external/vulkancts/modules/vulkan/spirv_assembly/vktSpvAsmLoopDepInfTests.cpp

   1 /*-------------------------------------------------------------------------
   2  * Vulkan Conformance Tests
   3  * ------------------------
   4  *
   5  * Copyright (c) 2017 The Khronos Group Inc.
   6  *
   7  * Licensed under the Apache License, Version 2.0 (the "License");
   8  * you may not use this file except in compliance with the License.
   9  * You may obtain a copy of the License at
  10  *
  11  *      http://www.apache.org/licenses/LICENSE-2.0
  12  *
  13  * Unless required by applicable law or agreed to in writing, software
  14  * distributed under the License is distributed on an "AS IS" BASIS,
  15  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  16  * See the License for the specific language governing permissions and
  17  * limitations under the License.
  18  *
  19  *//*!
  20  * \file
  21  * \brief SPIR-V Loop Control for DependencyInfinite qualifier tests
  22  *//*--------------------------------------------------------------------*/
  23
  24 #include "vkApiVersion.hpp"
  25
  26 #include "vktSpvAsmLoopDepInfTests.hpp"
  27 #include "vktTestCase.hpp"
  28 #include "vktSpvAsmComputeShaderCase.hpp"
  29
  30 #include "deRandom.hpp"
  31
  32 namespace vkt
  33 {
  34 namespace SpirVAssembly
  35 {
  36
  37 using namespace vk;
  38 using std::map;
  39 using std::string;
  40 using std::vector;
  41
  42 // Assembly code used for testing loop control with dependencies is based on GLSL source code:
  43 // #version 430
  44 //
  45 // layout(std140, set = 0, binding = 0) readonly buffer Input {
  46 //   float elements[];
  47 // } input_data;
  48 // layout(std140, set = 0, binding = 1) writeonly buffer Output {
  49 //   float elements[];
  50 // } output_data;
  51 //
  52 // void main() {
  53 //   const uint n = 12;
  54 //   float c[n];
  55 //   uint x = gl_GlobalInvocationID.x;
  56 //
  57 //   for (uint i = 0; i < n; ++i)
  58 //     c[i] = float(i) * input_data.elements[x];
  59 //
  60 //   output_data.elements[x] = 0.0f;
  61 //   for (uint i = 0; i < n; ++i)
  62 //     output_data.elements[x] += c[i];
  63 // }
  64 static void getComputeSourceCode (std::string& computeSourceCode)
  65 {
  66         computeSourceCode =
  67                 string(getComputeAsmShaderPreamble()) +
  68
  69                 "OpSource GLSL 430\n"
  70                 "OpName %main \"main\"\n"
  71                 "OpName %id \"gl_GlobalInvocationID\"\n"
  72
  73                 "OpDecorate %id BuiltIn GlobalInvocationId\n"
  74
  75                 + string(getComputeAsmInputOutputBufferTraits()) + string(getComputeAsmCommonTypes()) + string(getComputeAsmInputOutputBuffer()) +
  76
  77                 "%u32ptr        = OpTypePointer Function %u32\n"
  78
  79                 "%id            = OpVariable %uvec3ptr Input\n"
  80                 "%zero          = OpConstant %i32 0\n"
  81                 "%uzero         = OpConstant %u32 0\n"
  82                 "%fzero         = OpConstant %f32 0\n"
  83                 "%one           = OpConstant %i32 1\n"
  84                 "%twelve        = OpConstant %u32 12\n"
  85                 "%f32arr12_t    = OpTypeArray %f32 %twelve\n"
  86                 "%f32arr12ptr_t = OpTypePointer Function %f32arr12_t\n"
  87                 "%f32funcptr    = OpTypePointer Function %f32\n"
  88                 "%main          = OpFunction %void None %voidf\n"
  89                 "%entry         = OpLabel\n"
  90
  91                 "%f32arr12      = OpVariable %f32arr12ptr_t Function\n"
  92
  93                 "%i1            = OpVariable %u32ptr Function\n"
  94                 "                 OpStore %i1 %uzero\n"
  95                 "%i2            = OpVariable %u32ptr Function\n"
  96                 "                 OpStore %i2 %uzero\n"
  97
  98                 "%idval         = OpLoad %uvec3 %id\n"
  99                 "%x             = OpCompositeExtract %u32 %idval 0\n"
 100                 "%inloc         = OpAccessChain %f32ptr %indata %zero %x\n"
 101                 "%inval         = OpLoad %f32 %inloc\n"
 102
 103                 // for (uint i = 0; i < 12; ++i) c[i] = float(i) * input_data.elements[x];
 104                 "                 OpBranch %loop1_entry\n"
 105                 "%loop1_entry   = OpLabel\n"
 106                 "%i1_val        = OpLoad %u32 %i1\n"
 107                 "%cmp1_lt       = OpULessThan %bool %i1_val %twelve\n"
 108                 "                 OpLoopMerge %loop1_merge %loop1_body DependencyInfinite\n"
 109                 "                 OpBranchConditional %cmp1_lt %loop1_body %loop1_merge\n"
 110                 "%loop1_body    = OpLabel\n"
 111                 "%i1_valf32     = OpConvertUToF %f32 %i1_val\n"
 112                 "%mulf1         = OpFMul %f32 %i1_valf32 %inval\n"
 113                 "%outloc1       = OpAccessChain %f32funcptr %f32arr12 %i1_val\n"
 114                 "                 OpStore %outloc1 %mulf1\n"
 115                 "%new1_i        = OpIAdd %u32 %i1_val %one\n"
 116                 "                 OpStore %i1 %new1_i\n"
 117                 "                 OpBranch %loop1_entry\n"
 118                 "%loop1_merge   = OpLabel\n"
 119
 120                 //   output_data.elements[x] = 0.0f;
 121                 "%outloc        = OpAccessChain %f32ptr %outdata %zero %x\n"
 122                 "                 OpStore %outloc %fzero\n"
 123                 "                 OpBranch %loop2_entry\n"
 124
 125                 //   for (uint i = 0; i < n; ++i) output_data.elements[x] += c[i];
 126                 "%loop2_entry   = OpLabel\n"
 127                 "%i2_val        = OpLoad %u32 %i2\n"
 128                 "%cmp2_lt       = OpULessThan %bool %i2_val %twelve\n"
 129                 "                 OpLoopMerge %loop2_merge %loop2_body None\n"
 130                 "                 OpBranchConditional %cmp2_lt %loop2_body %loop2_merge\n"
 131                 "%loop2_body    = OpLabel\n"
 132                 "%arr1_i2loc    = OpAccessChain %f32funcptr %f32arr12 %i2_val\n"
 133                 "%arr1_i2val    = OpLoad %f32 %arr1_i2loc\n"
 134                 "%outval        = OpLoad %f32 %outloc\n"
 135                 "%addf1         = OpFAdd %f32 %outval %arr1_i2val\n"
 136                 "                 OpStore %outloc %addf1\n"
 137                 "%new_i2        = OpIAdd %u32 %i2_val %one\n"
 138                 "                 OpStore %i2 %new_i2\n"
 139                 "                 OpBranch %loop2_entry\n"
 140                 "%loop2_merge   = OpLabel\n"
 141
 142                 "                 OpReturn\n"
 143                 "                 OpFunctionEnd\n";
 144 }
 145
 146 static ComputeShaderSpec getComputeShaderSpec ()
 147 {
 148         de::Random                      rnd                             (0xABC);
 149         const int                       numElements             = 100;
 150         vector<float>           inputFloats             (numElements, 0);
 151         vector<float>           outputFloats    (numElements, 0);
 152         ComputeShaderSpec       spec;
 153
 154         for (size_t ndx = 0; ndx < numElements; ++ndx)
 155                 inputFloats[ndx] = rnd.getFloat(1.0f, 100.0f);
 156
 157         for (size_t ndx = 0; ndx < numElements; ++ndx)
 158         {
 159                 const deUint32 n = 12;
 160                 float c[n];
 161                 float result = 0.0f;
 162
 163                 for (deUint32 i = 0; i < n; ++i)
 164                         c[i] = float(i) * inputFloats[ndx];
 165
 166                 for (deUint32 i = 0; i < n; ++i)
 167                         result += c[i];
 168
 169                 outputFloats[ndx] = result;
 170         }
 171
 172         // Shader source code can be retrieved to complete definition of ComputeShaderSpec, though it is not required at this stage
 173         // getComputeSourceCode (spec.assembly);
 174
 175         spec.inputs.push_back(BufferSp(new Float32Buffer(inputFloats)));
 176         spec.outputs.push_back(BufferSp(new Float32Buffer(outputFloats)));
 177         spec.numWorkGroups      = tcu::IVec3(numElements, 1, 1);
 178         spec.verifyIO           = &verifyOutput;
 179
 180         return spec;
 181 }
 182
 183
 184 class SpvAsmLoopControlDependencyInfiniteInstance : public ComputeShaderSpec, public SpvAsmComputeShaderInstance
 185 {
 186 public:
 187         SpvAsmLoopControlDependencyInfiniteInstance     (Context& ctx);
 188 };
 189
 190 SpvAsmLoopControlDependencyInfiniteInstance::SpvAsmLoopControlDependencyInfiniteInstance (Context& ctx)
 191         : ComputeShaderSpec(getComputeShaderSpec())
 192         , SpvAsmComputeShaderInstance(ctx, *this, COMPUTE_TEST_USES_NONE)
 193 {
 194 }
 195
 196 SpvAsmLoopControlDependencyInfiniteCase::SpvAsmLoopControlDependencyInfiniteCase (tcu::TestContext& testCtx, const char* name, const char* description)
 197         : TestCase                      (testCtx, name, description)
 198 {
 199 }
 200
 201 void SpvAsmLoopControlDependencyInfiniteCase::initPrograms (SourceCollections& programCollection) const
 202 {
 203         std::string comp;
 204
 205         getComputeSourceCode(comp);
 206
 207         programCollection.spirvAsmSources.add("compute") << SpirVAsmBuildOptions(SPIRV_VERSION_1_3) << comp;
 208 }
 209
 210 TestInstance* SpvAsmLoopControlDependencyInfiniteCase::createInstance (Context& context) const
 211 {
 212         if (context.getUsedApiVersion() < VK_API_VERSION_1_1)
 213                 TCU_THROW(NotSupportedError, "SPIR-V higher than 1.3 is required for this test to run");
 214
 215         return new SpvAsmLoopControlDependencyInfiniteInstance(context);
 216 }
 217
 218 } // SpirVAssembly
 219 } // vkt