/*------------------------------------------------------------------------- * Vulkan Conformance Tests * ------------------------ * * Copyright (c) 2017 Google Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. * *//*! * \file * \brief SPIR-V Assembly Tests for the SPV_KHR_variable_pointers extension *//*--------------------------------------------------------------------*/ #include "tcuFloat.hpp" #include "tcuRGBA.hpp" #include "tcuStringTemplate.hpp" #include "tcuTestLog.hpp" #include "tcuVectorUtil.hpp" #include "vkDefs.hpp" #include "vkDeviceUtil.hpp" #include "vkMemUtil.hpp" #include "vkPlatform.hpp" #include "vkPrograms.hpp" #include "vkQueryUtil.hpp" #include "vkRef.hpp" #include "vkRefUtil.hpp" #include "vkStrUtil.hpp" #include "vkTypeUtil.hpp" #include "deRandom.hpp" #include "deStringUtil.hpp" #include "deUniquePtr.hpp" #include "deMath.h" #include "vktSpvAsmComputeShaderCase.hpp" #include "vktSpvAsmComputeShaderTestUtil.hpp" #include "vktSpvAsmGraphicsShaderTestUtil.hpp" #include "vktSpvAsmVariablePointersTests.hpp" #include "vktTestCaseUtil.hpp" #include "vktTestGroupUtil.hpp" #include "vktAmberTestCase.hpp" #include #include #include #include #include namespace vkt { namespace SpirVAssembly { using namespace vk; using std::map; using std::string; using std::vector; using tcu::IVec3; using tcu::IVec4; using tcu::RGBA; using tcu::TestLog; using tcu::TestStatus; using tcu::Vec4; using de::UniquePtr; using tcu::StringTemplate; using tcu::Vec4; namespace { template void fillRandomScalars (de::Random& rnd, T minValue, T maxValue, void* dst, int numValues, int offset = 0) { T* const typedPtr = (T*)dst; for (int ndx = 0; ndx < numValues; ndx++) typedPtr[offset + ndx] = de::randomScalar(rnd, minValue, maxValue); } // The following structure (outer_struct) is passed as a vector of 64 32-bit floats into some shaders. // // struct struct inner_struct { // vec4 x[2]; // vec4 y[2]; // }; // // struct outer_struct { // inner_struct r[2][2]; // }; // // This method finds the correct offset from the base of a vector given the indexes into the structure. // Returns the index in the inclusive range of 0 and 63. Each unit of the offset represents offset by the size of a 32-bit float. deUint32 getBaseOffset (deUint32 indexMatrixRow, deUint32 indexMatrixCol, deUint32 indexInnerStruct, deUint32 indexVec4Array, deUint32 indexVec4) { DE_ASSERT(indexMatrixRow < 2); DE_ASSERT(indexMatrixCol < 2); DE_ASSERT(indexInnerStruct < 2); DE_ASSERT(indexVec4Array < 2); DE_ASSERT(indexVec4 < 4); deUint32 offset = 0; // We have a matrix of 2 rows and 2 columns (total of 4 inner_structs). Each inner_struct contains 16 floats. // So, offset by 1 row means offset by 32 floats, and offset by 1 column means offset by 16 floats. offset += indexMatrixRow * 32; offset += indexMatrixCol * 16; // The inner structure contains 2 members, each having 8 floats. // So offset by 1 in the inner struct means offset by 8 floats. offset += indexInnerStruct * 8; // Each member (x|y) have 2 vectors of 4 floats. So, offset by 1 int the vec4 array means an offset by 4 floats. offset += indexVec4Array * 4; // Each vec4 contains 4 floats, so each offset in the vec4 means offset by 1 float. offset += indexVec4; return offset; } // The following structure (input_buffer) is passed as a vector of 128 32-bit floats into some shaders. // // struct struct inner_struct { // vec4 x[2]; // vec4 y[2]; // }; // // struct outer_struct { // inner_struct r[2][2]; // }; // // struct input_buffer { // outer_struct a; // outer_struct b; // } // // This method finds the correct offset from the base of a vector given the indexes into the structure. // Returns the index in the inclusive range of 0 and 127. deUint32 getBaseOffsetForSingleInputBuffer (deUint32 indexOuterStruct, deUint32 indexMatrixRow, deUint32 indexMatrixCol, deUint32 indexInnerStruct, deUint32 indexVec4Array, deUint32 indexVec4) { DE_ASSERT(indexOuterStruct < 2); DE_ASSERT(indexMatrixRow < 2); DE_ASSERT(indexMatrixCol < 2); DE_ASSERT(indexInnerStruct < 2); DE_ASSERT(indexVec4Array < 2); DE_ASSERT(indexVec4 < 4); // Get the offset assuming you have only one outer_struct. deUint32 offset = getBaseOffset(indexMatrixRow, indexMatrixCol, indexInnerStruct, indexVec4Array, indexVec4); // If the second outer structure (b) is chosen in the input_buffer, we need to add an offset of 64 since // each outer_struct contains 64 floats. if (indexOuterStruct == 1) offset += 64; return offset; } void addPhysicalOrVariablePointersComputeGroup (tcu::TestCaseGroup* group, bool physPtrs) { tcu::TestContext& testCtx = group->getTestContext(); de::Random rnd (deStringHash(group->getName())); const int seed = testCtx.getCommandLine().getBaseSeed(); const int numMuxes = 100; std::string inputArraySize = "200"; vector inputAFloats (2*numMuxes, 0); vector inputBFloats (2*numMuxes, 0); vector inputSFloats (numMuxes, 0); vector AmuxAOutputFloats (numMuxes, 0); vector AmuxBOutputFloats (numMuxes, 0); vector incrAmuxAOutputFloats (numMuxes, 0); vector incrAmuxBOutputFloats (numMuxes, 0); VulkanFeatures requiredFeatures; // Each output entry is chosen as follows: ( 0 <= i < numMuxes) // 1) For tests with one input buffer: output[i] = (s[i] < 0) ? A[2*i] : A[2*i+1]; // 2) For tests with two input buffers: output[i] = (s[i] < 0) ? A[i] : B[i]; fillRandomScalars(rnd, -100.f, 100.f, &inputAFloats[0], 2*numMuxes); fillRandomScalars(rnd, -100.f, 100.f, &inputBFloats[0], 2*numMuxes); // We want to guarantee that the S input has some positive and some negative values. // We choose random negative numbers for the first half, random positive numbers for the second half, and then shuffle. fillRandomScalars(rnd, -100.f, -1.f , &inputSFloats[0], numMuxes / 2); fillRandomScalars(rnd, 1.f , 100.f, &inputSFloats[numMuxes / 2], numMuxes / 2); de::Random(seed).shuffle(inputSFloats.begin(), inputSFloats.end()); for (size_t i = 0; i < numMuxes; ++i) { AmuxAOutputFloats[i] = (inputSFloats[i] < 0) ? inputAFloats[2*i] : inputAFloats[2*i+1]; AmuxBOutputFloats[i] = (inputSFloats[i] < 0) ? inputAFloats[i] : inputBFloats[i]; incrAmuxAOutputFloats[i] = (inputSFloats[i] < 0) ? 1 + inputAFloats[2*i] : 1 + inputAFloats[2*i+1]; incrAmuxBOutputFloats[i] = (inputSFloats[i] < 0) ? 1 + inputAFloats[i] : 1 + inputBFloats[i]; } std::string stringTemplate = "OpCapability Shader\n" "${ExtraCapability}\n"; stringTemplate += physPtrs ? "OpExtension \"SPV_KHR_physical_storage_buffer\"\n" : "OpExtension \"SPV_KHR_variable_pointers\"\n"; stringTemplate += "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" "OpMemoryModel " + string(physPtrs ? "PhysicalStorageBuffer64EXT" : "Logical") + " GLSL450\n" "OpEntryPoint GLCompute %main \"main\" %id\n" "OpExecutionMode %main LocalSize 1 1 1\n" "OpSource GLSL 430\n" "OpName %main \"main\"\n" "OpName %id \"gl_GlobalInvocationID\"\n" // Decorations "OpDecorate %id BuiltIn GlobalInvocationId\n" "OpDecorate %f32arr ArrayStride 4\n" "OpDecorate %sb_f32ptr ArrayStride 4\n" "OpDecorate %buf Block\n" "OpMemberDecorate %buf 0 Offset 0\n" "${ExtraDecorations}"; stringTemplate += physPtrs ? "OpDecorate %physPtrsStruct Block\n" "OpMemberDecorate %physPtrsStruct 0 Offset 0\n" "OpMemberDecorate %physPtrsStruct 1 Offset 8\n" "OpMemberDecorate %physPtrsStruct 2 Offset 16\n" "OpMemberDecorate %physPtrsStruct 3 Offset 24\n" "OpDecorate %indata_all DescriptorSet 0\n" "OpDecorate %indata_all Binding 0\n" "OpDecorate %first_ptr_param Restrict\n" "OpDecorate %second_ptr_param Restrict\n" : "OpDecorate %indata_a DescriptorSet 0\n" "OpDecorate %indata_a Binding 0\n" "OpDecorate %indata_b DescriptorSet 0\n" "OpDecorate %indata_b Binding 1\n" "OpDecorate %indata_s DescriptorSet 0\n" "OpDecorate %indata_s Binding 2\n" "OpDecorate %outdata DescriptorSet 0\n" "OpDecorate %outdata Binding 3\n"; stringTemplate += string(getComputeAsmCommonTypes()); stringTemplate += physPtrs ? "%sb_f32ptr = OpTypePointer PhysicalStorageBufferEXT %f32\n" "%buf = OpTypeStruct %f32arr\n" "%bufptrphys = OpTypePointer PhysicalStorageBufferEXT %buf\n" "%physPtrsStruct = OpTypeStruct %bufptrphys %bufptrphys %bufptrphys %bufptrphys\n" "%physPtrsStructPtr = OpTypePointer StorageBuffer %physPtrsStruct\n" "%indata_all = OpVariable %physPtrsStructPtr StorageBuffer\n" "%bufptrphysPtr = OpTypePointer StorageBuffer %bufptrphys\n" : "%sb_f32ptr = OpTypePointer StorageBuffer %f32\n" "%buf = OpTypeStruct %f32arr\n" "%bufptr = OpTypePointer StorageBuffer %buf\n" "%indata_a = OpVariable %bufptr StorageBuffer\n" "%indata_b = OpVariable %bufptr StorageBuffer\n" "%indata_s = OpVariable %bufptr StorageBuffer\n" "%outdata = OpVariable %bufptr StorageBuffer\n"; stringTemplate += "%id = OpVariable %uvec3ptr Input\n" "%zero = OpConstant %i32 0\n" "%one = OpConstant %i32 1\n" "%two = OpConstant %i32 2\n" "%three = OpConstant %i32 3\n" "%fzero = OpConstant %f32 0\n" "%fone = OpConstant %f32 1\n" "${ExtraTypes}" "${ExtraGlobalScopeVars}" // We're going to put the "selector" function here. // This function type is needed tests that use OpFunctionCall. "%selector_func_type = OpTypeFunction %sb_f32ptr %bool %sb_f32ptr %sb_f32ptr\n" "%choose_input_func = OpFunction %sb_f32ptr None %selector_func_type\n" "%is_neg_param = OpFunctionParameter %bool\n" "%first_ptr_param = OpFunctionParameter %sb_f32ptr\n" "%second_ptr_param = OpFunctionParameter %sb_f32ptr\n" "%selector_func_begin = OpLabel\n" "%result_ptr = OpSelect %sb_f32ptr %is_neg_param %first_ptr_param %second_ptr_param\n" "OpReturnValue %result_ptr\n" "OpFunctionEnd\n" // main function is the entry_point "%main = OpFunction %void None %voidf\n" "%label = OpLabel\n" "${ExtraFunctionScopeVars}"; if (physPtrs) { stringTemplate += "%indata_a_ptr = OpAccessChain %bufptrphysPtr %indata_all %zero\n" "%indata_a = OpLoad %bufptrphys %indata_a_ptr\n" "%indata_b_ptr = OpAccessChain %bufptrphysPtr %indata_all %one\n" "%indata_b = OpLoad %bufptrphys %indata_b_ptr\n" "%indata_s_ptr = OpAccessChain %bufptrphysPtr %indata_all %two\n" "%indata_s = OpLoad %bufptrphys %indata_s_ptr\n" "%outdata_ptr = OpAccessChain %bufptrphysPtr %indata_all %three\n" "%outdata = OpLoad %bufptrphys %outdata_ptr\n"; } stringTemplate += "%idval = OpLoad %uvec3 %id\n" "%i = OpCompositeExtract %u32 %idval 0\n" "%two_i = OpIAdd %u32 %i %i\n" "%two_i_plus_1 = OpIAdd %u32 %two_i %one\n" "%inloc_a_i = OpAccessChain %sb_f32ptr %indata_a %zero %i\n" "%inloc_b_i = OpAccessChain %sb_f32ptr %indata_b %zero %i\n" "%inloc_s_i = OpAccessChain %sb_f32ptr %indata_s %zero %i\n" "%outloc_i = OpAccessChain %sb_f32ptr %outdata %zero %i\n" "%inloc_a_2i = OpAccessChain %sb_f32ptr %indata_a %zero %two_i\n" "%inloc_a_2i_plus_1 = OpAccessChain %sb_f32ptr %indata_a %zero %two_i_plus_1\n" "%inval_s_i = OpLoad %f32 %inloc_s_i Aligned 4\n" "%is_neg = OpFOrdLessThan %bool %inval_s_i %fzero\n" "${ExtraSetupComputations}" "${ResultStrategy}" "%mux_output = OpLoad %f32 ${VarPtrName} Aligned 4\n" " OpStore %outloc_i %mux_output Aligned 4\n" " OpReturn\n" " OpFunctionEnd\n"; const StringTemplate shaderTemplate (stringTemplate); const bool singleInputBuffer[] = { true, false }; for (int inputBufferTypeIndex = 0 ; inputBufferTypeIndex < 2; ++inputBufferTypeIndex) { const bool isSingleInputBuffer = singleInputBuffer[inputBufferTypeIndex]; const string extraCap = string(physPtrs ? "OpCapability PhysicalStorageBufferAddressesEXT\n" : isSingleInputBuffer ? "OpCapability VariablePointersStorageBuffer\n" : "OpCapability VariablePointers\n"); const vector& expectedOutput = isSingleInputBuffer ? AmuxAOutputFloats : AmuxBOutputFloats; const vector& expectedIncrOutput = isSingleInputBuffer ? incrAmuxAOutputFloats : incrAmuxBOutputFloats; const string bufferType = isSingleInputBuffer ? "single_buffer" : "two_buffers"; const string muxInput1 = isSingleInputBuffer ? " %inloc_a_2i " : " %inloc_a_i "; const string muxInput2 = isSingleInputBuffer ? " %inloc_a_2i_plus_1 " : " %inloc_b_i "; // Set the proper extension features required for the test if (!physPtrs) { if (isSingleInputBuffer) requiredFeatures.extVariablePointers.variablePointersStorageBuffer = true; else requiredFeatures.extVariablePointers.variablePointers = true; } { // Variable Pointer Reads (using OpSelect) ComputeShaderSpec spec; map specs; string name = "reads_opselect_" + bufferType; specs["ExtraCapability"] = extraCap; specs["ExtraTypes"] = ""; specs["ExtraGlobalScopeVars"] = ""; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraSetupComputations"] = ""; specs["ExtraDecorations"] = ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = "%mux_output_var_ptr = OpSelect %sb_f32ptr %is_neg" + muxInput1 + muxInput2 + "\n"; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(numMuxes, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } { // Variable Pointer Reads (using OpFunctionCall) ComputeShaderSpec spec; map specs; string name = "reads_opfunctioncall_" + bufferType; specs["ExtraCapability"] = extraCap; specs["ExtraTypes"] = ""; specs["ExtraGlobalScopeVars"] = ""; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraSetupComputations"] = ""; specs["ExtraDecorations"] = ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = "%mux_output_var_ptr = OpFunctionCall %sb_f32ptr %choose_input_func %is_neg" + muxInput1 + muxInput2 + "\n"; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(numMuxes, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } { // Variable Pointer Reads (using OpPhi) ComputeShaderSpec spec; map specs; string name = "reads_opphi_" + bufferType; specs["ExtraCapability"] = extraCap; specs["ExtraTypes"] = ""; specs["ExtraGlobalScopeVars"] = ""; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraSetupComputations"] = ""; specs["ExtraDecorations"] = ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = " OpSelectionMerge %end_label None\n" " OpBranchConditional %is_neg %take_mux_input_1 %take_mux_input_2\n" "%take_mux_input_1 = OpLabel\n" " OpBranch %end_label\n" "%take_mux_input_2 = OpLabel\n" " OpBranch %end_label\n" "%end_label = OpLabel\n" "%mux_output_var_ptr = OpPhi %sb_f32ptr" + muxInput1 + "%take_mux_input_1" + muxInput2 + "%take_mux_input_2\n"; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(numMuxes, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } { // Variable Pointer Reads (using OpCopyObject) ComputeShaderSpec spec; map specs; string name = "reads_opcopyobject_" + bufferType; specs["ExtraCapability"] = extraCap; specs["ExtraTypes"] = ""; specs["ExtraGlobalScopeVars"] = ""; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraSetupComputations"] = ""; specs["ExtraDecorations"] = ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = "%mux_input_1_copy = OpCopyObject %sb_f32ptr" + muxInput1 + "\n" "%mux_input_2_copy = OpCopyObject %sb_f32ptr" + muxInput2 + "\n" "%mux_output_var_ptr = OpSelect %sb_f32ptr %is_neg %mux_input_1_copy %mux_input_2_copy\n"; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(numMuxes, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } { // Test storing into Private variables. const char* storageClasses[] = {"Private", "Function"}; for (int classId = 0; classId < 2; ++classId) { ComputeShaderSpec spec; map specs; std::string storageClass = storageClasses[classId]; std::string name = "stores_" + string(de::toLower(storageClass)) + "_" + bufferType; std::string extraVariable = "%mux_output_copy = OpVariable %sb_f32ptrptr " + storageClass + "\n"; specs["ExtraTypes"] = "%sb_f32ptrptr = OpTypePointer " + storageClass + " %sb_f32ptr\n"; specs["ExtraCapability"] = extraCap; specs["ExtraGlobalScopeVars"] = (classId == 0) ? extraVariable : ""; specs["ExtraFunctionScopeVars"] = (classId == 1) ? extraVariable : ""; specs["ExtraSetupComputations"] = ""; specs["ExtraDecorations"] = physPtrs ? "OpDecorate %mux_output_copy AliasedPointerEXT\n" : ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = "%opselect_result = OpSelect %sb_f32ptr %is_neg" + muxInput1 + muxInput2 + "\n" " OpStore %mux_output_copy %opselect_result\n" "%mux_output_var_ptr = OpLoad %sb_f32ptr %mux_output_copy Aligned 4\n"; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(numMuxes, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } } { // Variable Pointer Reads (Using OpPtrAccessChain) ComputeShaderSpec spec; map specs; std::string name = "reads_opptraccesschain_" + bufferType; std::string in_1 = isSingleInputBuffer ? " %a_2i_ptr " : " %a_i_ptr "; std::string in_2 = isSingleInputBuffer ? " %a_2i_plus_1_ptr " : " %b_i_ptr "; specs["ExtraTypes"] = ""; specs["ExtraCapability"] = extraCap; specs["ExtraGlobalScopeVars"] = ""; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraSetupComputations"] = ""; specs["ExtraDecorations"] = ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = "%a_ptr = OpAccessChain %sb_f32ptr %indata_a %zero %zero\n" "%b_ptr = OpAccessChain %sb_f32ptr %indata_b %zero %zero\n" "%s_ptr = OpAccessChain %sb_f32ptr %indata_s %zero %zero\n" "%out_ptr = OpAccessChain %sb_f32ptr %outdata %zero %zero\n" "%a_i_ptr = OpPtrAccessChain %sb_f32ptr %a_ptr %i\n" "%b_i_ptr = OpPtrAccessChain %sb_f32ptr %b_ptr %i\n" "%s_i_ptr = OpPtrAccessChain %sb_f32ptr %s_ptr %i\n" "%a_2i_ptr = OpPtrAccessChain %sb_f32ptr %a_ptr %two_i\n" "%a_2i_plus_1_ptr = OpPtrAccessChain %sb_f32ptr %a_ptr %two_i_plus_1\n" "%mux_output_var_ptr = OpSelect %sb_f32ptr %is_neg " + in_1 + in_2 + "\n"; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(numMuxes, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } { // Variable Pointer Writes ComputeShaderSpec spec; map specs; std::string name = "writes_" + bufferType; specs["ExtraCapability"] = extraCap; specs["ExtraTypes"] = ""; specs["ExtraGlobalScopeVars"] = ""; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraSetupComputations"] = ""; specs["ExtraDecorations"] = ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = "%mux_output_var_ptr = OpSelect %sb_f32ptr %is_neg" + muxInput1 + muxInput2 + "\n" + " %val = OpLoad %f32 %mux_output_var_ptr Aligned 4\n" " %val_plus_1 = OpFAdd %f32 %val %fone\n" " OpStore %mux_output_var_ptr %val_plus_1 Aligned 4\n"; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(numMuxes, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedIncrOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } // If we only have VariablePointersStorageBuffer, then the extension does not apply to Workgroup storage class. // Therefore the Workgroup tests apply to cases where the VariablePointers capability is used (when 2 input buffers are used). if (!physPtrs && !isSingleInputBuffer) { // VariablePointers on Workgroup ComputeShaderSpec spec; map specs; std::string name = "workgroup_" + bufferType; specs["ExtraCapability"] = extraCap; specs["ExtraTypes"] = "%c_i32_N = OpConstant %i32 " + inputArraySize + " \n" "%f32arr_N = OpTypeArray %f32 %c_i32_N\n" "%f32arr_wrkgrp_ptr = OpTypePointer Workgroup %f32arr_N\n" "%f32_wrkgrp_ptr = OpTypePointer Workgroup %f32\n"; specs["ExtraGlobalScopeVars"] = "%AW = OpVariable %f32arr_wrkgrp_ptr Workgroup\n" "%BW = OpVariable %f32arr_wrkgrp_ptr Workgroup\n"; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraSetupComputations"] = "%loc_AW_i = OpAccessChain %f32_wrkgrp_ptr %AW %i\n" "%loc_BW_i = OpAccessChain %f32_wrkgrp_ptr %BW %i\n" "%inval_a_i = OpLoad %f32 %inloc_a_i\n" "%inval_b_i = OpLoad %f32 %inloc_b_i\n" "%inval_a_2i = OpLoad %f32 %inloc_a_2i\n" "%inval_a_2i_plus_1 = OpLoad %f32 %inloc_a_2i_plus_1\n"; specs["ExtraDecorations"] = ""; specs["VarPtrName"] = "%output_var_ptr"; specs["ResultStrategy"] = " OpStore %loc_AW_i %inval_a_i\n" " OpStore %loc_BW_i %inval_b_i\n" "%output_var_ptr = OpSelect %f32_wrkgrp_ptr %is_neg %loc_AW_i %loc_BW_i\n"; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(numMuxes, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back (Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back (Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back (Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } } } void addVariablePointersComputeGroup(tcu::TestCaseGroup* group) { addPhysicalOrVariablePointersComputeGroup(group, false); } void addPhysicalPointersComputeGroup(tcu::TestCaseGroup* group) { addPhysicalOrVariablePointersComputeGroup(group, true); } void addComplexTypesPhysicalOrVariablePointersComputeGroup (tcu::TestCaseGroup* group, bool physPtrs) { tcu::TestContext& testCtx = group->getTestContext(); const int numFloats = 64; vector inputA (numFloats, 0); vector inputB (numFloats, 0); vector inputC (2*numFloats, 0); vector expectedOutput (1, 0); VulkanFeatures requiredFeatures; // These tests exercise variable pointers into various levels of the following data-structures. // // struct struct inner_struct { // vec4 x[2]; // array of 2 vectors. Each vector is 4 floats. // vec4 y[2]; // array of 2 vectors. Each vector is 4 floats. // }; // // struct outer_struct { // inner_struct r[2][2]; // }; // // struct input_buffer { // outer_struct a; // outer_struct b; // } // // inputA is of type outer_struct. // inputB is of type outer_struct. // inputC is of type input_buffer. // // inputA and inputB are of the same size. When testing variable pointers pointing to // two different input buffers, we use inputA and inputB. // // inputC is twice the size of inputA. When testing the VariablePointersStorageBuffer capability, // the variable pointer must be confined to a single buffer. These tests will use inputC. // // The inner_struct contains 16 floats. // The outer_struct contains 64 floats. // The input_buffer contains 128 floats. // Populate the first input (inputA) to contain: {0, 4, ... , 252} // Populate the second input (inputB) to contain: {3, 7, ... , 255} // Populate the third input (inputC) to contain: {0, 4, ... , 252, 3, 7, ... , 255} // Note that the first half of inputC is the same as inputA and the second half is the same as inputB. for (size_t i = 0; i < numFloats; ++i) { inputA[i] = 4*float(i) / 255; inputB[i] = ((4*float(i)) + 3) / 255; inputC[i] = inputA[i]; inputC[i + numFloats] = inputB[i]; } // In the following tests we use variable pointers to point to different types: // nested structures, matrices of structures, arrays of structures, arrays of vectors, vectors of scalars, and scalars. // outer_structure.inner_structure[?][?].x[?][?]; // ^ ^ ^ ^ ^ ^ ^ // // For tests with 2 input buffers: // 1. inputA or inputB = nested structure // 2. inputA.r or inputB.r = matrices of structures // 3. inputA.r[?] or inputB.r[?] = arrays of structures // 4. inputA.r[?][?] or inputB.r[?][?] = structures // 5. inputA.r[?][?].(x|y) or inputB.r[?][?].(x|y) = arrays of vectors // 6. inputA.r[?][?].(x|y)[?] or inputB.r[?][?].(x|y)[?] = vectors of scalars // 7. inputA.r[?][?].(x|y)[?][?] or inputB.r[?][?].(x|y)[?][?] = scalars // For tests with 1 input buffer: // 1. inputC.a or inputC.b = nested structure // 2. inputC.a.r or inputC.b.r = matrices of structures // 3. inputC.a.r[?] or inputC.b.r[?] = arrays of structures // 4. inputC.a.r[?][?] or inputC.b.r[?][?] = structures // 5. inputC.a.r[?][?].(x|y) or inputC.b.r[?][?].(x|y) = arrays of vectors // 6. inputC.a.r[?][?].(x|y)[?] or inputC.b.r[?][?].(x|y)[?] = vectors of scalars // 7. inputC.a.r[?][?].(x|y)[?][?] or inputC.b.r[?][?].(x|y)[?][?] = scalars const int numLevels = 7; // Decorations string commonDecorations = physPtrs ? "OpDecorate %physPtrsStruct Block\n" "OpMemberDecorate %physPtrsStruct 0 Offset 0\n" "OpMemberDecorate %physPtrsStruct 1 Offset 8\n" "OpMemberDecorate %physPtrsStruct 2 Offset 16\n" "OpMemberDecorate %physPtrsStruct 3 Offset 24\n" "OpDecorate %indata_all DescriptorSet 0\n" "OpDecorate %indata_all Binding 0\n" "OpDecorate %first_param Restrict\n" "OpDecorate %second_param Restrict\n" : "OpDecorate %outdata DescriptorSet 0 \n" "OpDecorate %outdata Binding 3 \n"; commonDecorations += "OpDecorate %id BuiltIn GlobalInvocationId \n" // Set the Block decoration "OpDecorate %output_buffer Block \n" // Set the Offsets "OpMemberDecorate %output_buffer 0 Offset 0 \n" "OpMemberDecorate %input_buffer 0 Offset 0 \n" "OpMemberDecorate %input_buffer 1 Offset 256 \n" "OpMemberDecorate %outer_struct 0 Offset 0 \n" "OpMemberDecorate %inner_struct 0 Offset 0 \n" "OpMemberDecorate %inner_struct 1 Offset 32 \n" // Set the ArrayStrides "OpDecorate %arr2_v4float ArrayStride 16 \n" "OpDecorate %arr2_inner_struct ArrayStride 64 \n" "OpDecorate %mat2x2_inner_struct ArrayStride 128 \n" "OpDecorate %outer_struct_ptr ArrayStride 256 \n" "OpDecorate %v4f32_ptr ArrayStride 16 \n"; string inputABDecorations = physPtrs ? "" : "OpDecorate %inputA DescriptorSet 0 \n" "OpDecorate %inputB DescriptorSet 0 \n" "OpDecorate %inputA Binding 0 \n" "OpDecorate %inputB Binding 1 \n"; // inputA and inputB have type outer_struct so it needs Block inputABDecorations += "OpDecorate %outer_struct Block \n"; string inputCDecorations = physPtrs ? "" : "OpDecorate %inputC DescriptorSet 0 \n" "OpDecorate %inputC Binding 2 \n"; inputCDecorations += physPtrs ? "" : // inputC has type input_buffer so it needs Block "OpDecorate %input_buffer Block \n"; string types = /////////////// // CONSTANTS // /////////////// "%c_bool_true = OpConstantTrue %bool \n" "%c_bool_false = OpConstantFalse %bool \n" "%c_i32_0 = OpConstant %i32 0 \n" "%c_i32_1 = OpConstant %i32 1 \n" "%c_i32_2 = OpConstant %i32 2 \n" "%c_i32_3 = OpConstant %i32 3 \n" "%c_u32_2 = OpConstant %u32 2 \n" /////////// // TYPES // /////////// "%v4f32 = OpTypeVector %f32 4 \n" // struct struct inner_struct { // vec4 x[2]; // array of 2 vectors // vec4 y[2]; // array of 2 vectors // }; "%arr2_v4float = OpTypeArray %v4f32 %c_u32_2 \n" "%inner_struct = OpTypeStruct %arr2_v4float %arr2_v4float \n" // struct outer_struct { // inner_struct r[2][2]; // }; "%arr2_inner_struct = OpTypeArray %inner_struct %c_u32_2 \n" "%mat2x2_inner_struct = OpTypeArray %arr2_inner_struct %c_u32_2 \n" "%outer_struct = OpTypeStruct %mat2x2_inner_struct \n" // struct input_buffer { // outer_struct a; // outer_struct b; // } "%input_buffer = OpTypeStruct %outer_struct %outer_struct \n" // struct output_struct { // float out; // } "%output_buffer = OpTypeStruct %f32 \n"; /////////////////// // POINTER TYPES // /////////////////// types += physPtrs ? "%output_buffer_ptr = OpTypePointer PhysicalStorageBufferEXT %output_buffer \n" "%input_buffer_ptr = OpTypePointer PhysicalStorageBufferEXT %input_buffer \n" "%outer_struct_ptr = OpTypePointer PhysicalStorageBufferEXT %outer_struct \n" "%mat2x2_ptr = OpTypePointer PhysicalStorageBufferEXT %mat2x2_inner_struct \n" "%arr2_ptr = OpTypePointer PhysicalStorageBufferEXT %arr2_inner_struct \n" "%inner_struct_ptr = OpTypePointer PhysicalStorageBufferEXT %inner_struct \n" "%arr_v4f32_ptr = OpTypePointer PhysicalStorageBufferEXT %arr2_v4float \n" "%v4f32_ptr = OpTypePointer PhysicalStorageBufferEXT %v4f32 \n" "%sb_f32ptr = OpTypePointer PhysicalStorageBufferEXT %f32 \n" "%physPtrsStruct = OpTypeStruct %outer_struct_ptr %outer_struct_ptr %input_buffer_ptr %output_buffer_ptr\n" "%physPtrsStructPtr = OpTypePointer StorageBuffer %physPtrsStruct\n" "%outer_struct_ptr_ptr = OpTypePointer StorageBuffer %outer_struct_ptr\n" "%input_buffer_ptr_ptr = OpTypePointer StorageBuffer %input_buffer_ptr\n" "%output_buffer_ptr_ptr = OpTypePointer StorageBuffer %output_buffer_ptr\n" : "%output_buffer_ptr = OpTypePointer StorageBuffer %output_buffer \n" "%input_buffer_ptr = OpTypePointer StorageBuffer %input_buffer \n" "%outer_struct_ptr = OpTypePointer StorageBuffer %outer_struct \n" "%mat2x2_ptr = OpTypePointer StorageBuffer %mat2x2_inner_struct \n" "%arr2_ptr = OpTypePointer StorageBuffer %arr2_inner_struct \n" "%inner_struct_ptr = OpTypePointer StorageBuffer %inner_struct \n" "%arr_v4f32_ptr = OpTypePointer StorageBuffer %arr2_v4float \n" "%v4f32_ptr = OpTypePointer StorageBuffer %v4f32 \n" "%sb_f32ptr = OpTypePointer StorageBuffer %f32 \n"; types += "${extra_types}\n"; /////////////// // VARIABLES // /////////////// types += physPtrs ? "%id = OpVariable %uvec3ptr Input \n" "%indata_all = OpVariable %physPtrsStructPtr StorageBuffer\n" : "%id = OpVariable %uvec3ptr Input \n" "%outdata = OpVariable %output_buffer_ptr StorageBuffer \n"; string inputABVariables = physPtrs ? "" : "%inputA = OpVariable %outer_struct_ptr StorageBuffer \n" "%inputB = OpVariable %outer_struct_ptr StorageBuffer \n"; string inputCVariables = physPtrs ? "" : "%inputC = OpVariable %input_buffer_ptr StorageBuffer \n"; const string inputCIntermediates ( // Here are the 2 nested structures within InputC. "%inputC_a = OpAccessChain %outer_struct_ptr %inputC %c_i32_0\n" "%inputC_b = OpAccessChain %outer_struct_ptr %inputC %c_i32_1\n" ); std::string stringTemplate = "OpCapability Shader\n" "${extra_capability}\n"; stringTemplate += physPtrs ? "OpExtension \"SPV_KHR_physical_storage_buffer\"\n" : "OpExtension \"SPV_KHR_variable_pointers\"\n"; stringTemplate += "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" "OpMemoryModel " + string(physPtrs ? "PhysicalStorageBuffer64EXT" : "Logical") + " GLSL450\n" "OpEntryPoint GLCompute %main \"main\" %id\n" "OpExecutionMode %main LocalSize 1 1 1\n" "OpSource GLSL 430\n" "OpName %main \"main\"\n" "OpName %id \"gl_GlobalInvocationID\"\n" + commonDecorations + "${input_decorations}\n" + string(getComputeAsmCommonTypes()) + types + "${input_variables}\n" // These selector functions return variable pointers. // These functions are used by tests that use OpFunctionCall to obtain the variable pointer "%selector_func_type = OpTypeFunction ${selected_type} %bool ${selected_type} ${selected_type}\n" "%choose_input_func = OpFunction ${selected_type} None %selector_func_type\n" "%choose_first_param = OpFunctionParameter %bool\n" "%first_param = OpFunctionParameter ${selected_type}\n" "%second_param = OpFunctionParameter ${selected_type}\n" "%selector_func_begin = OpLabel\n" "%result_ptr = OpSelect ${selected_type} %choose_first_param %first_param %second_param\n" "OpReturnValue %result_ptr\n" "OpFunctionEnd\n" // main function is the entry_point "%main = OpFunction %void None %voidf\n" "%label = OpLabel\n"; if (physPtrs) { stringTemplate += "%inputA_ptr = OpAccessChain %outer_struct_ptr_ptr %indata_all %c_i32_0\n" "%inputA = OpLoad %outer_struct_ptr %inputA_ptr\n" "%inputB_ptr = OpAccessChain %outer_struct_ptr_ptr %indata_all %c_i32_1\n" "%inputB = OpLoad %outer_struct_ptr %inputB_ptr\n" "%inputC_ptr = OpAccessChain %input_buffer_ptr_ptr %indata_all %c_i32_2\n" "%inputC = OpLoad %input_buffer_ptr %inputC_ptr\n" "%outdata_ptr = OpAccessChain %output_buffer_ptr_ptr %indata_all %c_i32_3\n" "%outdata = OpLoad %output_buffer_ptr %outdata_ptr\n"; } stringTemplate += "${input_intermediates}\n" // Define the 2 pointers from which we're going to choose one. "${a_loc} \n" "${b_loc} \n" // Choose between the 2 pointers / variable pointers. "${selection_strategy} \n" // OpAccessChain into the variable pointer until you get to the float. "%result_loc = OpAccessChain %sb_f32ptr %var_ptr ${remaining_indexes} \n" // Now load from the result_loc "%result_val = OpLoad %f32 %result_loc Aligned 4\n" // Store the chosen value to the output buffer. "%outdata_loc = OpAccessChain %sb_f32ptr %outdata %c_i32_0\n" " OpStore %outdata_loc %result_val Aligned 4\n" " OpReturn\n" " OpFunctionEnd\n"; const StringTemplate shaderTemplate (stringTemplate); for (int isSingleInputBuffer = 0 ; isSingleInputBuffer < 2; ++isSingleInputBuffer) { // Set the proper extension features required for the test if (!physPtrs) { if (isSingleInputBuffer) requiredFeatures.extVariablePointers.variablePointersStorageBuffer = true; else requiredFeatures.extVariablePointers.variablePointers = true; } for (int selectInputA = 0; selectInputA < 2; ++selectInputA) { const string extraCap = string(physPtrs ? "OpCapability PhysicalStorageBufferAddressesEXT\n" : isSingleInputBuffer ? "OpCapability VariablePointersStorageBuffer\n" : "OpCapability VariablePointers\n"); const string inputDecorations = isSingleInputBuffer ? inputCDecorations : inputABDecorations; const string inputVariables = isSingleInputBuffer ? inputCVariables : inputABVariables; const string inputIntermediates = isSingleInputBuffer ? inputCIntermediates : ""; const vector& selectedInput = isSingleInputBuffer ? inputC : (selectInputA ? inputA : inputB); const string bufferType = isSingleInputBuffer ? "single_buffer_" : "two_buffers_"; const string baseA = isSingleInputBuffer ? "%inputC_a" : "%inputA"; const string baseB = isSingleInputBuffer ? "%inputC_b" : "%inputB"; const string selectedInputStr = selectInputA ? "first_input" : "second_input"; const string spirvSelectInputA = selectInputA ? "%c_bool_true" : "%c_bool_false"; const int outerStructIndex = isSingleInputBuffer ? (selectInputA ? 0 : 1) : 0; // The indexes chosen at each level. At any level, any given offset is exercised. // outerStructIndex is 0 for inputA and inputB (because outer_struct has only 1 member). // outerStructIndex is 0 for member of inputC and 1 for member . const int indexesForLevel[numLevels][6]= {{outerStructIndex, 0, 0, 0, 0, 1}, {outerStructIndex, 1, 0, 1, 0, 2}, {outerStructIndex, 0, 1, 0, 1, 3}, {outerStructIndex, 1, 1, 1, 0, 0}, {outerStructIndex, 0, 0, 1, 1, 1}, {outerStructIndex, 1, 0, 0, 0, 2}, {outerStructIndex, 1, 1, 1, 1, 3}}; const string indexLevelNames[] = {"_outer_struct_", "_matrices_", "_arrays_", "_inner_structs_", "_vec4arr_", "_vec4_", "_float_"}; const string inputALocations[] = { "", "%a_loc = OpAccessChain %mat2x2_ptr " + baseA + " %c_i32_0", "%a_loc = OpAccessChain %arr2_ptr " + baseA + " %c_i32_0 %c_i32_0", "%a_loc = OpAccessChain %inner_struct_ptr " + baseA + " %c_i32_0 %c_i32_1 %c_i32_1", "%a_loc = OpAccessChain %arr_v4f32_ptr " + baseA + " %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%a_loc = OpAccessChain %v4f32_ptr " + baseA + " %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_0 %c_i32_0", "%a_loc = OpAccessChain %sb_f32ptr " + baseA + " %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_3"}; const string inputBLocations[] = { "", "%b_loc = OpAccessChain %mat2x2_ptr " + baseB + " %c_i32_0", "%b_loc = OpAccessChain %arr2_ptr " + baseB + " %c_i32_0 %c_i32_0", "%b_loc = OpAccessChain %inner_struct_ptr " + baseB + " %c_i32_0 %c_i32_1 %c_i32_1", "%b_loc = OpAccessChain %arr_v4f32_ptr " + baseB + " %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%b_loc = OpAccessChain %v4f32_ptr " + baseB + " %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_0 %c_i32_0", "%b_loc = OpAccessChain %sb_f32ptr " + baseB + " %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_3"}; const string inputAPtrAccessChain[] = { "", "%a_loc = OpPtrAccessChain %mat2x2_ptr " + baseA + " %c_i32_0 %c_i32_0", "%a_loc = OpPtrAccessChain %arr2_ptr " + baseA + " %c_i32_0 %c_i32_0 %c_i32_0", "%a_loc = OpPtrAccessChain %inner_struct_ptr " + baseA + " %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_1", "%a_loc = OpPtrAccessChain %arr_v4f32_ptr " + baseA + " %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%a_loc = OpPtrAccessChain %v4f32_ptr " + baseA + " %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_0 %c_i32_0", // Next case emulates: // %a_loc = OpPtrAccessChain %sb_f32ptr baseA %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_3 // But rewrite it to exercise OpPtrAccessChain with a non-zero first index: // %a_loc_arr is a pointer to an array that we want to index with 1. // But instead of just using OpAccessChain with first index 1, use OpAccessChain with index 0 to // get a pointer to the first element, then send that into OpPtrAccessChain with index 1. "%a_loc_arr = OpPtrAccessChain %arr_v4f32_ptr " + baseA + " %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 " "%a_loc_first_elem = OpAccessChain %v4f32_ptr %a_loc_arr %c_i32_0 " "%a_loc = OpPtrAccessChain %sb_f32ptr %a_loc_first_elem %c_i32_1 %c_i32_3"}; const string inputBPtrAccessChain[] = { "", "%b_loc = OpPtrAccessChain %mat2x2_ptr " + baseB + " %c_i32_0 %c_i32_0", "%b_loc = OpPtrAccessChain %arr2_ptr " + baseB + " %c_i32_0 %c_i32_0 %c_i32_0", "%b_loc = OpPtrAccessChain %inner_struct_ptr " + baseB + " %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_1", "%b_loc = OpPtrAccessChain %arr_v4f32_ptr " + baseB + " %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%b_loc = OpPtrAccessChain %v4f32_ptr " + baseB + " %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_0 %c_i32_0", // Next case emulates: // %b_loc = OpPtrAccessChain %sb_f32ptr basseB %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_3 // But rewrite it to exercise OpPtrAccessChain with a non-zero first index: // %b_loc_arr is a pointer to an array that we want to index with 1. // But instead of just using OpAccessChain with first index 1, use OpAccessChain with index 0 to // get a pointer to the first element, then send that into OpPtrAccessChain with index 1. "%b_loc_arr = OpPtrAccessChain %arr_v4f32_ptr " + baseB + " %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 " "%b_loc_first_elem = OpAccessChain %v4f32_ptr %b_loc_arr %c_i32_0 " "%b_loc = OpPtrAccessChain %sb_f32ptr %b_loc_first_elem %c_i32_1 %c_i32_3"}; const string remainingIndexesAtLevel[]= {"%c_i32_0 %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%c_i32_1 %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_2", "%c_i32_1 %c_i32_0 %c_i32_1 %c_i32_3", "%c_i32_1 %c_i32_0 %c_i32_0", "%c_i32_1 %c_i32_1", "%c_i32_2", ""}; const string pointerTypeAtLevel[] = {"%outer_struct_ptr", "%mat2x2_ptr", "%arr2_ptr", "%inner_struct_ptr", "%arr_v4f32_ptr", "%v4f32_ptr", "%sb_f32ptr"}; const string baseANameAtLevel[] = {baseA, "%a_loc", "%a_loc", "%a_loc", "%a_loc", "%a_loc", "%a_loc"}; const string baseBNameAtLevel[] = {baseB, "%b_loc", "%b_loc", "%b_loc", "%b_loc", "%b_loc", "%b_loc"}; for (int indexLevel = 0; indexLevel < numLevels; ++indexLevel) { const int baseOffset = getBaseOffsetForSingleInputBuffer(indexesForLevel[indexLevel][0], indexesForLevel[indexLevel][1], indexesForLevel[indexLevel][2], indexesForLevel[indexLevel][3], indexesForLevel[indexLevel][4], indexesForLevel[indexLevel][5]); // Use OpSelect to choose between 2 pointers { ComputeShaderSpec spec; map specs; string opCodeForTests = "opselect"; string name = opCodeForTests + indexLevelNames[indexLevel] + bufferType + selectedInputStr; specs["extra_types"] = ""; specs["extra_capability"] = extraCap; specs["input_decorations"] = inputDecorations; specs["input_variables"] = inputVariables; specs["input_intermediates"] = inputIntermediates; specs["selected_type"] = pointerTypeAtLevel[indexLevel]; specs["select_inputA"] = spirvSelectInputA; specs["a_loc"] = inputALocations[indexLevel]; specs["b_loc"] = inputBLocations[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = "%var_ptr = OpSelect " + pointerTypeAtLevel[indexLevel] + " " + spirvSelectInputA + " " + baseANameAtLevel[indexLevel] + " " + baseBNameAtLevel[indexLevel] + "\n"; expectedOutput[0] = selectedInput[baseOffset]; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(1, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputA)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputB)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputC)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } // Use OpFunctionCall to choose between 2 pointers { ComputeShaderSpec spec; map specs; string opCodeForTests = "opfunctioncall"; string name = opCodeForTests + indexLevelNames[indexLevel] + bufferType + selectedInputStr; specs["extra_types"] = ""; specs["extra_capability"] = extraCap; specs["input_decorations"] = inputDecorations; specs["input_variables"] = inputVariables; specs["input_intermediates"] = inputIntermediates; specs["selected_type"] = pointerTypeAtLevel[indexLevel]; specs["select_inputA"] = spirvSelectInputA; specs["a_loc"] = inputALocations[indexLevel]; specs["b_loc"] = inputBLocations[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = "%var_ptr = OpFunctionCall " + pointerTypeAtLevel[indexLevel] + " %choose_input_func " + spirvSelectInputA + " " + baseANameAtLevel[indexLevel] + " " + baseBNameAtLevel[indexLevel] + "\n"; expectedOutput[0] = selectedInput[baseOffset]; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(1, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputA)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputB)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputC)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } // Use OpPhi to choose between 2 pointers { ComputeShaderSpec spec; map specs; string opCodeForTests = "opphi"; string name = opCodeForTests + indexLevelNames[indexLevel] + bufferType + selectedInputStr; specs["extra_types"] = ""; specs["extra_capability"] = extraCap; specs["input_decorations"] = inputDecorations; specs["input_variables"] = inputVariables; specs["input_intermediates"] = inputIntermediates; specs["selected_type"] = pointerTypeAtLevel[indexLevel]; specs["select_inputA"] = spirvSelectInputA; specs["a_loc"] = inputALocations[indexLevel]; specs["b_loc"] = inputBLocations[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = " OpSelectionMerge %end_label None\n" " OpBranchConditional " + spirvSelectInputA + " %take_input_a %take_input_b\n" "%take_input_a = OpLabel\n" " OpBranch %end_label\n" "%take_input_b = OpLabel\n" " OpBranch %end_label\n" "%end_label = OpLabel\n" "%var_ptr = OpPhi " + pointerTypeAtLevel[indexLevel] + " " + baseANameAtLevel[indexLevel] + " %take_input_a " + baseBNameAtLevel[indexLevel] + " %take_input_b\n"; expectedOutput[0] = selectedInput[baseOffset]; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(1, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputA)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputB)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputC)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } // Use OpCopyObject to get variable pointers { ComputeShaderSpec spec; map specs; string opCodeForTests = "opcopyobject"; string name = opCodeForTests + indexLevelNames[indexLevel] + bufferType + selectedInputStr; specs["extra_types"] = ""; specs["extra_capability"] = extraCap; specs["input_decorations"] = inputDecorations; specs["input_variables"] = inputVariables; specs["input_intermediates"] = inputIntermediates; specs["selected_type"] = pointerTypeAtLevel[indexLevel]; specs["select_inputA"] = spirvSelectInputA; specs["a_loc"] = inputALocations[indexLevel]; specs["b_loc"] = inputBLocations[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = "%in_a_copy = OpCopyObject " + pointerTypeAtLevel[indexLevel] + " " + baseANameAtLevel[indexLevel] + "\n" "%in_b_copy = OpCopyObject " + pointerTypeAtLevel[indexLevel] + " " + baseBNameAtLevel[indexLevel] + "\n" "%var_ptr = OpSelect " + pointerTypeAtLevel[indexLevel] + " " + spirvSelectInputA + " %in_a_copy %in_b_copy\n"; expectedOutput[0] = selectedInput[baseOffset]; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(1, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputA)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputB)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputC)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } // Use OpPtrAccessChain to get variable pointers { ComputeShaderSpec spec; map specs; string opCodeForTests = "opptraccesschain"; string name = opCodeForTests + indexLevelNames[indexLevel] + bufferType + selectedInputStr; specs["extra_types"] = ""; specs["extra_capability"] = extraCap; specs["input_decorations"] = inputDecorations; specs["input_variables"] = inputVariables; specs["input_intermediates"] = inputIntermediates; specs["selected_type"] = pointerTypeAtLevel[indexLevel]; specs["select_inputA"] = spirvSelectInputA; specs["a_loc"] = inputAPtrAccessChain[indexLevel]; specs["b_loc"] = inputBPtrAccessChain[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = "%var_ptr = OpSelect " + pointerTypeAtLevel[indexLevel] + " " + spirvSelectInputA + " " + baseANameAtLevel[indexLevel] + " " + baseBNameAtLevel[indexLevel] + "\n"; expectedOutput[0] = selectedInput[baseOffset]; spec.usesPhysStorageBuffer = physPtrs; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(1, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputA)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputB)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputC)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); if (!physPtrs) spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } } } } } void addComplexTypesVariablePointersComputeGroup (tcu::TestCaseGroup* group) { addComplexTypesPhysicalOrVariablePointersComputeGroup(group, false); } void addComplexTypesPhysicalPointersComputeGroup (tcu::TestCaseGroup* group) { addComplexTypesPhysicalOrVariablePointersComputeGroup(group, true); } void addNullptrVariablePointersComputeGroup (tcu::TestCaseGroup* group) { tcu::TestContext& testCtx = group->getTestContext(); float someFloat = 78; vector input (1, someFloat); vector expectedOutput (1, someFloat); VulkanFeatures requiredFeatures; // Requires the variable pointers feature. requiredFeatures.extVariablePointers.variablePointers = true; const string decorations ( // Decorations "OpDecorate %id BuiltIn GlobalInvocationId \n" "OpDecorate %input DescriptorSet 0 \n" "OpDecorate %outdata DescriptorSet 0 \n" "OpDecorate %input Binding 0 \n" "OpDecorate %outdata Binding 1 \n" // Set the Block decoration "OpDecorate %float_struct Block \n" // Set the Offsets "OpMemberDecorate %float_struct 0 Offset 0 \n" ); const string types ( /////////// // TYPES // /////////// // struct float_struct { // float x; // }; "%float_struct = OpTypeStruct %f32 \n" /////////////////// // POINTER TYPES // /////////////////// "%float_struct_ptr = OpTypePointer StorageBuffer %float_struct \n" "%sb_f32ptr = OpTypePointer StorageBuffer %f32 \n" "%func_f32ptrptr = OpTypePointer Function %sb_f32ptr \n" /////////////// // CONSTANTS // /////////////// "%c_bool_true = OpConstantTrue %bool \n" "%c_i32_0 = OpConstant %i32 0 \n" "%c_null_ptr = OpConstantNull %sb_f32ptr \n" /////////////// // VARIABLES // /////////////// "%id = OpVariable %uvec3ptr Input \n" "%input = OpVariable %float_struct_ptr StorageBuffer \n" "%outdata = OpVariable %float_struct_ptr StorageBuffer \n" ); const StringTemplate shaderTemplate ( "OpCapability Shader\n" "OpCapability VariablePointers\n" "OpExtension \"SPV_KHR_variable_pointers\"\n" "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n" "OpMemoryModel Logical GLSL450\n" "OpEntryPoint GLCompute %main \"main\" %id\n" "OpExecutionMode %main LocalSize 1 1 1\n" "OpSource GLSL 430\n" "OpName %main \"main\"\n" "OpName %id \"gl_GlobalInvocationID\"\n" + decorations + string(getComputeAsmCommonTypes()) + types + // main function is the entry_point "%main = OpFunction %void None %voidf\n" "%label = OpLabel\n" // Note that the Variable Pointers extension allows creation // of a pointer variable with storage class of Private or Function. "%f32_ptr_var = OpVariable %func_f32ptrptr Function %c_null_ptr\n" "%input_loc = OpAccessChain %sb_f32ptr %input %c_i32_0\n" "%output_loc = OpAccessChain %sb_f32ptr %outdata %c_i32_0\n" "${NullptrTestingStrategy}\n" " OpReturn\n" " OpFunctionEnd\n"); // f32_ptr_var has been inintialized to NULL. // Now set it to point to the float variable that holds the input value { ComputeShaderSpec spec; map specs; string name = "opvariable_initialized_null"; specs["NullptrTestingStrategy"] = " OpStore %f32_ptr_var %input_loc \n" "%loaded_f32_ptr = OpLoad %sb_f32ptr %f32_ptr_var \n" "%loaded_f32 = OpLoad %f32 %loaded_f32_ptr \n" " OpStore %output_loc %loaded_f32 \n"; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(1, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(input)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } // Use OpSelect to choose between nullptr and valid pointer. Since we can't dereference nullptr, // it is forced to always choose the valid pointer. { ComputeShaderSpec spec; map specs; string name = "opselect_null_or_valid_ptr"; specs["NullptrTestingStrategy"] = "%selected_ptr = OpSelect %sb_f32ptr %c_bool_true %input_loc %c_null_ptr\n" "%loaded_var = OpLoad %f32 %selected_ptr\n" "OpStore %output_loc %loaded_var\n"; spec.assembly = shaderTemplate.specialize(specs); spec.numWorkGroups = IVec3(1, 1, 1); spec.requestedVulkanFeatures = requiredFeatures; spec.inputs.push_back(Resource(BufferSp(new Float32Buffer(input)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); spec.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)))); spec.extensions.push_back("VK_KHR_variable_pointers"); group->addChild(new SpvAsmComputeShaderCase(testCtx, name.c_str(), spec)); } } void addDynamicOffsetComputeGroup (tcu::TestCaseGroup* group) { #ifndef CTS_USES_VULKANSC tcu::TestContext &testCtx = group->getTestContext(); static const char dataDir[] = "spirv_assembly/instruction/compute/variable_pointer/dynamic_offset"; struct Case { string name; }; static const Case cases[] = { // Test accessing a descriptor array using a variable pointer from OpSelect { "select_descriptor_array"}, }; for (const auto& testCase : cases) { const string fileName = testCase.name + ".amber"; group->addChild(cts_amber::createAmberTestCase(testCtx, testCase.name.c_str(), dataDir, fileName, {"VK_KHR_variable_pointers", "VK_KHR_storage_buffer_storage_class", "VariablePointerFeatures.variablePointers", "VariablePointerFeatures.variablePointersStorageBuffer"})); } #else DE_UNREF(group); #endif // CTS_USES_VULKANSC } void addVariablePointersGraphicsGroup (tcu::TestCaseGroup* testGroup) { tcu::TestContext& testCtx = testGroup->getTestContext(); de::Random rnd (deStringHash(testGroup->getName())); map fragments; RGBA defaultColors[4]; vector extensions; const int seed = testCtx.getCommandLine().getBaseSeed(); const int numMuxes = 100; const std::string numMuxesStr = "100"; vector inputAFloats (2*numMuxes, 0); vector inputBFloats (2*numMuxes, 0); vector inputSFloats (numMuxes, 0); vector AmuxAOutputFloats (numMuxes, 0); vector AmuxBOutputFloats (numMuxes, 0); vector incrAmuxAOutputFloats (numMuxes, 0); vector incrAmuxBOutputFloats (numMuxes, 0); VulkanFeatures requiredFeatures; extensions.push_back("VK_KHR_variable_pointers"); getDefaultColors(defaultColors); // Each output entry is chosen as follows: ( 0 <= i < numMuxes) // 1) For tests with one input buffer: output[i] = (s[i] < 0) ? A[2*i] : A[2*i+1]; // 2) For tests with two input buffers: output[i] = (s[i] < 0) ? A[i] : B[i]; fillRandomScalars(rnd, -100.f, 100.f, &inputAFloats[0], 2*numMuxes); fillRandomScalars(rnd, -100.f, 100.f, &inputBFloats[0], 2*numMuxes); // We want to guarantee that the S input has some positive and some negative values. // We choose random negative numbers for the first half, random positive numbers for the second half, and then shuffle. fillRandomScalars(rnd, -100.f, -1.f , &inputSFloats[0], numMuxes / 2); fillRandomScalars(rnd, 1.f , 100.f, &inputSFloats[numMuxes / 2], numMuxes / 2); de::Random(seed).shuffle(inputSFloats.begin(), inputSFloats.end()); for (size_t i = 0; i < numMuxes; ++i) { AmuxAOutputFloats[i] = (inputSFloats[i] < 0) ? inputAFloats[2*i] : inputAFloats[2*i+1]; AmuxBOutputFloats[i] = (inputSFloats[i] < 0) ? inputAFloats[i] : inputBFloats[i]; incrAmuxAOutputFloats[i] = (inputSFloats[i] < 0) ? 1 + inputAFloats[2*i] : 1 + inputAFloats[2*i+1]; incrAmuxBOutputFloats[i] = (inputSFloats[i] < 0) ? 1 + inputAFloats[i] : 1 + inputBFloats[i]; } fragments["extension"] = "OpExtension \"SPV_KHR_variable_pointers\"\n" "OpExtension \"SPV_KHR_storage_buffer_storage_class\"\n"; const StringTemplate preMain ( "%c_i32_limit = OpConstant %i32 " + numMuxesStr + "\n" " %sb_f32 = OpTypePointer StorageBuffer %f32\n" " %ra_f32 = OpTypeRuntimeArray %f32\n" " %buf = OpTypeStruct %ra_f32\n" " %sb_buf = OpTypePointer StorageBuffer %buf\n" " ${ExtraTypes}" " ${ExtraGlobalScopeVars}" " %indata_a = OpVariable %sb_buf StorageBuffer\n" " %indata_b = OpVariable %sb_buf StorageBuffer\n" " %indata_s = OpVariable %sb_buf StorageBuffer\n" " %outdata = OpVariable %sb_buf StorageBuffer\n" " ${ExtraFunctions} "); const std::string selectorFunction ( // We're going to put the "selector" function here. // This function type is needed for tests that use OpFunctionCall. "%selector_func_type = OpTypeFunction %sb_f32 %bool %sb_f32 %sb_f32\n" "%choose_input_func = OpFunction %sb_f32 None %selector_func_type\n" "%is_neg_param = OpFunctionParameter %bool\n" "%first_ptr_param = OpFunctionParameter %sb_f32\n" "%second_ptr_param = OpFunctionParameter %sb_f32\n" "%selector_func_begin = OpLabel\n" "%result_ptr = OpSelect %sb_f32 %is_neg_param %first_ptr_param %second_ptr_param\n" "OpReturnValue %result_ptr\n" "OpFunctionEnd\n"); const StringTemplate decoration ( "OpMemberDecorate %buf 0 Offset 0\n" "OpDecorate %buf Block\n" "OpDecorate %ra_f32 ArrayStride 4\n" "OpDecorate %sb_f32 ArrayStride 4\n" "OpDecorate %indata_a DescriptorSet 0\n" "OpDecorate %indata_b DescriptorSet 0\n" "OpDecorate %indata_s DescriptorSet 0\n" "OpDecorate %outdata DescriptorSet 0\n" "OpDecorate %indata_a Binding 0\n" "OpDecorate %indata_b Binding 1\n" "OpDecorate %indata_s Binding 2\n" "OpDecorate %outdata Binding 3\n"); const StringTemplate testFunction ( "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n" "%param = OpFunctionParameter %v4f32\n" "%entry = OpLabel\n" "${ExtraFunctionScopeVars}" "%i = OpVariable %fp_i32 Function\n" "%should_run = OpFunctionCall %bool %isUniqueIdZero\n" " OpSelectionMerge %end_if None\n" " OpBranchConditional %should_run %run_test %end_if\n" "%run_test = OpLabel\n" " OpStore %i %c_i32_0\n" " OpBranch %loop\n" // loop header "%loop = OpLabel\n" "%15 = OpLoad %i32 %i\n" "%lt = OpSLessThan %bool %15 %c_i32_limit\n" " OpLoopMerge %merge %inc None\n" " OpBranchConditional %lt %write %merge\n" // loop body "%write = OpLabel\n" "%30 = OpLoad %i32 %i\n" "%two_i = OpIAdd %i32 %30 %30\n" "%two_i_plus_1 = OpIAdd %i32 %two_i %c_i32_1\n" "%loc_s_i = OpAccessChain %sb_f32 %indata_s %c_i32_0 %30\n" "%loc_a_i = OpAccessChain %sb_f32 %indata_a %c_i32_0 %30\n" "%loc_b_i = OpAccessChain %sb_f32 %indata_b %c_i32_0 %30\n" "%loc_a_2i = OpAccessChain %sb_f32 %indata_a %c_i32_0 %two_i\n" "%loc_a_2i_plus_1 = OpAccessChain %sb_f32 %indata_a %c_i32_0 %two_i_plus_1\n" "%loc_outdata_i = OpAccessChain %sb_f32 %outdata %c_i32_0 %30\n" "%val_s_i = OpLoad %f32 %loc_s_i\n" "%is_neg = OpFOrdLessThan %bool %val_s_i %c_f32_0\n" // select using a strategy. "${ResultStrategy}" // load through the variable pointer "%mux_output = OpLoad %f32 ${VarPtrName}\n" // store to the output vector. " OpStore %loc_outdata_i %mux_output\n" " OpBranch %inc\n" // ++i " %inc = OpLabel\n" " %37 = OpLoad %i32 %i\n" " %39 = OpIAdd %i32 %37 %c_i32_1\n" " OpStore %i %39\n" " OpBranch %loop\n" // Return and FunctionEnd "%merge = OpLabel\n" " OpBranch %end_if\n" "%end_if = OpLabel\n" "OpReturnValue %param\n" "OpFunctionEnd\n"); const bool singleInputBuffer[] = { true, false }; for (int inputBufferTypeIndex = 0 ; inputBufferTypeIndex < 2; ++inputBufferTypeIndex) { const bool isSingleInputBuffer = singleInputBuffer[inputBufferTypeIndex]; const string cap = isSingleInputBuffer ? "OpCapability VariablePointersStorageBuffer\n" : "OpCapability VariablePointers\n"; const vector& expectedOutput = isSingleInputBuffer ? AmuxAOutputFloats : AmuxBOutputFloats; const vector& expectedIncrOutput = isSingleInputBuffer ? incrAmuxAOutputFloats : incrAmuxBOutputFloats; const string bufferType = isSingleInputBuffer ? "single_buffer" : "two_buffers"; const string muxInput1 = isSingleInputBuffer ? " %loc_a_2i " : " %loc_a_i "; const string muxInput2 = isSingleInputBuffer ? " %loc_a_2i_plus_1 " : " %loc_b_i "; // Set the proper extension features required for the test if (isSingleInputBuffer) requiredFeatures.extVariablePointers.variablePointersStorageBuffer = true; else requiredFeatures.extVariablePointers.variablePointers = true; // All of the following tests write their results into an output SSBO, therefore they require the following features. requiredFeatures.coreFeatures.vertexPipelineStoresAndAtomics = DE_TRUE; requiredFeatures.coreFeatures.fragmentStoresAndAtomics = DE_TRUE; { // Variable Pointer Reads (using OpSelect) GraphicsResources resources; map specs; string name = "reads_opselect_" + bufferType; specs["ExtraTypes"] = ""; specs["ExtraGlobalScopeVars"] = ""; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraFunctions"] = ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = "%mux_output_var_ptr = OpSelect %sb_f32 %is_neg" + muxInput1 + muxInput2 + "\n"; fragments["capability"] = cap; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); createTestsForAllStages(name.c_str(), defaultColors, defaultColors, fragments, resources, extensions, testGroup, requiredFeatures); } { // Variable Pointer Reads (using OpFunctionCall) GraphicsResources resources; map specs; string name = "reads_opfunctioncall_" + bufferType; specs["ExtraTypes"] = ""; specs["ExtraGlobalScopeVars"] = ""; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraFunctions"] = selectorFunction; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = "%mux_output_var_ptr = OpFunctionCall %sb_f32 %choose_input_func %is_neg" + muxInput1 + muxInput2 + "\n"; fragments["capability"] = cap; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); createTestsForAllStages(name.c_str(), defaultColors, defaultColors, fragments, resources, extensions, testGroup, requiredFeatures); } { // Variable Pointer Reads (using OpPhi) GraphicsResources resources; map specs; string name = "reads_opphi_" + bufferType; specs["ExtraTypes"] = ""; specs["ExtraGlobalScopeVars"] = ""; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraFunctions"] = ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = " OpSelectionMerge %end_label None\n" " OpBranchConditional %is_neg %take_mux_input_1 %take_mux_input_2\n" "%take_mux_input_1 = OpLabel\n" " OpBranch %end_label\n" "%take_mux_input_2 = OpLabel\n" " OpBranch %end_label\n" "%end_label = OpLabel\n" "%mux_output_var_ptr = OpPhi %sb_f32" + muxInput1 + "%take_mux_input_1" + muxInput2 + "%take_mux_input_2\n"; fragments["capability"] = cap; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); createTestsForAllStages(name.c_str(), defaultColors, defaultColors, fragments, resources, extensions, testGroup, requiredFeatures); } { // Variable Pointer Reads (using OpCopyObject) GraphicsResources resources; map specs; string name = "reads_opcopyobject_" + bufferType; specs["ExtraTypes"] = ""; specs["ExtraGlobalScopeVars"] = ""; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraFunctions"] = ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = "%mux_input_1_copy = OpCopyObject %sb_f32" + muxInput1 + "\n" "%mux_input_2_copy = OpCopyObject %sb_f32" + muxInput2 + "\n" "%mux_output_var_ptr = OpSelect %sb_f32 %is_neg %mux_input_1_copy %mux_input_2_copy\n"; fragments["capability"] = cap; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); createTestsForAllStages(name.c_str(), defaultColors, defaultColors, fragments, resources, extensions, testGroup, requiredFeatures); } { // Test storing into Private variables. const char* storageClasses[] = {"Private", "Function"}; for (int classId = 0; classId < 2; ++classId) { GraphicsResources resources; map specs; std::string storageClass = storageClasses[classId]; std::string name = "stores_" + string(de::toLower(storageClass)) + "_" + bufferType; std::string extraVariable = "%mux_output_copy = OpVariable %sb_f32ptrptr " + storageClass + "\n"; specs["ExtraTypes"] = "%sb_f32ptrptr = OpTypePointer " + storageClass + " %sb_f32\n"; specs["ExtraGlobalScopeVars"] = (classId == 0) ? extraVariable : ""; specs["ExtraFunctionScopeVars"] = (classId == 1) ? extraVariable : ""; specs["ExtraFunctions"] = ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = "%opselect_result = OpSelect %sb_f32 %is_neg" + muxInput1 + muxInput2 + "\n" " OpStore %mux_output_copy %opselect_result\n" "%mux_output_var_ptr = OpLoad %sb_f32 %mux_output_copy\n"; fragments["capability"] = cap; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); createTestsForAllStages(name.c_str(), defaultColors, defaultColors, fragments, resources, extensions, testGroup, requiredFeatures); } } { // Variable Pointer Reads (using OpPtrAccessChain) GraphicsResources resources; map specs; std::string name = "reads_opptraccesschain_" + bufferType; std::string in_1 = isSingleInputBuffer ? " %a_2i_ptr " : " %a_i_ptr "; std::string in_2 = isSingleInputBuffer ? " %a_2i_plus_1_ptr " : " %b_i_ptr "; specs["ExtraTypes"] = ""; specs["ExtraGlobalScopeVars"] = ""; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraFunctions"] = ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = "%a_ptr = OpAccessChain %sb_f32 %indata_a %c_i32_0 %c_i32_0\n" "%b_ptr = OpAccessChain %sb_f32 %indata_b %c_i32_0 %c_i32_0\n" "%s_ptr = OpAccessChain %sb_f32 %indata_s %c_i32_0 %c_i32_0\n" "%out_ptr = OpAccessChain %sb_f32 %outdata %c_i32_0 %c_i32_0\n" "%a_i_ptr = OpPtrAccessChain %sb_f32 %a_ptr %30\n" "%b_i_ptr = OpPtrAccessChain %sb_f32 %b_ptr %30\n" "%s_i_ptr = OpPtrAccessChain %sb_f32 %s_ptr %30\n" "%a_2i_ptr = OpPtrAccessChain %sb_f32 %a_ptr %two_i\n" "%a_2i_plus_1_ptr = OpPtrAccessChain %sb_f32 %a_ptr %two_i_plus_1\n" "%mux_output_var_ptr = OpSelect %sb_f32 %is_neg " + in_1 + in_2 + "\n"; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); fragments["capability"] = cap; resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedOutput)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); createTestsForAllStages(name.c_str(), defaultColors, defaultColors, fragments, resources, extensions, testGroup, requiredFeatures); } { // Variable Pointer Writes GraphicsResources resources; map specs; std::string name = "writes_" + bufferType; specs["ExtraTypes"] = ""; specs["ExtraGlobalScopeVars"] = ""; specs["ExtraFunctionScopeVars"] = ""; specs["ExtraFunctions"] = ""; specs["VarPtrName"] = "%mux_output_var_ptr"; specs["ResultStrategy"] = "%mux_output_var_ptr = OpSelect %sb_f32 %is_neg" + muxInput1 + muxInput2 + "\n" + " %val = OpLoad %f32 %mux_output_var_ptr Aligned 4\n" " %val_plus_1 = OpFAdd %f32 %val %c_f32_1\n" " OpStore %mux_output_var_ptr %val_plus_1\n"; fragments["capability"] = cap; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputAFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputSFloats)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.outputs.push_back(Resource(BufferSp(new Float32Buffer(expectedIncrOutput)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); createTestsForAllStages(name.c_str(), defaultColors, defaultColors, fragments, resources, extensions, testGroup, requiredFeatures); } } } // Modifies the 'red channel' of the input color to the given float value. // Returns the modified color. void getExpectedOutputColor(RGBA (&inputColors)[4], RGBA (&expectedOutputColors)[4], float val) { Vec4 inColor0 = inputColors[0].toVec(); Vec4 inColor1 = inputColors[1].toVec(); Vec4 inColor2 = inputColors[2].toVec(); Vec4 inColor3 = inputColors[3].toVec(); inColor0[0] = val; inColor1[0] = val; inColor2[0] = val; inColor3[0] = val; expectedOutputColors[0] = RGBA(inColor0); expectedOutputColors[1] = RGBA(inColor1); expectedOutputColors[2] = RGBA(inColor2); expectedOutputColors[3] = RGBA(inColor3); } void addTwoInputBufferReadOnlyVariablePointersGraphicsGroup (tcu::TestCaseGroup* testGroup) { const int numFloatsPerInput = 64; vector inputA (numFloatsPerInput, 0); vector inputB (numFloatsPerInput, 0); deUint32 baseOffset = -1; VulkanFeatures requiredFeatures; map fragments; RGBA defaultColors[4]; RGBA expectedColors[4]; vector extensions; getDefaultColors(defaultColors); // Set the proper extension features required for the tests. requiredFeatures.extVariablePointers.variablePointers = true; // Set the required extension. extensions.push_back("VK_KHR_variable_pointers"); // These tests exercise variable pointers into various levels of the following data-structure: // struct struct inner_struct { // vec4 x[2]; // array of 2 vectors. Each vector is 4 floats. // vec4 y[2]; // array of 2 vectors. Each vector is 4 floats. // }; // // struct outer_struct { // inner_struct r[2][2]; // }; // // The inner_struct contains 16 floats, and the outer_struct contains 64 floats. // Therefore the input can be an array of 64 floats. // Populate the first input (inputA) to contain: {0, 4, ... , 252} / 255.f // Populate the second input (inputB) to contain: {3, 7, ... , 255} / 255.f for (size_t i = 0; i < numFloatsPerInput; ++i) { inputA[i] = 4*float(i) / 255; inputB[i] = ((4*float(i)) + 3) / 255; } // In the following tests we use variable pointers to point to different types: // nested structures, matrices of structures, arrays of structures, arrays of vectors, vectors of scalars, and scalars. // outer_structure.inner_structure[?][?].x[?][?]; // ^ ^ ^ ^ ^ ^ ^ // // 1. inputA or inputB = nested structure // 2. inputA.r or inputB.r = matrices of structures // 3. inputA.r[?] or inputB.r[?] = arrays of structures // 4. inputA.r[?][?] or inputB.r[?][?] = structures // 5. inputA.r[?][?].(x|y) or inputB.r[?][?].(x|y) = arrays of vectors // 6. inputA.r[?][?].(x|y)[?] or inputB.r[?][?].(x|y)[?] = vectors of scalars // 7. inputA.r[?][?].(x|y)[?][?] or inputB.r[?][?].(x|y)[?][?] = scalars const int numLevels = 7; fragments["capability"] = "OpCapability VariablePointers \n"; fragments["extension"] = "OpExtension \"SPV_KHR_variable_pointers\" \n" "OpExtension \"SPV_KHR_storage_buffer_storage_class\" \n"; const StringTemplate decoration ( // Set the Offsets "OpMemberDecorate %outer_struct 0 Offset 0 \n" "OpMemberDecorate %inner_struct 0 Offset 0 \n" "OpMemberDecorate %inner_struct 1 Offset 32 \n" // Set the ArrayStrides "OpDecorate %arr2_v4float ArrayStride 16 \n" "OpDecorate %arr2_inner_struct ArrayStride 64 \n" "OpDecorate %mat2x2_inner_struct ArrayStride 128 \n" "OpDecorate %mat2x2_ptr ArrayStride 128 \n" "OpDecorate %sb_buf ArrayStride 256 \n" "OpDecorate %v4f32_ptr ArrayStride 16 \n" "OpDecorate %outer_struct Block \n" "OpDecorate %in_a DescriptorSet 0 \n" "OpDecorate %in_b DescriptorSet 0 \n" "OpDecorate %in_a Binding 0 \n" "OpDecorate %in_b Binding 1 \n" "OpDecorate %in_a NonWritable \n" "OpDecorate %in_b NonWritable \n" ); const StringTemplate preMain ( /////////// // TYPES // /////////// // struct struct inner_struct { // vec4 x[2]; // array of 2 vectors // vec4 y[2]; // array of 2 vectors // }; "%arr2_v4float = OpTypeArray %v4f32 %c_u32_2 \n" "%inner_struct = OpTypeStruct %arr2_v4float %arr2_v4float \n" // struct outer_struct { // inner_struct r[2][2]; // }; "%arr2_inner_struct = OpTypeArray %inner_struct %c_u32_2 \n" "%mat2x2_inner_struct = OpTypeArray %arr2_inner_struct %c_u32_2 \n" "%outer_struct = OpTypeStruct %mat2x2_inner_struct \n" /////////////////// // POINTER TYPES // /////////////////// "%sb_buf = OpTypePointer StorageBuffer %outer_struct \n" "%mat2x2_ptr = OpTypePointer StorageBuffer %mat2x2_inner_struct \n" "%arr2_ptr = OpTypePointer StorageBuffer %arr2_inner_struct \n" "%inner_struct_ptr = OpTypePointer StorageBuffer %inner_struct \n" "%arr_v4f32_ptr = OpTypePointer StorageBuffer %arr2_v4float \n" "%v4f32_ptr = OpTypePointer StorageBuffer %v4f32 \n" "%sb_f32ptr = OpTypePointer StorageBuffer %f32 \n" /////////////// // VARIABLES // /////////////// "%in_a = OpVariable %sb_buf StorageBuffer \n" "%in_b = OpVariable %sb_buf StorageBuffer \n" /////////////// // CONSTANTS // /////////////// "%c_bool_true = OpConstantTrue %bool \n" "%c_bool_false = OpConstantFalse %bool \n" ////////////////////// // HELPER FUNCTIONS // ////////////////////// "${helper_functions} \n" ); const StringTemplate selectorFunctions ( // This selector function returns a variable pointer. // These functions are used by tests that use OpFunctionCall to obtain the variable pointer. "%selector_func_type = OpTypeFunction ${selected_type} %bool ${selected_type} ${selected_type}\n" "%choose_input_func = OpFunction ${selected_type} None %selector_func_type\n" "%choose_first_param = OpFunctionParameter %bool\n" "%first_param = OpFunctionParameter ${selected_type}\n" "%second_param = OpFunctionParameter ${selected_type}\n" "%selector_func_begin = OpLabel\n" "%result_ptr = OpSelect ${selected_type} %choose_first_param %first_param %second_param\n" " OpReturnValue %result_ptr\n" " OpFunctionEnd\n" ); const StringTemplate testFunction ( "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n" "%param = OpFunctionParameter %v4f32\n" "%entry = OpLabel\n" // Define base pointers for OpPtrAccessChain "%in_a_matptr = OpAccessChain %mat2x2_ptr %in_a %c_i32_0\n" "%in_b_matptr = OpAccessChain %mat2x2_ptr %in_b %c_i32_0\n" // Define the 2 pointers from which we're going to choose one. "${a_loc} \n" "${b_loc} \n" // Choose between the 2 pointers / variable pointers "${selection_strategy} \n" // OpAccessChain into the variable pointer until you get to the float. "%result_loc = OpAccessChain %sb_f32ptr %var_ptr ${remaining_indexes} \n" // Now load from the result_loc "%result_val = OpLoad %f32 %result_loc\n" // Modify the 'RED channel' of the output color to the chosen value "%output_color = OpCompositeInsert %v4f32 %result_val %param 0 \n" // Return and FunctionEnd "OpReturnValue %output_color\n" "OpFunctionEnd\n"); // When select is 0, the variable pointer should point to a value in the first input (inputA). // When select is 1, the variable pointer should point to a value in the second input (inputB). for (int selectInputA = 0; selectInputA < 2; ++selectInputA) { const string selectedInputStr = selectInputA ? "first_input" : "second_input"; const string spirvSelectInputA = selectInputA ? "%c_bool_true" : "%c_bool_false"; vector& selectedInput = selectInputA ? inputA : inputB; // The indexes chosen at each level. At any level, any given offset is exercised. // The first index is always zero as the outer structure has only 1 member. const int indexesForLevel[numLevels][6]= {{0, 0, 0, 0, 0, 1}, {0, 1, 0, 1, 0, 2}, {0, 0, 1, 0, 1, 3}, {0, 1, 1, 1, 0, 0}, {0, 0, 0, 1, 1, 1}, {0, 1, 0, 0, 0, 2}, {0, 1, 1, 1, 1, 3}}; const string indexLevelNames[] = {"_outer_struct_", "_matrices_", "_arrays_", "_inner_structs_", "_vec4arr_", "_vec4_", "_float_"}; const string inputALocations[] = { "", "%a_loc = OpAccessChain %mat2x2_ptr %in_a %c_i32_0", "%a_loc = OpAccessChain %arr2_ptr %in_a %c_i32_0 %c_i32_0", "%a_loc = OpAccessChain %inner_struct_ptr %in_a %c_i32_0 %c_i32_1 %c_i32_1", "%a_loc = OpAccessChain %arr_v4f32_ptr %in_a %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%a_loc = OpAccessChain %v4f32_ptr %in_a %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_0 %c_i32_0", "%a_loc = OpAccessChain %sb_f32ptr %in_a %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_3"}; const string inputBLocations[] = { "", "%b_loc = OpAccessChain %mat2x2_ptr %in_b %c_i32_0", "%b_loc = OpAccessChain %arr2_ptr %in_b %c_i32_0 %c_i32_0", "%b_loc = OpAccessChain %inner_struct_ptr %in_b %c_i32_0 %c_i32_1 %c_i32_1", "%b_loc = OpAccessChain %arr_v4f32_ptr %in_b %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%b_loc = OpAccessChain %v4f32_ptr %in_b %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_0 %c_i32_0", "%b_loc = OpAccessChain %sb_f32ptr %in_b %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_3"}; const string inputAPtrAccessChain[] = { "", "%a_loc = OpPtrAccessChain %mat2x2_ptr %in_a_matptr %c_i32_0", "%a_loc = OpPtrAccessChain %arr2_ptr %in_a_matptr %c_i32_0 %c_i32_0", "%a_loc = OpPtrAccessChain %inner_struct_ptr %in_a_matptr %c_i32_0 %c_i32_1 %c_i32_1", "%a_loc = OpPtrAccessChain %arr_v4f32_ptr %in_a_matptr %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%a_loc = OpPtrAccessChain %v4f32_ptr %in_a_matptr %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_0 %c_i32_0", // Next case emulates: // %a_loc = OpPtrAccessChain %sb_f32ptr %in_a_matptr %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_3 // But rewrite it to exercise OpPtrAccessChain with a non-zero first index: // %a_loc_arr is a pointer to an array that we want to index with 1. // But instead of just using OpAccessChain with first index 1, use OpAccessChain with index 0 to // get a pointer to the first element, then send that into OpPtrAccessChain with index 1. "%a_loc_arr = OpPtrAccessChain %arr_v4f32_ptr %in_a_matptr %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 " "%a_loc_first_elem = OpAccessChain %v4f32_ptr %a_loc_arr %c_i32_0 " "%a_loc = OpPtrAccessChain %sb_f32ptr %a_loc_first_elem %c_i32_1 %c_i32_3"}; const string inputBPtrAccessChain[] = { "", "%b_loc = OpPtrAccessChain %mat2x2_ptr %in_b_matptr %c_i32_0", "%b_loc = OpPtrAccessChain %arr2_ptr %in_b_matptr %c_i32_0 %c_i32_0", "%b_loc = OpPtrAccessChain %inner_struct_ptr %in_b_matptr %c_i32_0 %c_i32_1 %c_i32_1", "%b_loc = OpPtrAccessChain %arr_v4f32_ptr %in_b_matptr %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%b_loc = OpPtrAccessChain %v4f32_ptr %in_b_matptr %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_0 %c_i32_0", // Next case emulates: // %b_loc = OpPtrAccessChain %sb_f32ptr %in_b_matptr %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_3 // But rewrite it to exercise OpPtrAccessChain with a non-zero first index: // %b_loc_arr is a pointer to an array that we want to index with 1. // But instead of just using OpAccessChain with first index 1, use OpAccessChain with index 0 to // get a pointer to the first element, then send that into OpPtrAccessChain with index 1. "%b_loc_arr = OpPtrAccessChain %arr_v4f32_ptr %in_b_matptr %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 " "%b_loc_first_elem = OpAccessChain %v4f32_ptr %b_loc_arr %c_i32_0 " "%b_loc = OpPtrAccessChain %sb_f32ptr %b_loc_first_elem %c_i32_1 %c_i32_3"}; const string remainingIndexesAtLevel[]= {"%c_i32_0 %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%c_i32_1 %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_2", "%c_i32_1 %c_i32_0 %c_i32_1 %c_i32_3", "%c_i32_1 %c_i32_0 %c_i32_0", "%c_i32_1 %c_i32_1", "%c_i32_2", ""}; const string pointerTypeAtLevel[] = {"%sb_buf", "%mat2x2_ptr", "%arr2_ptr", "%inner_struct_ptr", "%arr_v4f32_ptr", "%v4f32_ptr", "%sb_f32ptr"}; const string baseANameAtLevel[] = {"%in_a", "%a_loc", "%a_loc", "%a_loc", "%a_loc", "%a_loc", "%a_loc"}; const string baseBNameAtLevel[] = {"%in_b", "%b_loc", "%b_loc", "%b_loc", "%b_loc", "%b_loc", "%b_loc"}; for (int indexLevel = 0; indexLevel < numLevels; ++indexLevel) { // index into the outer structure must be zero since the outer structure has only 1 member. DE_ASSERT(indexesForLevel[indexLevel][0] == 0); baseOffset = getBaseOffset(indexesForLevel[indexLevel][1], indexesForLevel[indexLevel][2], indexesForLevel[indexLevel][3], indexesForLevel[indexLevel][4], indexesForLevel[indexLevel][5]); // Use OpSelect to choose between 2 pointers { GraphicsResources resources; map specs; string opCodeForTests = "opselect"; string name = opCodeForTests + indexLevelNames[indexLevel] + selectedInputStr; specs["select_inputA"] = spirvSelectInputA; specs["helper_functions"] = ""; specs["a_loc"] = inputALocations[indexLevel]; specs["b_loc"] = inputBLocations[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = "%var_ptr = OpSelect " + pointerTypeAtLevel[indexLevel] + " " + spirvSelectInputA + " " + baseANameAtLevel[indexLevel] + " " + baseBNameAtLevel[indexLevel] + "\n"; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputA)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputB)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); getExpectedOutputColor(defaultColors, expectedColors, selectedInput[baseOffset]); createTestsForAllStages(name.c_str(), defaultColors, expectedColors, fragments, resources, extensions, testGroup, requiredFeatures); } // Use OpCopyObject to get variable pointers { GraphicsResources resources; map specs; string opCodeForTests = "opcopyobject"; string name = opCodeForTests + indexLevelNames[indexLevel] + selectedInputStr; specs["select_inputA"] = spirvSelectInputA; specs["helper_functions"] = ""; specs["a_loc"] = inputALocations[indexLevel]; specs["b_loc"] = inputBLocations[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = "%in_a_copy = OpCopyObject " + pointerTypeAtLevel[indexLevel] + " " + baseANameAtLevel[indexLevel] + "\n" "%in_b_copy = OpCopyObject " + pointerTypeAtLevel[indexLevel] + " " + baseBNameAtLevel[indexLevel] + "\n" "%var_ptr = OpSelect " + pointerTypeAtLevel[indexLevel] + " " + spirvSelectInputA + " %in_a_copy %in_b_copy\n"; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputA)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputB)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); getExpectedOutputColor(defaultColors, expectedColors, selectedInput[baseOffset]); createTestsForAllStages(name.c_str(), defaultColors, expectedColors, fragments, resources, extensions, testGroup, requiredFeatures); } // Use OpPhi to choose between 2 pointers { GraphicsResources resources; map specs; string opCodeForTests = "opphi"; string name = opCodeForTests + indexLevelNames[indexLevel] + selectedInputStr; specs["select_inputA"] = spirvSelectInputA; specs["helper_functions"] = ""; specs["a_loc"] = inputALocations[indexLevel]; specs["b_loc"] = inputBLocations[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = " OpSelectionMerge %end_label None\n" " OpBranchConditional " + spirvSelectInputA + " %take_input_a %take_input_b\n" "%take_input_a = OpLabel\n" " OpBranch %end_label\n" "%take_input_b = OpLabel\n" " OpBranch %end_label\n" "%end_label = OpLabel\n" "%var_ptr = OpPhi " + pointerTypeAtLevel[indexLevel] + " " + baseANameAtLevel[indexLevel] + " %take_input_a " + baseBNameAtLevel[indexLevel] + " %take_input_b\n"; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputA)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputB)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); getExpectedOutputColor(defaultColors, expectedColors, selectedInput[baseOffset]); createTestsForAllStages(name.c_str(), defaultColors, expectedColors, fragments, resources, extensions, testGroup, requiredFeatures); } // Use OpFunctionCall to choose between 2 pointers { GraphicsResources resources; map functionSpecs; map specs; string opCodeForTests = "opfunctioncall"; string name = opCodeForTests + indexLevelNames[indexLevel] + selectedInputStr; functionSpecs["selected_type"] = pointerTypeAtLevel[indexLevel]; specs["helper_functions"] = selectorFunctions.specialize(functionSpecs); specs["select_inputA"] = spirvSelectInputA; specs["a_loc"] = inputALocations[indexLevel]; specs["b_loc"] = inputBLocations[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = "%var_ptr = OpFunctionCall " + pointerTypeAtLevel[indexLevel] + " %choose_input_func " + spirvSelectInputA + " " + baseANameAtLevel[indexLevel] + " " + baseBNameAtLevel[indexLevel] + "\n"; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputA)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputB)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); getExpectedOutputColor(defaultColors, expectedColors, selectedInput[baseOffset]); createTestsForAllStages(name.c_str(), defaultColors, expectedColors, fragments, resources, extensions, testGroup, requiredFeatures); } // Use OpPtrAccessChain to get variable pointers { GraphicsResources resources; map specs; string opCodeForTests = "opptraccesschain"; string name = opCodeForTests + indexLevelNames[indexLevel] + selectedInputStr; specs["select_inputA"] = spirvSelectInputA; specs["helper_functions"] = ""; specs["a_loc"] = inputAPtrAccessChain[indexLevel]; specs["b_loc"] = inputBPtrAccessChain[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = "%var_ptr = OpSelect " + pointerTypeAtLevel[indexLevel] + " " + spirvSelectInputA + " " + baseANameAtLevel[indexLevel] + " " + baseBNameAtLevel[indexLevel] + "\n"; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputA)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputB)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); getExpectedOutputColor(defaultColors, expectedColors, selectedInput[baseOffset]); createTestsForAllStages(name.c_str(), defaultColors, expectedColors, fragments, resources, extensions, testGroup, requiredFeatures); } } } } void addSingleInputBufferReadOnlyVariablePointersGraphicsGroup (tcu::TestCaseGroup* testGroup) { const int numFloatsPerInnerStruct = 64; vector inputBuffer (2 * numFloatsPerInnerStruct, 0); deUint32 baseOffset = -1; VulkanFeatures requiredFeatures; map fragments; RGBA defaultColors[4]; RGBA expectedColors[4]; vector extensions; // Set the proper extension features required for the tests. // The following tests use variable pointers confined withing a single buffer. requiredFeatures.extVariablePointers.variablePointersStorageBuffer = true; // Set the required extension. extensions.push_back("VK_KHR_variable_pointers"); getDefaultColors(defaultColors); // These tests exercise variable pointers into various levels of the following data-structure: // struct struct inner_struct { // vec4 x[2]; // array of 2 vectors. Each vector is 4 floats. // vec4 y[2]; // array of 2 vectors. Each vector is 4 floats. // }; // // struct outer_struct { // inner_struct r[2][2]; // }; // // struct input_buffer { // outer_struct a; // outer_struct b; // } // // The inner_struct contains 16 floats, and the outer_struct contains 64 floats. // Therefore the input_buffer can be an array of 128 floats. // Populate input_buffer's first member (a) to contain: {0, 4, ... , 252} / 255.f // Populate input_buffer's second member (b) to contain: {3, 7, ... , 255} / 255.f for (size_t i = 0; i < numFloatsPerInnerStruct; ++i) { inputBuffer[i] = 4*float(i) / 255; inputBuffer[i + numFloatsPerInnerStruct] = ((4*float(i)) + 3) / 255; } // In the following tests we use variable pointers to point to different types: // nested structures, matrices of structures, arrays of structures, arrays of vectors, vectors of scalars, and scalars. // outer_struct.inner_struct[?][?].x[?][?]; // ^ ^ ^ ^ ^ ^ ^ // // 1. inputBuffer.a or inputBuffer.b = nested structure // 2. inputBuffer.a.r or inputBuffer.b.r = matrices of structures // 3. inputBuffer.a.r[?] or inputBuffer.b.r[?] = arrays of structures // 4. inputBuffer.a.r[?][?] or inputBuffer.b.r[?][?] = structures // 5. inputBuffer.a.r[?][?].(x|y) or inputBuffer.b.r[?][?].(x|y) = arrays of vectors // 6. inputBuffer.a.r[?][?].(x|y)[?] or inputBuffer.b.r[?][?].(x|y)[?] = vectors of scalars // 7. inputBuffer.a.r[?][?].(x|y)[?][?] or inputBuffer.b.r[?][?].(x|y)[?][?] = scalars const int numLevels = 7; fragments["capability"] = "OpCapability VariablePointersStorageBuffer \n"; fragments["extension"] = "OpExtension \"SPV_KHR_variable_pointers\" \n" "OpExtension \"SPV_KHR_storage_buffer_storage_class\" \n"; const StringTemplate decoration ( // Set the ArrayStrides "OpDecorate %arr2_v4float ArrayStride 16 \n" "OpDecorate %arr2_inner_struct ArrayStride 64 \n" "OpDecorate %mat2x2_inner_struct ArrayStride 128 \n" "OpDecorate %outer_struct_ptr ArrayStride 256 \n" "OpDecorate %v4f32_ptr ArrayStride 16 \n" // Set the Offsets "OpMemberDecorate %input_buffer 0 Offset 0 \n" "OpMemberDecorate %input_buffer 1 Offset 256 \n" "OpMemberDecorate %outer_struct 0 Offset 0 \n" "OpMemberDecorate %inner_struct 0 Offset 0 \n" "OpMemberDecorate %inner_struct 1 Offset 32 \n" "OpDecorate %input_buffer Block \n" "OpDecorate %input DescriptorSet 0 \n" "OpDecorate %input Binding 0 \n" "OpDecorate %input NonWritable \n" ); const StringTemplate preMain ( /////////// // TYPES // /////////// // struct struct inner_struct { // vec4 x[2]; // array of 2 vectors // vec4 y[2]; // array of 2 vectors // }; "%arr2_v4float = OpTypeArray %v4f32 %c_u32_2 \n" "%inner_struct = OpTypeStruct %arr2_v4float %arr2_v4float \n" // struct outer_struct { // inner_struct r[2][2]; // }; "%arr2_inner_struct = OpTypeArray %inner_struct %c_u32_2 \n" "%mat2x2_inner_struct = OpTypeArray %arr2_inner_struct %c_u32_2 \n" "%outer_struct = OpTypeStruct %mat2x2_inner_struct \n" // struct input_buffer { // outer_struct a; // outer_struct b; // } "%input_buffer = OpTypeStruct %outer_struct %outer_struct \n" /////////////////// // POINTER TYPES // /////////////////// "%input_buffer_ptr = OpTypePointer StorageBuffer %input_buffer \n" "%outer_struct_ptr = OpTypePointer StorageBuffer %outer_struct \n" "%mat2x2_ptr = OpTypePointer StorageBuffer %mat2x2_inner_struct \n" "%arr2_ptr = OpTypePointer StorageBuffer %arr2_inner_struct \n" "%inner_struct_ptr = OpTypePointer StorageBuffer %inner_struct \n" "%arr_v4f32_ptr = OpTypePointer StorageBuffer %arr2_v4float \n" "%v4f32_ptr = OpTypePointer StorageBuffer %v4f32 \n" "%sb_f32ptr = OpTypePointer StorageBuffer %f32 \n" /////////////// // VARIABLES // /////////////// "%input = OpVariable %input_buffer_ptr StorageBuffer \n" /////////////// // CONSTANTS // /////////////// "%c_bool_true = OpConstantTrue %bool \n" "%c_bool_false = OpConstantFalse %bool \n" ////////////////////// // HELPER FUNCTIONS // ////////////////////// "${helper_functions} \n" ); const StringTemplate selectorFunctions ( // These selector functions return variable pointers. // These functions are used by tests that use OpFunctionCall to obtain the variable pointer "%selector_func_type = OpTypeFunction ${selected_type} %bool ${selected_type} ${selected_type}\n" "%choose_input_func = OpFunction ${selected_type} None %selector_func_type\n" "%choose_first_param = OpFunctionParameter %bool\n" "%first_param = OpFunctionParameter ${selected_type}\n" "%second_param = OpFunctionParameter ${selected_type}\n" "%selector_func_begin = OpLabel\n" "%result_ptr = OpSelect ${selected_type} %choose_first_param %first_param %second_param\n" "OpReturnValue %result_ptr\n" "OpFunctionEnd\n" ); const StringTemplate testFunction ( "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n" "%param = OpFunctionParameter %v4f32\n" "%entry = OpLabel\n" // Here are the 2 nested structures: "%in_a = OpAccessChain %outer_struct_ptr %input %c_i32_0\n" "%in_b = OpAccessChain %outer_struct_ptr %input %c_i32_1\n" // Define the 2 pointers from which we're going to choose one. "${a_loc} \n" "${b_loc} \n" // Choose between the 2 pointers / variable pointers "${selection_strategy} \n" // OpAccessChain into the variable pointer until you get to the float. "%result_loc = OpAccessChain %sb_f32ptr %var_ptr ${remaining_indexes} \n" // Now load from the result_loc "%result_val = OpLoad %f32 %result_loc\n" // Modify the 'RED channel' of the output color to the chosen value "%output_color = OpCompositeInsert %v4f32 %result_val %param 0 \n" // Return and FunctionEnd "OpReturnValue %output_color\n" "OpFunctionEnd\n"); // When select is 0, the variable pointer should point to a value in the first input_buffer member (a). // When select is 1, the variable pointer should point to a value in the second input_buffer member (b). // Since the 2 members of the input_buffer (a and b) are of type outer_struct, we can conveniently use // the same indexing scheme that we used for the 2-input-buffer tests. for (int selectInputA = 0; selectInputA < 2; ++selectInputA) { const string selectedInputStr = selectInputA ? "first_input" : "second_input"; const string spirvSelectInputA = selectInputA ? "%c_bool_true" : "%c_bool_false"; const int outerStructIndex = selectInputA ? 0 : 1; // The indexes chosen at each level. At any level, any given offset is exercised. // outerStructIndex is 0 for member (a) and 1 for member (b). const int indexesForLevel[numLevels][6]= {{outerStructIndex, 0, 0, 0, 0, 1}, {outerStructIndex, 1, 0, 1, 0, 2}, {outerStructIndex, 0, 1, 0, 1, 3}, {outerStructIndex, 1, 1, 1, 0, 0}, {outerStructIndex, 0, 0, 1, 1, 1}, {outerStructIndex, 1, 0, 0, 0, 2}, {outerStructIndex, 1, 1, 1, 1, 3}}; const string indexLevelNames[] = {"_outer_struct_", "_matrices_", "_arrays_", "_inner_structs_", "_vec4arr_", "_vec4_", "_float_"}; const string inputALocations[] = { "", "%a_loc = OpAccessChain %mat2x2_ptr %in_a %c_i32_0", "%a_loc = OpAccessChain %arr2_ptr %in_a %c_i32_0 %c_i32_0", "%a_loc = OpAccessChain %inner_struct_ptr %in_a %c_i32_0 %c_i32_1 %c_i32_1", "%a_loc = OpAccessChain %arr_v4f32_ptr %in_a %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%a_loc = OpAccessChain %v4f32_ptr %in_a %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_0 %c_i32_0", "%a_loc = OpAccessChain %sb_f32ptr %in_a %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_3"}; const string inputBLocations[] = { "", "%b_loc = OpAccessChain %mat2x2_ptr %in_b %c_i32_0", "%b_loc = OpAccessChain %arr2_ptr %in_b %c_i32_0 %c_i32_0", "%b_loc = OpAccessChain %inner_struct_ptr %in_b %c_i32_0 %c_i32_1 %c_i32_1", "%b_loc = OpAccessChain %arr_v4f32_ptr %in_b %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%b_loc = OpAccessChain %v4f32_ptr %in_b %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_0 %c_i32_0", "%b_loc = OpAccessChain %sb_f32ptr %in_b %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_3"}; const string inputAPtrAccessChain[] = { "", "%a_loc = OpPtrAccessChain %mat2x2_ptr %in_a %c_i32_0 %c_i32_0", "%a_loc = OpPtrAccessChain %arr2_ptr %in_a %c_i32_0 %c_i32_0 %c_i32_0", "%a_loc = OpPtrAccessChain %inner_struct_ptr %in_a %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_1", "%a_loc = OpPtrAccessChain %arr_v4f32_ptr %in_a %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%a_loc = OpPtrAccessChain %v4f32_ptr %in_a %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_0 %c_i32_0", // Next case emulates: // %a_loc = OpPtrAccessChain %sb_f32ptr %in_a %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_3 // But rewrite it to exercise OpPtrAccessChain with a non-zero first index: // %a_loc_arr is a pointer to an array that we want to index with 1. // But instead of just using OpAccessChain with first index 1, use OpAccessChain with index 0 to // get a pointer to the first element, then send that into OpPtrAccessChain with index 1. "%a_loc_arr = OpPtrAccessChain %arr_v4f32_ptr %in_a %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 " "%a_loc_first_elem = OpAccessChain %v4f32_ptr %a_loc_arr %c_i32_0 " "%a_loc = OpPtrAccessChain %sb_f32ptr %a_loc_first_elem %c_i32_1 %c_i32_3"}; const string inputBPtrAccessChain[] = { "", "%b_loc = OpPtrAccessChain %mat2x2_ptr %in_b %c_i32_0 %c_i32_0", "%b_loc = OpPtrAccessChain %arr2_ptr %in_b %c_i32_0 %c_i32_0 %c_i32_0", "%b_loc = OpPtrAccessChain %inner_struct_ptr %in_b %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_1", "%b_loc = OpPtrAccessChain %arr_v4f32_ptr %in_b %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%b_loc = OpPtrAccessChain %v4f32_ptr %in_b %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_0 %c_i32_0", // Next case emulates: // %b_loc = OpPtrAccessChain %sb_f32ptr %in_b %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_1 %c_i32_3 // But rewrite it to exercise OpPtrAccessChain with a non-zero first index: // %b_loc_arr is a pointer to an array that we want to index with 1. // But instead of just using OpAccessChain with first index 1, use OpAccessChain with index 0 to // get a pointer to the first element, then send that into OpPtrAccessChain with index 1. "%b_loc_arr = OpPtrAccessChain %arr_v4f32_ptr %in_b %c_i32_0 %c_i32_0 %c_i32_1 %c_i32_1 %c_i32_1 " "%b_loc_first_elem = OpAccessChain %v4f32_ptr %b_loc_arr %c_i32_0 " "%b_loc = OpPtrAccessChain %sb_f32ptr %b_loc_first_elem %c_i32_1 %c_i32_3"}; const string remainingIndexesAtLevel[]= {"%c_i32_0 %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_0 %c_i32_1", "%c_i32_1 %c_i32_0 %c_i32_1 %c_i32_0 %c_i32_2", "%c_i32_1 %c_i32_0 %c_i32_1 %c_i32_3", "%c_i32_1 %c_i32_0 %c_i32_0", "%c_i32_1 %c_i32_1", "%c_i32_2", ""}; const string pointerTypeAtLevel[] = {"%outer_struct_ptr", "%mat2x2_ptr", "%arr2_ptr", "%inner_struct_ptr", "%arr_v4f32_ptr", "%v4f32_ptr", "%sb_f32ptr"}; const string baseANameAtLevel[] = {"%in_a", "%a_loc", "%a_loc", "%a_loc", "%a_loc", "%a_loc", "%a_loc"}; const string baseBNameAtLevel[] = {"%in_b", "%b_loc", "%b_loc", "%b_loc", "%b_loc", "%b_loc", "%b_loc"}; for (int indexLevel = 0; indexLevel < numLevels; ++indexLevel) { // Use OpSelect to choose between 2 pointers { GraphicsResources resources; map specs; string opCodeForTests = "opselect"; string name = opCodeForTests + indexLevelNames[indexLevel] + selectedInputStr; specs["select_inputA"] = spirvSelectInputA; specs["helper_functions"] = ""; specs["a_loc"] = inputALocations[indexLevel]; specs["b_loc"] = inputBLocations[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = "%var_ptr = OpSelect " + pointerTypeAtLevel[indexLevel] + " " + spirvSelectInputA + " " + baseANameAtLevel[indexLevel] + " " + baseBNameAtLevel[indexLevel] + "\n"; baseOffset = getBaseOffsetForSingleInputBuffer(indexesForLevel[indexLevel][0], indexesForLevel[indexLevel][1], indexesForLevel[indexLevel][2], indexesForLevel[indexLevel][3], indexesForLevel[indexLevel][4], indexesForLevel[indexLevel][5]); fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBuffer)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); getExpectedOutputColor(defaultColors, expectedColors, inputBuffer[baseOffset]); createTestsForAllStages(name.c_str(), defaultColors, expectedColors, fragments, resources, extensions, testGroup, requiredFeatures); } // Use OpCopyObject to get variable pointers { GraphicsResources resources; map specs; string opCodeForTests = "opcopyobject"; string name = opCodeForTests + indexLevelNames[indexLevel] + selectedInputStr; specs["select_inputA"] = spirvSelectInputA; specs["helper_functions"] = ""; specs["a_loc"] = inputALocations[indexLevel]; specs["b_loc"] = inputBLocations[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = "%in_a_copy = OpCopyObject " + pointerTypeAtLevel[indexLevel] + " " + baseANameAtLevel[indexLevel] + "\n" "%in_b_copy = OpCopyObject " + pointerTypeAtLevel[indexLevel] + " " + baseBNameAtLevel[indexLevel] + "\n" "%var_ptr = OpSelect " + pointerTypeAtLevel[indexLevel] + " " + spirvSelectInputA + " %in_a_copy %in_b_copy\n"; baseOffset = getBaseOffsetForSingleInputBuffer(indexesForLevel[indexLevel][0], indexesForLevel[indexLevel][1], indexesForLevel[indexLevel][2], indexesForLevel[indexLevel][3], indexesForLevel[indexLevel][4], indexesForLevel[indexLevel][5]); fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBuffer)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); getExpectedOutputColor(defaultColors, expectedColors, inputBuffer[baseOffset]); createTestsForAllStages(name.c_str(), defaultColors, expectedColors, fragments, resources, extensions, testGroup, requiredFeatures); } // Use OpPhi to choose between 2 pointers { GraphicsResources resources; map specs; string opCodeForTests = "opphi"; string name = opCodeForTests + indexLevelNames[indexLevel] + selectedInputStr; specs["select_inputA"] = spirvSelectInputA; specs["helper_functions"] = ""; specs["a_loc"] = inputALocations[indexLevel]; specs["b_loc"] = inputBLocations[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = " OpSelectionMerge %end_label None\n" " OpBranchConditional " + spirvSelectInputA + " %take_input_a %take_input_b\n" "%take_input_a = OpLabel\n" " OpBranch %end_label\n" "%take_input_b = OpLabel\n" " OpBranch %end_label\n" "%end_label = OpLabel\n" "%var_ptr = OpPhi " + pointerTypeAtLevel[indexLevel] + " " + baseANameAtLevel[indexLevel] + " %take_input_a " + baseBNameAtLevel[indexLevel] + " %take_input_b\n"; baseOffset = getBaseOffsetForSingleInputBuffer(indexesForLevel[indexLevel][0], indexesForLevel[indexLevel][1], indexesForLevel[indexLevel][2], indexesForLevel[indexLevel][3], indexesForLevel[indexLevel][4], indexesForLevel[indexLevel][5]); fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBuffer)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); getExpectedOutputColor(defaultColors, expectedColors, inputBuffer[baseOffset]); createTestsForAllStages(name.c_str(), defaultColors, expectedColors, fragments, resources, extensions, testGroup, requiredFeatures); } // Use OpFunctionCall to choose between 2 pointers { GraphicsResources resources; map functionSpecs; map specs; string opCodeForTests = "opfunctioncall"; string name = opCodeForTests + indexLevelNames[indexLevel] + selectedInputStr; //string selectedType = "%mat2x2_ptr"; functionSpecs["selected_type"] = pointerTypeAtLevel[indexLevel]; specs["helper_functions"] = selectorFunctions.specialize(functionSpecs); specs["select_inputA"] = spirvSelectInputA; specs["a_loc"] = inputALocations[indexLevel]; specs["b_loc"] = inputBLocations[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = "%var_ptr = OpFunctionCall " + pointerTypeAtLevel[indexLevel] + " %choose_input_func " + spirvSelectInputA + " " + baseANameAtLevel[indexLevel] + " " + baseBNameAtLevel[indexLevel] + "\n"; baseOffset = getBaseOffsetForSingleInputBuffer(indexesForLevel[indexLevel][0], indexesForLevel[indexLevel][1], indexesForLevel[indexLevel][2], indexesForLevel[indexLevel][3], indexesForLevel[indexLevel][4], indexesForLevel[indexLevel][5]); fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBuffer)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); getExpectedOutputColor(defaultColors, expectedColors, inputBuffer[baseOffset]); createTestsForAllStages(name.c_str(), defaultColors, expectedColors, fragments, resources, extensions, testGroup, requiredFeatures); } // Use OpPtrAccessChain to get variable pointers { GraphicsResources resources; map specs; string opCodeForTests = "opptraccesschain"; string name = opCodeForTests + indexLevelNames[indexLevel] + selectedInputStr; specs["select_inputA"] = spirvSelectInputA; specs["helper_functions"] = ""; specs["a_loc"] = inputAPtrAccessChain[indexLevel]; specs["b_loc"] = inputBPtrAccessChain[indexLevel]; specs["remaining_indexes"] = remainingIndexesAtLevel[indexLevel]; specs["selection_strategy"] = "%var_ptr = OpSelect " + pointerTypeAtLevel[indexLevel] + " " + spirvSelectInputA + " " + baseANameAtLevel[indexLevel] + " " + baseBNameAtLevel[indexLevel] + "\n"; baseOffset = getBaseOffsetForSingleInputBuffer(indexesForLevel[indexLevel][0], indexesForLevel[indexLevel][1], indexesForLevel[indexLevel][2], indexesForLevel[indexLevel][3], indexesForLevel[indexLevel][4], indexesForLevel[indexLevel][5]); fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(inputBuffer)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); getExpectedOutputColor(defaultColors, expectedColors, inputBuffer[baseOffset]); createTestsForAllStages(name.c_str(), defaultColors, expectedColors, fragments, resources, extensions, testGroup, requiredFeatures); } } } } void addNullptrVariablePointersGraphicsGroup (tcu::TestCaseGroup* testGroup) { float someFloat = 78 / 255.f; vector input (1, someFloat); vector expectedOutput (1, someFloat); VulkanFeatures requiredFeatures; map fragments; RGBA defaultColors[4]; RGBA expectedColors[4]; vector extensions; getDefaultColors(defaultColors); getExpectedOutputColor(defaultColors, expectedColors, someFloat); // Set the required extension. extensions.push_back("VK_KHR_variable_pointers"); // Requires the variable pointers feature. requiredFeatures.extVariablePointers.variablePointers = true; fragments["capability"] = "OpCapability VariablePointers \n"; fragments["extension"] = "OpExtension \"SPV_KHR_variable_pointers\" \n" "OpExtension \"SPV_KHR_storage_buffer_storage_class\" \n"; const StringTemplate decoration ( // Decorations "OpDecorate %input DescriptorSet 0 \n" "OpDecorate %input Binding 0 \n" "OpDecorate %input NonWritable \n" // Set the Block decoration "OpDecorate %float_struct Block \n" // Set the Offsets "OpMemberDecorate %float_struct 0 Offset 0 \n" ); const StringTemplate preMain ( // struct float_struct { // float x; // }; "%float_struct = OpTypeStruct %f32 \n" // POINTER TYPES "%float_struct_ptr = OpTypePointer StorageBuffer %float_struct \n" "%sb_f32ptr = OpTypePointer StorageBuffer %f32 \n" "%func_f32ptrptr = OpTypePointer Function %sb_f32ptr \n" // CONSTANTS "%c_bool_true = OpConstantTrue %bool \n" "%c_null_ptr = OpConstantNull %sb_f32ptr \n" // VARIABLES "%input = OpVariable %float_struct_ptr StorageBuffer \n" ); const StringTemplate testFunction ( "%test_code = OpFunction %v4f32 None %v4f32_v4f32_function\n" "%param = OpFunctionParameter %v4f32\n" "%entry = OpLabel\n" // Note that the Variable Pointers extension allows creation // of a pointer variable with storage class of Private or Function. "%f32_ptr_var = OpVariable %func_f32ptrptr Function %c_null_ptr\n" "%input_loc = OpAccessChain %sb_f32ptr %input %c_i32_0\n" // Null testing strategy "${NullptrTestingStrategy}\n" // Modify the 'RED channel' of the output color to the chosen value "%output_color = OpCompositeInsert %v4f32 %result_val %param 0 \n" // Return and FunctionEnd "OpReturnValue %output_color\n" "OpFunctionEnd\n"); // f32_ptr_var has been inintialized to NULL. // Now set it to the input variable and return it as output { GraphicsResources resources; map specs; specs["NullptrTestingStrategy"] = " OpStore %f32_ptr_var %input_loc \n" "%loaded_f32_ptr = OpLoad %sb_f32ptr %f32_ptr_var \n" "%result_val = OpLoad %f32 %loaded_f32_ptr \n"; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(input)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); createTestsForAllStages("opvariable_initialized_null", defaultColors, expectedColors, fragments, resources, extensions, testGroup, requiredFeatures); } // Use OpSelect to choose between nullptr and a valid pointer. Since we can't dereference nullptr, // it is forced to always choose the valid pointer. { GraphicsResources resources; map specs; specs["NullptrTestingStrategy"] = "%selected_ptr = OpSelect %sb_f32ptr %c_bool_true %input_loc %c_null_ptr\n" "%result_val = OpLoad %f32 %selected_ptr\n"; fragments["decoration"] = decoration.specialize(specs); fragments["pre_main"] = preMain.specialize(specs); fragments["testfun"] = testFunction.specialize(specs); resources.inputs.push_back(Resource(BufferSp(new Float32Buffer(input)), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER)); createTestsForAllStages("opselect_null_or_valid_ptr", defaultColors, expectedColors, fragments, resources, extensions, testGroup, requiredFeatures); } } } // anonymous tcu::TestCaseGroup* createVariablePointersComputeGroup (tcu::TestContext& testCtx) { // Compute tests for SPV_KHR_variable_pointers extension de::MovePtr group (new tcu::TestCaseGroup(testCtx, "variable_pointers")); // Test the variable pointer extension using a compute shader addTestGroup(group.get(), "compute", addVariablePointersComputeGroup); // Testing Variable Pointers pointing to various types in different input buffers addTestGroup(group.get(), "complex_types_compute", addComplexTypesVariablePointersComputeGroup); // Test the usage of nullptr using the variable pointers extension in a compute shader addTestGroup(group.get(), "nullptr_compute", addNullptrVariablePointersComputeGroup); // Testing variable pointers referring to descriptors using dynamic offset addTestGroup(group.get(), "dynamic_offset", addDynamicOffsetComputeGroup); return group.release(); } tcu::TestCaseGroup* createPhysicalPointersComputeGroup (tcu::TestContext& testCtx) { de::MovePtr group (new tcu::TestCaseGroup(testCtx, "physical_pointers", "Compute tests for SPV_KHR_physical_storage_buffer extension")); // Test the physical storage buffer extension using a compute shader addTestGroup(group.get(), "compute", addPhysicalPointersComputeGroup); // Testing physical pointers pointing to various types in different input buffers addTestGroup(group.get(), "complex_types_compute", addComplexTypesPhysicalPointersComputeGroup); return group.release(); } tcu::TestCaseGroup* createVariablePointersGraphicsGroup (tcu::TestContext& testCtx) { de::MovePtr group (new tcu::TestCaseGroup(testCtx, "variable_pointers", "Graphics tests for SPV_KHR_variable_pointers extension")); addTestGroup(group.get(), "graphics", addVariablePointersGraphicsGroup); // Testing Variable Pointers pointing to different input buffers in graphics pipeline (no SSBO writes) addTestGroup(group.get(), "multi_buffer_read_only_graphics", addTwoInputBufferReadOnlyVariablePointersGraphicsGroup); // Testing Variable Pointers confined to a single input buffer in graphics pipeline (no SSBO writes) addTestGroup(group.get(), "single_buffer_read_only_graphics", addSingleInputBufferReadOnlyVariablePointersGraphicsGroup); // Test the usage of nullptr using the variable pointers extension in graphics pipeline addTestGroup(group.get(), "nullptr_graphics", addNullptrVariablePointersGraphicsGroup); return group.release(); } } // SpirVAssembly } // vkt