Enable HLSL legalization

Also added known-good mechanism to fetch latest validated spirv-tools. Also added -Od and -Os to disable optimizer and optimize for size. Fetching spirv-tools is optional for both glsl and hlsl. Legalization of hlsl is done by default if spirv-opt is present at cmake time. Optimization for glsl is currently done through the option -Os. Legalization testing is currently only done on four existing shaders. A separate baseLegalResults directory holds those results. All previous testing is done with the optimizer disabled.
BabylonJS · Sep 27, 2017 · cd1f169 · cd1f169
1 parent 44dd6a0
commit cd1f169
Show file tree

Hide file tree

Showing 16 changed files with 535 additions and 22 deletions.
diff --git a/.appveyor.yml b/.appveyor.yml
@@ -25,6 +25,7 @@ matrix:
 # scripts that run after cloning repository
 install:
   - git clone https://github.com/google/googletest.git External/googletest
+  - update_glslang_sources.py
 
 build:
   parallel: true  # enable MSBuild parallel builds

diff --git a/.gitignore b/.gitignore
@@ -7,3 +7,4 @@ TAGS
 build/
 Test/localResults/
 External/googletest
+External/spirv-tools
diff --git a/.travis.yml b/.travis.yml
@@ -55,6 +55,7 @@ install:
 
 before_script:
   - git clone --depth=1 https://github.com/google/googletest.git External/googletest
+  - update_glslang_sources.py
 
 script:
   - mkdir build && cd build

diff --git a/README.md b/README.md
@@ -74,6 +74,14 @@ cd <the directory glslang was cloned to, "External" will be a subdirectory>
 git clone https://github.com/google/googletest.git External/googletest
 ```
 
+If you wish to assure that SPIR-V generated from HLSL is legal for Vulkan,
+or wish to invoke -Os to reduce SPIR-V size from HLSL or GLSL, install
+spirv-tools with this:
+
+```bash
+./update_glslang_sources.py
+```
+
 #### 3) Configure
 
 Assume the source directory is `$SOURCE_DIR` and

diff --git a/SPIRV/GlslangToSpv.cpp b/SPIRV/GlslangToSpv.cpp
@@ -52,6 +52,16 @@ namespace spv {
 #endif
 }
 
+#ifdef ENABLE_OPT
+    #include "spirv-tools/optimizer.hpp"
+    #include "message.h"
+    #include "SPVRemapper.h"
+#endif
+
+#ifdef ENABLE_OPT
+using namespace spvtools;
+#endif
+
 // Glslang includes
 #include "../glslang/MachineIndependent/localintermediate.h"
 #include "../glslang/MachineIndependent/SymbolTable.h"
@@ -5960,6 +5970,12 @@ void OutputSpvHex(const std::vector<unsigned int>& spirv, const char* baseName,
     out.close();
 }
 
+#ifdef ENABLE_OPT
+void errHandler(const std::string& str) {
+    std::cerr << str << std::endl;
+}
+#endif
+
 //
 // Set up the glslang traversal
 //
@@ -5988,6 +6004,49 @@ void GlslangToSpv(const glslang::TIntermediate& intermediate, std::vector<unsign
     it.finishSpv();
     it.dumpSpv(spirv);
 
+#ifdef ENABLE_OPT
+    // If from HLSL, run spirv-opt to "legalize" the SPIR-V for Vulkan
+    // eg. forward and remove memory writes of opaque types.
+    if ((intermediate.getSource() == EShSourceHlsl ||
+                options->optimizeSize) &&
+            !options->disableOptimizer) {
+        spv_target_env target_env = SPV_ENV_UNIVERSAL_1_2;
+
+        spvtools::Optimizer optimizer(target_env);
+        optimizer.SetMessageConsumer([](spv_message_level_t level,
+                                         const char* source,
+                                         const spv_position_t& position,
+                                         const char* message) {
+            std::cerr << StringifyMessage(level, source, position, message)
+                      << std::endl;
+        });
+
+        optimizer.RegisterPass(CreateInlineExhaustivePass());
+        optimizer.RegisterPass(CreateLocalAccessChainConvertPass());
+        optimizer.RegisterPass(CreateLocalSingleBlockLoadStoreElimPass());
+        optimizer.RegisterPass(CreateLocalSingleStoreElimPass());
+        optimizer.RegisterPass(CreateInsertExtractElimPass());
+        optimizer.RegisterPass(CreateAggressiveDCEPass());
+        optimizer.RegisterPass(CreateDeadBranchElimPass());
+        optimizer.RegisterPass(CreateBlockMergePass());
+        optimizer.RegisterPass(CreateLocalMultiStoreElimPass());
+        optimizer.RegisterPass(CreateInsertExtractElimPass());
+        optimizer.RegisterPass(CreateAggressiveDCEPass());
+        // TODO(greg-lunarg): Add this when AMD driver issues are resolved
+        // if (options->optimizeSize)
+        //     optimizer.RegisterPass(CreateCommonUniformElimPass());
+
+        if (!optimizer.Run(spirv.data(), spirv.size(), &spirv))
+            return;
+
+        // Remove dead module-level objects: functions, types, vars
+        // TODO(greg-lunarg): Switch to spirv-opt versions when available
+        spv::spirvbin_t Remapper(0);
+        Remapper.registerErrorHandler(errHandler);
+        Remapper.remap(spirv, spv::spirvbin_t::DCE_ALL);
+    }
+#endif
+
     glslang::GetThreadPoolAllocator().pop();
 }
 

diff --git a/SPIRV/GlslangToSpv.h b/SPIRV/GlslangToSpv.h
@@ -48,8 +48,11 @@
 namespace glslang {
 
 struct SpvOptions {
-    SpvOptions() : generateDebugInfo(false) { }
+    SpvOptions() : generateDebugInfo(false), disableOptimizer(true),
+        optimizeSize(false) { }
     bool generateDebugInfo;
+    bool disableOptimizer;
+    bool optimizeSize;
 };
 
 void GetSpirvVersion(std::string&);

diff --git a/StandAlone/StandAlone.cpp b/StandAlone/StandAlone.cpp
@@ -95,6 +95,8 @@ enum TOptions {
     EOptionAutoMapLocations     = (1 << 25),
     EOptionDebug                = (1 << 26),
     EOptionStdin                = (1 << 27),
+    EOptionOptimizeDisable      = (1 << 28),
+    EOptionOptimizeSize         = (1 << 29),
 };
 
 //
@@ -528,6 +530,18 @@ void ProcessArguments(std::vector<std::unique_ptr<glslang::TWorkItem>>& workItem
             case 'I':
                 IncludeDirectoryList.push_back(getStringOperand("-I<dir> include path"));
                 break;
+            case 'O':
+                if (argv[0][2] == 'd')
+                    Options |= EOptionOptimizeDisable;
+                else if (argv[0][2] == 's')
+#ifdef ENABLE_OPT
+                    Options |= EOptionOptimizeSize;
+#else
+                    Error("-Os not available; optimizer not linked");
+#endif
+                else
+                    Error("unknown -O option");
+                break;
             case 'S':
                 if (argc <= 1)
                     Error("no <stage> specified for -S");
@@ -882,6 +896,8 @@ void CompileAndLinkShaderUnits(std::vector<ShaderCompUnit> compUnits)
                     glslang::SpvOptions spvOptions;
                     if (Options & EOptionDebug)
                         spvOptions.generateDebugInfo = true;
+                    spvOptions.disableOptimizer = Options & EOptionOptimizeDisable;
+                    spvOptions.optimizeSize = Options & EOptionOptimizeSize;
                     glslang::GlslangToSpv(*program.getIntermediate((EShLanguage)stage), spirv, &logger, &spvOptions);
 
                     // Dump the spv to a file or stdout, etc., but only if not doing
@@ -1201,6 +1217,8 @@ void usage()
            "  -H          print human readable form of SPIR-V; turns on -V\n"
            "  -I<dir>     add dir to the include search path; includer's directory\n"
            "              is searched first, followed by left-to-right order of -I\n"
+           "  -Od         disables optimization. May cause illegal SPIR-V for HLSL.\n"
+           "  -Os         optimizes SPIR-V to minimize size.\n"
            "  -S <stage>  uses specified stage rather than parsing the file extension\n"
            "              choices for <stage> are vert, tesc, tese, geom, frag, or comp\n"
            "  -U<macro>   undefine a pre-processor macro\n"

diff --git a/Test/baseLegalResults/hlsl.aliasOpaque.frag.out b/Test/baseLegalResults/hlsl.aliasOpaque.frag.out
@@ -0,0 +1,46 @@
+hlsl.aliasOpaque.frag
+// Module Version 10000
+// Generated by (magic number): 80001
+// Id's are bound by 61
+
+                              Capability Shader
+               1:             ExtInstImport  "GLSL.std.450"
+                              MemoryModel Logical GLSL450
+                              EntryPoint Fragment 4  "main" 46
+                              ExecutionMode 4 OriginUpperLeft
+                              Source HLSL 500
+                              Name 4  "main"
+                              Name 36  "gss"
+                              Name 37  "gtex"
+                              Name 46  "@entryPointOutput"
+                              Decorate 36(gss) DescriptorSet 0
+                              Decorate 37(gtex) DescriptorSet 0
+                              Decorate 46(@entryPointOutput) Location 0
+               2:             TypeVoid
+               3:             TypeFunction 2
+               6:             TypeSampler
+               7:             TypePointer UniformConstant 6
+               8:             TypeFloat 32
+              10:             TypeImage 8(float) 2D sampled format:Unknown
+              11:             TypePointer UniformConstant 10
+              12:             TypeVector 8(float) 4
+              25:             TypeSampledImage 10
+              27:             TypeVector 8(float) 2
+              28:    8(float) Constant 1045220557
+              29:    8(float) Constant 1050253722
+              30:   27(fvec2) ConstantComposite 28 29
+         36(gss):      7(ptr) Variable UniformConstant
+        37(gtex):     11(ptr) Variable UniformConstant
+              39:    8(float) Constant 1077936128
+              45:             TypePointer Output 12(fvec4)
+46(@entryPointOutput):     45(ptr) Variable Output
+         4(main):           2 Function None 3
+               5:             Label
+              56:          10 Load 37(gtex)
+              57:           6 Load 36(gss)
+              58:          25 SampledImage 56 57
+              59:   12(fvec4) ImageSampleImplicitLod 58 30
+              60:   12(fvec4) VectorTimesScalar 59 39
+                              Store 46(@entryPointOutput) 60
+                              Return
+                              FunctionEnd
diff --git a/Test/baseLegalResults/hlsl.flattenOpaque.frag.out b/Test/baseLegalResults/hlsl.flattenOpaque.frag.out
@@ -0,0 +1,65 @@
+hlsl.flattenOpaque.frag
+// Module Version 10000
+// Generated by (magic number): 80001
+// Id's are bound by 118
+
+                              Capability Shader
+               1:             ExtInstImport  "GLSL.std.450"
+                              MemoryModel Logical GLSL450
+                              EntryPoint Fragment 4  "main" 83
+                              ExecutionMode 4 OriginUpperLeft
+                              Source HLSL 500
+                              Name 4  "main"
+                              Name 37  "tex"
+                              Name 68  "s.s2D"
+                              Name 73  "s2.s2D"
+                              Name 74  "s2.tex"
+                              Name 83  "@entryPointOutput"
+                              Decorate 37(tex) DescriptorSet 0
+                              Decorate 68(s.s2D) DescriptorSet 0
+                              Decorate 73(s2.s2D) DescriptorSet 0
+                              Decorate 74(s2.tex) DescriptorSet 0
+                              Decorate 83(@entryPointOutput) Location 0
+               2:             TypeVoid
+               3:             TypeFunction 2
+               6:             TypeSampler
+               7:             TypePointer UniformConstant 6
+               8:             TypeFloat 32
+               9:             TypeVector 8(float) 4
+              14:             TypeVector 8(float) 2
+              21:             TypeImage 8(float) 2D sampled format:Unknown
+              22:             TypePointer UniformConstant 21
+         37(tex):     22(ptr) Variable UniformConstant
+              40:             TypeSampledImage 21
+              42:    8(float) Constant 1045220557
+              43:    8(float) Constant 1050253722
+              44:   14(fvec2) ConstantComposite 42 43
+       68(s.s2D):      7(ptr) Variable UniformConstant
+      73(s2.s2D):      7(ptr) Variable UniformConstant
+      74(s2.tex):     22(ptr) Variable UniformConstant
+              82:             TypePointer Output 9(fvec4)
+83(@entryPointOutput):     82(ptr) Variable Output
+         4(main):           2 Function None 3
+               5:             Label
+              97:          21 Load 37(tex)
+              98:           6 Load 68(s.s2D)
+              99:          40 SampledImage 97 98
+             100:    9(fvec4) ImageSampleImplicitLod 99 44
+             102:          21 Load 37(tex)
+             103:           6 Load 68(s.s2D)
+             104:          40 SampledImage 102 103
+             106:    9(fvec4) ImageSampleImplicitLod 104 44
+              91:    9(fvec4) FAdd 100 106
+             108:          21 Load 74(s2.tex)
+             109:           6 Load 73(s2.s2D)
+             110:          40 SampledImage 108 109
+             111:    9(fvec4) ImageSampleImplicitLod 110 44
+              93:    9(fvec4) FAdd 91 111
+             113:          21 Load 74(s2.tex)
+             114:           6 Load 73(s2.s2D)
+             115:          40 SampledImage 113 114
+             117:    9(fvec4) ImageSampleImplicitLod 115 44
+              95:    9(fvec4) FAdd 93 117
+                              Store 83(@entryPointOutput) 95
+                              Return
+                              FunctionEnd
diff --git a/Test/baseLegalResults/hlsl.flattenOpaqueInit.vert.out b/Test/baseLegalResults/hlsl.flattenOpaqueInit.vert.out
@@ -0,0 +1,70 @@
+hlsl.flattenOpaqueInit.vert
+WARNING: 0:20: '=' : cannot do member-wise aliasing for opaque members with this initializer 
+
+// Module Version 10000
+// Generated by (magic number): 80001
+// Id's are bound by 76
+
+                              Capability Shader
+               1:             ExtInstImport  "GLSL.std.450"
+                              MemoryModel Logical GLSL450
+                              EntryPoint Vertex 4  "main" 58
+                              Source HLSL 500
+                              Name 4  "main"
+                              Name 17  "FxaaTex"
+                              MemberName 17(FxaaTex) 0  "smpl"
+                              MemberName 17(FxaaTex) 1  "tex"
+                              Name 36  "g_tInputTexture_sampler"
+                              Name 37  "g_tInputTexture"
+                              Name 39  "t"
+                              Name 43  "flattenTemp"
+                              Name 45  "tex2.smpl"
+                              Name 50  "tex2.tex"
+                              Name 58  "@entryPointOutput"
+                              Decorate 36(g_tInputTexture_sampler) DescriptorSet 0
+                              Decorate 37(g_tInputTexture) DescriptorSet 0
+                              Decorate 58(@entryPointOutput) Location 0
+               2:             TypeVoid
+               3:             TypeFunction 2
+               6:             TypeSampler
+               7:             TypePointer UniformConstant 6
+               8:             TypeFloat 32
+               9:             TypeImage 8(float) 2D sampled format:Unknown
+              10:             TypePointer UniformConstant 9
+              11:             TypeVector 8(float) 4
+     17(FxaaTex):             TypeStruct 6 9
+              26:             TypeSampledImage 9
+              28:             TypeVector 8(float) 2
+              29:    8(float) Constant 1050253722
+              30:    8(float) Constant 1053609165
+              31:   28(fvec2) ConstantComposite 29 30
+              32:    8(float) Constant 0
+36(g_tInputTexture_sampler):      7(ptr) Variable UniformConstant
+37(g_tInputTexture):     10(ptr) Variable UniformConstant
+              38:             TypePointer UniformConstant 17(FxaaTex)
+           39(t):     38(ptr) Variable UniformConstant
+ 43(flattenTemp):     38(ptr) Variable UniformConstant
+   45(tex2.smpl):      7(ptr) Variable UniformConstant
+              46:             TypeInt 32 1
+              47:     46(int) Constant 0
+    50(tex2.tex):     10(ptr) Variable UniformConstant
+              51:     46(int) Constant 1
+              57:             TypePointer Output 11(fvec4)
+58(@entryPointOutput):     57(ptr) Variable Output
+         4(main):           2 Function None 3
+               5:             Label
+              70: 17(FxaaTex) Load 39(t)
+                              Store 43(flattenTemp) 70
+              63:      7(ptr) AccessChain 43(flattenTemp) 47
+              64:           6 Load 63
+                              Store 45(tex2.smpl) 64
+              65:     10(ptr) AccessChain 43(flattenTemp) 51
+              66:           9 Load 65
+                              Store 50(tex2.tex) 66
+              72:           9 Load 37(g_tInputTexture)
+              73:           6 Load 36(g_tInputTexture_sampler)
+              74:          26 SampledImage 72 73
+              75:   11(fvec4) ImageSampleExplicitLod 74 31 Lod 32
+                              Store 58(@entryPointOutput) 75
+                              Return
+                              FunctionEnd