_0, sc_1)
6228 "%sc_ext_0 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 0\n" // sc_2
6229 "%sc_ext_1 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 1\n" // sc_0
6230 "%sc_ext_2 = OpSpecConstantOp %i32 CompositeExtract %sc_vec3_012 2\n" // sc_1
6231 "%sc_sub = OpSpecConstantOp %i32 ISub %sc_ext_0 %sc_ext_1\n" // (sc_2 - sc_0)
6232 "%sc_final = OpSpecConstantOp %i32 IMul %sc_sub %sc_ext_2\n"; // (sc_2 - sc_0) * sc_1
6234 const char function2[] =
6235 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6236 "%param = OpFunctionParameter %v4f32\n"
6237 "%label = OpLabel\n"
6238 "%result = OpVariable %fp_v4f32 Function\n"
6239 " OpStore %result %param\n"
6240 "%loc = OpAccessChain %fp_f32 %result %sc_final\n"
6241 "%val = OpLoad %f32 %loc\n"
6242 "%add = OpFAdd %f32 %val %c_f32_0_5\n"
6243 " OpStore %loc %add\n"
6244 "%ret = OpLoad %v4f32 %result\n"
6245 " OpReturnValue %ret\n"
6248 map<string, string> fragments;
6249 vector<deInt32> specConstants;
6251 fragments["decoration"] = decorations2;
6252 fragments["pre_main"] = typesAndConstants2;
6253 fragments["testfun"] = function2;
6255 specConstants.push_back(56789);
6256 specConstants.push_back(-2);
6257 specConstants.push_back(56788);
6259 createTestsForAllStages("vector_related", inputColors, outputColors2, fragments, specConstants, group.get());
6261 return group.release();
6264 tcu::TestCaseGroup* createOpPhiTests(tcu::TestContext& testCtx)
6266 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "opphi", "Test the OpPhi instruction"));
6267 RGBA inputColors[4];
6268 RGBA outputColors1[4];
6269 RGBA outputColors2[4];
6270 RGBA outputColors3[4];
6271 map<string, string> fragments1;
6272 map<string, string> fragments2;
6273 map<string, string> fragments3;
6275 const char typesAndConstants1[] =
6276 "%c_f32_p2 = OpConstant %f32 0.2\n"
6277 "%c_f32_p4 = OpConstant %f32 0.4\n"
6278 "%c_f32_p6 = OpConstant %f32 0.6\n"
6279 "%c_f32_p8 = OpConstant %f32 0.8\n";
6281 // vec4 test_code(vec4 param) {
6282 // vec4 result = param;
6283 // for (int i = 0; i < 4; ++i) {
6286 // case 0: operand = .2; break;
6287 // case 1: operand = .6; break;
6288 // case 2: operand = .4; break;
6289 // case 3: operand = .0; break;
6290 // default: break; // unreachable
6292 // result[i] += operand;
6296 const char function1[] =
6297 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6298 "%param1 = OpFunctionParameter %v4f32\n"
6300 "%iptr = OpVariable %fp_i32 Function\n"
6301 " OpStore %iptr %c_i32_0\n"
6302 "%result = OpVariable %fp_v4f32 Function\n"
6303 " OpStore %result %param1\n"
6307 "%ival = OpLoad %i32 %iptr\n"
6308 "%lt_4 = OpSLessThan %bool %ival %c_i32_4\n"
6309 " OpLoopMerge %exit %loop None\n"
6310 " OpBranchConditional %lt_4 %entry %exit\n"
6312 "%entry = OpLabel\n"
6313 "%loc = OpAccessChain %fp_f32 %result %ival\n"
6314 "%val = OpLoad %f32 %loc\n"
6315 " OpSelectionMerge %phi None\n"
6316 " OpSwitch %ival %default 0 %case0 1 %case1 2 %case2 3 %case3\n"
6318 "%case0 = OpLabel\n"
6320 "%case1 = OpLabel\n"
6322 "%case2 = OpLabel\n"
6324 "%case3 = OpLabel\n"
6327 "%default = OpLabel\n"
6331 "%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
6332 "%add = OpFAdd %f32 %val %operand\n"
6333 " OpStore %loc %add\n"
6334 "%ival_next = OpIAdd %i32 %ival %c_i32_1\n"
6335 " OpStore %iptr %ival_next\n"
6339 "%ret = OpLoad %v4f32 %result\n"
6340 " OpReturnValue %ret\n"
6344 fragments1["pre_main"] = typesAndConstants1;
6345 fragments1["testfun"] = function1;
6347 getHalfColorsFullAlpha(inputColors);
6349 outputColors1[0] = RGBA(178, 180, 229, 255);
6350 outputColors1[1] = RGBA(178, 153, 102, 255);
6351 outputColors1[2] = RGBA(51, 180, 102, 255);
6352 outputColors1[3] = RGBA(51, 153, 229, 255);
6354 createTestsForAllStages("out_of_order", inputColors, outputColors1, fragments1, group.get());
6356 const char typesAndConstants2[] =
6357 "%c_f32_p2 = OpConstant %f32 0.2\n";
6359 // Add .4 to the second element of the given parameter.
6360 const char function2[] =
6361 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6362 "%param = OpFunctionParameter %v4f32\n"
6363 "%entry = OpLabel\n"
6364 "%result = OpVariable %fp_v4f32 Function\n"
6365 " OpStore %result %param\n"
6366 "%loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6367 "%val = OpLoad %f32 %loc\n"
6371 "%step = OpPhi %i32 %c_i32_0 %entry %step_next %phi\n"
6372 "%accum = OpPhi %f32 %val %entry %accum_next %phi\n"
6373 "%step_next = OpIAdd %i32 %step %c_i32_1\n"
6374 "%accum_next = OpFAdd %f32 %accum %c_f32_p2\n"
6375 "%still_loop = OpSLessThan %bool %step %c_i32_2\n"
6376 " OpLoopMerge %exit %phi None\n"
6377 " OpBranchConditional %still_loop %phi %exit\n"
6380 " OpStore %loc %accum\n"
6381 "%ret = OpLoad %v4f32 %result\n"
6382 " OpReturnValue %ret\n"
6386 fragments2["pre_main"] = typesAndConstants2;
6387 fragments2["testfun"] = function2;
6389 outputColors2[0] = RGBA(127, 229, 127, 255);
6390 outputColors2[1] = RGBA(127, 102, 0, 255);
6391 outputColors2[2] = RGBA(0, 229, 0, 255);
6392 outputColors2[3] = RGBA(0, 102, 127, 255);
6394 createTestsForAllStages("induction", inputColors, outputColors2, fragments2, group.get());
6396 const char typesAndConstants3[] =
6397 "%true = OpConstantTrue %bool\n"
6398 "%false = OpConstantFalse %bool\n"
6399 "%c_f32_p2 = OpConstant %f32 0.2\n";
6401 // Swap the second and the third element of the given parameter.
6402 const char function3[] =
6403 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6404 "%param = OpFunctionParameter %v4f32\n"
6405 "%entry = OpLabel\n"
6406 "%result = OpVariable %fp_v4f32 Function\n"
6407 " OpStore %result %param\n"
6408 "%a_loc = OpAccessChain %fp_f32 %result %c_i32_1\n"
6409 "%a_init = OpLoad %f32 %a_loc\n"
6410 "%b_loc = OpAccessChain %fp_f32 %result %c_i32_2\n"
6411 "%b_init = OpLoad %f32 %b_loc\n"
6415 "%still_loop = OpPhi %bool %true %entry %false %phi\n"
6416 "%a_next = OpPhi %f32 %a_init %entry %b_next %phi\n"
6417 "%b_next = OpPhi %f32 %b_init %entry %a_next %phi\n"
6418 " OpLoopMerge %exit %phi None\n"
6419 " OpBranchConditional %still_loop %phi %exit\n"
6422 " OpStore %a_loc %a_next\n"
6423 " OpStore %b_loc %b_next\n"
6424 "%ret = OpLoad %v4f32 %result\n"
6425 " OpReturnValue %ret\n"
6429 fragments3["pre_main"] = typesAndConstants3;
6430 fragments3["testfun"] = function3;
6432 outputColors3[0] = RGBA(127, 127, 127, 255);
6433 outputColors3[1] = RGBA(127, 0, 0, 255);
6434 outputColors3[2] = RGBA(0, 0, 127, 255);
6435 outputColors3[3] = RGBA(0, 127, 0, 255);
6437 createTestsForAllStages("swap", inputColors, outputColors3, fragments3, group.get());
6439 return group.release();
6442 tcu::TestCaseGroup* createNoContractionTests(tcu::TestContext& testCtx)
6444 de::MovePtr<tcu::TestCaseGroup> group (new tcu::TestCaseGroup(testCtx, "nocontraction", "Test the NoContraction decoration"));
6445 RGBA inputColors[4];
6446 RGBA outputColors[4];
6448 // With NoContraction, (1 + 2^-23) * (1 - 2^-23) - 1 should be conducted as a multiplication and an addition separately.
6449 // For the multiplication, the result is 1 - 2^-46, which is out of the precision range for 32-bit float. (32-bit float
6450 // only have 23-bit fraction.) So it will be rounded to 1. Then the final result is 0. On the contrary, the result will
6451 // be 2^-46, which is a normalized number perfectly representable as 32-bit float.
6452 const char constantsAndTypes[] =
6453 "%c_vec4_0 = OpConstantComposite %v4f32 %c_f32_0 %c_f32_0 %c_f32_0 %c_f32_1\n"
6454 "%c_vec4_1 = OpConstantComposite %v4f32 %c_f32_1 %c_f32_1 %c_f32_1 %c_f32_1\n"
6455 "%c_f32_1pl2_23 = OpConstant %f32 0x1.000002p+0\n" // 1 + 2^-23
6456 "%c_f32_1mi2_23 = OpConstant %f32 0x1.fffffcp-1\n" // 1 - 2^-23
6459 const char function[] =
6460 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6461 "%param = OpFunctionParameter %v4f32\n"
6462 "%label = OpLabel\n"
6463 "%var1 = OpVariable %fp_f32 Function %c_f32_1pl2_23\n"
6464 "%var2 = OpVariable %fp_f32 Function\n"
6465 "%red = OpCompositeExtract %f32 %param 0\n"
6466 "%plus_red = OpFAdd %f32 %c_f32_1mi2_23 %red\n"
6467 " OpStore %var2 %plus_red\n"
6468 "%val1 = OpLoad %f32 %var1\n"
6469 "%val2 = OpLoad %f32 %var2\n"
6470 "%mul = OpFMul %f32 %val1 %val2\n"
6471 "%add = OpFAdd %f32 %mul %c_f32_n1\n"
6472 "%is0 = OpFOrdEqual %bool %add %c_f32_0\n"
6473 "%ret = OpSelect %v4f32 %is0 %c_vec4_0 %c_vec4_1\n"
6474 " OpReturnValue %ret\n"
6477 struct CaseNameDecoration
6484 CaseNameDecoration tests[] = {
6485 {"multiplication", "OpDecorate %mul NoContraction"},
6486 {"addition", "OpDecorate %add NoContraction"},
6487 {"both", "OpDecorate %mul NoContraction\nOpDecorate %add NoContraction"},
6490 getHalfColorsFullAlpha(inputColors);
6492 for (deUint8 idx = 0; idx < 4; ++idx)
6494 inputColors[idx].setRed(0);
6495 outputColors[idx] = RGBA(0, 0, 0, 255);
6498 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(CaseNameDecoration); ++testNdx)
6500 map<string, string> fragments;
6502 fragments["decoration"] = tests[testNdx].decoration;
6503 fragments["pre_main"] = constantsAndTypes;
6504 fragments["testfun"] = function;
6506 createTestsForAllStages(tests[testNdx].name, inputColors, outputColors, fragments, group.get());
6509 return group.release();
6512 tcu::TestCaseGroup* createMemoryAccessTests(tcu::TestContext& testCtx)
6514 de::MovePtr<tcu::TestCaseGroup> memoryAccessTests (new tcu::TestCaseGroup(testCtx, "opmemoryaccess", "Memory Semantics"));
6517 const char constantsAndTypes[] =
6518 "%c_a2f32_1 = OpConstantComposite %a2f32 %c_f32_1 %c_f32_1\n"
6519 "%fp_a2f32 = OpTypePointer Function %a2f32\n"
6520 "%stype = OpTypeStruct %v4f32 %a2f32 %f32\n"
6521 "%fp_stype = OpTypePointer Function %stype\n";
6523 const char function[] =
6524 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6525 "%param1 = OpFunctionParameter %v4f32\n"
6527 "%v1 = OpVariable %fp_v4f32 Function\n"
6528 " OpStore %v1 %c_v4f32_1_1_1_1\n"
6529 "%v2 = OpVariable %fp_a2f32 Function\n"
6530 " OpStore %v2 %c_a2f32_1\n"
6531 "%v3 = OpVariable %fp_f32 Function\n"
6532 " OpStore %v3 %c_f32_1\n"
6534 "%v = OpVariable %fp_stype Function\n"
6535 "%vv = OpVariable %fp_stype Function\n"
6536 "%vvv = OpVariable %fp_f32 Function\n"
6538 "%p_v4f32 = OpAccessChain %fp_v4f32 %v %c_u32_0\n"
6539 "%p_a2f32 = OpAccessChain %fp_a2f32 %v %c_u32_1\n"
6540 "%p_f32 = OpAccessChain %fp_f32 %v %c_u32_2\n"
6541 "%v1_v = OpLoad %v4f32 %v1 ${access_type}\n"
6542 "%v2_v = OpLoad %a2f32 %v2 ${access_type}\n"
6543 "%v3_v = OpLoad %f32 %v3 ${access_type}\n"
6545 " OpStore %p_v4f32 %v1_v ${access_type}\n"
6546 " OpStore %p_a2f32 %v2_v ${access_type}\n"
6547 " OpStore %p_f32 %v3_v ${access_type}\n"
6549 " OpCopyMemory %vv %v ${access_type}\n"
6550 " OpCopyMemory %vvv %p_f32 ${access_type}\n"
6552 "%p_f32_2 = OpAccessChain %fp_f32 %vv %c_u32_2\n"
6553 "%v_f32_2 = OpLoad %f32 %p_f32_2\n"
6554 "%v_f32_3 = OpLoad %f32 %vvv\n"
6556 "%ret1 = OpVectorTimesScalar %v4f32 %param1 %v_f32_2\n"
6557 "%ret2 = OpVectorTimesScalar %v4f32 %ret1 %v_f32_3\n"
6558 " OpReturnValue %ret2\n"
6561 struct NameMemoryAccess
6568 NameMemoryAccess tests[] =
6571 { "volatile", "Volatile" },
6572 { "aligned", "Aligned 1" },
6573 { "volatile_aligned", "Volatile|Aligned 1" },
6574 { "nontemporal_aligned", "Nontemporal|Aligned 1" },
6575 { "volatile_nontemporal", "Volatile|Nontemporal" },
6576 { "volatile_nontermporal_aligned", "Volatile|Nontemporal|Aligned 1" },
6579 getHalfColorsFullAlpha(colors);
6581 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameMemoryAccess); ++testNdx)
6583 map<string, string> fragments;
6584 map<string, string> memoryAccess;
6585 memoryAccess["access_type"] = tests[testNdx].accessType;
6587 fragments["pre_main"] = constantsAndTypes;
6588 fragments["testfun"] = tcu::StringTemplate(function).specialize(memoryAccess);
6589 createTestsForAllStages(tests[testNdx].name, colors, colors, fragments, memoryAccessTests.get());
6591 return memoryAccessTests.release();
6593 tcu::TestCaseGroup* createOpUndefTests(tcu::TestContext& testCtx)
6595 de::MovePtr<tcu::TestCaseGroup> opUndefTests (new tcu::TestCaseGroup(testCtx, "opundef", "Test OpUndef"));
6596 RGBA defaultColors[4];
6597 map<string, string> fragments;
6598 getDefaultColors(defaultColors);
6600 // First, simple cases that don't do anything with the OpUndef result.
6601 fragments["testfun"] =
6602 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6603 "%param1 = OpFunctionParameter %v4f32\n"
6604 "%label_testfun = OpLabel\n"
6605 "%undef = OpUndef %type\n"
6606 "OpReturnValue %param1\n"
6609 struct NameCodePair { string name, code; };
6610 const NameCodePair tests[] =
6612 {"bool", "%type = OpTypeBool"},
6613 {"vec2uint32", "%type = OpTypeVector %u32 2"},
6614 {"image", "%type = OpTypeImage %f32 2D 0 0 0 0 Unknown"},
6615 {"sampler", "%type = OpTypeSampler"},
6616 {"sampledimage", "%img = OpTypeImage %f32 2D 0 0 0 0 Unknown\n" "%type = OpTypeSampledImage %img"},
6617 {"pointer", "%type = OpTypePointer Function %i32"},
6618 {"runtimearray", "%type = OpTypeRuntimeArray %f32"},
6619 {"array", "%c_u32_100 = OpConstant %u32 100\n" "%type = OpTypeArray %i32 %c_u32_100"},
6620 {"struct", "%type = OpTypeStruct %f32 %i32 %u32"}};
6621 for (size_t testNdx = 0; testNdx < sizeof(tests) / sizeof(NameCodePair); ++testNdx)
6623 fragments["pre_main"] = tests[testNdx].code;
6624 createTestsForAllStages(tests[testNdx].name, defaultColors, defaultColors, fragments, opUndefTests.get());
6628 fragments["testfun"] =
6629 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6630 "%param1 = OpFunctionParameter %v4f32\n"
6631 "%label_testfun = OpLabel\n"
6632 "%undef = OpUndef %f32\n"
6633 "%zero = OpFMul %f32 %undef %c_f32_0\n"
6634 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6635 "%b = OpFAdd %f32 %a %zero\n"
6636 "%ret = OpVectorInsertDynamic %v4f32 %param1 %b %c_i32_0\n"
6637 "OpReturnValue %ret\n"
6640 createTestsForAllStages("float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6642 fragments["testfun"] =
6643 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6644 "%param1 = OpFunctionParameter %v4f32\n"
6645 "%label_testfun = OpLabel\n"
6646 "%undef = OpUndef %i32\n"
6647 "%zero = OpIMul %i32 %undef %c_i32_0\n"
6648 "%a = OpVectorExtractDynamic %f32 %param1 %zero\n"
6649 "%ret = OpVectorInsertDynamic %v4f32 %param1 %a %c_i32_0\n"
6650 "OpReturnValue %ret\n"
6653 createTestsForAllStages("sint32", 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 %u32\n"
6660 "%zero = OpIMul %u32 %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("uint32", 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 %v4f32\n"
6673 "%vzero = OpVectorTimesScalar %v4f32 %undef %c_f32_0\n"
6674 "%zero_0 = OpVectorExtractDynamic %f32 %vzero %c_i32_0\n"
6675 "%zero_1 = OpVectorExtractDynamic %f32 %vzero %c_i32_1\n"
6676 "%zero_2 = OpVectorExtractDynamic %f32 %vzero %c_i32_2\n"
6677 "%zero_3 = OpVectorExtractDynamic %f32 %vzero %c_i32_3\n"
6678 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6679 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6680 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6681 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6682 "%sum_0 = OpFAdd %f32 %param1_0 %zero_0\n"
6683 "%sum_1 = OpFAdd %f32 %param1_1 %zero_1\n"
6684 "%sum_2 = OpFAdd %f32 %param1_2 %zero_2\n"
6685 "%sum_3 = OpFAdd %f32 %param1_3 %zero_3\n"
6686 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6687 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6688 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6689 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6690 "OpReturnValue %ret\n"
6693 createTestsForAllStages("vec4float32", defaultColors, defaultColors, fragments, opUndefTests.get());
6695 fragments["pre_main"] =
6696 "%v2f32 = OpTypeVector %f32 2\n"
6697 "%m2x2f32 = OpTypeMatrix %v2f32 2\n";
6698 fragments["testfun"] =
6699 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6700 "%param1 = OpFunctionParameter %v4f32\n"
6701 "%label_testfun = OpLabel\n"
6702 "%undef = OpUndef %m2x2f32\n"
6703 "%mzero = OpMatrixTimesScalar %m2x2f32 %undef %c_f32_0\n"
6704 "%zero_0 = OpCompositeExtract %f32 %mzero 0 0\n"
6705 "%zero_1 = OpCompositeExtract %f32 %mzero 0 1\n"
6706 "%zero_2 = OpCompositeExtract %f32 %mzero 1 0\n"
6707 "%zero_3 = OpCompositeExtract %f32 %mzero 1 1\n"
6708 "%param1_0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6709 "%param1_1 = OpVectorExtractDynamic %f32 %param1 %c_i32_1\n"
6710 "%param1_2 = OpVectorExtractDynamic %f32 %param1 %c_i32_2\n"
6711 "%param1_3 = OpVectorExtractDynamic %f32 %param1 %c_i32_3\n"
6712 "%sum_0 = OpFAdd %f32 %param1_0 %zero_0\n"
6713 "%sum_1 = OpFAdd %f32 %param1_1 %zero_1\n"
6714 "%sum_2 = OpFAdd %f32 %param1_2 %zero_2\n"
6715 "%sum_3 = OpFAdd %f32 %param1_3 %zero_3\n"
6716 "%ret3 = OpVectorInsertDynamic %v4f32 %param1 %sum_3 %c_i32_3\n"
6717 "%ret2 = OpVectorInsertDynamic %v4f32 %ret3 %sum_2 %c_i32_2\n"
6718 "%ret1 = OpVectorInsertDynamic %v4f32 %ret2 %sum_1 %c_i32_1\n"
6719 "%ret = OpVectorInsertDynamic %v4f32 %ret1 %sum_0 %c_i32_0\n"
6720 "OpReturnValue %ret\n"
6723 createTestsForAllStages("matrix", defaultColors, defaultColors, fragments, opUndefTests.get());
6725 return opUndefTests.release();
6728 void createOpQuantizeSingleOptionTests(tcu::TestCaseGroup* testCtx)
6730 const RGBA inputColors[4] =
6733 RGBA(0, 0, 255, 255),
6734 RGBA(0, 255, 0, 255),
6735 RGBA(0, 255, 255, 255)
6738 const RGBA expectedColors[4] =
6740 RGBA(255, 0, 0, 255),
6741 RGBA(255, 0, 0, 255),
6742 RGBA(255, 0, 0, 255),
6743 RGBA(255, 0, 0, 255)
6746 const struct SingleFP16Possibility
6749 const char* constant; // Value to assign to %test_constant.
6751 const char* condition; // Must assign to %cond an expression that evaluates to true after %c = OpQuantizeToF16(%test_constant + 0).
6757 -constructNormalizedFloat(1, 0x300000),
6758 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6763 constructNormalizedFloat(7, 0x000000),
6764 "%cond = OpFOrdEqual %bool %c %test_constant\n"
6766 // SPIR-V requires that OpQuantizeToF16 flushes
6767 // any numbers that would end up denormalized in F16 to zero.
6771 std::ldexp(1.5f, -140),
6772 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6777 -std::ldexp(1.5f, -140),
6778 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6783 std::ldexp(1.0f, -16),
6784 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6785 }, // too small positive
6787 "negative_too_small",
6789 -std::ldexp(1.0f, -32),
6790 "%cond = OpFOrdEqual %bool %c %c_f32_0\n"
6791 }, // too small negative
6795 -std::ldexp(1.0f, 128),
6797 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6798 "%inf = OpIsInf %bool %c\n"
6799 "%cond = OpLogicalAnd %bool %gz %inf\n"
6804 std::ldexp(1.0f, 128),
6806 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6807 "%inf = OpIsInf %bool %c\n"
6808 "%cond = OpLogicalAnd %bool %gz %inf\n"
6811 "round_to_negative_inf",
6813 -std::ldexp(1.0f, 32),
6815 "%gz = OpFOrdLessThan %bool %c %c_f32_0\n"
6816 "%inf = OpIsInf %bool %c\n"
6817 "%cond = OpLogicalAnd %bool %gz %inf\n"
6822 std::ldexp(1.0f, 16),
6824 "%gz = OpFOrdGreaterThan %bool %c %c_f32_0\n"
6825 "%inf = OpIsInf %bool %c\n"
6826 "%cond = OpLogicalAnd %bool %gz %inf\n"
6831 std::numeric_limits<float>::quiet_NaN(),
6833 // Can't use %c, because NaN+0 isn't necessarily a NaN (Vulkan spec A.4).
6834 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6835 "%nan = OpIsNan %bool %direct_quant\n"
6836 "%as_int = OpBitcast %i32 %direct_quant\n"
6837 "%positive = OpSGreaterThan %bool %as_int %c_i32_0\n"
6838 "%cond = OpLogicalAnd %bool %nan %positive\n"
6843 std::numeric_limits<float>::quiet_NaN(),
6845 // Can't use %c, because NaN+0 isn't necessarily a NaN (Vulkan spec A.4).
6846 "%direct_quant = OpQuantizeToF16 %f32 %test_constant\n"
6847 "%nan = OpIsNan %bool %direct_quant\n"
6848 "%as_int = OpBitcast %i32 %direct_quant\n"
6849 "%negative = OpSLessThan %bool %as_int %c_i32_0\n"
6850 "%cond = OpLogicalAnd %bool %nan %negative\n"
6853 const char* constants =
6854 "%test_constant = OpConstant %f32 "; // The value will be test.constant.
6856 StringTemplate function (
6857 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6858 "%param1 = OpFunctionParameter %v4f32\n"
6859 "%label_testfun = OpLabel\n"
6860 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6861 "%b = OpFAdd %f32 %test_constant %a\n"
6862 "%c = OpQuantizeToF16 %f32 %b\n"
6864 "%retval = OpSelect %v4f32 %cond %c_v4f32_1_0_0_1 %param1"
6865 " OpReturnValue %retval\n"
6869 const char* specDecorations = "OpDecorate %test_constant SpecId 0\n";
6870 const char* specConstants =
6871 "%test_constant = OpSpecConstant %f32 0.\n"
6872 "%c = OpSpecConstantOp %f32 QuantizeToF16 %test_constant\n";
6874 StringTemplate specConstantFunction(
6875 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6876 "%param1 = OpFunctionParameter %v4f32\n"
6877 "%label_testfun = OpLabel\n"
6879 "%retval = OpSelect %v4f32 %cond %c_v4f32_1_0_0_1 %param1"
6880 " OpReturnValue %retval\n"
6884 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6886 map<string, string> codeSpecialization;
6887 map<string, string> fragments;
6888 codeSpecialization["condition"] = tests[idx].condition;
6889 fragments["testfun"] = function.specialize(codeSpecialization);
6890 fragments["pre_main"] = string(constants) + tests[idx].constant + "\n";
6891 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
6894 for (size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx)
6896 map<string, string> codeSpecialization;
6897 map<string, string> fragments;
6898 vector<deInt32> passConstants;
6899 deInt32 specConstant;
6901 codeSpecialization["condition"] = tests[idx].condition;
6902 fragments["testfun"] = specConstantFunction.specialize(codeSpecialization);
6903 fragments["decoration"] = specDecorations;
6904 fragments["pre_main"] = specConstants;
6906 memcpy(&specConstant, &tests[idx].valueAsFloat, sizeof(float));
6907 passConstants.push_back(specConstant);
6909 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
6913 void createOpQuantizeTwoPossibilityTests(tcu::TestCaseGroup* testCtx)
6915 RGBA inputColors[4] = {
6917 RGBA(0, 0, 255, 255),
6918 RGBA(0, 255, 0, 255),
6919 RGBA(0, 255, 255, 255)
6922 RGBA expectedColors[4] =
6924 RGBA(255, 0, 0, 255),
6925 RGBA(255, 0, 0, 255),
6926 RGBA(255, 0, 0, 255),
6927 RGBA(255, 0, 0, 255)
6930 struct DualFP16Possibility
6935 const char* possibleOutput1;
6936 const char* possibleOutput2;
6939 "positive_round_up_or_round_down",
6941 constructNormalizedFloat(8, 0x300300),
6946 "negative_round_up_or_round_down",
6948 -constructNormalizedFloat(8, 0x600800),
6955 constructNormalizedFloat(8, 0x01e000),
6960 "carry_to_exponent",
6962 constructNormalizedFloat(8, 0xfee000),
6967 StringTemplate constants (
6968 "%input_const = OpConstant %f32 ${input}\n"
6969 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6970 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6973 StringTemplate specConstants (
6974 "%input_const = OpSpecConstant %f32 0.\n"
6975 "%possible_solution1 = OpConstant %f32 ${output1}\n"
6976 "%possible_solution2 = OpConstant %f32 ${output2}\n"
6979 const char* specDecorations = "OpDecorate %input_const SpecId 0\n";
6981 const char* function =
6982 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
6983 "%param1 = OpFunctionParameter %v4f32\n"
6984 "%label_testfun = OpLabel\n"
6985 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
6986 // For the purposes of this test we assume that 0.f will always get
6987 // faithfully passed through the pipeline stages.
6988 "%b = OpFAdd %f32 %input_const %a\n"
6989 "%c = OpQuantizeToF16 %f32 %b\n"
6990 "%eq_1 = OpFOrdEqual %bool %c %possible_solution1\n"
6991 "%eq_2 = OpFOrdEqual %bool %c %possible_solution2\n"
6992 "%cond = OpLogicalOr %bool %eq_1 %eq_2\n"
6993 "%retval = OpSelect %v4f32 %cond %c_v4f32_1_0_0_1 %param1"
6994 " OpReturnValue %retval\n"
6997 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
6998 map<string, string> fragments;
6999 map<string, string> constantSpecialization;
7001 constantSpecialization["input"] = tests[idx].input;
7002 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7003 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7004 fragments["testfun"] = function;
7005 fragments["pre_main"] = constants.specialize(constantSpecialization);
7006 createTestsForAllStages(tests[idx].name, inputColors, expectedColors, fragments, testCtx);
7009 for(size_t idx = 0; idx < (sizeof(tests)/sizeof(tests[0])); ++idx) {
7010 map<string, string> fragments;
7011 map<string, string> constantSpecialization;
7012 vector<deInt32> passConstants;
7013 deInt32 specConstant;
7015 constantSpecialization["output1"] = tests[idx].possibleOutput1;
7016 constantSpecialization["output2"] = tests[idx].possibleOutput2;
7017 fragments["testfun"] = function;
7018 fragments["decoration"] = specDecorations;
7019 fragments["pre_main"] = specConstants.specialize(constantSpecialization);
7021 memcpy(&specConstant, &tests[idx].inputAsFloat, sizeof(float));
7022 passConstants.push_back(specConstant);
7024 createTestsForAllStages(string("spec_const_") + tests[idx].name, inputColors, expectedColors, fragments, passConstants, testCtx);
7028 tcu::TestCaseGroup* createOpQuantizeTests(tcu::TestContext& testCtx)
7030 de::MovePtr<tcu::TestCaseGroup> opQuantizeTests (new tcu::TestCaseGroup(testCtx, "opquantize", "Test OpQuantizeToF16"));
7031 createOpQuantizeSingleOptionTests(opQuantizeTests.get());
7032 createOpQuantizeTwoPossibilityTests(opQuantizeTests.get());
7033 return opQuantizeTests.release();
7036 struct ShaderPermutation
7038 deUint8 vertexPermutation;
7039 deUint8 geometryPermutation;
7040 deUint8 tesscPermutation;
7041 deUint8 tessePermutation;
7042 deUint8 fragmentPermutation;
7045 ShaderPermutation getShaderPermutation(deUint8 inputValue)
7047 ShaderPermutation permutation =
7049 static_cast<deUint8>(inputValue & 0x10? 1u: 0u),
7050 static_cast<deUint8>(inputValue & 0x08? 1u: 0u),
7051 static_cast<deUint8>(inputValue & 0x04? 1u: 0u),
7052 static_cast<deUint8>(inputValue & 0x02? 1u: 0u),
7053 static_cast<deUint8>(inputValue & 0x01? 1u: 0u)
7058 tcu::TestCaseGroup* createModuleTests(tcu::TestContext& testCtx)
7060 RGBA defaultColors[4];
7061 RGBA invertedColors[4];
7062 de::MovePtr<tcu::TestCaseGroup> moduleTests (new tcu::TestCaseGroup(testCtx, "module", "Multiple entry points into shaders"));
7064 const ShaderElement combinedPipeline[] =
7066 ShaderElement("module", "main", VK_SHADER_STAGE_VERTEX_BIT),
7067 ShaderElement("module", "main", VK_SHADER_STAGE_GEOMETRY_BIT),
7068 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7069 ShaderElement("module", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7070 ShaderElement("module", "main", VK_SHADER_STAGE_FRAGMENT_BIT)
7073 getDefaultColors(defaultColors);
7074 getInvertedDefaultColors(invertedColors);
7075 addFunctionCaseWithPrograms<InstanceContext>(moduleTests.get(), "same_module", "", createCombinedModule, runAndVerifyDefaultPipeline, createInstanceContext(combinedPipeline, map<string, string>()));
7077 const char* numbers[] =
7082 for (deInt8 idx = 0; idx < 32; ++idx)
7084 ShaderPermutation permutation = getShaderPermutation(idx);
7085 string name = string("vert") + numbers[permutation.vertexPermutation] + "_geom" + numbers[permutation.geometryPermutation] + "_tessc" + numbers[permutation.tesscPermutation] + "_tesse" + numbers[permutation.tessePermutation] + "_frag" + numbers[permutation.fragmentPermutation];
7086 const ShaderElement pipeline[] =
7088 ShaderElement("vert", string("vert") + numbers[permutation.vertexPermutation], VK_SHADER_STAGE_VERTEX_BIT),
7089 ShaderElement("geom", string("geom") + numbers[permutation.geometryPermutation], VK_SHADER_STAGE_GEOMETRY_BIT),
7090 ShaderElement("tessc", string("tessc") + numbers[permutation.tesscPermutation], VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7091 ShaderElement("tesse", string("tesse") + numbers[permutation.tessePermutation], VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7092 ShaderElement("frag", string("frag") + numbers[permutation.fragmentPermutation], VK_SHADER_STAGE_FRAGMENT_BIT)
7095 // If there are an even number of swaps, then it should be no-op.
7096 // If there are an odd number, the color should be flipped.
7097 if ((permutation.vertexPermutation + permutation.geometryPermutation + permutation.tesscPermutation + permutation.tessePermutation + permutation.fragmentPermutation) % 2 == 0)
7099 addFunctionCaseWithPrograms<InstanceContext>(moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline, createInstanceContext(pipeline, defaultColors, defaultColors, map<string, string>()));
7103 addFunctionCaseWithPrograms<InstanceContext>(moduleTests.get(), name, "", createMultipleEntries, runAndVerifyDefaultPipeline, createInstanceContext(pipeline, defaultColors, invertedColors, map<string, string>()));
7106 return moduleTests.release();
7109 tcu::TestCaseGroup* createLoopTests(tcu::TestContext& testCtx)
7111 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "loop", "Looping control flow"));
7112 RGBA defaultColors[4];
7113 getDefaultColors(defaultColors);
7114 map<string, string> fragments;
7115 fragments["pre_main"] =
7116 "%c_f32_5 = OpConstant %f32 5.\n";
7118 // A loop with a single block. The Continue Target is the loop block
7119 // itself. In SPIR-V terms, the "loop construct" contains no blocks at all
7120 // -- the "continue construct" forms the entire loop.
7121 fragments["testfun"] =
7122 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7123 "%param1 = OpFunctionParameter %v4f32\n"
7125 "%entry = OpLabel\n"
7126 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7129 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7131 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7132 "%delta = OpPhi %f32 %c_f32_1 %entry %minus_delta %loop\n"
7133 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7134 "%val = OpFAdd %f32 %val1 %delta\n"
7135 "%minus_delta = OpFSub %f32 %c_f32_0 %delta\n"
7136 "%count__ = OpISub %i32 %count %c_i32_1\n"
7137 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7138 "OpLoopMerge %exit %loop None\n"
7139 "OpBranchConditional %again %loop %exit\n"
7142 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7143 "OpReturnValue %result\n"
7147 createTestsForAllStages("single_block", defaultColors, defaultColors, fragments, testGroup.get());
7149 // Body comprised of multiple basic blocks.
7150 const StringTemplate multiBlock(
7151 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7152 "%param1 = OpFunctionParameter %v4f32\n"
7154 "%entry = OpLabel\n"
7155 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7158 ";adds and subtracts 1.0 to %val in alternate iterations\n"
7160 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %gather\n"
7161 "%delta = OpPhi %f32 %c_f32_1 %entry %delta_next %gather\n"
7162 "%val1 = OpPhi %f32 %val0 %entry %val %gather\n"
7163 // There are several possibilities for the Continue Target below. Each
7164 // will be specialized into a separate test case.
7165 "OpLoopMerge %exit ${continue_target} None\n"
7169 ";delta_next = (delta > 0) ? -1 : 1;\n"
7170 "%gt0 = OpFOrdGreaterThan %bool %delta %c_f32_0\n"
7171 "OpSelectionMerge %gather DontFlatten\n"
7172 "OpBranchConditional %gt0 %even %odd ;tells us if %count is even or odd\n"
7175 "OpBranch %gather\n"
7178 "OpBranch %gather\n"
7180 "%gather = OpLabel\n"
7181 "%delta_next = OpPhi %f32 %c_f32_n1 %even %c_f32_1 %odd\n"
7182 "%val = OpFAdd %f32 %val1 %delta\n"
7183 "%count__ = OpISub %i32 %count %c_i32_1\n"
7184 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7185 "OpBranchConditional %again %loop %exit\n"
7188 "%result = OpVectorInsertDynamic %v4f32 %param1 %val %c_i32_0\n"
7189 "OpReturnValue %result\n"
7193 map<string, string> continue_target;
7195 // The Continue Target is the loop block itself.
7196 continue_target["continue_target"] = "%loop";
7197 fragments["testfun"] = multiBlock.specialize(continue_target);
7198 createTestsForAllStages("multi_block_continue_construct", defaultColors, defaultColors, fragments, testGroup.get());
7200 // The Continue Target is at the end of the loop.
7201 continue_target["continue_target"] = "%gather";
7202 fragments["testfun"] = multiBlock.specialize(continue_target);
7203 createTestsForAllStages("multi_block_loop_construct", defaultColors, defaultColors, fragments, testGroup.get());
7205 // A loop with continue statement.
7206 fragments["testfun"] =
7207 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7208 "%param1 = OpFunctionParameter %v4f32\n"
7210 "%entry = OpLabel\n"
7211 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7214 ";adds 4, 3, and 1 to %val0 (skips 2)\n"
7216 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7217 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
7218 "OpLoopMerge %exit %continue None\n"
7222 ";skip if %count==2\n"
7223 "%eq2 = OpIEqual %bool %count %c_i32_2\n"
7224 "OpSelectionMerge %continue DontFlatten\n"
7225 "OpBranchConditional %eq2 %continue %body\n"
7228 "%fcount = OpConvertSToF %f32 %count\n"
7229 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7230 "OpBranch %continue\n"
7232 "%continue = OpLabel\n"
7233 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
7234 "%count__ = OpISub %i32 %count %c_i32_1\n"
7235 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7236 "OpBranchConditional %again %loop %exit\n"
7239 "%same = OpFSub %f32 %val %c_f32_8\n"
7240 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7241 "OpReturnValue %result\n"
7243 createTestsForAllStages("continue", defaultColors, defaultColors, fragments, testGroup.get());
7245 // A loop with early exit. May be specialized with either break or return.
7246 StringTemplate earlyExitLoop(
7247 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7248 "%param1 = OpFunctionParameter %v4f32\n"
7250 "%entry = OpLabel\n"
7251 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7252 "%dot = OpDot %f32 %param1 %param1\n"
7253 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7254 "%zero = OpConvertFToU %u32 %div\n"
7255 "%two = OpIAdd %i32 %zero %c_i32_2\n"
7256 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7259 ";adds 4 and 3 to %val0 (exits early)\n"
7261 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %continue\n"
7262 "%val1 = OpPhi %f32 %val0 %entry %val %continue\n"
7263 "OpLoopMerge %exit %continue None\n"
7267 ";end loop if %count==%two\n"
7268 "%above2 = OpSGreaterThan %bool %count %two\n"
7269 "OpSelectionMerge %continue DontFlatten\n"
7270 // This can either branch to %exit or to another block with OpReturnValue %param1.
7271 "OpBranchConditional %above2 %body ${branch_destination}\n"
7274 "%fcount = OpConvertSToF %f32 %count\n"
7275 "%val2 = OpFAdd %f32 %val1 %fcount\n"
7276 "OpBranch %continue\n"
7278 "%continue = OpLabel\n"
7279 "%val = OpPhi %f32 %val2 %body %val1 %if\n"
7280 "%count__ = OpISub %i32 %count %c_i32_1\n"
7281 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7282 "OpBranchConditional %again %loop %exit\n"
7285 "%same = OpFSub %f32 %val %c_f32_7\n"
7286 "%result = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7287 "OpReturnValue %result\n"
7290 map<string, string> branch_destination;
7292 // A loop with break.
7293 branch_destination["branch_destination"] = "%exit";
7294 fragments["testfun"] = earlyExitLoop.specialize(branch_destination);
7295 createTestsForAllStages("break", defaultColors, defaultColors, fragments, testGroup.get());
7297 // A loop with return.
7298 branch_destination["branch_destination"] = "%early_exit\n"
7299 "%early_exit = OpLabel\n"
7300 "OpReturnValue %param1\n";
7301 fragments["testfun"] = earlyExitLoop.specialize(branch_destination);
7302 createTestsForAllStages("return", defaultColors, defaultColors, fragments, testGroup.get());
7304 return testGroup.release();
7307 // Adds a new test to group using custom fragments for the tessellation-control
7308 // stage and passthrough fragments for all other stages. Uses default colors
7309 // for input and expected output.
7310 void addTessCtrlTest(tcu::TestCaseGroup* group, const char* name, const map<string, string>& fragments)
7312 RGBA defaultColors[4];
7313 getDefaultColors(defaultColors);
7314 const ShaderElement pipelineStages[] =
7316 ShaderElement("vert", "main", VK_SHADER_STAGE_VERTEX_BIT),
7317 ShaderElement("tessc", "main", VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT),
7318 ShaderElement("tesse", "main", VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT),
7319 ShaderElement("frag", "main", VK_SHADER_STAGE_FRAGMENT_BIT),
7322 addFunctionCaseWithPrograms<InstanceContext>(group, name, "", addShaderCodeCustomTessControl,
7323 runAndVerifyDefaultPipeline, createInstanceContext(
7324 pipelineStages, defaultColors, defaultColors, fragments, StageToSpecConstantMap()));
7327 // A collection of tests putting OpControlBarrier in places GLSL forbids but SPIR-V allows.
7328 tcu::TestCaseGroup* createBarrierTests(tcu::TestContext& testCtx)
7330 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "barrier", "OpControlBarrier"));
7331 map<string, string> fragments;
7333 // A barrier inside a function body.
7334 fragments["pre_main"] =
7335 "%Workgroup = OpConstant %i32 2\n"
7336 "%SequentiallyConsistent = OpConstant %i32 0x10\n";
7337 fragments["testfun"] =
7338 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7339 "%param1 = OpFunctionParameter %v4f32\n"
7340 "%label_testfun = OpLabel\n"
7341 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7342 "OpReturnValue %param1\n"
7344 addTessCtrlTest(testGroup.get(), "in_function", fragments);
7346 // Common setup code for the following tests.
7347 fragments["pre_main"] =
7348 "%Workgroup = OpConstant %i32 2\n"
7349 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7350 "%c_f32_5 = OpConstant %f32 5.\n";
7351 const string setupPercentZero = // Begins %test_code function with code that sets %zero to 0u but cannot be optimized away.
7352 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7353 "%param1 = OpFunctionParameter %v4f32\n"
7354 "%entry = OpLabel\n"
7355 ";param1 components are between 0 and 1, so dot product is 4 or less\n"
7356 "%dot = OpDot %f32 %param1 %param1\n"
7357 "%div = OpFDiv %f32 %dot %c_f32_5\n"
7358 "%zero = OpConvertFToU %u32 %div\n";
7360 // Barriers inside OpSwitch branches.
7361 fragments["testfun"] =
7363 "OpSelectionMerge %switch_exit None\n"
7364 "OpSwitch %zero %switch_default 0 %case0 1 %case1 ;should always go to %case0\n"
7366 "%case1 = OpLabel\n"
7367 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7368 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7369 "%wrong_branch_alert1 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7370 "OpBranch %switch_exit\n"
7372 "%switch_default = OpLabel\n"
7373 "%wrong_branch_alert2 = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7374 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7375 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7376 "OpBranch %switch_exit\n"
7378 "%case0 = OpLabel\n"
7379 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7380 "OpBranch %switch_exit\n"
7382 "%switch_exit = OpLabel\n"
7383 "%ret = OpPhi %v4f32 %param1 %case0 %wrong_branch_alert1 %case1 %wrong_branch_alert2 %switch_default\n"
7384 "OpReturnValue %ret\n"
7386 addTessCtrlTest(testGroup.get(), "in_switch", fragments);
7388 // Barriers inside if-then-else.
7389 fragments["testfun"] =
7391 "%eq0 = OpIEqual %bool %zero %c_u32_0\n"
7392 "OpSelectionMerge %exit DontFlatten\n"
7393 "OpBranchConditional %eq0 %then %else\n"
7396 ";This barrier should never be executed, but its presence makes test failure more likely when there's a bug.\n"
7397 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7398 "%wrong_branch_alert = OpVectorInsertDynamic %v4f32 %param1 %c_f32_0_5 %c_i32_0\n"
7402 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7406 "%ret = OpPhi %v4f32 %param1 %then %wrong_branch_alert %else\n"
7407 "OpReturnValue %ret\n"
7409 addTessCtrlTest(testGroup.get(), "in_if", fragments);
7411 // A barrier after control-flow reconvergence, tempting the compiler to attempt something like this:
7412 // http://lists.llvm.org/pipermail/llvm-dev/2009-October/026317.html.
7413 fragments["testfun"] =
7415 "%thread_id = OpLoad %i32 %BP_gl_InvocationID\n"
7416 "%thread0 = OpIEqual %bool %thread_id %c_i32_0\n"
7417 "OpSelectionMerge %exit DontFlatten\n"
7418 "OpBranchConditional %thread0 %then %else\n"
7421 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7425 "%val1 = OpVectorExtractDynamic %f32 %param1 %zero\n"
7429 "%val = OpPhi %f32 %val0 %else %val1 %then\n"
7430 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7431 "%ret = OpVectorInsertDynamic %v4f32 %param1 %val %zero\n"
7432 "OpReturnValue %ret\n"
7434 addTessCtrlTest(testGroup.get(), "after_divergent_if", fragments);
7436 // A barrier inside a loop.
7437 fragments["pre_main"] =
7438 "%Workgroup = OpConstant %i32 2\n"
7439 "%SequentiallyConsistent = OpConstant %i32 0x10\n"
7440 "%c_f32_10 = OpConstant %f32 10.\n";
7441 fragments["testfun"] =
7442 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7443 "%param1 = OpFunctionParameter %v4f32\n"
7444 "%entry = OpLabel\n"
7445 "%val0 = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7448 ";adds 4, 3, 2, and 1 to %val0\n"
7450 "%count = OpPhi %i32 %c_i32_4 %entry %count__ %loop\n"
7451 "%val1 = OpPhi %f32 %val0 %entry %val %loop\n"
7452 "OpControlBarrier %Workgroup %Workgroup %SequentiallyConsistent\n"
7453 "%fcount = OpConvertSToF %f32 %count\n"
7454 "%val = OpFAdd %f32 %val1 %fcount\n"
7455 "%count__ = OpISub %i32 %count %c_i32_1\n"
7456 "%again = OpSGreaterThan %bool %count__ %c_i32_0\n"
7457 "OpLoopMerge %exit %loop None\n"
7458 "OpBranchConditional %again %loop %exit\n"
7461 "%same = OpFSub %f32 %val %c_f32_10\n"
7462 "%ret = OpVectorInsertDynamic %v4f32 %param1 %same %c_i32_0\n"
7463 "OpReturnValue %ret\n"
7465 addTessCtrlTest(testGroup.get(), "in_loop", fragments);
7467 return testGroup.release();
7470 // Test for the OpFRem instruction.
7471 tcu::TestCaseGroup* createFRemTests(tcu::TestContext& testCtx)
7473 de::MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "frem", "OpFRem"));
7474 map<string, string> fragments;
7475 RGBA inputColors[4];
7476 RGBA outputColors[4];
7478 fragments["pre_main"] =
7479 "%c_f32_3 = OpConstant %f32 3.0\n"
7480 "%c_f32_n3 = OpConstant %f32 -3.0\n"
7481 "%c_f32_4 = OpConstant %f32 4.0\n"
7482 "%c_f32_p75 = OpConstant %f32 0.75\n"
7483 "%c_v4f32_p75_p75_p75_p75 = OpConstantComposite %v4f32 %c_f32_p75 %c_f32_p75 %c_f32_p75 %c_f32_p75 \n"
7484 "%c_v4f32_4_4_4_4 = OpConstantComposite %v4f32 %c_f32_4 %c_f32_4 %c_f32_4 %c_f32_4\n"
7485 "%c_v4f32_3_n3_3_n3 = OpConstantComposite %v4f32 %c_f32_3 %c_f32_n3 %c_f32_3 %c_f32_n3\n";
7487 // The test does the following.
7488 // vec4 result = (param1 * 8.0) - 4.0;
7489 // return (frem(result.x,3) + 0.75, frem(result.y, -3) + 0.75, 0, 1)
7490 fragments["testfun"] =
7491 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7492 "%param1 = OpFunctionParameter %v4f32\n"
7493 "%label_testfun = OpLabel\n"
7494 "%v_times_8 = OpVectorTimesScalar %v4f32 %param1 %c_f32_8\n"
7495 "%minus_4 = OpFSub %v4f32 %v_times_8 %c_v4f32_4_4_4_4\n"
7496 "%frem = OpFRem %v4f32 %minus_4 %c_v4f32_3_n3_3_n3\n"
7497 "%added = OpFAdd %v4f32 %frem %c_v4f32_p75_p75_p75_p75\n"
7498 "%xyz_1 = OpVectorInsertDynamic %v4f32 %added %c_f32_1 %c_i32_3\n"
7499 "%xy_0_1 = OpVectorInsertDynamic %v4f32 %xyz_1 %c_f32_0 %c_i32_2\n"
7500 "OpReturnValue %xy_0_1\n"
7504 inputColors[0] = RGBA(16, 16, 0, 255);
7505 inputColors[1] = RGBA(232, 232, 0, 255);
7506 inputColors[2] = RGBA(232, 16, 0, 255);
7507 inputColors[3] = RGBA(16, 232, 0, 255);
7509 outputColors[0] = RGBA(64, 64, 0, 255);
7510 outputColors[1] = RGBA(255, 255, 0, 255);
7511 outputColors[2] = RGBA(255, 64, 0, 255);
7512 outputColors[3] = RGBA(64, 255, 0, 255);
7514 createTestsForAllStages("frem", inputColors, outputColors, fragments, testGroup.get());
7515 return testGroup.release();
7518 tcu::TestCaseGroup* createInstructionTests (tcu::TestContext& testCtx)
7520 de::MovePtr<tcu::TestCaseGroup> instructionTests (new tcu::TestCaseGroup(testCtx, "instruction", "Instructions with special opcodes/operands"));
7521 de::MovePtr<tcu::TestCaseGroup> computeTests (new tcu::TestCaseGroup(testCtx, "compute", "Compute Instructions with special opcodes/operands"));
7522 de::MovePtr<tcu::TestCaseGroup> graphicsTests (new tcu::TestCaseGroup(testCtx, "graphics", "Graphics Instructions with special opcodes/operands"));
7524 computeTests->addChild(createOpNopGroup(testCtx));
7525 computeTests->addChild(createOpLineGroup(testCtx));
7526 computeTests->addChild(createOpNoLineGroup(testCtx));
7527 computeTests->addChild(createOpConstantNullGroup(testCtx));
7528 computeTests->addChild(createOpConstantCompositeGroup(testCtx));
7529 computeTests->addChild(createOpConstantUsageGroup(testCtx));
7530 computeTests->addChild(createSpecConstantGroup(testCtx));
7531 computeTests->addChild(createOpSourceGroup(testCtx));
7532 computeTests->addChild(createOpSourceExtensionGroup(testCtx));
7533 computeTests->addChild(createDecorationGroupGroup(testCtx));
7534 computeTests->addChild(createOpPhiGroup(testCtx));
7535 computeTests->addChild(createLoopControlGroup(testCtx));
7536 computeTests->addChild(createFunctionControlGroup(testCtx));
7537 computeTests->addChild(createSelectionControlGroup(testCtx));
7538 computeTests->addChild(createBlockOrderGroup(testCtx));
7539 computeTests->addChild(createMultipleShaderGroup(testCtx));
7540 computeTests->addChild(createMemoryAccessGroup(testCtx));
7541 computeTests->addChild(createOpCopyMemoryGroup(testCtx));
7542 computeTests->addChild(createOpCopyObjectGroup(testCtx));
7543 computeTests->addChild(createNoContractionGroup(testCtx));
7544 computeTests->addChild(createOpUndefGroup(testCtx));
7545 computeTests->addChild(createOpUnreachableGroup(testCtx));
7546 computeTests ->addChild(createOpQuantizeToF16Group(testCtx));
7547 computeTests ->addChild(createOpFRemGroup(testCtx));
7549 RGBA defaultColors[4];
7550 getDefaultColors(defaultColors);
7552 de::MovePtr<tcu::TestCaseGroup> opnopTests (new tcu::TestCaseGroup(testCtx, "opnop", "Test OpNop"));
7553 map<string, string> opNopFragments;
7554 opNopFragments["testfun"] =
7555 "%test_code = OpFunction %v4f32 None %v4f32_function\n"
7556 "%param1 = OpFunctionParameter %v4f32\n"
7557 "%label_testfun = OpLabel\n"
7566 "%a = OpVectorExtractDynamic %f32 %param1 %c_i32_0\n"
7567 "%b = OpFAdd %f32 %a %a\n"
7569 "%c = OpFSub %f32 %b %a\n"
7570 "%ret = OpVectorInsertDynamic %v4f32 %param1 %c %c_i32_0\n"
7573 "OpReturnValue %ret\n"
7576 createTestsForAllStages("opnop", defaultColors, defaultColors, opNopFragments, opnopTests.get());
7579 graphicsTests->addChild(opnopTests.release());
7580 graphicsTests->addChild(createOpSourceTests(testCtx));
7581 graphicsTests->addChild(createOpSourceContinuedTests(testCtx));
7582 graphicsTests->addChild(createOpLineTests(testCtx));
7583 graphicsTests->addChild(createOpNoLineTests(testCtx));
7584 graphicsTests->addChild(createOpConstantNullTests(testCtx));
7585 graphicsTests->addChild(createOpConstantCompositeTests(testCtx));
7586 graphicsTests->addChild(createMemoryAccessTests(testCtx));
7587 graphicsTests->addChild(createOpUndefTests(testCtx));
7588 graphicsTests->addChild(createSelectionBlockOrderTests(testCtx));
7589 graphicsTests->addChild(createModuleTests(testCtx));
7590 graphicsTests->addChild(createSwitchBlockOrderTests(testCtx));
7591 graphicsTests->addChild(createOpPhiTests(testCtx));
7592 graphicsTests->addChild(createNoContractionTests(testCtx));
7593 graphicsTests->addChild(createOpQuantizeTests(testCtx));
7594 graphicsTests->addChild(createLoopTests(testCtx));
7595 graphicsTests->addChild(createSpecConstantTests(testCtx));
7596 graphicsTests->addChild(createSpecConstantOpQuantizeToF16Group(testCtx));
7597 graphicsTests->addChild(createBarrierTests(testCtx));
7598 graphicsTests->addChild(createDecorationGroupTests(testCtx));
7599 graphicsTests->addChild(createFRemTests(testCtx));
7601 instructionTests->addChild(computeTests.release());
7602 instructionTests->addChild(graphicsTests.release());
7604 return instructionTests.release();
projects / platform / upstream / VK-GL-CTS.git / blob